Highlights:

• Ralph Baric confirmed that DEFUSE proposed inserting novel furin cleavage sites into live viruses, inspired by feline coronaviruses
• SARS-CoV-2’s furin cleavage sites is identical to the one found in several lethal feline coronavirus strains
• Ralph Baric strongly believes WIV had unpublished viruses and viral reverse genetics systems
• Most plausible Covid origin is the result of research on identifying new SARS-like viruses and developing broad vaccines against them

Quick Summary:

Ralph Baric’s January 2024 testimony provided crucial insights into the DEFUSE grant proposal, a collaborative project involving EcoHealth and the Wuhan Institute of Virology (WIV), and the research interests of WIV prior to the Covid outbreak. The pre-Covid landscape of coronavirology research saw intense US-China collaboration aimed at creating broad-spectrum vaccines for SARS- and MERS-like viruses. Key players like Baric, along with researchers Lanying Du, Yusen Zhou, Fang Li, and Shibo Jiang, focused on overcoming challenges such as antibody-dependent enhancement (ADE). By 2018–2019, as collaboration between Baric and his counterparts shifted, WIV became central in SARS-like virus research.

Drawing on Baric’s testimony, I argue that the emergence of SARS-CoV-2 in Wuhan wasn’t coincidental but a likely result of research at WIV. In particular, I outline how WIV researchers, after a post-2018 shift in focus, concentrated on identifying SARS-like viruses with spike proteins 10–25% different from SARS1, capable of evading SARS1-based antibodies, and potentially causing ADE. They were plausibly inspired by Baric’s ideas on furin cleavage sites and his work with Fang Li on SARS and MERS spike cleavage, leading them to engineer furin cleavage sites into novel SARS-like strains. Baric’s testimony suggests that feline coronaviruses inspired this suggestion, and the identical furin cleavage sites in lethal feline coronavirus strains and SARS-CoV-2 strongly indicate that the WIV could have inserted such a site into a SARS-CoV-2 precursor to extend their research from MERS to novel SARS-like coronaviruses. This suggests the emergence of SARS-CoV-2 was not a coincidence but a potential result of this focused research.

In his recently published testimony from January 2024, Ralph Baric, the preeminent coronavirus expert, dispelled several talking points that lab leak critics have used repeatedly. He confirmed that the DEFUSE grant proposal — coauthored by him, EcoHealth, and the Wuhan Institute of Virology — was indeed suggesting to insert novel furin cleavage sites into SARS-like viruses, and do so in live full-length viruses rather than just pseudoviruses or chimeras. Importantly, he also said that the WIV not only had unpublished viruses but almost certainly had unpublished molecular clones (i.e. viral reverse genetics systems).

Moreover, Dr. Baric strengthened a plausible explanation for why a genetic engineer might have chosen to create a furin cleavage site like that in SARS2 — to mimic one often seen in pathogenic strains of a feline coronavirus. As I will show in this article, this explanation fits well with one plausible lab leak scenario: that SARS2 leaked during work on novel SARS vaccines conducted by Yusen Zhou — a SARS and MERS vaccine expert (and reportedly a decorated PLA officer) who filed a patent for a Covid vaccine in February 2020 and died a few months later under mysterious circumstances.

Yusen Zhou was married to Lanying Du, a frequent collaborator of Shibo Jiang, Zhengli Shi, and Fang Li. As far back as 2016, the WIV website called Fang Li a visiting researcher at WIV when showcasing their paper on MERS vaccines where Lanying Du is the first author, while Shibo Jiang, Yusen Zhou, and Fang Li are all senior and corresponding authors.

Feline coronaviruses, their mutating furin cleavage sites, and how such mutations enable viruses to infect immune cells and cause antibody-dependent enhancement of viral entry (ADE) is the thread that connects the research described in DEFUSE and EcoHealth grants to the research that Yusen Zhou, Lanying Du, Shibo Jiang, Zhengli Shi, and Fang Li were doing independently — without Ralph Baric or EcoHealth — in the context of MERS and SARS in late 2019.

But before I dive into their research to outline how a genetic engineer might have arrived at the idea to insert a furin cleavage site from a feline coronavirus into a SARS2 progenitor and potentially study its impact on ADE, let me first turn to Dr. Baric to set the stage. I believe his next few quotes are very important to understanding the key coronavirology research areas in the years leading up to the Covid outbreak:

1. Developing broadly active coronavirus vaccines and therapeutics

Dr. Baric: “One of the things that drove the 2015 paper was that SARS coronavirus emerged in 2003. It was controlled by public health intervention strategies because it didn’t transmit until you got clinical disease. People thought it was a fluke, one-off, it’s not going to happen again. Then MERS coronavirus emerged in 2012, again; highly pathogenic, 35 percent mortality rate, but it didn’t transmit very well. So that data made us ask the fundamental question: What is the risk level that exists in nature? This paper, in essence, said the risk in nature — that risk existed in nature. And then the mouse models were then used to develop countermeasures. So almost immediately in parallel with this paper, we started working with Gilead Scientific to evaluate nucleoside inhibitors that might work against the coronavirus family. After testing a bunch of things, we eventually got down to remdesivir, demonstrating that it worked against the MERS coronavirus and the SARS coronavirus. That led to a companion paper that included these viruses in 2017 that said these are broad spectrum antivirals that work in robust animal models of disease. And the preclinical data was now available to move into the clinical trials. So that’s why animal models are so important. Ultimately, remdesivir, molnupiravir, the Moderna vaccine, I don’t know if we ever did the Janssen vaccine. But several therapeutic antibodies had all made it through the FDA and into the clinic, went through our lab, and many of them touched these viruses that were developed in the 2015 paper. These same viruses are being used for universal vaccine design for all sarbecoviruses and all betacoronaviruses [here and subsequently all emphasis is added by me — Y.D.]. So if you want to really protect the public, you have to have the appropriate virologic reagents that challenge the effectiveness of either your drug or your antibody or your vaccine and prove performance. So ultimately, the goal of what resulted from this paper was the idea that we had to develop drugs, we had to develop immunotherapeutics that were broadly active. And we had to develop vaccines that were broadly active. And that paper, including the viruses, the human viruses that occurred, were included in studies that were used with the Moderna vaccine as well. So, again, animal model development is key to this. It’s, again, very, very easy to make drugs that work against something that barely replicates, but then when they get into the humans, they fail. So that’s the basis for it. That’s probably a little longwinded. I apologize. Anyway, that’s the thought process.”

2. Broad coronavirus vaccines and “vaccine-associated deleterious outcomes” (ADE — antibody-dependent enhancement of viral entry)

Q: “I appreciate that science lesson. I’m going to change topics a bit. We have heard from multiple witnesses that the creation of a vaccine for COVID-19 happened almost miraculously fast, and they credit this speed to the fact that coronavirus research and mRNA research had been going on for years prior to the COVID-19 pandemic. You were a part of this process, both with ongoing research and active involvement in the COVID-19 vaccine testing, correct?”

A: “That’s correct.”

Q: “In terms of the development and testing of a COVID-19 vaccine, in 2020, your involvement was running safety and efficacy trials for Moderna’s vaccine using your lab’s chimeric coronavirus strains, human respiratory cell cultures, and lab mice. Is that accurate?”

A: “For the COVID-19 vaccine, I don’t think we tried any — we used any chimeras. The only thing we really used was the mouse-adapted SARS2 coronavirus, the MA10, which was called MA10 in this case. It was ten passages in mice that produced a lethal infection. But I can tell you that our involvement with mRNA technology started in 2016, early 2017, in collaboration with Barney Graham and Kizzmekia Corbett at the NIH VRC, where they had just worked. Well, Jason McLellan and Barney had really worked out the technology to freeze the coronavirus spike glycoprotein in what was called the prefusion state, which had all the big, juicy neutralization epitopes in the right context. So they wanted to evaluate mRNA vaccine performance, and so they contacted us and we worked with them on mRNA vaccines for MERS coronavirus mostly, but also SARS coronavirus in 2003, and were actually writing the paper in December 2019 when COVID hit. And so we stopped writing the paper. When they received the sequence, they ordered the constructs. I was told that I had to have a mouse model available by the end of April, so my job was to make a robust mouse model in sufficient time to test that vaccine in April and May, so that the final reports could be compiled, including some studies that were designed to look for what are called variant phenotype vaccine associated — oh, crap, I forget the name. Do you have to type everything that I say? Great.”

Q: “We’re all allowed to have those moments.”

A: “I’m having a moment. But they’re probably going to become more frequent over the next hour, I have to admit. But it’s vaccine associated deleterious outcome. In this case, there’s something, either the vaccine enhances the availability of the virus to grow or it causes some kind of pathology. And it needed to be tested for that, because, earlier, it had been shown with earlier vaccines with the SARS strain that you’ve got those phenotypes. My job was to make the mouse model and design those experiments and have them all done by April.”

Q: “And we’ve heard from multiple people that this was all on a timeline that was way faster than any other vaccine.”

A: “It was very stressful.”

Q: “I’m sure.”

A: “It was very stressful.”

Q: “You mentioned that you had been working on this, on vaccines, prior to 2016. I know, reading articles and research that you’ve done, it seems like you’ve been working on a pan-coronavirus vaccine for many years, and that’s been one of your research focuses; is that right?”

A: “Well, again, the discovery work we did said that there was a zoonotic virus. There are animal viruses out there that are high risk. You don’t know which one will evolve. So the only kind of countermeasure you can make is broad spectrum. It either has to be a broad spectrum drug, or you have to have a vaccine that provides like an umbrella of breadth to many strains. And so what you try to do with your discovery work is to find the strains that are the most different, and then some in the middle. So then you can say, well, it works on the bookends, it works in the middle, I hope it works against the new thing, right?”

Q: “Sure.”

A: “That’s the only way to do it.”

Q: “You mentioned a little bit throughout today some therapeutics that you were testing before and other research that was sort of useful for the pandemic. Can you elaborate on what pieces or findings from research prior to the pandemic were useful in determining and finding vaccines and therapeutics once the pandemic was widespread?”

A: “Well, certainly having isolates and robust mouse models of human disease, using the human strain of MERS and the SARS strain that caused human disease were really important. But that captured this much of the variation, like a paper-thin sliver of the variation that exists in the family. So you need to have natural, other zoonotic isolates with robust mouse models, so you’ll be able to really evaluate the performance of the vaccine when it’s not a perfect match, because when the vaccine’s not a perfect match is when all these adverse reactions can occur, or you have this because you have a breakthrough. So we did discovery work. That discovery work is important because it gave us breadth both with MERS and with SARS. In addition, at the same time, we were part of a grant that was funded to try to develop drugs against coronaviruses, with Mark Denison at Vanderbilt and Gilead were collaborators. And so Gilead was gracious enough to provide a fairly robust panel of nucleoside inhibitors that we screened working down to remdesivir, that we then moved from — the classic approach was, you know, cells, continuous cells in culture, to primary human cells, to the animal models, and demonstrated that it not only worked against SARS and MERS but it worked against all these other bat coronaviruses, other human coronaviruses, other animal coronaviruses, 12 different viruses. So we knew it had broad spectrum. So now the hypothesis is, you have a broad spectrum drug. Any new virus comes along, you immediately test the hypothesis and evaluate remdesivir, molnupiravir, Paxlovid, therapeutic antibodies, vaccines, to see if they provide breadth. And simultaneously, you use that information in a reiterative fashion now to develop broader-based vaccine platforms. So one of the innovations that we did was to take spike glycoproteins across the phylogenetic tree, blend them together as a chimera, delivered on mRNA vaccine that would provide neutralizing breadth against a greater percentage of the strains.”

Q: “So would it be accurate to say that research on a pathogen that’s not yet infecting people gives scientists a basis to make their hypotheses for how a pathogen that is infecting people may react to therapeutics or a vaccine?”

A: “It’s more than that, It’s absolutely essential. You have no idea of the breadth of performance of your product if you don’t have natural isolates available in the virus family. So, for example, calls to shut down discovery work in the natural world will basically mean that the U.S. is at greater risk for future emerging diseases because we don’t know what’s there, and we can’t test products against it.”

3. Looking for novel SARS-like viruses whose spike are 20–25% different from SARS1

Q: “And then on the other aspect of looking — and this may relate to sort of the search for a broad spectrum coronavirus vaccine. What was the rationale between looking for a SARS-related coronavirus that sort of a 10 to 20 percent divergent in the spike from SARS1?”

Dr. Baric: “Sure. So SARS 2003 is the bookend, right? You know how much variation. WIV1 and SHCO14 have about 8 to 12 percent variation in the spike or the RBD. The clade 2 strains like HKU3 have 30 to 35 percent variation in the spike, they’ve got deletions in the RBD, they can’t use human ACE2 receptors. If you take those two numbers, subtract 10 or 12 from 35, divided by 2, added to 12, you get a number between 20 and 25. And that was our prediction, that there would be strains with that much variation that could still use human ACE2 receptors. It turns out SARS2 had 22 percent variation, so we were within the range, but we were really not completely right. In MERS, there are strains with 35 percent variation in the RBD that could still use the human. So in reality, it’s probably much greater than 20, 25 percent.”

Q: “Really?”

A: “That was our estimate. And the reason we’re interested in that, the strains with the most variation become important for developing countermeasures in vaccines. So if you have a strain that’s really different than therapeutic antibodies, you can look for broadly neutralizing antibodies. They may not work. Your vaccine, if you have an animal model, you can ask, does it cover this much variation? And if it doesn’t, it gives you the starting material to develop a second generation vaccine that can capture it. So again, that variation — I have no interest in simply resurrecting every single coronavirus.”

Q: “Sure,”

A: “I’m interested in the bookends and a couple intermediate ones because that’s what’s best for countermeasure development.”

Key research areas in coronavirology prior to the Covid outbreak

I hope you read the above quotes fully, because they shed light on what Ralph Baric and, by extension, his collabo-competitors were interested in in the years preceding the outbreak: working on broad vaccines and therapeutics that could protect against many coronaviruses, especially SARS- and MERS-like ones. To get the right breadth of SARS-like CoVs, by 2018, Baric, along with WIV and EcoHealth, were now looking for novel strains that were 20–25% different from SARS1 in their spike protein (SARS2 is 24% different).

However, one challenge that broad vaccines faced was antibody-dependent enhancement of viral entry (ADE, which Baric calls “vaccine-associated deleterious outcome”). In a nutshell, ADE is when vaccines instead of protecting from a virus, can actually make a subsequent infection worse because the antibodies they elicit are not only not neutralizing enough to prevent infection, but they can actually help the virus get into cells, often even immune system cells. One hypothetical way such “bad” antibodies can do this is by changing protease cleavage (e.g. by furin) via physically changing the spike protein structure due to their binding. In fact, this is something WIV and its collaborators were actively researching in late 2019, but without Ralph Baric.

Furin cleavage sites in the DEFUSE proposal

Speaking of furin cleavage sites, Dr. Baric has finally cleared up all uncertainty about what exactly the DEFUSE grant was proposing to do:

“Now, the way the furin cleavage site was built in that grant, at least in the earlier versions, some of that may have been lost as they tried to condense it to get it to fit, was that the first part was that we were fundamentally interested in why didn’t sarbecoviruses have a furin cleavage site. There had been studies done in 2010, 2011, 2012 using pseudotypes. Catherine Holmes published one in JB, there was a Chinese group that published it, where they dropped the furin cleavage site into the SARS from 2003. There was no increased infectivity, there was just a little bit more fusion between the cells. So no really big phenotype. Another example of furin cleavage sites with coronaviruses, a researcher at University of Pennsylvania knocks out the furin cleavage sites in mouse hepatitis. No change in pathogenesis for the ability of the virus to replicate. Feline infectious peritonitis virus, it’s an enteric form, it’s got a furin cleavage site, it replicates, and it got very mild infection. When the furin cleavage site is lost, it kills the cat. So it’s a flip, right? Furin cleavage site is the loss of — it’s protecting from virulent disease. So the data going into that proposal, the exact role of furin cleavage site was not clear. We were interested in it because most other coronaviruses in family had those sites. Why didn’t sarbecovirus? So the way the grant was designed was that the discovery group would look, as they did discovery, if they found one with the furin cleavage site, we would first study the pseudotypes. The second thing we would do is move it into the chimeras to see what the effect on applicants was. The third thing was we would probably build virulent viruses and study pathogenesis, and then we would knock out the furin cleavage site.”

I emphasized the part about feline coronaviruses in the quote above because not only could it explain the unusual FCS seen in SARS2, but also because it has a potential connection to ADE. But first let me elaborate on the role of furin cleavage sites in feline coronavirus pathogenicity that Dr. Baric is referring to.

Put simply, not all furin cleavage sites are created equal. Some are cleaved (i.e. cut) more efficiently, some less. The canonical pattern of an efficient cleavage site is RxRR, where R denotes arginine and ‘x’ stands for any amino acid residue (for sticklers, RxKR is also considered canonical but that’s of minor relevance to us). Now, if instead of RxRR the FCS has RxyR (where ‘x’ could be anything, while ‘y’ could be anything but R or K; for example, RRAR, RSVR, or RRSR), then it is cleaved less efficiently than RxRR, and sometimes, depending on what ‘x’ and ‘y’ actually are, it can stop getting cleaved by furin at all. But less efficient cleavage does not automatically mean a less efficient virus, as the case of feline coronaviruses illustrates.

Coronavirologists have been puzzling over feline coronaviruses for decades. Initially, they even thought that the mild enteric coronavirus and the lethal peritonitis one are two different viruses. However, step by step, they realized that it is the same virus, but one that establishes a persistent infection in a given cat and, if it then mutates in a certain way, it can become lethal. (Quick aside — new fear unlocked: how many human persistent viruses can do the same? Could shingles be caused by something similar?)

Strikingly, the mutation that kills cats is a furin cleavage site mutation that makes it less efficient! I should note that I was surprised to hear Dr. Baric describe this mutation as “knocking out” a furin cleavage site, because that is not what happens — the FCS is almost always still there, it just mutates from a more efficient, canonical RxRR form to a less efficient RxyR one.

I first learned about this aspect of feline CoVs from a Twitter thread describing the following study of feline coronaviruses that provides extremely compelling evidence that what turns a benign enteric (FECV) infection into a lethal peritonitis one (FIPV) is indeed the mutation of a highly cleavable RRSRR or RRARR furin cleavage site into something less efficient, like PRRAR:

As you can see from the figure above, that study reported three different FIPV strains that had the same PRRAR furin cleavage site as in SARS2. It also mentioned an example of two cats living in the same household who were both carrying the benign enteric version of the virus but when one of the cats died, it turned out that its virus mutated its FCS from RxRR to a less efficient RxyR. This is the observation Ralph Baric was likely referring to in his testimony.

By the way, Zhengli Shi must have also known about this peculiarity of feline CoVs, and not just through conversations with Ralph Baric. In 2015, she coauthored a joint paper between WIV and Wuhan University that tested a pan-coronavirus inhibitor against several CoVs, including SARS1, MERS, and FIPV. The FIPV strain they used had the same FCS mutation as above, which turned a canonical RxRR FCS into a less efficient RRSR:

Other examples of FCS mutations changing coronavirus properties

A similar change in virulence due to an FCS mutation was observed in a human coronavirus, OC43: when it had the less efficient RRSR furin cleavage site instead of the canonical RRSRR one, it was more lethal and more neurovirulent when tested in a mouse model.

Possibly, the RRSR furin cleavage site in OC43 inspired a genetic engineer to introduce this “suboptimal” FCS, as lab leak critics call the SARS2 FCS, into another human coronavirus, NL63, which does not normally have one.

Besides feline coronaviruses, Ralph Baric also mentioned murine ones, MHV, as the inspiration behind his FCS suggestions in DEFUSE. There are two notable MHV strains when it comes to our analysis: A59 and JHM. JHM has the canonical RRARR furin cleavage site, while A59 has the less efficient RRAHR. Notably, in this case, JHM is the more lethal one and more neurotropic, while A59 is milder but does cause more severe hepatitis. The difference in their protease cleavage was suggested by University of Pennsylvania researchers as one potential factor that could explain the disparity between these strains. Notably, MHV strains generally have interesting variability in their protease cleavage sites, somewhat reminiscent of the patterns seen in feline CoVs:

Speaking of MHV, Ralph Baric and Fang Li worked on a joint paper on MHV-A59 just before the pandemic (submitted in October 2019), including investigating its protease cleavage:

It is worth noting that compared with the unliganded S-e, the proteolysis sites (at the S1/S2 region and S2’ site) do not become more exposed in the receptor-bound S-e (S4 Fig). Overall, receptor binding does not trigger dramatic structural changes in MHV S-e, which still stays in the pre-fusion conformation.

Prior research into MERS and SARS spike cleavage

The DEFUSE authors’ interest in looking for existing furin cleavage sites in SARS-like viruses and potentially engineering novel ones was preceded by similar joint work on MERS: in 2015, Ralph Baric, Shi Zhengli, Shibo Jiang and Lanying Du coauthored a paper (with Fang Li, who was a visiting researcher at WIV, as the senior and corresponding author) showing that engineering a novel FCS in a MERS-like HKU4 virus and ablating a single N-linked glycan enabled it to enter human cells.

Long before that, in 2007, Shibo Jiang, Yusen Zhou, and Lanying Du studied the role of S1/S2 cleavage in SARS1 viral entry:

We further explored if the cleavage of the S protein indeed plays a critical role in SARS-CoV infection. … here we have provided the cleavage of the S protein as another crucial target for the development of vaccines and antiviral agents. Inhibition of the cleavage of the S protein into functional S1 and S2 subunits using agents such as Ben-HCl can effectively block viral entry.

Then in 2009, Shibo Jiang, Lanying Du, and Yusen Zhou, in the paper titled “The spike protein of SARS-CoV — a target for vaccine and therapeutic development” again talked about SARS1 cleavage, and this time contrasted it with other coronaviruses cleaved by furin. In the same paper the authors mention prior work coauthored by Lanying Du on peptides that interfere with S1/S2 cleavage:

Cleavage of the S protein trimer is an important event in infection, making the potential cleavage site between S1 and S2 domains another target for development of anti-SARS-CoV agents.

Notably, two years prior to the 2015 MERS-HKU4 FCS study, Shibo Jiang coauthored a paper (for which Lanying Du is listed as the editor) that created a novel RIRR furin cleavage site (albeit not in a coronavirus) via a 12-nt insertion reminiscent of the one that has endowed SARS2 with its RRAR furin cleavage site.

Therefore when SARS2 emerged with its distinctive furin cleavage site — introduced via a very odd 12-nt insertion that also included a leading proline, just as in the MERS FCS — it is very surprising that despite all their prior work, Shibo Jiang, Lanying Du, and Zhengli Shi failed to mention the FCS in their January 2020 paper, even as they were comparing SARS2 to other SARS-like strains precisely at the S1/S2 cleavage site:

By aligning 2019-nCoV S protein sequence with those of SARS-CoV and several bat-SL-CoVs, we predicted that the cleavage site for generating S1 and S2 subunits is located at R694/S695

What is even more remarkable is that Zhengli Shi’s next major paper in early February 2020, which disclosed the RaTG13 virus, also failed to mention the novel FCS — and that paper had 29 authors. It even claimed to provide a protein sequence alignment of the S1 subunit of SARS2’s spike protein relative to RaTG13 and other SARS-like CoVs, but instead of including the full S1 protein, the authors chose to cut their alignment a few residues short of the novel furin cleavage site.

Research focus on spike cleavage, vaccines and ADE in 2019

In late 2018, Ralph Baric and Fang Li coauthored a paper on MERS and SARS spike cleavage titled “Lysosomal Proteases Are a Determinant of Coronavirus Tropism”. They were interested in how differences in protease efficiency between bats and humans shape the ability of MERS- and SARS-like viruses to jump between species. While that study focused on lysosomal proteases like cathepsin L, the authors’ clear interest in the differences between MERS and SARS spike processing is highly suggestive that investigating the role of proprotein convertases like furin in the context of SARS-like viruses was also on their mind — especially given that work on this study was done around the time that the DEFUSE proposal was being drafted, as Fang Li’s FOIA emails indicate.

In 2019, Ralph Baric’s group continued looking into proteolytic cleavage, and reported investigating two new MERS-like CoVs that both had an FCS at the S1/S2 junction. Their findings again confirmed that spike cleavage played a very important role in tissue and species tropism.

Surprisingly, Fang Li was not involved in that study, despite it being quite similar to the previous joint work he did with Ralph Baric on HKU4, MERS, and SARS. Neither was Zhengli Shi’s group, despite having coauthored the DEFUSE proposal with Ralph Baric just a year before.

In turn, Ralph Baric’s prior coauthors on the 2015 MERS-HKU4 furin cleavage site paper, Fang Li, Zhengli Shi and Lanying Du, did not include his group in their latest 2019 research, but instead were joined by Yusen Zhou to again look into the cleavage of the MERS spike while studying antibody-dependent enhancement of viral entry.

That paper was co-led by Fang Li and Lanying Du. Besides MERS, the authors also looked into how SARS1 entry into immune cells (e.g. macrophages) can be enhanced via an RBD-specific antibody that Shibo Jiang and Yusen Zhou developed previously:

To expand the above-described observations to another coronavirus, we investigated ADE of SARS-CoV entry. We previously identified a SARS-CoV RBD-specific MAb, named 33G4, which binds to the ACE2-binding region of SARS-CoV RBD (49, 50); this MAb was examined for its potential capability to mediate ADE of SARS-CoV entry (Fig. 3E). The result showed that 33G4 mediated SARS-CoV pseudovirus entry into CD32A-expressing cells but blocked SARS-CoV pseudovirus entry into ACE2-expressing cells. Therefore, both the MERS-CoV RBD-specific MAb and the SARS-CoV RBD-specific MAb can mediate the entry of the respective coronavirus into Fc receptor-expressing human cells while blocking the entry of the respective coronavirus into viral-receptor-expressing human cells.

Shibo Jiang, Yusen Zhou, and Lanying Du have been close colleagues for over a decade, having initially worked on SARS vaccines as far back as at least 2007, when both Lanying Du and Yusen Zhou still held positions with the State Key Laboratory of Pathogen and Biosecurity at the Beijing Institute of Microbiology & Epidemiology. Shibo Jiang’s focus on SARS vaccines and mAbs goes back even farther, as his 2005 and 2006 patents show.

The three researchers have also written extensively on spike-based MERS vaccines from 2013 to 2018, and coauthored joint patents on vaccines against MERS and influenza. By 2019, they have been collaborating for many years under a grant R01AI098775 titled “RBD recombinant protein-based SARS vaccine for biodefense” and have published many papers on SARS and MERS spike-based vaccines. For example, in 2017, they published a paper on a MERS vaccine that used multiple RBDs from several MERS-CoVs to generate broader immunity.

While the Primary Investigator on the above grant was Peter Hotez, in 2018, Lanying Du has gotten her own grant R01AI139092 titled “Structure-based design of coronavirus subunit vaccines”. It was a joint effort with Fang Li and Shibo Jiang:

The description of that grant indicates that the grantees were interested in novel SARS and MERS vaccines, as well as in modulating spike glycosylation, which is notable in connection with SARS2’s missing N370 glycan and the prior research of Lanying Du and Shibo Jiang, in which they removed the same glycan in a SARS1 vaccine candidate.

Under the new Lanying Du grant, in 2019, Yusen Zhou coauthored four papers: three on MERS and one on Zika vaccines:

In light of that grant and a 2017 joint paper with WIV, coauthored by Shibo Jiang, Lanying Du, and Zhengli Shi, on SARS cross-neutralization, it would be logical to assume that the frequent collaborators were also interested in broad SARS vaccines. In fact, based on a manuscript submitted to biorXiv in May 2020, Lanying Du and Shibo Jiang were indeed working on a SARS vaccine in 2019.

Fang Li too was interested in these topics, as his March 2019 talk titled “SARS and MERS coronaviruses: cross-species transmission and vaccine design” indicates:

The ADE and feline coronaviruses connection

There are notable parallels between Covid and feline infectious peritonitis. Recall that the latter is caused by persistent feline coronaviruses that mutate their furin cleavage sites to be less efficient. In light of that, and the fact that SARS2 shares the same FCS (PRRAR) as some known strains of FIPV, those parallels are even more notable. Let me quote a paper titled “Clinical and Molecular Relationships between COVID-19 and Feline Infectious Peritonitis (FIP)” about some particular similarities:

6. Molecular Similarities between the FCoV and SARS-CoV-2 Spike Proteins
…
SARS-CoV-2 is similar to FCoV-1 (and currently unique for SARS-related viruses) in that there are two identified cleavage sites (S1/S2 and S2′), with the former, the furin cleavage site or FCS, thought to be a significant factor in pandemic spread [157,158,159]. In both cases, the presence of the S1/S2 cleavage sites sets FCoV-1 and SARS-CoV-2 apart from their close family members. The importance of the cleavage activation site appears to link directly to the proteases necessary for viral infection and thus, to an additional component of tissue tropism. In type I FCoV, the transition from FECV to macrophage-tropic FIPV was first shown with amino acid substitutions at the S1/S2 cleavage site on FIP-confirmed pathology samples, which were predicted to downregulate proteolytic priming by furin-like proteases prior to S2′-mediated fusion activation [72,160,161].

The authors also mention ADE:

Antibody-dependent enhancement (ADE), the process by which viral–antibody complexes enhance infection, was of particular concern during the SARS-CoV-2 vaccine development process. FIPV has been shown to exhibit ADE in the presence of anti-FIPV antibodies [146]. This enhancement of infection appears to be specific to serotype, with passive immunization of cats against type I or type II FIPV resulting in ADE only after challenging with the same serotype for which immunization was performed [147]. As a result, ADE has been a significant challenge toward the development of FIP vaccines. In human coronavirus diseases, ADE is yet to be fully understood. In SARS-CoV, higher concentrations of anti-spike antibodies were found to have a greater neutralizing effect, whereas more dilute concentrations were suggested to contribute to ADE in vitro [148]. In SARS-CoV-2, ADE was observed in monocyte lineages but was not associated with upregulation of proinflammatory cytokines [149]. Modeling of spike protein sequences identified possible mechanisms for ADE, involving interaction with Fc receptors on monocytes and mast cells [150].

Recall that the Twitter thread that pointed out the 2013 paper on FCS mutations like RRARR -> PRRAR turning benign FECV into lethal FIPV did so in connection with the 2019 WIV paper on MERS ADE coauthored by Zhengli Shi, Yusen Zhou, Lanying Du, and Fang Li. That paper looked into how antibody binding could enhance MERS viral entry into immune cells (“Fc receptor-expressing cells”, e.g. monocytes and macrophages) by modulating spike cleavage, and referenced three papers on feline CoV ADE:

ADE has been observed for coronaviruses. Several studies have shown that sera induced by SARS-CoV spike enhance viral entry into Fc receptor-expressing cells (42,–44). Further, one study demonstrated that unlike receptor-dependent viral entry, serum-dependent SARS-CoV entry does not go through the endosome pathway (44). Additionally, it has long been known that immunization of cats with feline coronavirus spike leads to worsened future infection due to the induction of infection-enhancing antibodies (45,–47). However, detailed molecular mechanisms for ADE of coronavirus entry are still unknown. We previously discovered a monoclonal antibody (MAb) (named Mersmab1) which has strong binding affinity for MERS-CoV RBD and efficiently neutralizes MERS-CoV entry by outcompeting DPP4 (48); this discovery allowed us to comparatively study the molecular mechanisms for antibody-dependent and receptor-dependent viral entries.

The 2013 paper on the FCS mutations turning FECV into FIPV nicely summarizes that the lethality of FIPV is associated with its newfound ability to infect immune cells like monocytes and macrophages:

A long-standing hypothesis is that FIP viruses arise from internal mutation of endemic FECVs (12), which is believed to occur in approximately 1%–5% of enteric infections, resulting in the ability of the virus to infect blood monocytes and tissue macrophages. The resulting productive infection of these cells, a hallmark of FIP, enables systemic spread and results in macrophage activation, with concomitant immune-mediated events leading to death.

Recall that the authors of the 2019 MERS ADE paper led by Fang Li and Lanying Du were also interested in SARS ADE, and claimed that an antibody they previously developed can help SARS get into immune cells.

With that interest in mind, and considering their additional interest in engineering novel furin cleavage sites in SARS-like viruses, as well as searching for novel SARS-like viruses that are 10–25% different from SARS1 and likely to evade SARS1-based antibodies, which is a recipe for ADE, it is completely plausible for researchers who have just studied how MERS spike cleavage affects ADE in MERS, when they come across a MERS-like FCS in a feline CoV hypothesized to cause ADE, might want to try putting such an FCS in a SARS-like CoV to see if SARS1-based antibodies could not only fail to neutralize such a virus but could also supercharge it via similar mechanisms that they just demonstrated causes ADE in MERS.

Addressing Ralph Baric’s skepticism about FCS being engineered

In his testimony, a few hours before explaining his rationale behind suggesting to look for and engineer furin cleavage sites in SARS-like viruses in DEFUSE, Ralph Baric voiced typical arguments for why he thinks the FCS in SARS2 is unlikely to have been engineered. In short, “why use an inefficient FCS”, “why include a leading proline?”, “why create it via an insertion?”, and “why is the insertion not in frame?”:

Q: “I want to ask about the furin site a little bit. Dr. Garry, after the call, in the notes, expressed concern over — he called it a 13 nucleotide insertion that was created at the site, and said ‘I just can’t figure out how this gets accomplished in nature, but in a lab, it would be easy.’ How would you kind of refute Dr. Garry’s points there?”

A: “The sequence, you only need to insert three amino acids to make a furin cleavage site. Four amino acids went in asymmetrically. Why would anybody engineer that and do it that way, putting in an extra residue which is a proline, which puts kinks in proteins, it usually screws things up. And ultimately, that proline changed within a few — within one or two variants. So that didn’t make a lot of sense to me. But if you were going to engineer it, I guess the question would be, you don’t need to put four amino acids in, it’s easier to put three amino acids in, in the frame. And also, you’d probably want to put one in that was efficient. The sequence in SARS2 is not a very efficient cleavage site.”

Q: “So Dr. Garry was just kind of wrong?”

A: “You can make — no, I’m not saying he’s wrong. I’m just saying that means if it went in that way, then it was nefarious purposes to begin with, right? Because you’re basically trying to cover up what you did. I don’t think — I mean, when I looked at it, when it went in asymmetrically, that was more akin to recombination for me. Because recombination is not always perfect. Sometimes you have perfect recombination, but oftentimes, it’s offset and it introduces additional residue. One nucleotide or two nucleotides, depending on how it goes in, it’s sort of the random process of recombination.”

Let me address them below.

Why use an inefficient FCS?

This one is the easiest to answer, as Ralph Baric has answered it himself later in his testimony when he pointed to feline coronaviruses, where the change from an efficient to an inefficient FCS turns a benign enteric infection into a lethal one, potentially via ADE mechanisms. Thus, for researchers studying ADE and the role spike protein cleavage plays in it, picking an inefficient FCS — like PRRAR seen in multiple lethal strains of feline viruses where it causes ADE — to use for their research seems like a logical choice.

In fact, it seems that Dr. Baric himself suggested that reducing the FCS efficiency in a live virus would have been the final step in the research he proposed in DEFUSE:

The third thing was we would probably build virulent viruses and study pathogenesis, and then we would knock out the furin cleavage site.

As when speaking about the FCS mutations that turn benign feline CoV strains into lethal FIPV ones Dr. Baric referred to those mutations as “when the cleavage site is lost”, then “knocking out” an FCS could well include turning the canonical efficient FCS like RRARR into the inefficient PRRAR.

In fact, inspired by the same observations in feline and murine coronaviruses as Dr. Baric, a genetic engineer choosing to use PRRAR to “knock out” an FCS in a SARS-like strain, might well have started with RRARR as an initial efficient FCS — both feline enteric CoVs and MHV-JVH have RRARR as their FCS. For their inefficient FCS counterparts, MHV-A59 has RRAHR, while several FIPV strains have PRRAR. The latter also has a leading proline just like MERS, so a genetic engineer previously working on MERS could thus give preference to PRRAR.

Why include a leading proline?

I was actually surprised to hear this question from Dr. Baric, since he worked on MERS cleavage extensively, including creating a novel FCS in HKU4, a MERS-like bat strain. MERS, of course, also has a leading proline, as does the PRRAR furin cleavage site seen in multiple lethal feline coronavirus strains, as I just mentioned above.

Why is the insertion not in frame?

While the reading frame was obviously preserved by the PRRA insertion, the serine codon immediately preceding it is different in SARS2 than in its close relatives like BANAL-52 or RaTG13 (TCT vs. TCA):

This means that at the nucleotide level, relative to those strains, the insertion in SARS2 looks as if it has split the ancestral TCA codon with the 12 inserted nucleotides.

However, there are at least two other alternatives. One is that the direct progenitor of SARS2 prior to the insertion already had TCT for its serine, and the insertion of 12 nucleotides was fully in frame. Another possibility is that the direct progenitor of SARS2 prior to the insertion had TCA for its serine just like BANAL-52 or RaTG13, then the in-frame insertion happened, and then at some point between that and getting noticed in the first human patients, the TCA codon mutated to TCT.

Both scenarios are plausible, as SARS2 is hundreds of mutations away from its closest relatives like BANAL-52 or RaTG13, and several of those mutations are present in close proximity to the TCT codon in question, as can be seen in the figure above. In fact, the S1/S2 junction is generally a mutational hotspot, and under a lab leak scenario, the engineered virus would subsequently be subjected to testing in cells and/or animals, acquiring additional mutations before spilling over to humans.

Why use an insertion to create the FCS?

In line with the scenario where researchers studying the role of spike protein cleavage in MERS ADE were now looking to extend their research to SARS-like strains, the desire to equip a SARS2 progenitor with a MERS-like FCS can also explain why a genetic engineer might have chosen to do so via an insertion rather than mutation of existing nucleotides: the MERS spike protein is four residues longer at the S1/S2 junction than SARS-like viruses, and an insertion like PRRA removes this imbalance:

To see what impact the insertion has on the 3D structure of the protein, I aligned 3D structures of the MERS spike with two versions of the SARS2 spike: one where the FCS would have been created without an insertion (i.e. via mutation of existing residues) and another one as found in SARS2 normally (i.e. where the FCS arose via the PRRA insertion). It turned out, thanks to the PRRA insertion, the S1/S2 loop in SARS2 has its cleavage point ~3x closer to the MERS one than if the FCS in SARS2 was engineered without an insertion (i.e. via mutation of existing residues).

Here is a snapshot of the three spikes and the distances between the S1/S2 junctions relative to MERS (orange) between SARS2 (green) and a SARS2 mutant that has its FCS arise without an insertion (pink):

For a detailed explanation, see my 3D protein modeling video in this Twitter thread:

So if a genetic engineer was interested in creating a MERS-like FCS in a SARS2 progenitor, doing it via an insertion makes sense, especially in connection with studying ADE in SARS-like viruses.

In line with this hypothesis, SARS2 has two interesting unique mutations in its receptor binding domain (RBD) not seen in any other known SARS-like strains: R346 and A372. The first one is reminiscent of a SARS1 escape mutation from the S109.8 antibody, while the second removes a highly conserved N-linked glycan previously implicated in DC-SIGN binding (a receptor on immune cells involved in ADE) and ablated in a SARS1 vaccine candidate in a 2013 paper coauthored by Lanying Du and Shibo Jiang on RBD-based SARS1 vaccines. I described both of those unique SARS2 mutations in detail in my previous article:

SARS-CoV-2 is precisely the virus WIV was hunting for in 2019

In that article, I also review how in line with Ralph Baric’s explanation about the necessity to broaden the panel of viruses to test novel vaccines and therapies against, EcoHealth and WIV updated their virus hunting criteria to now look for SARS-like viruses that were 10–25% different from SARS1 in their spike but could still enter human cells. To accomplish that, they proposed sampling 5,000 bats in order to “almost fully characterize the expected natural diversity of SARSr- and other β-CoVs in the region”.

SARS2 fits those criteria like a glove: its spike is 24% different from that of SARS1, and yet it binds to the human ACE2 receptor even better than SARS1. Moreover, since SARS2 or its BANAL-like progenitor would have been able to evade SARS1-based vaccines and antibodies, such strains would have been prioritized and likely turned into full-genome synthetic backbones, as the joint EcoHealth/WIV grant renewal application suggested.

Pangolin coronaviruses in 2019

All of the lab leak puzzle pieces relating to relevant research done in 2019 in Wuhan are in place. The only missing piece is a SARS2 progenitor strain at WIV. But there are other known puzzle pieces that strongly suggest that, given the interests of WIV and its vaccine designing collaborators, they could have been actively searching for a BANAL-like progenitor of SARS2 in 2019. Those puzzle pieces are two SARS-like coronaviruses found in pangolin samples that were being actively researched in 2019. One such coronavirus is now known as GX/2017, and the other as MP789 or GD/2019.

GD/2019 is the most interesting one because it has a 97% identical receptor binding domain (RBD) to SARS2. It was found in tissue samples of confiscated contraband pangolins, whose potential infection by SARS-like viruses was first reported by the Guangdong Institute of Applied Biological Resources (GIABR), a close collaborator of WIV.

GIABR submitted their paper in September 2019. In it they described sequencing tissue samples collected earlier in 2019 from confiscated pangolins, and reported finding coronavirus genome fragments that matched to a number of SARS-like viruses. Most notably, in the “lung08” sample they reported a 500-nt fragment that aligned to the spike of the HKU3 virus, with a 82.6% identity. HKU3 was listed as one of EcoHealth/WIV’s viruses of interest in the DEFUSE proposal and the November 2018 NIH grant renewal application.

After the outbreak, metagenomic analysis of the sequencing reads from the same “lung08” sample showed that it contained reads matching the SARS2 receptor binding motif (RBM, the part of the RBD that comes into contact with the ACE2 receptor), and pooled reads from lung07, lung08, and lung09 samples would later produce a near-complete genome of the MP789 virus that shares 97% identity with SARS2’s RBD but, more importantly, has a near-identical RBM with SARS2: they share 77 out of 78 residues, including all six residues critical for efficient binding to ACE2.

The senior author of that 2019 GIABR paper on pangolin CoVs, Jinping Chen, collaborated with WIV and EcoHealth before — for example, on a 2014 paper sampling multiple bat species at the Xishuangbanna Tropical Botanical Garden in Yunnan, near the border with Laos, within a few kilometers of where a very notable SARS-like virus, RmYN02, was found in 2019. So it is quite possible that WIV was aware of GIABR’s findings regarding identifying a number of SARS-like viruses in pangolin samples. In that case WIV would have likely had access to those samples and/or sequencing data in 2019.

The GIABR analysis of its pangolin samples could have been spurred on by the earlier discovery of GX/2017 by Yigang Tong of the “State Key Laboratory of Pathogen and Biosecurity” at the Beijing Institute of Microbiology and Epidemiology — the same lab as Yusen Zhou’s. In fact, in their February 2020 paper, Dr. Tong’s group claims to have isolated GX/2017 long before the Covid outbreak. On May 1, 2024, Peter Daszak testified that he was familiar with Yigang Tong’s discovery even before publication, as they have discussed this during their in-person meeting:

Dr. Tong did collaborate with WIV and EcoHealth previously, e.g. in 2018 on a SADS-CoV paper titled “Fatal swine acute diarrhoea syndrome caused by an HKU2-related coronavirus of bat origin”. Libiao Zhang of GIABR, a longtime bat sampling partner of WIV, is also on that paper. But I am more intrigued about the collaboration between Yigang Tong and Yusen Zhou, as they both hail from the same lab. For example, in 2016 they published a joint paper.

What is even more intriguing is that Yigang Tong has filed his own patent for a SARS2 vaccine, and he did so on January 22, 2020 — a whole month prior to Yusen Zhou’s patent! That’s some speed: the SARS2 genome was only made public on January 10, and Wuhan didn’t even go into lockdown until January 23, and yet, by January 20 — judging by the text of the patent — Dr. Tong has already finalized his application.

If in 2019 Peter Dazsak was aware of a new SARS-like CoV found in pangolin samples by Yigang Tong, then Zhengli Shi, Yusen Zhou, Lanying Du, Shibo Jiang, and Fang Li must have been aware of it as well. They were also likely aware of the GIABR discovery. If they had access to even partial spike sequences of those viruses or raw sequencing data, WIV and its collaborators could have searched for RBD fragments of the spike that fit their “10–25% different from SARS1" criterion.

In that case, they would have seen that GX/2017 and GD/2019 RBDs not only fit that bill but they also show high similarity to another interesting strain that WIV sequenced in 2018, RaTG13 (or Ra4991, as it was known at the time). All of these strains are also antigenically different enough from SARS1 to escape SARS1-based antibodies, even more so than WIV1 and SHC014:

Moreover, in terms of the ability to bind to the human ACE2 receptor, computer modeling would have shown a progressive increase from RaTG13 to GX/2017 and finally to GD/2019.

That would have likely prompted a further search for viruses that shared the same RBD as GD/2019 — most likely via deep sequencing of pangolin samples and/or bat samples. Such search could well have produced a BANAL-52-like SARS2 progenitor which would then have been subjected to the kind of research described in DEFUSE and EcoHealth grants, including engineering a novel FCS and studying its ability to evade SARS1-based antibodies, as well as the propensity of those antibodies to cause ADE.

In fact, according to the declassified US Intelligence (ODNI) report on Covid origins, in 2019, WIV was analyzing at least some pangolin samples:

• Since 2019, some WIV researchers analyzed pangolin samples to better understand disease outbreaks in these animals.
…
• As of January 2019, WIV researchers performed SARS-like coronavirus experiments in BSL-2 laboratories, despite acknowledgements going back to 2017 of these virus’ ability to directly infect humans through their spike protein and early 2019 warnings of the danger of this practice. Separately, the WIV’s plan to conduct analysis of potential epidemic viruses from pangolin samples in fall 2019, suggests the researchers sought to isolate live viruses.

Thus, judging by the puzzle pieces around the only missing piece of a direct SARS2 progenitor, it seems that WIV could also have been looking for it back in 2019.

Addressing Ralph Baric’s arguments for natural origin

While Dr. Baric does not rule out a lab leak, he leans strongly towards a natural origin. Before I address his reasons for it, I want to highlight that despite his views, Dr. Baric does not think the Huanan market was the source of the outbreak, and he does not subscribe to the two-spillover hypothesis:

Q: “The 177 official WHO China corona reported cases, if you put the molecular clock to mid-October, then all of the activities around that — the market in Wuhan is actually two months or so?”

A: “It’s a major problem with that Wuhan study — that market study, yes.”

Q: “Can you just elaborate on that a little bit? I don’t have the expertise.”

A: “Okay, so keep it in context. The context is, what do you have data for? And the only thing we have really solid data for is that the market was the site of amplification in late December, January. That’s still two months from the origin date, based on a molecular clock, which means it was circulating somewhere before it got there. And the question is, where was it?”

Q: “To that point, I guess without getting too far away from our next set of questions, how hard — you’re talking about several hundred, if not several thousand human cases by the time you’re getting into January — early January, late December?”

A: “Remember that 90 percent of those cases are asymptomatic.”

Q: “Right.”

A: “85, 90 percent. So imagine trying to chase a transmission cycle. Early cases are almost impossible, because most — many asymptomatics are in the middle of it. So now you have a case here and a case here, but they’re actually truly linked by someone in the middle.”

Q: “Who just walked around with it.”

A: “Yeah. And you can’t unravel that transmission cycle until you do deep sequencing on both of them. And then you look for SNPs, and you can say, this patient is linked to this patient. It had to go through somebody else because there’s another marker. So all that — so it’s a fundamental problem with the papers that are reported to prove — they write it too strong, I think, but they’re very passionate about their data. And to be fair to them, it is the best data that’s out there, that they can’t — they don’t have the early cases. What they have, they have the cluster in the market and they have two SNPs, which they argue are indicative of two different zoonotic introductions, which other people argue with. It’s one nucleotide that’s making that call, so it — actually claimed there were two independent introductions.”

Q: “And they had some — “

A: “It’s a stretch. It’s a stretch. There are a lot of virologists that look at that data and go, mmm.”

Q: “Because it looks like, as I understand those two differences between the two lineages, it’s one looks marginally more like an ancestral bat virus?”

A: “Yes.”

Q: “And one looks a little more humanized?”

A: “At one nucleotide level. And they don’t know what the ancestral bat virus really was.”

Q: “Sure.”

A: “So from my perspective, clearly, the open market was a conduit for expansion of the disease. Is that where it started? I don’t think so.”

Q: “Keeping in mind the Chinese government’s ability to cover things up, is it at all worrisome to you or notable to you that we don’t have a second market or a third market or additional lineages coming out of nearby cities, like we saw with SARS1, where you had sort of a wave of spillover into the human population?”

A: “Remember that the Chinese Health Minister, I think on like the 24th of January, said community spread was rampant and asymptomatic spread was rampant. And they quarantined.”

Q: “A lot of people.”

A: “Within a few days of that, they quarantined 65 million. They came in and cleaned the market in Wuhan on, like, the 30th of December. What I don’t know is whether they went to every other market in Wuhan and other surrounding large metropolitan areas, or when they found them, they just wiped out — they cleaned those out. I don‘t think — I don’t have any information on it. I don’t know if you have any information on it.”

Q: “Not that we’ve seen.”

Incidentally, mid- or late-October spillover is well within range of the purported hospitalization in early November of the three members of the Zhengli Shi’s lab, reported as potential initial patients. Here is what Dr. Baric said about this:

Q: “The last major point that ODNI states is that there were Wuhan Institute researchers that were ill in the fall of 2019. The illness doesn’t necessarily support or refute either hypothesis or prove that it came from a lab. Did you have any awareness of any Wuhan Institute researchers being sick in the fall of 2019?”

A: “I’ve heard this report, but I’m not — and I’ve heard that they’ve been named, but I haven’t actually seen any of the data that supports that. So I don’t know how authentic it is. I mean, there’s, what, 5, 600 people who work in the Wuhan Institute of Virology. I don’t know the full number, but — and there was flu going on at the time, so it wouldn’t surprise me if they got sick. And I believe they — if they’re just getting physicals, they go to the hospital. So that’s their medical care system. So looking at it from that point of view, that doesn’t tell me anything.”

Q: “Okay.”

A: “I will also note one other thing. If you look at the molecular clock of the virus, it emerged in the middle of October, late October, not the middle or end of November. So people who say that those were the first cases, no chance. There were five or six transmission cycles at least before they would have been infected.”

The WIV staffers in question are actually reported to have gone to the hospital in early November:

Ben Hu and his colleagues Yu Ping and Yan Zhu were identified in media reports in June 2023 as the three workers who fell sick. All were from Shi Zhengli’s laboratory at the Wuhan Institute of Virology. Intelligence has suggested they were hospitalised in the first and second week of November — well before a cluster of cases in December were traced back to the Wuhan wet market.

So the timing of their illness does not contradict Dr. Baric’s views on the timing of SARS2’s spillover.

Also, if the Covid outbreak didn’t start at the Huanan market in Wuhan, where else could it start naturally, and get to Wuhan without infecting anyone else along the way? The nearest bat habitat is over a thousand miles away. Here is Dr. Baric’s explanation for why he feels zoonosis is much more likely than a lab leak:

Q: “So for a little bit of historical context, for zoonotic jumps with coronaviruses or even other viruses in general, could you just talk a little bit about how zoonotic jumps would happen or have happened?”

A: “In the context of coronaviruses?”

Q: “Or any other viruses, if that makes it easier for you to talk about.”

A: “Well, the first thing that has to happen is that human populations have to come into close contact with animals that encode these viruses. So that’s obviously the first thing. So there are, like, people in the extractive industry who may be loggers or hunters or, you know, gathers or collects bushmeat, those kind of people are the most likely to come into contact with zoonotic viruses and become infected. Now, the vast majority of contacts where zoonotic viruses actually are introduced into a human being, most of those don’t progress. The recent data with coronaviruses, for example, that was published in Southeast Asia argues that there’s somewhere between 50 to 60,000 exposures where people working with bats come in contact with bat coronaviruses, and actually seroconvert. That means they get infected, probably had very mild disease and recovered, 50,000. So if you think about how many — well, let’s put it in the context of coronaviruses. So 2002, SARS emerged; 2019, SARS2 emerged. That’s 17 years times 50,000 exposures a year, it’s actually a little higher. So about a million exposures between human disease outbreaks. So the vast majority of exposures are self-contained and do not transmit to another person, and then do not establish or colonize the new population. But this is occurring all the time. And so when you get to origins, for example, and you ask the question, what’s more likely, is it a lab leak or is it natural processes? You’re looking at one in a million, a million exposures occurring over 17 years versus what happens in a laboratory setting. No chance it’s even close. And the diversity in nature, hundreds of millions of times more diverse than what was in the Wuhan Institute of Virology. So that gradient is huge. And if you consider that, it’s more likely to be a natural event than it is to come out of the laboratory. The data — that’s what the data screams.”

…

Q: “And when we’re looking at the SARS-CoV-2 or COVID-19 pandemic, it sounds like you feel strongly that it was a zoonotic or natural origin. But would you say that it’s not settled yet what the origin of the COVID-19 pandemic was?”

A: “Again, I have at different times speculated on three possibilities. The first is natural origin. The second is accidental escape from the laboratory setting, which can also include collection, which you can ask about if you’d like more details on that. And then the third would be the possibility of engineering. There is no hard evidence to support engineering. Initially, for example, the receptor binding domain was argued to be completely unique and perfectly positioned, perfectly designed to bind the human ACE2 receptor. Well, no, there are virtually identical strains in bat strains that are found in nature. So it’s not been engineered. In addition, that spike has undergone successive sets of — the RBD has gone successive adaptive changes that increase bind infinity for the ACE2 over a thousand fold. It is not perfectly designed. It’s just like the origin SARS1, which underwent specific changes that enhanced its transmissibility as it was spreading. The exact same process. So the RBD is out. The second idea that it was engineered, there was a very bad bioinformatic paper, for example, that said — it came from the HIV — which was total nonsense. The better argument was that there might be a super antigen site, but there was a paper that was just published that said, no, there’s no super antigen site. So, in essence, the scientific process says, okay, if this is the hypothesis, let’s do experiments to see if we can disprove it. If we can’t disprove it, then it’s likely. So far there’s no backbone genome that’s close enough to have been engineered in the SARS2. Most of the components that were originally argued as being engineered failed. The only one that’s left is the furin cleavage site, which has multiple explanations. So that leaves two possibilities. The first is escape from the laboratory. And you can’t rule that out, because they do work at BSL-2. You just can’t. But for the reasons I talked about earlier, just on the frequency and the exposure level in nature versus lab, it’s massively — what’s that called, massive — the scales are massively weighted to natural origins, yes, sorry.”

Q: “Sure. And taking out bioengineered, I think there’s much consensus that that is not what we’re looking at here. But with the lab leak and zoonotic, there would be possibilities for it to be somewhat more of a combination of the two. I’m thinking about, specifically, you said researchers go out and collect samples, they bring them back to the lab. Maybe they do no manipulation on it, so it’s just whatever they collected out in nature. Something happens, there’s a lab accident, and somebody is exposed to a virus and gets infected. While I understand this would be very rare, that would sort of be a combo of a lab accident with a natural virus, correct?”

A: “Yes, and still be a natural virus that inadvertently escaped the laboratory, because biosafety practices weren’t sufficiently robust. Now, when you think about collection, at least the group at EcoHealth and the groups that they collaborate with, again, I haven’t been in the cave with them, but the pictures that I have seen is they’re fully dressed in Tyvek suits and with all the protective gear. So, in essence, they are collecting — in essence, in laboratory appropriate conditions, and then bringing the samples back. Their weakness is trying to culture the viruses at BSL-2. It’s just the chance of an accident is increased under BSL-2 conditions, as compared to BSL-3.”

In a nutshell, Dr. Baric feels that a natural origin is much more likely because he believes there are 50–60 thousand people quietly getting infected with SARS-like viruses in Southeast Asia every year. Where did he get this number? Turns out, from Peter Daszak and Zhengli Shi.

I won’t go too deep into their analysis, although I do feel that their methodology is flawed. For example, the key metric of their entire paper, how often contact with bats results in a detectable SARS-like infection, is based on poorly substantiated data — just a single observation of seropositivity out of 199 tested people who had contact with bats. Moreover, there is some weirdness with that number, tucked away in Supplementary Table 4:

The reference 19 used to substantiate it points to a 2019 paper by the same authors which previously reported:

Serological testing of serum samples from 1,497 local residents revealed that 9 individuals (0.6%) in four study sites were positive for bat coronaviruses, indicating exposure at some point in their life to bat-borne SARSr-CoVs (n=7, Yunnan), HKU10-CoV (n=2, Guangxi), or other coronaviruses that are phylogenetically closely related to these. All individuals who tested positive (male=6, female=3) were over 45 years old, and most (n=8) were making a living from crop production.

How the authors extracted a subset of 199 tests of the people who had contacts with bats from the above data is unclear. The only plausible explanation is a subset of 201 people who reported having bats in their house, of whom 1 tested positive for some coronavirus (as indicated by the red number 1 in their table):

However, the 2022 paper citing those 2019 data has two instances of positive tests — one for a SARS-like virus and another one for HKU10-CoV, while the 2019 paper does not report any co-infections, and both HKU10-CoV positive tests are reported to have been from Guangxi, while all seven SARS-like CoV positive ones from Yunnan.

But the biggest observation from the 2019 data is that 8 of the 9 people (out of 1,497) testing positive for exposure to SARS-like or HKU10-CoVs reported no contact with bats! This undermines the 2022 paper’s entire thesis that it is contact with bats that is the primary source of human exposure to these viruses. Clearly, that wasn’t the case in their 2019 paper.

Curiously, before the Covid outbreak, Peter Daszak and Zhengli Shi were claiming the risk of SARS-like spillover was quite low. In their 2018 paper titled “Serological Evidence of Bat SARS-Related Coronavirus Infection in Humans, China” they reported that only 6 of the 218 villagers living within 1–6 km of the caves WIV sampled in (including Shitou cave where the SHC014 virus was found) were seropositive for SARS-like viruses. Ironically, 240 Wuhan residents were used as control.

In a wider follow-up study published in September 2019, just before the outbreak, Peter Daszak and Zhengli Shi reported even lower seropositivity prevalence — 0.6% (9 out of 1,497) that I already quoted above.

However, after the outbreak happened, Peter Daszak was pointing to their earlier, higher seropositivity prevalence (6 out of 218 villagers, who were originally called a “high risk group of residents living in close proximity to bat colonies” in the 2018 paper), and extrapolating their findings to all of SE Asia, arriving at “1–7 million people per year exposed to bat origin SARS-related CoVs”:

Ultimately, “1–7 million people per year exposed to” evolved into “66,280 people … infected with SARSr-CoVs annually in Southeast Asia”, and that became the pro-zoonosis talking point that Ralph Baric based his assessment of the likelihood of Covid’s natural origin on.

It is interesting how the authors also expanded their focus from China to all of Southeast Asia. This is another problem with using their 2022 paper to infer Covid origin: out of the 499 million people they included in their calculation, less than 10% live in China, while the majority live thousands of kilometers away from Wuhan — e.g. 151 million live on the island of Java alone. Thus, the relevance of such distant bat exposure to an outbreak in Wuhan is almost as moot as it would have been if the Covid outbreak happened near Ralph Baric’s lab in North Carolina — just because such a virus could have spilled over naturally thousands of kilometers away doesn’t explain why it actually spilled over in an urban center near a lab studying viruses just like it, 1500 km away from its bat habitat.

Coming back to Ralph Baric’s argument, it is also odd that he brackets the timeline of SARS2 emergence by 17 years since the emergence of SARS1, because SARS2 didn’t have just 17 years to emerge naturally to cause a Covid-like pandemic, it had decades if not centuries. Yet, it emerged only a year after WIV and EcoHealth, inspired by Baric, started to look for SARS-like viruses just like it: with a spike 10–25% different from SARS1 that is still able to infect human cells while evading SARS1-based antibodies.

Moreover, miraculously, the virus that WIV planned to hunt for in remote bat caves in Yunnan, just happened to emerge on its doorstep. Should I even mention that the virus also had a unique feature that WIV and its DEFUSE coauthors wanted to look for and engineer, a furin cleavage site? Which no other known SARS-like virus has. All of these incredible coincidences, as well as their timing synchronized with a shift in WIV’s research interests should skew probabilities much more towards a lab leak than zoonosis.

The lab leak case

Let me outline a potential lab leak scenario based on all of the above. In 2018 and 2019, US and Chinese coronavirologists were engaged in intense research focused on vaccine development and ADE in the context of developing broad-spectrum vaccines for SARS-like and MERS-like viruses. Ralph Baric was working on a US-based vaccine platform, while Lanying Du, Yusen Zhou, Fang Li, and Shibo Jiang were doing a lot of vaccine research in China.

Despite previously working closely with Ralph Baric and EcoHealth, by 2019, Lanying Du, Fang Li, and Shibo Jiang have significantly decreased their collaboration. Conversely, their joint work with WIV increased, and some research plans outlined in DEFUSE were likely now being done by WIV in collaboration with those researchers rather than Ralph Baric or EcoHealth. On the vaccine research front, WIV was now working with Lanying Du’s husband, Yusen Zhou.

By 2019, WIV held all the cards, as by then they had humanized mouse models and genetic engineering expertise, and, most importantly, they had access to bat caves, while their foreign collaborators were at their mercy when it came to finding novel “interesting” viruses, as Dr. Baric testified:

A: “It would have been going in — if year 6 was around 2019 or 2020, that’s when I would have been a part of [the EcoHealth/WIV grant]. And my role was to study a couple of the viruses that the Wuhan Institute of Virology found that they were willing to share with me. So I always viewed that as not number one or number two on the list, maybe number five or number six on the list.”

Q: “I understand, I think I understand what you’re saying. But when you say not one or two on the list, but number five on the list, is that as far as they are giving you the fifth most interesting virus that they had found?”

A: “Well, to be fair to them, they did the discovery work and they’re going to choose the priority of what they want to work on first. And so I’m not going to get the dregs, that would be an unfair characterization, but I’m not going to get number one. I’m going to get somewhere down the list, which is okay, and I understand that process. Hopefully, it would be something that they felt would be interesting as well.”

Needing to expand their diverse panel of viruses to evaluate the effectiveness of new broad vaccines, WIV and its collaborators kept looking for novel viruses. In particular, for SARS vaccines, they were interested in novel SARS-like viruses, differing from SARS1 by 10–25% in its spike protein. They were also interested in strains that could escape SARS1-based antibodies, which at the same time could lead to ADE.

Inspired by Ralph Baric’s ideas about furin cleavage sites in SARS-like viruses, as well as his joint work with Fang Li on SARS and MERS spike cleavage, WIV and its collaborators were also likely looking to engineer furin cleavage sites in novel SARS-like strains that fit their search criteria. Based on their interest in vaccines and ADE, extending their investigation of MERS ADE to novel SARS-like CoVs that could escape SARS1-based antibodies seemed like a logical next step, as did using an FCS from feline coronavirus strains known to exhibit ADE, especially since that FCS was similar to the one in MERS.

So when in late 2019 SARS2 emerges against this backdrop in Wuhan — with its spike 24% different from SARS1 and yet extremely adept at binding the human ACE2 receptor, and with a MERS-like furin cleavage site identical to the one found in several lethal feline coronaviruses — it is much more plausible that this virus materialized in Wuhan through the research that was being conducted there, rather than by a series of incredible coincidences in nature.

Lab leak critics often say that before the pandemic, the Wuhan Institute of Virology would not have considered a SARS2 progenitor interesting. “WIV was only interested in SARS-like viruses that were very close to SARS1,” they say.

In actuality, the exact opposite is true — in late 2018, EcoHealth and WIV outlined their updated virus hunting criteria for 2019–2023, which show that they were now looking for SARS-like viruses that were 10–25% different from SARS1 in their spike but could still enter human cells. SARS2 fits those criteria like a glove: its spike is 24% different from that of SARS1, and yet it binds to the human ACE2 receptor even better than SARS1.

Moreover, as I will show below, since SARS2 or its BANAL-like progenitor would have been able to evade SARS1-based vaccines and antibodies, such strains would have been prioritized and likely turned into full-genome synthetic backbones.

The good old days of 2018

Let’s travel back in time to November 2018. By this time, EcoHealth has been working closely with WIV for many years on a number of projects involving MERS-like and SARS-like viruses. On the SARS-like front, they had a 5-year NIH grant R01AI110964 (2014–2019) that was up for renewal at the time.

Just a few months prior, in March 2018, EcoHealth pitched a $14M mega-grant proposal for DARPA suggesting to sample thousands of bats in three Yunnan caves in search of novel SARS-like viruses, which the grant proposal authors then promised to “DEFUSE” by immunizing bats against them with “novel chimeric polyvalent recombinant spike proteins” (emphasis added):

Our goal is to defuse the potential for spillover of novel bat-origin high-zoonotic risk SARS-related coronaviruses in Asia. In TA1 we will intensively sample bats at our field sites where we have identified high spillover risk SARSr-CoVs. We will sequence their spike proteins, reverse engineer them to conduct binding assays, and insert them into bat SARSr-CoV (WIV1, SHC014) backbones (these use bat-SARSr-CoV backbones, not SARS-CoV, and are exempt from dual-use and gain of function concerns) to infect humanized mice and assess capacity to cause SARS-like disease.
…
In TA2, we will evaluate two approaches to reduce SARSr-CoV shedding in cave bats: (1) Broadscale immune boosting, in which we will inoculate bats with immune modulators to upregulate their innate immune response and downregulate viral replication; (2) Targeted immune boosting, in which we will inoculate bats with novel chimeric polyvalent recombinant spike proteins plus the immune modulator to enhance innate immunity against specific, high-risk viruses. … The most effective biologicals will be trialed in our test cave sites in Yunnan Province, with reduction in viral shedding as proof-of-concept.

As the DEFUSE grant proposal was not funded, parts of it made it to EcoHealth’s November 2018 grant renewal application for their NIH grant R01AI110964. Now EcoHealth and WIV were proposing to sample 5,000 bats in an attempt to “almost fully characterize the expected natural diversity of SARSr- and other β-CoVs in the region”. Their new focus on a particular subset of SARS-like strains was clear:

1.4.b Viral strain prioritization: Of the expected 100–200 novel SARSr-CoV strains, we will down-select to prioritize for further characterization based on S genes that are: i) different from SHC014, WIV1, SARS-CoV with diversity ranges of 10–25%; ii) have virus S RBD that could use human/bat receptors; iii) have recombinant chimeric spikes indicative of gene flow between clade I and II strains; iv) have bat ACE2 receptors that might select for spike RBDs that can use human receptors for entry (15/18 conserved residues in human/bat ACE2 molecules that bind SARSs-CoV S RBD domains are likely more efficient receptors than 3/18 conserved sites).

SARS2’s spike is 23% different from those of SCH014 or WIV1, and 24% different from that of SARS1. Similarly, the spikes of BANAL-52/103/236 strains are all 22% different from those of SCH014 or WIV1, and 23% different from SARS1's. But in terms of the ability to bind to the human ACE2 receptor, the RBD of SARS2-like strains is even better than SARS1, which would have been apparent even from computer modeling.

Here is what the grant authors planned to do with these strains (emphasis added):

We will use S protein sequences to select a range of viral strains that cover the 10-
25% S protein divergence we predict as high public health potential and construct chimeric SARSr-CoVs using the WIV1 backbone and these S genes as done previously (12, 18, 38). We will rescue of full-length clones and assess infection of non-permissive cells expressing human, bat and civet ACE2 receptors, Vero cells, primary human airway epithelial cells, and CaCo cells for HKU3r-CoVs (which have not been cultured and may use intestinal epithelium in nature). We will conduct experimental infections in hACE2 transgenic mice to
assess pathogenicity and clinical signs (18). Finally, using a panel of mAbs that neutralize SARS-CoV infection in vitro and in vivo, and vaccine against SARS-CoV S protein, we will examine the capacity of strains with divergent S protein sequences to evade therapeutics, revealing strains with high public health potential.

The last bit is particularly interesting. The grant renewal application elaborates (emphasis added):

3.3.b Primary human airway epithelial cell culture: …SARSr-CoVs that differ significantly in S protein sequence (11–24%) from epidemic SARS-CoV yet replicate in vitro, will also be evaluated for sensitivity to neutralization in Vero cells by PRNT50 assays using broadly SARS-CoV cross reactive human mAbs S227.14, S230.15, S215.17, and S109.8 (49, 55).
…
3.3.c Humanized mouse infection experiments: …hACE2 transgenic mice will be injected with SARS-CoV mAbs, and infected with chimeric bat SARSr-CoVs. Clinical signs and morbidity will be assessed and tissue pathology examined and compared with mice without treatment of mAbs to determine the therapeutic effect on SARSr-CoV infection, and protection of SARSr-CoV by wildtype SARS-S based vaccines assessed as described (56, 66). We will sequence full length genomes of high risk strains that are antigenically distinct and escape SARS cross neutralization, synthetically reconstruct a small subset (1–2) and evaluate the ability of nucleoside analogues to inhibit growth in HAE cultures and/or in vivo (55, 56).

The last sentence shows just how interested the grant authors were in novel SARS-like viruses that could escape SARS1-based neutralization. So interested that they planned to create synthetic backbones out of them for further testing in cells and/or mice to see if nucleoside analogs can inhibit their growth.

Previously, EcoHealth pitched this type of work to DARPA in their DEFUSE mega-grant (while promising WIV that they too will get to do some of this work if the grant gets funded, despite the proposal saying that this work was to be done at UNC):

Note that some nucleoside analogs (e.g. ribavirin or molnupiravir) are mutagens which are meant to inhibit viral growth by introducing a lot of mutations. The flip side of this, as the molnupiravir saga shows, is that mutagens can also accelerate viral evolution. When such mutagens are used during viral passaging in the lab (in vitro or in vivo), this could phylogenetically “age” passaged strains at a much quicker pace than what would have been observed in nature.

Escape from SARS1-based immunity

So why do I say that SARS2 or its BANAL-like progenitors would have grabbed the grant authors’ attention? Because not only do they fit the “10–25% spike difference while still binding to human ACE2" criteria, but they are also antigenically very distinct from SARS1 and are practically certain to evade SARS1-based mAbs or vaccines. Here’s why.

The “broadly SARS-CoV cross reactive human mAbs S227.14, S230.15, S215.17, and S109.8” that the grant renewal application names as the neutralization panel to test novel SARS-like CoVs against are the key to understanding this. In 2008, Ralph Baric’s group investigated these mAbs and has produced the following figure nicely summarizing which epitopes they target on the SARS1 spike protein:

Now, if we compare the spikes of SARS2 and its related strains to those of SARS1 in the above figure, we see something quite striking:

Namely, all of the epitopes in the receptor binding domain (RBD) and almost all other epitopes in S1 that Baric’s group has identified as being the targets of those mAbs are different in SARS2-like viruses.

Moreover, in 2010, the Baric group followed up on this work by creating SARS1 escape mutants against these and other mAbs. They’ve observed the following escape mutations in the RBD:

Again, we see that SARS2-like strains already have different amino acid residues than SARS1 in all but one location relevant for these mAbs. Taken together, these observations suggest those antibodies would almost certainly be ineffective against SARS2-like strains. Coupled with how well SARS2 or BANAL-52/236/103 bind to the human ACE2 receptor, this would have made such strains the top priority for the grant authors as per their November 2018 proposal.

Could the unique SARS2 R346 mutation be an escape mutation?

As an aside, I was intrigued to see that one of the SARS1 mAb S109.8 escape mutants from the 2008 Baric group paper had a K333N mutation in the RBD:

Four out of six plaques of the S109.8 escape mutants contained a single amino acid change at T332I, while two plaques contained a single amino acid change in an adjacent residue at position K333N.

That SARS1 escape mutation corresponds to a notable SARS2 RBD mutation (R346) relative to its ancestral bat strains. Moreover, that SARS2 location is the most frequent reversion (R346T) — i.e. mutation back to the amino acid residue present in ancestral strains — observed in over 11% of circulating SARS2 strains:

Based on this, I wonder whether the SARS2 R346 mutation could be a sign of SARS2 having encountered selective pressure similar to that exerted by the S109.8 antibody on the SARS1 escape mutant in the 2008 Baric group paper, because in 2016, the Baric group — in the paper titled “SARS-like WIV1-CoV poised for human emergence” — subjected WIV1 to neutralization testing with several antibodies, including S109.8, and then talked about the very same AA position 333 of SARS1/WIV1 spikes: the authors noted the different AA residues between SARS1 and WIV1 as the likely reason for reduced neutralization efficiency of S109.8 against WIV1 relative to SARS1 (emphasis added):

Having established a potential threat based on replication in primary human cells and preference for the human ACE2 receptor in vivo, we next sought to determine if monoclonal antibody therapies could be used to lessen disease similar to ZMApp for Ebola (13). We first tested a SARS-CoV monoclonal derived via phage display and antibody escape (Fm6) (14) and found both wild-type SARS-CoV Urbani and WIV1-MA15 were strongly neutralized at low antibody concentrations (Fig. 4A). Similarly, a panel of monoclonal antibodies derived from B cells from SARS-CoV–infected patients also prevented virus infection via WIV1-CoV spike (15, 16). Both antibodies 230.15 and 227.14 robustly inhibited WIV1-MA15 replication with kinetics similar to or exceeding SARS-CoV Urbani (Fig. 4 B and C). In contrast, antibody 109.8, which maps outside the receptor binding domain, produced only marginal neutralization of WIV1-MA15 (Fig. 4D). Whereas the residue associated with prior escape mutants was conserved at position 332, the adjacent residue had a significant change (K33[3]T) [332 instead of 333 was a typo in the paper — Y.D.] in WIV1-CoV, possibly contributing to reduced efficacy of this antibody.

S109.8 is even more relevant in the context of SARS2 because, as I have shown above, SARS2 has a very different RBD neutralization profile than SARS1 and thus antibodies like S230.15 and S227.14 that are very good at neutralizing SARS1 and SARS1-like strains (like WIV1) would almost certainly be ineffective against SARS2. But S109.8, on the other hand, “which maps outside the receptor binding domain”, as the Baric group noted, could still provide at least some “marginal neutralization” as it did for WIV1. After all, WIV1 and SARS2-like strains have identical residues in positions 332 and 333 (T and T; numbering is relative to the SARS1 spike). SARS2 itself, however, is unique among SARS2-like strains in having an R in that position (346 relative to SARS2 numbering), reminiscent of the 2008 S109.8 SARS1 escape mutants.

Was this work already underway in late 2019?

Now, fast forward a year to November 2019, and Peter Daszak, EcoHealth’s President, implies that some of the work described in their grant renewal application has already been done:

A few weeks later, in a December 9, 2019 interview, he repeated this claim:

“we’ve now found — after, you know, six or seven years of doing this — over a hundred new SARS-related coronaviruses very close to SARS. Some of them get into human cells in the lab, some of them can cause SARS disease in humanized mouse models and are untreatable with therapeutic monoclonals and you can’t vaccinate against them with the vaccine”

Interestingly, in the above interview Daszak mentions having found over a hundred new SARS-related coronaviruses, while in the November 2018 grant renewal application EcoHealth only mentioned 52:

We sampled and PCR-screened >16,000 individual bats from 6 families (16 genera) in southern China, finding 9 species positive (5,730 individuals screened) for SARSr-CoVs (Table 1, Fig. 1). We identified 178 novel β-CoVs, of which 172 were novel (52 novel SARSr-CoVs).

It would be interesting to know whether Daszak was just exaggerating or whether EcoHealth has truly identified an additional ~50 novel SARS-like CoVs between November 2018 and December 2019. Judging by the increase in the numbers of sampled bats between the March 2018 DEFUSE proposal and the November 2018 NIH grant renewal application, EcoHealth and WIV have been sampling new bats quite actively:

In particular, they report additionally sampling 1,023 R. pusillus bats, which are the host species of BANAL-103.

It would also be very interesting to know what were the novel SARS-like strains they tested which were able to evade mAbs and vaccines and what they ended up doing with them. Because in the very same December 2019 interview Peter Daszak seems to imply they wanted to use the spike proteins from these strains for SARS vaccines (emphasis added):

[Interviewer]: you’re saying these are diverse coronaviruses and you can’t vaccinate against them, and there are no antivirals, so what do we do?

[Daszak]: …you can manipulate [coronaviruses] in the lab pretty easily — it’s just the spike protein that drives a lot of what happens with zoonotic risk. So you can get the sequence, you can build the protein — and we work with Ralph Baric at UNC to do this — insert it into the backbone of another virus and do some work in the lab. So you can get more predictive when you find a sequence, you’ve got this diversity. Now the logical progression for vaccines is — if you’re going to develop a vaccine for SARS, people are going to use pandemic SARS, but let’s try and insert some of these other related [spikes] and get a better vaccine.

[Interviewer:] I guess also the knowledge of what’s out there, if you see something emerging, it can give it a head start on making a vaccine or therapeutic?

[Daszak]: that’s true

WIV’s interest in SARS1-based immunity escape was quite recent

Against the backdrop of Baric group’s investigation into SARS1-based mAbs and vaccines in 2008, its analysis of SARS1 escape mutants in 2010, and its evaluation of how novel SARS-like CoVs like WIV1 could escape neutralization in 2016, their perennial scientific “coopetitor”, WIV, was actively trying to keep up.

In December 2017, Zhengli Shi co-authored a study titled Cross-neutralization of SARS coronavirus-specific antibodies against bat SARS-like coronaviruses which looked into whether SARS1-based antibodies can protect against WIV1 and SHC014 (essentially, the reference SARS-like bat viruses named in DEFUSE and the NIH grant). It seems that their findings were the inspiration behind the new focus for 2019–2023 on novel SARS-like CoVs whose spikes are 10–25% different from that of SARS1, as those were the strains most likely to evade SARS1-based protection:

In summary, our results have demonstrated that most SARS-CoV RBD-specific antibodies tested in this study could cross-neutralize SL-CoV strain WIV1, but not SHC014. While SARS-CoV and WIV1 have comparable RBD, the RBD of SHC014 is much more variable, as previously noted (Figure S1 in Supporting Information). More specifically, the RBD of SHC014 has a difference of 24 amino acids (aa) compared to that of SARS-CoV, while the RBD of WIV1 only has a difference of 8 aa. This may explain why SHC014 could not be cross-neutralized effectively by most antibodies (mAbs or pAbs) targeting SARS-CoV RBD. The fact that SHC014 retains its ability to infect human cells implies that the available antibodies and vaccines based on SARS-CoV RBD will not protect the next SARS-like disease caused by bat SL-CoV SHC014 strain.

It seems that such strains were sometimes referred to as “highly variable strains” by the grant authors — at least that’s my read based on the quote above (“the RBD of SHC014 is much more variable”) and the description in one of the DEFUSE drafts (emphasis added):

Low abundant mutations, especially in RBD residues that interface with ACE2 receptors, would alter risk assessment calculations as strains identified as low risk, might actually encode high risk, but low abundant variants. To test this hypothesis, we will closely with the modeling core and Dr. Shi’s laboratory to identify highly variable residue changes in the SARSr-CoV S RBD, and use commercial gene blocks to introduce these changes singly and then in combination into the S glycoprotein gene of the parental low risk, high abundant strain parent. We will evaluate the ability of these low abundant chimeric viruses to use human, bat, civet and mouse ACE2 receptors, and more importantly, replicate efficiently in human primary cells.
…
we will synthesize full length rs4237, a highly variable SARSr-CoV that encodes the SHC014 RBD contact interface residues at 442, 487 and 491 but also encodes mutation at 479 (N479S) and has the 432–437 and 458–472 deletions

In light of that, it is quite interesting that based on their January 13, 2020 email exchange — just days after the SARS2 genome was made public — both Ralph Baric and Peter Daszak realized immediately that SARS2 is the “highly variable SARS-like CoV” they set out to hunt for in late 2018:

I wonder if it gave them pause that exactly the virus they decided to hunt for just a year before has now emerged on WIV’s doorstep — and nowhere else — instead of near its natural habitat 1500 km away.

Familiar faces

Coming back to the above 2017 WIV antibody study, it is notable that besides Zhengli Shi its coauthors include Shibo Jiang and Lanying Du. The latter is the widow and close colleague of Yusen Zhou — the SARS and MERS vaccine expert (and reportedly a decorated PLA officer) who filed a patent for a Covid vaccine in February 2020 and then died a few months later under mysterious circumstances.

Shibo Jiang, Lanying Du, and Yusen Zhou have been close colleagues for over a decade, having initially worked on SARS vaccines as far back as at least 2007, when both Lanying Du and Yusen Zhou still held positions with the State Key Laboratory of Pathogen and Biosecurity at the Beijing Institute of Microbiology & Epidemiology. Shibo Jiang’s focus on SARS vaccines and mAbs goes back even farther, as his 2005 and 2006 patents show.

The three researchers have also written extensively on spike-based MERS vaccines from 2013 to 2018, and coauthored joint patents on vaccines against MERS and influenza. By 2019, they have been collaborating for many years under a grant R01AI098775 titled “RBD recombinant protein-based SARS vaccine for biodefense” and have published many papers on SARS and MERS spike-based vaccines:

In 2017, the trio published a paper on a MERS vaccine titled “Recombinant Receptor-Binding Domains of Multiple Middle East Respiratory Syndrome Coronaviruses (MERS-CoVs) Induce Cross-Neutralizing Antibodies against Divergent Human and Camel MERS-CoVs and Antibody Escape Mutants” which implemented the exact strategy that Peter Dazsak talked about in his interview above: the vaccine used multiple RBDs from several MERS-CoVs to generate broader immunity.

In light of the previously mentioned 2017 joint paper with WIV on SARS cross-neutralization, it would be logical to assume that the frequent collaborators were next planning to try the same approach for broad SARS vaccines. In fact, based on this manuscript submitted to biorXiv in May 2020, Lanying Du and Shibo Jiang were indeed working on a SARS vaccine in 2019, under their usual R01AI098775 grant. In that manuscript, the authors mention their interest in the multi-RBD approach for a broad coronavirus vaccine.

Their 2020 paper was a continuation of the work started as far back as 2013, as described in this paper, in which the authors have designed and evaluated several RBD-based SARS vaccine candidates. Intriguingly, in that study led by Shibo Jiang, the authors mention deleting three N-linked glycosylation sites found in the SARS1 RBD:

This is notable because one of those sites — N-40 in the diagram, N357 in the SARS1 spike, mutated to an A in the glycan-ablated mutants RBD193-N3 and RBD219-N3 in the paper above — corresponds to the N370 glycan in SARS2 which mysteriously appears to have been ablated via a T372A mutation not seen in any ancestral SARS2-like strains.

I find this coincidence intriguing because the DEFUSE proposal also suggested ablating some N-linked glycosylation sites in SARS1 and SCH014 to evaluate their role in cross-species transmission and impact on pathogenesis:

SARS S has 23 potential N-linked glycosylation sites N-linked glycosylation sites (NX(S/T); X is anything but proline) and 13 of these have been confirmed using biochemical approaches (e.g., positions: 118, 119, 227, 269, 318, 330, 357, 783, 1056, 1080, 1140, 1155, and 1176). Importantly, several N-linked glycosylation sites regulate SARS particle binding DC-SIGN/L-SIGN, alternative entry receptors for SARS-CoV infections (positions: N109, N118, N119, and especially N227, N699)( PMC1641789, PMC2168787, PMC2168787) and may protect critical sites for antibody neutralization(PMC5515730). Importantly, the emergence of human SARS-CoV from civet and raccoon dog reservoir strains was associated with the evolution of mutations that introduced two N-linked glycosylation sites that promote DC-SIGN/L-SIGN binding (N227, N699), suggesting a role in the expanding human 2003 epidemic (PMC2168787). Interesting[ly], these N-linked glycosylation sites are absent from civet, raccoon dog strains and clade 2 SARSr-CoV, but are present in WIV1, WIV16 and SCH014 as well as human epidemic [SARS-CoV] strains, supporting a potential role in host jumping (Fig C).

To evaluate the role of these mutations in cross species transmission and pathogenesis, we will sequentially introduce clade 2 residues at positions N227 and N699 of SARS-CoV and SCH014 and evaluate virus growth in Vero, Huh7 cells (nonpermissive) expressing ectopically expressed DC-SIGN and HAE cultures, anticipating reduced virus growth efficiency. Using the clade 2 rs4237 molecular clone described above, we will introduce the clade I mutations that introduce N-linked glycosylation sites at positons 227 and N699 and in rs4237 RBD deletion repaired strains, evaluating virus growth efficiency on Vero, HAE or Hela cells ± ectopically expressing DC-Sign (PMC2168787). In vivo, we will evaluate pathogenesis in transgenic ACE2 mice. Experimental outcomes from these studies will be incorporated into risk management models to identify high and low risk SARS-like CoV.

DC-SIGN is a receptor found on the surface of dendritic cells and macrophages, and DC-SIGN binding is relevant in the context of antibody-dependent enhancement of viral entry (e.g. in MERS), which, as I will show later, was a hot topic for WIV and its collaborators in 2019. While the DEFUSE quote above highlights glycans N227 and N699 as important in DC-SIGN binding, the same quote also references another paper (PMC1641789) which instead points to glycans N330 and N357:

We therefore suggest that residues N330 and N357 of the S protein may play important roles for interaction between the S glycoprotein and DC-SIGN.

Thus, it seems that the N357 glycan has been on the radar of the DEFUSE authors and their longtime collaborators, and seeing it ablated in SARS2 is certainly notable.

Of course, the DEFUSE proposal is most notable for mentioning another unique feature of SARS2 — a furin cleavage site, which none of the other known hundreds of SARS-like viruses have. Here is a detailed description of the authors’ interest from one of the earlier drafts of DEFUSE (emphasis added):

After receptor binding, cell surface or endosomal proteases cleave the SARS S glycoprotein to activation fusion mediated entry (PMC4151778). Massive changes in Spike structure occur to mediate membrane fusion and entry (PMC5651768). The absence of S cleavage prevents SARS-coV entry (PMC5457962). A variety of proteases, including TMPRSS2, TMPRSS11a, HAT, trypsin and cathepsin L carry out these processes on the SARS S glycoprotein (PMC3233180, PMC3889862, PMC5479546, PMID:26206723)(Fig C). In some instances, tissue culture adaptations introduce a furin cleavage site, which can direct entry processes as well, usually by cleaving S at positions 757 and 900 in S2 of other coronaviruses, but not SARS (PMID:26206723). For SARS-CoV, a variety key cleavage sites in S have been identified including R667/S668, R678/M679 for trypsin and cathepsin L, respectively, R667 and R792 (and other unidentified sites) for TMPRSS2, and R667 for HAT. Therefore, all SARSr-CoV S gene sequences will be analyzed for the presence of these appropriately conserved proteolytic cleavage sites in S2 and for the presence of potential furin cleavage sites (R-X-[K/R]-R↓) and which can be predicted computationally (PMC3281273). Importantly, SARr-CoV with mismatches in proteolytic cleavage sites can be activated by exogenous trypsin or cathepsin L (Fig D), providing another strategy to recover non-cultivatable viruses. In instances where clear mismatches occur in these S2 proteolytic cleavage sites of SARSr-CoV, we will introduce the appropriate human-specific cleavage sites and evaluate growth potential in Vero and HAE cultures.

The DEFUSE authors’ interest in looking for existing furin cleavage sites in SARS-like viruses and potentially engineering novel ones likely stems from their previous work with MERS: in 2015 Ralph Baric, Shi Zhengli, Shibo Jiang and Lanying Du coauthored a paper showing that engineering a novel FCS in a MERS-like HKU4 virus and ablating a single N-linked glycan enabled it to enter human cells:

To evaluate the potential genetic changes required for HKU4 to infect human cells, we reengineered HKU4 spike, aiming to build its capacity to mediate viral entry into human cells. To this end, we introduced two single mutations, S746R and N762A, into HKU4 spike. The S746R mutation was expected to restore the hPPC motif in HKU4 spike, whereas the N762A mutation likely disrupted the potential N-linked glycosylation site in the hECP motif in HKU4 spike.

“Restoring the hPPC motif” in the above quote describes restoring the RSVR furin cleavage site found in MERS but absent in HKU4:

It seems that spike protein cleavage, in general, and furin cleavage sites, in particular, were a hot topic in coronavirology in 2018-2019. Other groups were joining in: for example, a Beijing group reported introducing a novel FCS into a chicken CoV in September 2019.

At around the same time, Ralph Baric’s group reported looking into proteolytic cleavage of two new MERS-like CoVs that both had an FCS at the S1/S2 junction. Their findings showed that spike cleavage played a very important role in tissue and species tropism:

Together, these results indicate that proteolytic cleavage of the spike, not receptor binding, is the primary infection barrier for these two group 2c CoVs. Coupled with receptor binding, proteolytic activation offers a new parameter to evaluate the emergence potential of bat CoVs and offers a means to recover previously unrecoverable zoonotic CoV strains.

Surprisingly, despite having coauthored the DEFUSE proposal with Ralph Baric just a year before, Zhengli Shi’s group was not included in the above work. In turn, Ralph Baric’s prior coauthors on the 2015 MERS-HKU4 FCS paper, Zhengli Shi and Lanying Du, did not include his group in their latest 2019 study, but instead were joined by Yusen Zhou to again look into the cleavage of the MERS spike while studying antibody-dependent enhancement of viral entry:

To understand the molecular mechanism of ADE, we investigated whether Mersmab1 triggers any conformational change of MERS-CoV spike. It was shown previously that DPP4 binds to MERS-CoV spike and stabilizes the RBD in the standing-up position, resulting in a weakened spike structure and allowing the S2′ site to become exposed to proteases (51). We repeated this experiment: MERS-CoV pseudoviruses were incubated with DPP4 and then subjected to trypsin cleavage (Fig. 4A). The results showed that during the viral packaging process, virus surface-anchored MERS-CoV spike molecules were cleaved at the S1/S2 site by proprotein convertases; in the absence of DPP4, the spike molecules could not be cleaved further at the S2′ site by trypsin. These data suggest that only the S1/S2 site, and not the S2′ site, was accessible to proteases in the free form of the spike trimer. In the presence of DPP4, a significant amount of MERS-CoV spike molecules were cleaved at the S2′ site by trypsin, indicating that DPP4 binding triggered a conformational change of MERS-CoV spike to expose the S2′ site. Interestingly, we found that Mersmab1 binding also allowed MERS-CoV spike to be cleaved at the S2′ site by trypsin. As a negative control, the SARS-CoV RBD-specific MAb did not trigger MERS-CoV spike to be cleaved at the S2′ site by trypsin. Hence, like DPP4, Mersmab1 triggers a conformational change of MERS-CoV spike to expose the S2′ site for proteolysis.

The authors also looked into how SARS1 entry into immune cells (e.g. macrophages) can be enhanced via an RBD-specific antibody that Shibo Jiang and Yusen Zhou developed previously (emphasis added):

To expand the above-described observations to another coronavirus, we investigated ADE of SARS-CoV entry. We previously identified a SARS-CoV RBD-specific MAb, named 33G4, which binds to the ACE2-binding region of SARS-CoV RBD (49, 50); this MAb was examined for its potential capability to mediate ADE of SARS-CoV entry (Fig. 3E). The result showed that 33G4 mediated SARS-CoV pseudovirus entry into CD32A-expressing cells but blocked SARS-CoV pseudovirus entry into ACE2-expressing cells. Therefore, both the MERS-CoV RBD-specific MAb and the SARS-CoV RBD-specific MAb can mediate the entry of the respective coronavirus into Fc receptor-expressing human cells while blocking the entry of the respective coronavirus into viral-receptor-expressing human cells. For the remainder of this study, we selected the MERS-CoV RBD-specific MAb Mersmab1 for in-depth analysis of ADE.

In the process, the authors ablated the MERS FCS for some of their recombinant spikes:

To stabilize S-e in the prefusion conformation, we followed the procedure from a previous study by introducing mutations to the S1/S2 protease cleavage site (RSVR748–751ASVA) and the S2 region (V1060P and L1061P)

Before that, in 2007 Shibo Jiang, Yusen Zhou, and Lanying Du studied the role of S1/S2 cleavage in SARS1 viral entry (emphasis added):

We further explored if the cleavage of the S protein indeed plays a critical role in SARS-CoV infection. Our results demonstrate that: (1) Factor Xa can effectively cleave the S protein in the pseudovirus SARS-CoV/HIV into S1 and S2 subunits and this cleavage is inhibited by Ben-HCl (Fig. 2B); (2) the S protein can be cleaved into S1 and S2 subunits when the pseudovrus is incubated with its target cells (293T/ACE2) and the level of the cleavage correlates with viral infectivity (Fig. 3); and (3) the target cells express Factor Xa (Fig. 4), a membrane-bound protease [18]. … Thus, the infection of SARS-CoV not only involves the binding of RBD with its receptor ACE2 and the fusion between viral envelope and host cell membrane, but also is associated with the cleavage of the S protein by proteases on the cell membrane, such as Factor Xa.
…
Previous studies have explored several targets for vaccine and antiviral agent developments. For example, vaccine candidates targeting the RBD are able to induce strong neutralizing antibodies and thus provide protection against SARS-CoV infection [19], [20]. Antiviral peptides targeting the HR2 region of the S protein can block the entry of SARS-CoV into target cells [21]. Small molecules targeting SARS-CoV proteases, such as main proteases 3C-like protease (3CLpro) [22], papain-like protease 2 (PLP2) [23] and helicase [24], can inhibit viral replication. In addition to these targets, here we have provided the cleavage of the S protein as another crucial target for the development of vaccines and antiviral agents. Inhibition of the cleavage of the S protein into functional S1 and S2 subunits using agents such as Ben-HCl can effectively block viral entry.

In 2009, Shibo Jiang, Lanying Du, and Yusen Zhou, in the paper titled “The spike protein of SARS-CoV — a target for vaccine and therapeutic development” again talked about SARS1 cleavage, and this time contrasted it with other coronaviruses cleaved by furin:

Unlike the S proteins of coronaviruses cleaved by furin-like proteases, the S protein of SARS-CoV can be cleaved by cathepsin L at position T678 or by trypsin at R667. In contrast to the entrance mechanism of HIV, SARS-CoV can enter cells from an acidic environment of the endosome. Nevertheless, SARS-CoV can also enter the target cell surface, which is mediated by proteases on the cell surface through a non-endosomal-dependent pathway.

In the same paper the authors mention prior work coauthored by Lanying Du on peptides that interfere with S1/S2 cleavage:

Peptides that interfere with the cleavage of S protein. Cleavage of the S protein trimer is an important event in infection, making the potential cleavage site between S1 and S2 domains another target for development of anti-SARS-CoV agents. Synthetic peptides, including P6 (amino acids 598–617) and P8 (amino acids 737–756), both of which are close to the S1–S2 connection and cleavage site, exhibit potent inhibitory activity against the GZ50 strain of SARS-CoV infection in fetal rhesus kidney (FRhK4) cells, and have IC90 values of approximately 100 and 25 μM. This suggests that binding of the peptides to the S protein interferes with the cleavage of S1 and S2, inhibiting the production of functional S1 and S2 subunits and subsequent fusion of the viral envelope and the host cell membrane.

Notably, two years prior to the 2015 MERS-HKU4 FCS study, Shibo Jiang coauthored a paper (for which Lanying Du is listed as the editor) that created a novel RIRR furin cleavage site (albeit not in a coronavirus) via a 12-nt insertion reminiscent of the one that has endowed SARS2 with its RRAR furin cleavage site.

Therefore when SARS2 emerged with its distinctive furin cleavage site —introduced via a very odd 12-nt insertion that also included a leading proline, just as in the MERS FCS — it is very surprising that despite all their prior work, Shibo Jiang, Lanying Du, and Zhengli Shi failed to mention the FCS in their January 2020 paper, despite comparing SARS2 to other SARS-like strains precisely at the S1/S2 cleavage site:

By aligning 2019-nCoV S protein sequence with those of SARS-CoV and several bat-SL-CoVs, we predicted that the cleavage site for generating S1 and S2 subunits is located at R694/S695

What is even more remarkable is that Zhengli Shi’s next major paper in early February 2020, which disclosed RaTG13, also failed to mention the novel FCS — and that paper had 29 authors. It even claimed to provide a protein sequence alignment of the S1 subunit of SARS2’s spike protein relative to RaTG13 and other SARS-like CoVs, but instead of including the full S1 protein, the authors chose to cut their alignment a few residues short of the novel furin cleavage site.

Incredible foresight or a self-fulfilling prophecy?

Looking back at the SARS2 spillover in Wuhan in late 2019, it’s very odd that the virus typically found so far away — in Yunnan or even further south in Laos or Myanmar — decided to spill over right near the Wuhan Institute of Virology. Doubly odd that this happened not long after the institute decided to look for viruses just like it. Triply odd considering that SARS2 has unique features that were never found in SARS-like viruses in nature before or after its outbreak, but were the focus of research interest of WIV and its close collaborators.

So either those researchers had a remarkable gift of foresight and perfect timing in anticipating that a virus like SARS2 is overdue for spillover, which, in a cruel twist of fate, just happened to occur exclusively at their doorstep, or… Or it’s a lab leak.

The latter would be the most parsimonious explanation, as, first of all, WIV was the virus sampling hub for the above researchers, boasting the biggest collection of coronavirus samples in the world: “We sampled and PCR-screened >16,000 individual bats from 6 families (16 genera) in southern China,” said their November 2018 NIH grant renewal application.

Secondly, WIV was also an animal testing hub — they even kept live bats, which at some point Zhengli Shi cared for personally:

Zhang Huajun said: “The research team captured several bats from the wild to prepare as experimental animals, and they need to be fed every day. During the Spring Festival this year, the students went home for the holidays, and Teacher Shi silently took on the task of raising bats.”

WIV also had a flair for doing exotic experiments, such as trying to infect civets by SARS-like viruses, as Zhengli Shi mentioned in her interview:

Q: Has your lab done any animal experiments with SARS-related viruses recently? If so, can you provide any details?

A: We performed in vivo experiments in transgenic (human ACE2 expressing) mice and civets in 2018 and 2019 in the Institute’s biosafety laboratory. The viruses we used were bat SARSr-CoV close to SARS-CoV. Operation of this work was undertaken strictly following the regulations on biosafety management of pathogenic microbes in laboratories in China. The results suggested that bat SARSr-CoV can directly infect civets and can also infect mice with human ACE2 receptors. Yet it showed low pathogenicity in mice and no pathogenicity in civets. These data are being sorted and will be published soon.

Speaking of biosafety regulations, prior to the Covid outbreak, they were pretty lax. In fact, those lax standards at WIV were initially pitched as a selling point for DEFUSE:

Ralph Baric’s comment on this cost-saving measure shows his uneasiness:

[In] the US, these recombinant SARS CoV are studied under BSL3, not BSL2, especially important for those that are able to bind and replicate in primary human cells. In [C]hina, might be growin[g] these virus under [BSL2].

US resea[r]chers will likely freak out.

Heeding Baric’s hint, BLS-2 was changed to BSL-3 in later drafts:

In 2016, however, WIV had no qualms about disclosing in academic papers that work with live SARS-like viruses was done at BSL2 (emphasis added):

Virus and cells
The SL-CoV WIV1 strain (GenBank accession number KF367457) and other viruses were propagated as described previously (2). Sendai virus (SeV) strain Cantell (kindly provided by Hanzhong Wang) was propagated in 10-day-old embryonated chicken eggs at 37°C for 48 h (24). All experiments using live virus was conducted under biosafety level 2 (BSL2) conditions.

They were not alone. Before the pandemic, other labs in China also handled coronaviruses at BSL2. On notable example is a Beijing lab which claims to have isolated the SARS2-like pangolin coronavirus GX/2017 (GX_P2V/pangolin/2017/Guangxi strain) in 2017 and then “routinely” culturing the live virus under BSL2 conditions:

Our reasoning of this isolate having low or no pathogenicity in humans was based on the fact that, back in 2017, no suspected infections were found in those having close contacts with pangolins; and our pangolin coronavirus isolate was routinely cultured in biosafety level 2 facilities.

Finally, as the US instituted a pause on gain-of-function work in 2014, WIV gained another competitive advantage, since it was not subject to US regulations. More and more gain-of-function work kept being done at WIV, culminating in late 2017 with the creation of 8 chimeric SARS-like viruses.

Ancestral habitat of SARS2

The closest strain to SARS2 with a near-identical RBD, BANAL-52, has been found in R. malayanus bats in Laos (incidentally, EcoHealth reported sampling bats in Laos in 2015). Based on extensive bat sampling done after the outbreak, which, in addition to BANAL-52, has yielded multiple SARS2-like strains, the prevailing consensus is that SARS2’s bat progenitor comes from Yunnan or countries bordering it to the south. As Pekar et al. put it:

The closest-inferred ancestors of SARS-CoV-2 likely circulated in Yunnan, China or northern Laos, overlapping with a set of contiguous cave structures extending through these regions.

The above authors then try to draw a parallel between the emergence of SARS1 and SARS2:

Our results indicate that both SARS-CoV-1 and SARS-CoV-2 emerged in humans over a thousand kilometers from where their closest-inferred bat virus ancestors likely circulated.

But even according to their analysis, in contrast to SARS2, the bat reservoir of SARS1 was considerably closer to the site of the initial SARS1 spillover in Foshan, just west of Guangzhou (the capital of the Guangdong province):

Moreover, the entire Guangdong province lies within a diverse horseshoe bat habitat with many different species carrying SARS-like viruses, while Wuhan is, at best, on the outskirts of that ecosystem:

Also, at the time of SARS1 spillover, Guangdong was the center of wildlife trade in China, as well as the capital of exotic culinary practices, as the following 2006 Chinese paper describes (emphasis added):

‘The people take food as their heaven’. ‘Beijing people talk about everything, Shanghai people buy everything and Guangdong people eat everything’. These traditional Chinese sayings illustrate the importance of ‘having good and delicious food’ among the Chinese, especially those from southern China, where the severe acute respiratory syndrome (SARS) epidemic originated. In China, food is considered most nutritive and delicious if it is freshly prepared from live animals, whereas frozen meat is considered inferior. Therefore, wet-markets are located in the vicinity of residential areas in most parts of China, which allows frequent contacts between human and live food animals. In addition to ordinary food animals, many people in southern China, such as Guangdong Province, have the habits of eating a wide range of exotic wild animals, including civets and bats, as this is traditionally believed to improve health and sexual performance.

Wuhan cuisine is much less exotic, and China has cracked down on wildlife trade after the SARS1 outbreak. By the time EcoHealth and WIV were writing their NIH grant renewal application in 2018, they have noted this shift in spillover risk for SARS-like viruses (emphasis added):

Our previous R01 hypothesis was that SARSr-CoV spillover would most likely occur through the trade in bats for food, via the same market chains that [led] to the emergence of SARS (20). To test this, we conducted an exploratory study using standardized one-on-one semi-structured ethnographic interviews and observational data in southern China among 88 people involved in trading wild bats, to assess local social and cultural norms and individual attitudes underlying contact with bats (publication in prep.). Our results suggest that in the 11 years since the emergence of SARS, there have been substantial changes to the wildlife trade: 1) Former wildlife markets are now predominantly selling captive-bred species (poultry, livestock, farmed wildlife); and 2) few bats are now sold through markets. We identified other risk factors for spillover, including people living near to bat roosts, and those visiting bat caves for hunting or recreation.

I can’t help but wonder whether the biggest spillover risk for novel SARS-like viruses in 2019 was now people visiting bat caves for virus hunting.

The SARS2 outbreak is not like that of SARS1

But the most important difference between SARS1 and SARS2 is how they spilled over to humans: the emergence pattern of SARS1 is in stark contrast to that of SARS2. While SARS2 had a single spillover in Wuhan, there were multiple independent spillovers of SARS1 over several months in a number of cities:

In zoonosis, it is much more common for a virus to spill over multiple times, and those spillovers often feature notable genetic differences between strains. Other viruses are also known to have had multiple spillovers. For example, HIV had at least four independent spillovers, while MERS had 50 zoonotic introductions.

During the SARS1 outbreak, scientists quickly found infected intermediate host animals — palm civets and raccoon dogs — and isolated SARS1 strains from them that had many mutations relative to human strains. With SARS2, however, even after 4 years of looking, not a single infected animal has been found.

Zoonosis proponents usually claim that such an animal hasn’t been found for the lack of looking but that is untrue. The search efforts have been considerable. The WHO report on Covid origins goes on for pages and pages detailing such efforts of testing “over 80,000 wildlife, livestock and poultry samples … collected from 31 provinces in China” that were all negative for SARS2:

The Chinese CDC started collecting susceptible animals, including 15 raccoon dogs, around Wuhan as early as January 7, 2020, and in 2022 published their testing results — all negative for SARS2 or its progenitor:

Coronaviruses in wild animals sampled in and around Wuhan at the beginning of COVID-19 emergence

In addition, the now famous Gao et al. study that did detailed environmental sampling of the the Huanan market tested hundreds of samples from many animals sold there but none tested positive for SARS2:

The 457 animal samples mainly collected between January 1 st and March 2 nd , 2020 included 188 individuals belonging to 18 species (with some stray animals sampled until March 30 th ) (Table 2). The sources of the samples include unsold goods kept in refrigerators and freezers in the stalls of HSM, and goods kept in warehouses and refrigerators related to the HSM. Samples from stray animals in the market were also collected, i.e. swab samples from 10 stray cats, 27 cat feces, one dog, one weasel, and 10 rats. All the 457 animal samples tested negative for SARS-CoV-2 nucleic acid, suggesting that the animal infections with SARS-CoV-2 might be rare in the market.

Another study looked at game animals from across China and found 45 new mammalian viruses but no SARS2 progenitor:

Game animals are wildlife species often traded and consumed as exotic food, and are potential reservoirs for SARS-CoV and SARS-CoV-2. We performed a meta-transcriptomic analysis of 1725 game animals, representing 16 species and five mammalian orders, sampled across China. From this we identified 71 mammalian viruses, with 45 described for the first time.

That study also included some pangolin samples:

Previously, a 2020 study focusing on pangolins entering the wildlife trade didn’t find SARS2 either:

No Evidence of Coronaviruses or Other Potentially Zoonotic Viruses in Sunda pangolins (Manis javanica) Entering the Wildlife Trade via Malaysia - EcoHealth

Pangolins

Pangolins were an early suspect because the closest case of a pre-outbreak animal carrying a SARS2-like virus — with a near-identical RBD but without an FCS — is a case of infected pangolins reported in 2019. The only caveat is that this case, as well as the presence of SARS-like viruses in those pangolins, were first reported by the Guangdong Institute of Applied Biological Resources (GIABR), a close collaborator of the Wuhan Institute of Virology.

GIABR submitted their paper in September 2019. In it they described sequencing tissue samples collected earlier in 2019 from confiscated pangolins, and reported finding coronavirus genome fragments that matched to a number of SARS-like viruses. Most notably, in the “lung08” sample they reported a 500-nt fragment that aligned to the spike of the HKU3 virus, with a 82.6% identity. HKU3, of course, was listed as one of EcoHealth/WIV’s viruses of interest in the DEFUSE proposal and the November 2018 NIH grant renewal application.

After the outbreak, metagenomic analysis of the sequencing reads from the same “lung08” sample would show that it contained reads matching the SARS2 receptor binding motif (RBM, the part of the RBD that comes into contact with the ACE2 receptor), and pooled reads from lung07, lung08, and lung09 samples would later produce a near-complete genome of the MP789 virus that shares 97% identity with SARS2’s RBD but, more importantly, has a near-identical RBM with SARS2: they share 77 out of 78 residues, including all six residues critical for efficient binding to ACE2.

The senior author of that 2019 GIABR paper on pangolin CoVs, Jinping Chen, collaborated with WIV and EcoHealth before — for example, on a 2014 paper sampling multiple bat species at the Xishuangbanna Tropical Botanical Garden in Yunnan, near the border with Laos, within a few kilometers of where a very notable SARS-like virus, RmYN02, would be found in 2019. So it is quite possible that WIV was aware of GIABR’s findings regarding identifying a number of SARS-like viruses in pangolin samples. In that case WIV would have likely had access to those samples and/or sequencing data in 2019.

According to the declassified US Intelligence (ODNI) report on Covid origins, in 2019, WIV was analyzing at least some pangolin samples (emphasis added):

• Since 2019, some WIV researchers analyzed pangolin samples to better understand disease outbreaks in these animals.
…
• As of January 2019, WIV researchers performed SARS-like coronavirus experiments in BSL-2 laboratories, despite acknowledgements going back to 2017 of these virus’ ability to directly infect humans through their spike protein and early 2019 warnings of the danger of this practice. Separately, the WIV’s plan to conduct analysis of potential epidemic viruses from pangolin samples in fall 2019, suggests the researchers sought to isolate live viruses.

Speaking of the ODNI report, while it seemed to downplay the observations that some WIV researchers “were ill in Fall 2019 with symptoms… consistent with but not diagnostic of COVID-19”, it is still notable that news sources named some members of the Zhengli Shi lab as the ill researchers in question:

The current and former U.S. officials told the Journal the three who fell ill were [Ben] Hu; Yu Ping, a Chinese scientist who wrote a 2019 thesis on SARS-related coronaviruses found in bats; and another scientist named Yan Zhu.

The researchers’ names were noted last week in an article in Public, which publishes on the Substack platform, and were independently confirmed by the [Wall Street] Journal.

In December 2017, Ben Hu — who was a part of the DEFUSE proposal and was the first author of the 2017 WIV paper that reported creating several chimeric SARS-like viruses — gave a detailed interview that echoed themes common with the March 2018 DEFUSE proposal and the November 2018 NIH grant renewal application:

Bioinformatics: After the publication of this research work which lasted for 5 years, what is the next research plan of the group?

Dr. Ben Hu: In addition to the two aspects mentioned above, namely serological investigation for exposed populations and assessment of cross-species transmission of SARS-like coronavirus through animal models, we will not stop the surveillance carried out for bat coronavirus. There is a need to investigate whether unknown strains of SARS-like coronaviruses that pose a potential threat to public health are also present in other bat caves in Yunnan or in other provinces.

Ben Hu previously worked with GIABR on a number of projects. In the same interview he thanked Libiao Zhang of GIABR “who collected a large number of samples for us across the country”:

Intriguingly, in 2019, Libiao Zhang’s team was sequencing bat samples they previously collected in Mengla County in Yunnan to show the presence of Rhinolophus malayanus bats in China, which previously have not been observed there:

The very same bat species would later turn out to be the host of the closest SARS2 relative bat virus found so far, BANAL-52, sampled in Laos, which, along with the MP789/GD-2019 pangolin CoV found in the sequences reported by GIABR, has a near-identical RBD to SARS2. At present, in China, R. malayanus bats are believed to only be found in the south of Yunnan:

As I mentioned previously, EcoHealth and WIV did sample bats in Laos in 2015, about 200 km from where BANAL-52 was found in 2020. In 2021, WIV published a joint paper about this with Libiao Zhang:

While BANAL viruses were found in Laos in 2020 (about 500 km from Mengla County but within the same bat ecosystem), in 2019, another relative of SARS2, called RmYN02, was found in R. malayanus bats in Mengla County just 15 minutes away from the site where Libiao Zhang’s team collected their samples for their 2019 R. malayanus paper:

RmYN02 is notable because it contains a somewhat similar amino acid sequence (PAAR) at the same S1/S2 spot where SARS2 contains a novel FCS (PRRAR). RmYN02’s discoverers even erroneously claimed that, like the PRRA insertion in SARS2, the PAA fragment is also an insertion in RmYN02. However, as more viruses with the same fragment (e.g. BANAL-116 and BANAL-247) were found, we have subsequently shown that the RmYN02 discoverers were incorrect, and PAA in RmYN02 arose via point mutations rather than an insertion, and, judging by the diversity in that fragment between geographically dispersed strains, it arose in their ancestor dozens if not hundreds of years ago.

How did SARS2 get its furin cleavage site?

However, the PAA fragment of RmYN02 is notable because it could have potentially served as an inspiration for a genetic engineer looking to study mutations which could lead to cross-species transmission — in particular, to model what would happen if PAAR mutates into PRRAR to create a novel furin cleavage site that is well-known to expand viral species tropism.

What lends credence to this hypothesis is the fact that the PAA fragment in RmYN02, as well as in BANAL-116 and BANAL-247 strains, is coded by CCT GCA GCG codons, while the PRRA insertion in SARS2 is coded by CCT CGG CGG GCA — i.e. the codons in the SARS2 insertion coding for proline (CCT) and alanine (GCA) are identical to those in RmYN02 and the BANAL strains.

In other words, the PRRA SARS2 insertion looks as if someone took PA from the naturally found PAAR fragment at the S1/S2 spike junction of RmYN02/BANAL strains and inserted two CGG-coded arginines to yield a PRRA insertion, which has, in turn, created a PRRAR furin cleavage site reminiscent of the PRVSR one in MERS.

Some think the above steps could happen naturally, but that is astronomically unlikely. Now, PAAR could theoretically mutate into PRRAR naturally in RmYN02-like strains — although we still haven’t found a single SARS-like bat CoV with an FCS, indicating that there is a strong selective pressure against them in bats — but for the resulting strain to then recombine with a SARS2 progenitor in a way that would only result in the 12-nt PRRA insertion is a statistical impossibility.

In any case, by now, even staunch zoonosis proponents agree that the FCS is extremely unlikely to have arisen in bats, as it is most likely detrimental to enteric tropism, which is favored by these viruses in their natural hosts. However, rather than concede that a genetic engineer might have chosen to engineer the FCS in SARS2, lab leak critics usually respond with some variation of the “but why RRAR?” argument implying that it is less efficient than a canonical RxRR used in some previous FCS genetic engineering efforts.

However, in the past, some researchers actually did choose to engineer non-canonical furin cleavage sites in coronaviruses, for example, RRSR in 2011. Also, in addition to the potential PAAR->PRRAR modeling scenario above, other reasons for RRAR include having observed an RRAR FCS in other sampled coronaviruses, as Zhengli Shi’s team did in 2017 with AcCoV-JC34. Also, WIV has previously identified RRAR as a nuclear localization signal, and coincidentally the first two arginines in that fragment are also coded by CGG CGG.

The potential link to MERS and feline CoV

As to why anyone would engineer a less efficient FCS than RxRR — it turns out, a highly cleavable FCS could actually be detrimental to viral pathogenicity and/or fitness. There is an intriguing Twitter thread about this, which describes the following study of feline coronaviruses that suggests that what turns a benign FECV infection into a lethal FIPV one is the mutation of a highly cleavable RRSRR FCS into something less efficient, like PRRAR:

The study mentions a striking example of two cats living in the same household who were both carrying the benign version of the virus but when one of the cats developed FIP, it turned out that the virus it carried mutated its FCS from RxRR to a less efficient RxxR:

Zhengli Shi must have been familiar with this aspect of feline CoVs. In 2015, she coauthored a joint paper between WIV and Wuhan University that tested a pan-coronavirus inhibitor against several CoVs, including SARS1, MERS, and FIPV. The FIPV strain they used had the same FCS mutation as above, which turned a canonical RxRR FCS into a less efficient RRSR :

The author of the Twitter thread who pointed out the paper on FCS mutations turning FECV into FIPV did so in connection with the 2019 WIV paper on MERS antibody-dependent enhancement (ADE) of viral entry that I alluded to before. The connection is that feline CoVs are also known to exhibit ADE, and the 2019 WIV paper references three papers on feline CoV ADE (emphasis added):

ADE has been observed for coronaviruses. Several studies have shown that sera induced by SARS-CoV spike enhance viral entry into Fc receptor-expressing cells (42,–44). Further, one study demonstrated that unlike receptor-dependent viral entry, serum-dependent SARS-CoV entry does not go through the endosome pathway (44). Additionally, it has long been known that immunization of cats with feline coronavirus spike leads to worsened future infection due to the induction of infection-enhancing antibodies (45,–47). However, detailed molecular mechanisms for ADE of coronavirus entry are still unknown. We previously discovered a monoclonal antibody (MAb) (named Mersmab1) which has strong binding affinity for MERS-CoV RBD and efficiently neutralizes MERS-CoV entry by outcompeting DPP4 (48); this discovery allowed us to comparatively study the molecular mechanisms for antibody-dependent and receptor-dependent viral entries.

As you might recall, in that 2019 WIV paper, Zhengli Shi, Lanying Du, and Yusen Zhou looked into how antibody binding could enhance MERS viral entry into immune cells via modulating spike cleavage. In light of that, parallels between the furin cleavage sites of MERS (PRSVR), SARS2 (PRRAR) and some of the above FIPV CoVs (PRRAR) are quite interesting.

The desire to equip a SARS2 progenitor with a MERS-like FCS can also explain why a genetic engineer might have chosen to do so via an insertion rather than mutation of existing nucleotides: the MERS spike protein is four residues longer at the S1/S2 junction than SARS-like viruses, and an insertion like PRRA removes this imbalance, and potentially helps fully align the 3D structure of the two proteins at this junction:

Recall that as part of their 2019 MERS ADE study, WIV and colleagues were already looking at antibody-dependent enhancement of SARS1 and claimed that a mAb they previously developed can help SARS1 get into immune cells:

To expand the above-described observations to another coronavirus, we investigated ADE of SARS-CoV entry. We previously identified a SARS-CoV RBD-specific MAb, named 33G4, which binds to the ACE2-binding region of SARS-CoV RBD (49, 50); this MAb was examined for its potential capability to mediate ADE of SARS-CoV entry (Fig. 3E). The result showed that 33G4 mediated SARS-CoV pseudovirus entry into CD32A-expressing cells but blocked SARS-CoV pseudovirus entry into ACE2-expressing cells. Therefore, both the MERS-CoV RBD-specific MAb and the SARS-CoV RBD-specific MAb can mediate the entry of the respective coronavirus into Fc receptor-expressing human cells while blocking the entry of the respective coronavirus into viral-receptor-expressing human cells.

A researcher looking to expand that research to novel SARS-like CoVs could then do something similar to a SARS2 progenitor, which could explain the unique SARS2 mutations like R346 and A372. The first one is reminiscent of a SARS1 escape mutation from the S109.8 antibody, while the second removes an N-linked glycan previously implicated in DC-SIGN binding (a receptor on immune cells involved in ADE).

Coming back to the FCS itself, a genetic engineer with a recently stated interest — as per the March 2018 DEFUSE proposal — to look for existing furin cleavage sites in novel SARS-like viruses and create new ones, could have had a number of reasons to create an FCS like the one seen in SARS2. Moreover, such an engineer might have then had several different FCS candidates of varying furin cleavage efficiency, and evaluated how such efficiency affected tropism, pathogenicity, transmissibility and ADE.

Mengla County

How likely was WIV to have come across a BANAL-like and/or RmYN02-like strain prior to the SARS2 outbreak? Quite likely. Not only did WIV sample in Laos, but they have also done extensive sampling in Mengla County, as evidenced by a 2017 paper, as well as a table below from a 2018 paper coauthored by Ben Hu and Zhengli Shi:

In 2019, Zhengli Shi even reported a novel virus they discovered in Mengla County which they named “Mengla Virus”. In 2019, they fully sequenced that virus from samples collected in December 2015. Ben Hu is not on that paper but Yan Zhu (one of the 3 WIV staffers named as sick with Covid-like symptoms in Fall 2019) is, as is Libiao Zhang.

Also, in January 2019, Shi’s team and Libiao Zhang returned to Mengla County to sample more bats in 2 caves and track them via GPS. They published their results in 2022, showing cross-border incursions of Mengla bats into neighboring Laos and Myanmar:

Overall, the above observations paint a compelling picture of extensive sampling by WIV and its collaborators of the bat ecosystem widely believed to house the bat progenitor of SARS2:

Curiously, before carrying out the GPS bat tracking research of Mengla bats in 2019, WIV pitched putting GPS trackers on bats in the DEFUSE proposal:

ICARUS satellite transmitters (1g) wil be attached to 12 Rhinolophus spp. bats from each study roost (36 bats total) to determine nightly foraging dispersal patterns

Which goes to show that just because DEFUSE wasn’t funded doesn’t mean WIV couldn’t do the proposed research on their own. After all, WIV had ample funding from Chinese sources. For example, in 2018, Ben Hu was awarded a three-year grant starting from January 2019 titled “Pathogenicity of Two New Bat SARS-Related Coronaviruses to Transgenic Mice Expressing Human ACE2 Receptor.” His boss, Zhengli Shi, also had a number of local grants, including a large 8.5mm RMB (~$1.2mm) grant to study “Genetic evolution and transmission mechanism of important bat-borne viruses,” among others:

But what about the Huanan Seafood Market?

The entire case for the natural origin of Covid — which posits that the proximity of the SARS2 outbreak to WIV is a mere coincidence — is built on the claim that SARS2 spilled over to humans at the Huanan Seafood Market via some unknown intermediate animal. “Why are the early Wuhan Covid cases centered around the market if the virus didn’t emerge there?” is the essence of the zoonotic hypothesis.

However, a wildlife market is the perfect location for a novel virus to have gotten noticed in China, even if that virus came to the market from a nearby lab. First, wildlife markets are great superspreading locations for SARS2, as we learned during the course of the pandemic: in several instances such markets sparked Covid outbreaks in places previously Covid-free: in Thailand, Singapore, and twice in China. The Huanan market was a very busy place — 10,000 daily visitors spiking up to 100,000 on the weekends — near a key transportation hub in Wuhan.

But more importantly, after the SARS1 outbreak, contact with wildlife was a big red flag in the Chinese surveillance system for novel “unexplained pneumonias”. Annual pneumonia outbreaks are common, and without additional criteria it would be impractical to treat every cluster of pneumonia cases as a signal of a novel coronavirus. So if a SARS-like virus leaks from a lab during a flu season, it could well remain undetected before arriving in a wildlife market and infecting several elderly people severely enough to end up in a hospital, which would then report this cluster of suspicious cases to the centralized surveillance system.

After that, of course, the market would become the prime suspect, which is exactly what happened. So much so that initially a market link was a part of the diagnostic and reporting criteria for Covid. Unfortunately, even after the initial fog of war cleared, a proper retrospective search for the earliest patients was never carried out. If you are interested in diving deeper into the history of early cases, as well as learning about a potential connection of some market patients with a Mahjong room at the market, I recommend this excellent presentation by Gilles Demaneuf. As things stand right now, the current subset of known early cases exhibits — quite understandably — signs of a pro-market ascertainment bias.

The market was dominated by a later lineage of the virus

The biggest problem for the market origin hypothesis is that all 16 of the earliest Covid patients linked to the market were infected by lineage B of the virus — a phylogenetically later version than ancestral lineage A also circulating in Wuhan at the time. This is problematic for the zoonotic hypothesis because in a typical spillover event one would expect the earliest strain of the virus dominating the site of its spillover.

Additionally, nearly all environmental samples from the market that were sequenced extensively enough to determine their lineage were identified as lineage B. The only lineage A sample later detected in the market did not come directly from an environmental swab but rather from a cultured laboratory experiment attempting to grow the virus from that swab. Despite the original environmental sample having a very high Ct PCR value and containing only 22 SARS2 genomic sequencing reads, indicating a low viral load in that sample, a virus was successfully cultured from it. Surprisingly, the cultured virus had additional mutations not present in early genomes but observed later in the pandemic, suggesting possible laboratory contamination rather than genuine market origin.

In any case, the overwhelming dominance of lineage B in both early human cases and environmental samples from the market strongly implies that the market was not where SARS2 initially jumped to humans. Instead, the market likely amplified a lineage B strain sometime after lineage A was already in circulation in Wuhan. This scenario is further strengthened by the observation that lineage B potentially had an edge over lineage A in terms of transmissibility, as experimentally lineage A samples replicated to lower titers than lineage B ones.

The double spillover hypothesis

To explain the dominance of lineage B in the market that undermines the zoonotic spillover hypothesis, one pro-zoonosis research group proposed that the two lineages actually represent two separate spillover events from wildlife. In that scenario, an animal with a phylogenetically later strain, lineage B, was brought to the market first, started the Covid outbreak, and only then, a second animal carrying an earlier, ancestral strain of the virus arrived and started infecting humans with lineage A. Of course, even that scenario doesn’t explain why all 16 early patients from the market had lineage B.

Before detailing how the Pekar et al. two-spillover hypothesis turned out to be based on incorrect assumptions and erroneous modeling code, I should say that even if the two SARS2 lineages truly stemmed from two spillovers, that does not preclude a lab leak, as it is entirely possible for a virus to leak from a lab more than once. In fact, SARS1 escaped from the same lab twice in a span of two weeks in 2004, if not thrice.

The Pekar et al. hypothesis was predicated on two things: first, the authors disregarded all known intermediate genomes between A and B under the explanation that they all stemmed from sequencing errors, rather than were true descendants of lineage A. Then the authors developed a Bayesian model to simulate viral evolution, providing probability estimates for different ancestral scenarios.

In their initial paper, this model indicated a fairly strong Bayes factor in favor of two separate spillovers. But thanks to some remarkable citizen science, it turned out the Pekar et al. code had a number of bugs. Once those bugs were fixed, the Bayes factor supporting two spillovers materially decreased:

That alone severely undermined the double spillover hypothesis. However, what provided dispositive evidence for rejecting it, was confirming the existence of true intermediate early genomes between lineages A and B (called B0 in the quote and figure below):

Notably, two B0 genomes recovered here and seven genomes recovered from Wuhan, Singapore, and the UAE were at the basal position of lineage B and its descendants (abbreviated as Lineage Bs).
…
Lineage B0 viruses (haplotype ‘T/T’) were reported in Wuhan, Singapore, and the UAE (Supplementary Table S3) and occupied the intermediate position of lineage A and lineage B. Herein, lineage B0 genomes were recovered from two patients, who were infected at different regions (Henan and Shanghai) and hospitalized on 4 and 8 February, respectively. Interestingly, our genomes share 100 per cent similarity with four genomes sampled from Wuhan and Singapore. Overall, these genomes have no difference from the reference genome of lineage A or B with the exception of the nucleotide at the site 8,782 or 28,144, but it should be noted that there was an indeterminate C/A nucleotide assignment (1,125 counts of ‘C’ and 2,008 counts of ‘A’) at the position 27,230 of SH-P37–2-Shanghai genome. Finally, either lineage A0 or B0 genomes were recovered and confirmed by both RT-PCR and transcriptomic protocol (SRR25229357, SRR25229358, SRR25229360, and SRR25229361). Hence, these data indicate the true existence and evolution of lineages A0 and B0 viruses (Supplementary Table S8) in human populations.

Furthermore, even lineage A might not have been the original lineage that spilled over to humans. Some of the earliest genomes were found to be even closer to the bat ancestor than lineage A — for example, A0 (with a C29095T mutation) in the above paper. The same mutation was observed in one of the early Wuhan patients, in a sequencing dataset subsequently removed from NCBI but recovered by Jesse Bloom. That dataset also contained sequences from several more patients that likely had lineage A.

In conclusion, there is no reason to believe there were two separate spillovers. The scenario most consistent with known observations is that lineage A evolved into lineage B already in humans. This strongly implies that the Huanan market was exposed to a subset of genomes that emerged after the initial spillover, and therefore was not the initial spillover site but rather a location where an already circulating lineage B was amplified.

Steelmanning zoonosis vs. lab leak

Let us now digest the observations described above and produce two strongest scenarios for each origin hypothesis. Then you, dear reader, can decide for yourself which one sounds more plausible.

For that I will defer to Will Van Treuren who served as a judge in the Rootclaim Covid origins debate. The two judges in that debate ruled against the lab leak hypothesis, so I am unbiased here in endorsing the scenarios Will has constructed — even if I believe that he picked the wrong one (possibly because at the time of the debate we didn’t yet have the DEFUSE drafts or conclusive evidence of intermediate genomes ruling out double spillover). Here are the two scenarios, extracted from his debate decision:

My view

While I appreciate that through the prism of their knowledge, experience or biases, different people might weigh known evidence differently in favor of one or the other hypothesis, I think a lab leak is the more parsimonious explanation. I find zoonosis much less likely because I just can’t get over the series of incredible coincidences it requires to explain the location and the timing of the outbreak.

Not only is there no convincing evidence of zoonosis four years after the outbreak, even the best hypothetical scenario outlined by Will of how an intermediate host would get to Wuhan, and only Wuhan, via wildlife trade is implausible, in my view. Exclusivity of Wuhan aside, years before getting to Wuhan, that animal host needed to have picked up an enteric bat virus, only transmissible via an oral-fecal route, sustained it in its population long enough to develop an FCS for that virus to become respiratory, and thus highly transmissible, but then waited until that animal traveled to Wuhan to only infect people once it got there but not along the way — and have that exclusive journey happen twice in a row in a span of a week or two, as postulated by the double spillover scenario that is required to reconcile the contradiction of lineage B dominating the market.

That scenario requires hundreds if not thousands of animals sustaining a reservoir of the virus for years, so we would expect to have discovered numerous infected animals by now, much like during the SARS1 outbreak where several strains were identified in palm civets and raccoon dogs with sequencing technology from 20 years ago. The failure to detect any such infected animals today, despite vastly improved sequencing capabilities, severely undermines zoonosis.

Conversely, on the lab leak side, there is a whole range of plausible scenarios. A lab leak doesn’t even require a single lab member knowing about it — the virus could have leaked during virus culturing experiments. If someone in Wuhan was culturing pangolin samples from the same batch as the one sequenced by GIABR in October 2019, later found to contain the MP789 virus, or even Mengla samples in search of the Mengla virus, a SARS2 progenitor contaminating those samples could have been — unknowingly — successfully cultured, and then leaked due to the culturing being done under BSL2.

Alternatively, as Will’s scenario describes, a lab leak could have occurred from the research of a single grad student pursuing the directions outlined in DEFUSE and the EcoHealth/WIV NIH grant, while everyone else including his/her superiors could plausibly plead ignorance. Finally, the research resulting in a lab leak could have stemmed from the research into antibody-dependent enhancement in connection with broad RBD-based vaccines for SARS-like and MERS-like viruses, as well as other antibody-based and drug-based pan-coronavirus therapies.

It is that type of research that WIV was actively pursuing in late 2019, with several collaborators in China and the US, including EcoHealth. And it is that type of research for which SARS2 fits the profile of a novel SARS-like virus that WIV has been on the hunt for since at least 2018.

So when a virus fitting that profile spills over exclusively in Wuhan and contains unique features that WIV talked about in research proposals shortly before, and four years later we have not found its animal progenitor or a single infected animal, the most plausible explanation to me is that it was a lab leak.

Epilogue

Four years later — to the day — since I published my first Medium article exploring Covid origins, the case for zoonosis only got weaker, while the case for lab leak got stronger.

Several factors have weakened the zoonotic hypothesis. Probably the biggest factor is not finding any infected intermediate animals. On the academic side, the paper hailed as a “lab leak killer” in 2020 is now facing calls for retraction (there is even a petition) after internal Slack messages have shown that its authors held very different views in private. In intermediate years, claims of “dispositive evidence” of the Huanan market being the site of spillover came and went, as did the double spillover hypothesis.

Strengthening the lab leak side, the DEFUSE proposal was revealed in mid-2021, and its early drafts were later released via FOIA. In addition, the EcoHealth/WIV NIH grant renewal application provided additional color as to the type of research WIV was doing in 2019, as well as the kind of SARS-like viruses they were searching for.

So today, while we do not yet have certainty in any particular Covid origin scenario, a lab leak origin looks highly likely. Hopefully, within another four years, this mystery will be solved completely. Until then, we can keep debating about raccoon dogs, pangolins, and stuffy Mahjong rooms at the Huanan Seafood Market.

“My love don’t cost a thing.”
—Jennifer Lopez

A long-standing hypothesis about aging suggests that it stems from an inherent biophysical limitation that demands the allocation of an organism’s resources between reproduction and longevity, inevitably leading to a trade-off between the two. This notion is based on the belief that some vital resource is never sufficiently abundant to fully support both objectives at once, and so organisms have to balance between extending their lifespan or reproducing more abundantly.

However, advocates of this concept fail to clearly define these essential resources. Are we talking food? Water? Oxygen? No, they concede, external resources like these cannot be the limiting factor; instead, the problem arises when these external resources get converted into internal resources, which then require the aforementioned trade-off that ultimately leads to aging.

I won’t dwell on this for too long; let me state upfront that this hypothesis of limited resources has been debunked numerous times. The most obvious way to test it experimentally was to take animals and selectively breed long-lived ones, expecting their fertility to decrease as they allocate more resources to longevity. Which is exactly what was done in this work. In violation of the trade-off hypothesis, the experiments showed that flies which began to live longer also began to breed more — both their early and overall fertility increased:

Similar results were found in other studies on various types of flies. For example, in this work, researchers analyzed a thousand Mediterranean fruit flies and found no correlation between fertility and longevity:

An unexpected result in the study was the complete absence of any relationship of early reproduction to either subsequent reproduction or to future life expectancy. This was surprising because, according to Roff (19), conventional life history theory holds that because egg laying is stressful and requires a major expenditure of energy, females that are reproductively active at young ages should be more frail and less fecund at older ages than females that were reproductively less active at young ages. Based on this reasoning, this weakening effect should be manifested as either decreased fecundity at later ages or as increased mortality (i.e., demographic cost of reproduction; see refs. 20–22).

The graph below from this work clearly shows how a prolific breeder (65 eggs per day at the peak on day 10 and 2000 eggs in a lifetime) and a mediocre breeder (25 eggs per day at the peak on day 10 and 500 eggs in a lifetime) have pretty much the same lifespan:

The graph above is a 2D slice from an even cooler 3D graph:

The authors did not stop there and next studied reproduction patterns in another 500 flies, confirming the absence of a trade-off between fertility and lifespan:

…our findings do not support the idea that there is a direct cost of reproduction. The link between mortality and reproduction is carried by the dynamics of reproduction and not by the absolute magnitude of reproduction, as measured in the number of eggs produced. For example, a high reproduction rate with slowly declining reproductive potential is associated with a longer life span according to our findings. In contrast, the classical cost-of-reproduction hypothesis would associate high reproduction rates with shortened life spans.

The only correlation between lifespan and reproduction that they observed was the correlation between the rate of decline in reproduction and aging: the faster the relative (not absolute!) fertility falls, the earlier the flies died. It did not matter when a fly began to breed or what its average egg laying rate was. The only thing that correlated with lifespan was the rate of decline in its fertility. The researchers went on to conclude that it looks like the flies had a sort of a reproduction clock that ticked at a certain speed proportional to the rate of aging of the body — i.e. those who aged faster lost fertility faster:

In particular, our analysis provides a detailed description of the nature of the linkage between the dynamics of the reproductive trajectory and subsequent mortality. We have established the primacy of the rate of reproductive decline over absolute levels of reproduction regarding this link. It is quite amazing that, based solely on knowledge of early reproductive patterns, our approach allows a reasonable prediction of the increase in subsequent death rates at the level of the individual. A possible interpretation of this finding is that the rate of reproductive decline is a good indicator of the speed of aging of an organism. In this sense, the reproductive clock is synchronized with an individual’s biological age as contrasted to chronological age.

Next, one of the above coauthors, James Carey, conducted another study on fruit flies where he also showed that there is no negative relationship between fertility and life expectancy:

To find out which parameters are associated with longevity, we take the most long-lived Drosophila, which are in the 10th cluster of flies. It is surprising that these long-lived flies do not differ from the rest of the population in their reproductive capacity, RC (60.3 ± 5.5 vs 60.47 ± 4.3 eggs/day), or in the length of their maturity stage, T (14.3 ± 3.6 vs 13 ± 4.2 days).
…
We can see that, indeed, neither RC nor T is related to prolongation of life span in Drosophila. Only the exponent α-sen is clearly related to prolong senescence. The smaller this exponent is, the slower is the decrease in egg-laying after the onset of senescence, and the longer is life span. The maturity period and the reproductive capacity tend to be constant in all clusters. This means that longevity in Drosophila is associated only with the rate of senescence.

Below is a clear demonstration that the fertility is actually higher in longer-lived flies. The graphs show the dynamics of egg laying in flies with very different lifespans — ranging from 20 to 60 days, and we see that the longer living flies were also the most prolific:

One would think that by now the absence of a trade-off in flies was clear, but, alas, the Lord of the Flies, James Carey, has not yet had enough! He went on to study Mexican flies and yet again showed the absence of a connection between fertility and longevity:

[K]nowledge of early egg laying provides no information for predicting either remaining life span or the number of eggs laid during the subsequent one-month period. Thus, at the level of the individual in non-manipulative studies such as the current one, there appears to be no obvious cost of reproduction.

Ok, enough about flies — let’s look at other insects. Namely, social insects: ants, termites, bees, etc. They don’t just undermine the trade-off hypothesis, they turn it on its head completely! Social insects have queens who live 10–20 times longer than their non-breeding identical twins, while laying thousands of eggs per day. In honeybees, for example, the queens can lay their own weight in eggs per day, and yet they live for up to 8 years, while the workers live for a few weeks in the summer of a few months if they get to stick around the hive for overwintering.

A similar trend holds in ants, including the ant species that holds the lifespan record for insects — the black garden ant — whose queens are known to live for ~30 years, a lifespan unattainable even for most species of mammals:

Another interesting lifespan factoid about these black garden ants is that their worker lifespan seems to get shorter as the colony grows — and we’re talking a sizeable 1.5x decrease in median lifespan — which could be a possible population control mechanism for the colony:

The curious case of Indian jumping ants (Harpegnathos saltator) offers another compelling argument against the trade-off hypothesis. These ants have a special social role called “gamergate” reserved for worker ants that can take on a reproductive role in the absence of a queen. In these species, queens live for about 5 years, whereas workers have a much shorter lifespan of ~7 months. But if a queen dies or is removed, some workers morph into gamergates — pseudo-queens that engage in duels for dominance. The triumphant gamergates collectively assume the queen’s egg-laying duties, and their lifespans skyrockets to 3–4 years:

This further refutes the idea that there’s an inherent “resource” allocated at birth, as some could have argued is done for queens. Because for a worker ant to become a breeder would logically consume more of that inborn “resource,” but in actuality such a transformation results in a marked lifespan extension. So this observation further debunks the notion of an inherent trade-off between reproduction and longevity. Oh, and if a gamergate is moved to a colony with an active queen, it stops breeding and reverts to a worker, with the requisite shrinkage in lifespan. Life’s twists and turns can be so cruel.

Another refutation of the trade-off hypothesis comes from the Monarch butterfly. In the summer, Monarch adults live for only a few weeks, but in the fall they migrate south for wintering, and such migrants can live for up to 9 months. Moreover, not only do these migrants first have to fly for thousands of miles to Mexico, and only then mate in the following spring, but then the females have to make the return journey while pregnant. Talk about the disparity in the “life force resource” between the summer and winter individuals!

Intriguingly, a single JH1 hormone from the family of juvenile hormones seems to play a key role in the disparity in lifespan between migrants and non-migrants in Monarchs. When it is introduced to butterflies under laboratory conditions, they live 2x shorter, but if, on the contrary, its production is blocked, then the butterflies live 1.5x longer. The graphs below show the resulting 3x difference in median life expectancy between those who received additional JH1 (blue curves) and those in whom its production was blocked (red). Top row is females, bottom row — males.

By the way, juvenile hormones are responsible for regulating development in many insect species, including social insects. For example, they were shown to determine social roles in bees during development.

Speaking of social animals, it is not only social insects that display this “anti-trade-off” disparity where reproduction actually prolongs lifespan, but some social mammals as well. In particular, the longest-lived naked mole rat who was 39 years old in 2021 (hopefully, he is still alive!) is a breeding male, and generally breeders have lower mortality risk than non-breeders:

Also, in a close relative of naked mole rats, another social African rodent, Ansell’s mole rat, both male and female breeders have considerably longer lifespans than their non-breeding counterparts:

While we’re on the topic of interesting mammalian species, there is one that has a curious seasonal lifespan disparity highly reminiscent of Monarch butterflies — the Montane vole (Microtus montanus). If it is born in the spring, it quickly matures (within 3–4 weeks), breeds, and dies in the same year. But if it is born in the fall, then just like in Monarch butterflies, its reproductive ability is put on pause until the next spring, which greatly increases its lifespan:

If all of the observations above haven’t yet been convincing, here is a study in which the authors surveyed 12 species of birds and 17 species of mammals living in zoos, and also concluded that there is no reason to believe that early or general fertility negatively affects life expectancy.

Moreover, further counterevidence to the trade-off hypothesis is provided by turtles, in whom fertility only increases with age, as it does in rabbits:

There are yet more species that are characterized by an increase in fertility (the thick line in the graph below) with age:

Finally, human females also do not seem to have any negative effects from childbearing on their lifespan:

So Jennifer Lopez was right — truly, love don’t cost a thing.

“Zero to one” or 0.5 to 0.51?

Dr. Robert Malone is back from obscurity on the news that he is suing the NYT over their recent article:

The Latest Covid Misinformation Star Says He Invented the Vaccines

I have been pointing out Malone’s dubious claims about the safety and efficacy of Covid vaccines for many months now [1, 2], but until recently I was never really curious enough to investigate his original claim to fame, namely that he has invented mRNA vaccines to begin with. So let’s do it. Let’s do a deep dive on Malone’s inventorship claims.

In particular, at various times he has repeatedly claimed to have invented mRNA vaccines, the field of “messenger mRNA [sic] therapeutics”, or even “RNA as a drug”. Intriguingly, until Dec 2020, Malone’s own website only modestly called him “one of the original inventors of ‘DNA vaccination’” and did not even mention RNA, but then in May 2021 it was changed to start calling Malone “the inventor of mRNA vaccines”, and by June 2021 the website claimed Malone invented the entire field:

In support of his inventorship claims, Malone cites Vical Inc. patents where he is one of the co-authors. However, the first author on all those patents is Dr. Phillip Felgner who was the one who actually invented the lipid encapsulation methodology, which is the methodology that makes it possible to deliver DNA and RNA to cells (the process is often called ‘lipofection’ which stands for lipid-mediated transfection).

When elaborating on the topic, Malone usually concedes that he didn’t invent any actual vaccines (i.e. approved therapies) but rather invented the underlying “mRNA vaccine technology”. However, as we just saw above, the lipid-mediated delivery technology that Malone used in late 1980s to deliver RNA into cells was not invented by him. Moreover, none of that particular technology is used in today’s mRNA vaccines.

How It All Began

So how close was Malone to the original invention of that delivery technology? Well, Dr. Phillip Felgner started working on his lipofection methodology in the early 1980s long before he knew Malone and by 1987 has demonstrated that it can be used to successfully deliver DNA and RNA into cells — as disclosed in one of Felgner’s patents filed in 1987, which provided examples of doing so for both DNA and RNA. Note that the above patent predates any patents that Felgner would later file with Malone as a co-author.

Felgner’s most famous lipid molecule is DOTMA, and it has since become a key ingredient of one widely used commercial lipofection reagent, Lipofectin™. DOTMA is what Felgner would later send to Malone to experiment with RNA in late 1987.

Note that DOTMA-mediated lipofection is just one of many possible lipofection methods — others have used liposomes to accomplish the goal of RNA delivery into cells almost a decade prior. For example, one group has demonstrated successful liposomal delivery of mRNA into murine cells in 1978, while another did the same for human cells. However, canonical liposomes were suboptimal for gene delivery purposes, particularly due to their negative charge, and it took Felgner’s breakthrough in creating positively charged lipids to catalyze the lipofection gene delivery approach.

Malone first started experimenting with using Felgner’s DOTMA lipid formulation to deliver RNA into cells in late 1987 — as part of his graduate work at Salk. In early 1988, Malone wrote a letter to Felgner, in which he thanked Felgner for the chance to contribute data to “[Felgner’s] DOTMA RNA transfer project” and for sending over the actual lipids for Malone to do so. In the same letter Malone fully acknowledged that Felgner has already tried his own method on RNA but was having problems:

So based on his own words, Malone’s contribution was figuring out how to use the DOTMA methodology for mRNA lipofection and diagnosing why Felgner’s earlier attempts to do so were unsuccessful: judging by the above letter, Felgner used mRNA that didn’t have a poly(A) tail — a feature that is important for mRNA stability in eukaryotes. Also, the reporter gene Felgner used to verify that his mRNA was actually being translated into protein was hard to detect at low levels.

Why did Felgner collaborate with Malone on the DOTMA RNA experiments in the first place? It seems that Malone had a skill that was quite rare at the time — he mastered a recently developed method of RNA synthesis. It was developed in 1984 by the groups of Krieg and Melton at Harvard, and turned custom RNA synthesis from a prohibitively complex process into something even a grad student could do. Of course, being one of the first to master a novel method someone else has invented is laudable but not an invention in itself.

So, in a nutshell, it seems that Malone’s contribution to Felgner’s DOTMA project was synthesizing the right mRNA that would be stable and detectable in eukaryotic cells, mixing that mRNA with the DOTMA lipids provided by Felgner, and confirming that the same cellular uptake and expression of lipid-encapsulated nucleotides which Felgner has previously observed with DNA and RNA was also happening with mRNA synthesized by Malone.

An Inventor is (Not) Born

Excited by his success, Malone then wrote an internal Salk memo in which he mused that DOTMA-mediated mRNA delivery could provide “a way to treat RNA as a drug”. It seems that Malone believes that this letter marks the point at which he has invented “RNA as a drug” in 1988:

Does this letter really grant Malone the title of having invented RNA as a drug? Does his contribution to Felgner’s DOTMA project really make Malone the inventor of mRNA vaccines?

Speaking of vaccines, they were an interest of Felgner long before he met Malone. Felgner was working on vaccine adjuvants at Syntex since the mid-1980s and one of the applications of the DOTMA methodology was to deliver vaccine adjuvants — in fact, the DOTMA RNA lipofection example provided in Felgner’s 1987 DOTMA patent demonstrates delivering RNA for a known vaccine adjuvant, poly I:C. So it is only natural to assume that it was just as obvious to Felgner, as it was to Malone, that the DOTMA technology (or any other technology) to deliver DNA or RNA into animals can be used to deliver genes on which the animals’ immune system can be trained to produce protective immunity, i.e. as a vaccine.

Of course, in biotech, an idea is nothing without execution, and for patenting purposes Felgner needed to provide animal data which would have demonstrated that the DNA or RNA they administer to animals do, in fact, produce the desired therapeutic response, be it the production of a needed protein, or an immune response to one.

To generate such animal data Felgner engaged several collaborators, one of which was Dr. Jon Wolff, who ultimately ran some of Vical’s animal studies at his fledgling lab at the University of Wisconsin-Madison. Wolff’s initial animal studies did not investigate any potential therapeutic effects but just tried to confirm that the reporter gene (CAT) encoded by DNA or RNA liposomes administered to animals does end up being successfully taken up by their cells and translated into protein.

That work was by then done under the umbrella of Vical — a startup which Felgner joined in late 1988 from Syntex and brought over his lipofection platform. In early 1989 Felgner also recruited Malone, who by then had a falling out with his supervisors at both Salk and UCSD, and decided to drop out of his PhD program. However, at Vical, Malone also lasted only a few months before resigning in August of 1989 due to perceived unjust claiming of credit by Felgner for Malone’s “intellectual contributions to the gene therapy project”:

In his brief time at Vical, one of Malone’s roles relevant to our analysis was to produce lipid-encapsulated DNA and RNA formulations used by Wolff in his animal experiments, as well as “naked” (i.e. not encapsulated) DNA and RNA to be used as experimental controls. The data from those experiments was then used for Vical’s only patent application 07/326,305 filed while Malone was still with the company. That application would go on to provide the priority date for several of Vical’s subsequent lipofection gene delivery patents.

Intriguingly, as Vical’s patent application was being written, Malone wrote a competing patent application for Salk, which was filed on the same day (March 21, 1989) as the Vical patent application, and which tried to shoot down Felgner’s previous claims of successful RNA transfection via his DOTMA method:

It is interesting to see Malone’s above dig at Felgner — “no experimental results or even theoretical considerations are disclosed in support of this speculation”. It looks like Malone was unaware that DOTMA-mediated RNA transfection was actually disclosed in Felgner’s patent US4897355 in 1987.

The Salk application written by Malone looks even more ethically and legally dubious when one considers that Malone (who was already employed by Vical at the time) was carbon-copied by Felgner on a Vical memo to its patent lawyers regarding Vical’s own application:

The very same memo by Felgner spelled out vaccine applications of the lipofection platform in question:

Moreover, whereas Malone’s application did not even mention vaccination or provide any in vivo data on gene delivery (in vitro RNA transfection of embryos which were then lysed hours after RNA delivery doesn’t really qualify as in vivo data), Vical’s application provided actual examples of vaccinating animals by mRNA:

I believe Vical’s focus on HIV vaccines was inspired by its cofounder Dennis Carson (who is probably most known for having co-invented an anti-cancer drug). Prior to co-founding Vical, Carson was at Scripps where he worked on anti-HIV treatments for several years. He also looked into the Epstein-Barr virus’ gp110 envelope protein with Gary Rhodes who would later also join Vical and lead its DNA vaccine efforts targeted at HIV’s gp120 envelope protein. In any case, only mRNA (as per above) and not DNA vaccination was part of the original 1989 Vical application, and Rhodes would only be added as a co-inventor to DNA vaccination patents in 1990. Curiously, as Vical’s internal meeting notes show, gp120 mRNA for those studies was produced by Jon Wolf’s own lab and not by Malone (see bullet VI):

So it was Wolff’s lab that both prepared the HIV mRNA formulations and ran the HIV vaccination animal studies. As Malone’s lab notes show, the only animal experiments he was involved in (via preparing DNA/RNA formulations for them) were reporter gene experiments (CAT and luciferase) and not any clinically relevant animal studies. For the latter studies — like experiments involving HIV (gp120 or NEF genes), polio or EMC viruses — it was Wolff’s lab that was making its own DNA/RNA formulations.

Thus, as Malone was not involved in any vaccination studies during his time at Vical, it is unclear how he can claim any credit as even a co-inventor of mRNA vaccination, let alone call himself “the inventor of mRNA vaccines”.

So What Did Malone Invent?

Putting aside the bad look (and legal aspects) of Malone trying to scoop his own company’s patent application for Felgner’s lipofection platform, let us come back to the potential novelty of what Malone has done in using lipofection to deliver RNA into cells.

Even back then it was clear that Malone was far from the first to do so: just a few months after the Salk application, in his joint paper with Felgner and his former Salk PhD supervisor, Dr. Inder Verma, Malone acknowledged that other groups have previously succeeded at delivering RNA into cells: “progress in introducing RNA molecules into cells has been very slow and restricted to a few cases (1–4).” Curiously, in those 4 references of prior works Malone did not include the two 1978 papers (i.e. a decade prior) that used liposomes to deliver mRNA into human and murine cells, which I cited earlier.

The animal experiments conducted by Wolff which showed that DOTMA-mediated lipofection can be used to deliver mRNA not just into cells but into adult animals — and, more importantly, to get that mRNA translated into detectable quantities of protein — were subsequently published in 1990 in a joint paper with Malone, Felgner and four other authors. By then Malone had already left Vical and went back to Northwestern to complete his medical degree.

That paper was to be his last collaboration with Felgner or Wolff. However, lipofection work at Vical was only beginning. In fact, all of Vical’s patents [1, 2, 3, 4, 5, 6, 7, 8, 9] on which Malone is named as a co-inventor were filed after he left, in some cases many years after. The original patent application 07/326,305 filed during Malone’s tenure at Vical was ultimately abandoned and only used as a priority date anchor for subsequent divisional or continuation-in-part applications.

Scientific publications on DNA and RNA lipofection by Felgner and his team also continued. Judging by the signatures on the relevant Vical lab notebook pages posted by Malone, his responsibilities on the DNA and RNA lipofection project were taken over by Felgner’s wife, Dr. Jiin Felgner:

She was one of Vical’s first employees, and has stayed with the company for a decade. In the 1990s, she has also co-authored several of Vical liposomal patents [1, 2, 3] that do not have Malone as a co-inventor, and has also co-authored several academic papers on the topic [1, 2, 3, 4].

Coming back to Phil Felgner, I think one of the best measuring sticks of the relative impact of Malone vs. Felgner on the field of RNA delivery is provided in the 1998 paper by Katalin Kariko, who many believe might get a Nobel Prize for her role in the development of mRNA therapeutics. In that paper Kariko et al. cite Felgner 10 times in connection with mRNA lipofection, while only citing Malone once and even then in co-authorship with Felgner. I should mention that Kariko has herself managed to successfully deliver RNA into cells long before Malone — as early as 1985. Notably, both Felgner and Kariko have been recognized in 2021 for their contributions to COVID-19 mRNA vaccines by Spain’s Princess of Asturias Award for Scientific Research — a very prestigious scientific award sometimes dubbed as “Spain’s Nobel Prize”.

Moreover, most of Malone’s academic citations come from papers and patents in co-authorship with Felgner, namely 2 papers and 9 patents, which, as I mentioned above, were all filed after Malone left the company in 1989. Google Scholar shows 12,847 citations for Malone, with 5172 of those coming from the Wolff et al. paper on in vivo RNA and DNA delivery using Felgner’s lipofection. Another 3607 of Malone’s citations come from Vical patents filed between 1996 and 2007. For comparison, Felgner has 41,355 citations in Google Scholar, with his pre-Vical 1987 lipofection paper having the most citations at 6258.

Naked DNA and RNA Delivery

The story of Malone and Felgner would be incomplete without the naked DNA/RNA delivery saga. During animal experiments that Wolff was running for Vical on in vivo delivery of liposomal DNA and RNA formulations, it was accidentally discovered that you could actually deliver naked DNA and RNA i.e. one not encapsulated in lipids.

Initially such naked DNA and RNA was meant to be used as control by Wolff, i.e. no activity was expected to be observed from such formulations. However, Wolff’s experimental readouts have shown that even naked DNA/RNA get into animal cells and result in protein expression. This was so unexpected that it was initially dismissed as an experimental error. However, after repeating the experiment, Vical was confident they had indeed stumbled upon a monumental discovery. They have filed appropriate patents and have made an announcement in the press in March of 1990:

The experiment was so elementary, and the results so surprising, that researchers working with San Diego’s Vical Inc. couldn’t really believe what they were seeing. It all seemed too simple.

They had been injecting submicroscopic fatty globules containing DNA or RNA into mice to see what would happen. The idea was that the fat globules, called liposomes, would be taken up by cells. The cells would use the genetic material inside to make proteins they couldn’t otherwise make.

The researchers found moderate success with that, but the rigors of science demanded that the experiment have a “control” portion — injecting the raw DNA or RNA into the mice to show that the liposomes themselves were making it possible for the new genes to be incorporated into the cell’s processes.

It turned out the cells like the raw material even better and began making the new proteins for as long as six months.

“This was a big surprise, and that’s really what you’re looking for in this area,” said Philip L. Felgner, director of product development at Vical. Felgner worked on the experiment with Dr. Jon Wolff and others at the University of Wisconsin at Madison.

…

Vical hopes that the results of this checking and double-checking, reported in today’s issue of the journal Science, will convert the company from a bare-bones start-up to a major player in the ranks of San Diego’s biotechnology community.

The above article barely mentioned Malone as an also-ran, and this must have caused him great irritation. In fact, years later, in 2001, he even wrote to Vical’s CEO making a case that it was Malone and not Jon Wolff who meets the inventorship criteria for discovering the naked DNA/RNA delivery methodology:

In the process, Malone seems to shoot down the very validity of the discovery:

As the 1990 LA Times article mentioned, Vical considered the naked DNA delivery discovery as very important. In particular, it was thought to pave the way for “DNA vaccination” — a term that became quite popular in the 1990s, and which Malone originally claimed to have co-invented, before he converted himself into “the inventor of mRNA vaccines” or “RNA as a drug”. The 1996 Lancet article explains why Vical felt the naked DNA delivery held so much potential:

In fact, what Felgner and Wolf originally thought was a laboratory error turned out to be a seminal observation — one that may lead to a revolutionary new type of vaccine — the naked DNA vaccine.

Only a small amount of protein was produced after DNA injection, but Felgner and Wolf immediately realised that it might be enough to generate an immune response to an antigenic protein. They tested this idea by injecting a plasmid containing the gene for the HIV envelope protein gp120 and found that they induced both antibody and cytotoxic immune responses in mice.

Vical’s co-founder and co-author of its gene delivery patents, Dennis Carson, in a 2006 interview claims to have recognized the potential of naked DNA delivery for “DNA vaccination immediately”:

In our initial experiments, it was discovered serendipitously that naked DNA, without the liposome delivery system, gave better expression in the muscle than the DNA encapsulated in a liposome. I wrote a letter — I still have it — in 1988, to the CEO of the company, saying this was a major discovery and it could have very interesting applications in immunology. … I realized that injecting naked DNA with a vector simply mimicked what a virus did.

As I mentioned above, Carson was an immunologist previously working on HIV therapies, and Felgner has worked on vaccines long before he met Malone. The original March 1989 Vical patent application not only mentioned vaccination as one of the applications of DNA/RNA delivery but provided a concrete example of vaccinating mice by mRNA coding for HIV’s gp120 protein. Those experiments did not involve Malone at all. Moreover, after Malone left, Vical has entered into a collaboration with Chiron to test their gp120 DNA (rather than mRNA) formulation. Gary Rhodes led those animal studies at Vical in-house and has earned a spot as a co-inventor on relevant DNA vaccination patents. In fact, it seems that at around the same time DNA vaccines became the focus of the industry and have remained so for many years, as evidenced by Merck licensing the Vical patents with the goal of producing their own DNA vaccines.

One possible reason for the popularity of DNA vaccines over RNA ones is RNA cytotoxicity due to excessive inflammation — an issue that seemed prohibitive to the use of “RNA as a drug” until in 2008 Katalin Kariko and Drew Weissman have figured out how to circumvent it with pseudo-uridine modRNA.

Maybe that relative obscurity of RNA therapies over DNA ones until essentially 2020 is why until May 2021 Malone has referred to himself only as a co-inventor of the DNA vaccine technology, as evidenced by his website metamorphosis which I highlighted in the beginning of this article, as well as his 2017 CV:

As I’ve already mentioned, the original Vical patent application only provides examples of mRNA vaccination, and DNA vaccination work was done by Vical after Malone had left the company. Of course, his name was still retained on DNA vaccination patents which were filed by Vical after Malone left because those patents were continuations of the original Vical application, but that doesn’t mean Malone did any actual work on DNA vaccination. Even the mRNA vaccination experiments — i.e. experiments trying to elicit expression of a therapeutically relevant protein rather than just a reporter gene like CAT — that are provided in the original Vical application were done without Malone’s involvement.

So why in May 2021 did Malone precipitously abandon his self-appointed title of co-inventor of “DNA vaccination” in favour of “the inventor of mRNA vaccines”? Does that mean he no longer believes he has invented DNA vaccination? He could have easily claimed to have invented both DNA and mRNA vaccination. Was his sudden rebranding brought on by the success of mRNA Covid vaccines? After all, Malone’s wife claimed that they were “now saving the world from COVID-19”. That is quite ironic in itself, given that Malone is one of the loudest voices producing all sorts of anti-vaccine rhetoric about these very vaccines: from claims that the spike protein encoded by them is cytotoxic and will often “cause permanent damage in children’s critical organs” to claims that these vaccines are unsafe for women or that they cause ADE.

As a bit of an aside but a relevant point on Malone’s character, I found it quite interesting that one recurring theme in recollections by Malone or his wife is how often he felt that he was treated unfairly by nearly everyone he worked with. One quote that I found particularly illustrative on this point comes from Malone’s letter to UCSD that he sent a few weeks before resigning from Vical. In that letter he was “begging for reinstatement back into the PhD Program”, as per Malone’s website, and had this to say about his former colleagues:

That seems like quite a long list of people Malone felt victimized by in his short time at UCSD.

In any case, Malone’s own motivations for making his claims aside, I hope my analysis has provided enough evidence for people to draw their own conclusions about the role he has played in the invention of mRNA vaccines or RNA as a drug.

The backdrop

Sometime in April 2021 Bret Weinstein started flirting with antivaxx misinformation. Their first date was when he invited Geert Vanden Bosche on his podcast and instead of challenging Geert’s outlandish claims seemed to fawn over them.

In a few short weeks, Bret turned full antivaxxer and an avid ivermectin fan, which culminated in his “How to save the world” podcast with known antivaxxers Steve Kirsch and Robert Malone, as well as podcasts with an ivermectin superfan Pierre Kory: one ominously titled “COVID, Ivermectin, and the Crime of the Century”, and another “Emergency podcast” with Joe Rogan.

Those podcasts got millions of views, and emboldened Bret to further amplify his antivaxx and ivermectin messaging. Considering that Bret and Heather have weekly podcasts where they have been repeating the same points week to week, they just might have been the biggest source of antivaxx and ivermectin misinformation in the world during this pandemic.

Who knows how many people they managed to persuade to follow their example of forgoing vaccinations in favor of “prophylactic ivermectin”. After all, Bret sounded extremely convinced in the protective powers of ivermectin against Covid, and even took it live on air:

Yuri Deigin on Twitter: "And yeah, Bret totally was not doing this to lead by example, and totally was not recommending that any of his fans follow his example. That's why he did it live on air, right? Wink-wink. pic.twitter.com/7LBrcQRZjn / Twitter"

And yeah, Bret totally was not doing this to lead by example, and totally was not recommending that any of his fans follow his example. That's why he did it live on air, right? Wink-wink. pic.twitter.com/7LBrcQRZjn

https://twitter.com/ydeigin/status/1425613829497663498

Ironically, a few short months later Bret would find himself live-eating his words rather than ivermectin, as he had to distance himself from the Carvallo study which alone formed the basis for his conviction in ivermectin’s 100% efficacy at preventing Covid (23:48 timestamp):

However, back at the start of Bret’s descent into antivaxxery, many people who were more qualified to assess biomedical claims have tried to tell Bret he is wrong, but to no avail. He ignored criticism and plowed straight into promoting deadly misinformation that almost certainly resulted in deaths of some of those he managed to persuade.

The pushback and the anti-pushback

In July, Claire Berlinski and I came out with a public criticism of Bret’s medical misinformation campaign in our Quillette article. That’s when a hitherto unknown DarkHorse superfan leapt to Bret’s defense: Alexandros Marinos wrote a long Twitter thread about our article that tried to debunk it “line-by-line” .

Marinos’ chivalry did not go unnoticed by Bret and Heather, as they mentioned him and his thread quite favorably on their podcasts, agreeing that they do, in fact, need better skeptics — better than Claire and I, presumably. Alas, Marinos’ BetterSkeptics project was born.

By the way, thanks to his efforts, Marinos would later gain favor with another antivaxxer, Robert Malone, whose opinion he seemed to value very highly:

Here is how Marinos personally describes the idea behind the project:

The BS project went live 10 days after Marinos’ original Twitter thread. The crux of it was as follows: it nominated 3 referees to assess any submitted claims from 4 podcasts featuring Bret: the three podcasts I already mentioned above, as well as DarkHorse podcast #87 where Bret and Heather addressed the criticisms in our Quillette article.

The claims submitted to BS were meant to point out falsehoods or unsupported statements made by Bret or his guests. The referees would then give a grade to each claim, and claims that got enough points would make it to the next round, where they could be “appealed” by others with counter-evidence, after which the referees would render their final decisions.

Ultimately, 126 claims would be submitted, of which 44 were deemed to be “assessable” by referees, i.e. worthy to be analyzed by the refs and receive a validity score from 0 to 5. That was the “S1 score”, or the grade after the first round.

12 of the 44 claims made it to the next round, which then mimicked an appeals court where defenders of Bret and Heather could challenge initially validated claims in an attempt to overturn them. If they succeeded, the referees would lower the score of a previously validated claim rendering it no longer so. The “S2 score” after this round was the final score. Whatever claims would survive the two rounds with an aggregate score of 9 or above would be deemed as validated by the BS project.

Of the 12 initially validated claims, 9 would ultimately be overturned, resulting in just 3 claims of 44 to be ruled false or unsupported by the referees. Following a public outcry, the organizers then surprisingly re-reviewed, and then “over-overturned” 2 of those 9 overturned claims, for a final total of 5 claims of 44 to ultimately be ruled false or unsupported.

Doomed from the start

The biggest problem of the BS project was as glaring as it was obvious: only 1 of the 3 referees had the requisite expertise to assess biomedical claims: Dr. Jocelynn Pearl who has a PhD in Molecular and Cellular Biology and actual drug development experience. The other two referees were Max Borders and Paul Crowley. Max is a philosopher and writer, and Paul is a software engineer. Obviously, peer review is only as good as the peers doing the reviewing.

The difference in referee expertise becomes obvious if one compares the sheer number of claims that Max or Paul would deem worthy of qualifying for the next round (by giving claims a score of 3 or above) to that of Jocelynn.

Jocelynn gave a passing score to 36 of the 44 claims:

While Paul and Max gave passing scores to just 12 and 14 claims, respectively:

The second issue is that the evaluation process was not blinded: all referees saw each other’s grades and comments in real time. Also, Max and Paul were usually a couple of days behind Jocelynn in their analysis, and thus had the benefit of seeing Jocelyn’s marks so they could adjust theirs to potentially ensure the desired outcome.

Another very important point that deserves mention is the following: for a claim to qualify for the next round, it needed to receive a score of 9 or above aggregated from the 3 referees, and in the beginning Jocelynn was giving scores of 5, 4 or 3 to the claims she considered valid. But within a day or two of the start of the project, Paul has decided that 3 should really be the maximum score for any valid claim, and subsequently convinced the other referees of this:

This would now imply that for any claim to make it to the next round, all 3 referees would need to unanimously agree on that. Is there any wonder then that just 12 claims made it through?

Even so, 12 potential falsehoods must have seemed like too much to the DarkHorse superfan BS organizers, and 9 of the 12 were quickly overturned after an emotional outburst by Marinos:

Yuri Deigin on Twitter: "And yes, I can even point out the exact time you decided to walk back on those 12 initially validated claims: when I mentioned the fact that you validated one of Heathers falsehoods.Your bias is obvious, as is obvious that the judges backtracked on their marks after this tweet: https://t.co/9kIFAnTgsF / Twitter"

And yes, I can even point out the exact time you decided to walk back on those 12 initially validated claims: when I mentioned the fact that you validated one of Heathers falsehoods.Your bias is obvious, as is obvious that the judges backtracked on their marks after this tweet: https://t.co/9kIFAnTgsF

https://twitter.com/ydeigin/status/1425720854386823170

The absurdity of the overturning process did not go unnoticed, and ultimately resulted in the project sheepishly reinstating 2 of the overturned claims, contrary to the project’s own rules:

Better Skeptics on Twitter: "Do vaccines "normally" stay in the shoulder area after injection?This was one of the challenges submitted as part of our Ground Truth Challenge.The submission didn't make it past rebuttals, but when putting together our review we felt the evidence both ways was inconclusive. / Twitter"

Do vaccines "normally" stay in the shoulder area after injection?This was one of the challenges submitted as part of our Ground Truth Challenge.The submission didn't make it past rebuttals, but when putting together our review we felt the evidence both ways was inconclusive.

https://twitter.com/BetterSkeptics/status/1428421637624406021

Claims

Let us now dive into the actual claims that were submitted to the BS project.

Validated falsehoods

First, here are the 5 claims the project verified as false or unsupported.

Claim #1 (IVM 100%)

Target quote: [00:03:40] Bret: All right. I’m unvaccinated but I am on Prophylactic Ivermectin, and the data actually shocking as this will be to some people, the data suggests that Prophylactic Ivermectin is something like a 100% effective at preventing people from contracting COVID when taken properly.

This one is pretty self-explanatory. The only source supporting ivermectin’s 100% efficacy against Covid was the deeply flawed Carvallo study, and even Bret has now distanced himself from it.

In some ways, it was quite vindicating to see the BS project validate the most important falsehood on which Bret’s entire narrative was built: that ivermectin is 100% effective at preventing Covid. Take that claim out and Bret’s entire house of cards comes crashing down.

Claim #2 (Vaccine spread)

Target quote: [00:12:21] Steve: And so, if it was really a safe vaccine, then what I just saw was impossible. So that’s kind of what triggered me off on this, and then I had been on this Canadian of physicians call, uh, with Dr. um, Byram, uh, Bridle. And he, uh, showed this chart and he said, “Normally, when you vaccinate someone, the vaccine goes in the shoulder, and it stays in the shoulder, in-in the shoulder area. And what happens is the antibodies are generated and then they attack the, um, this antigen, in the shoulder, and it wipes it out, and-and we’re good.”

This is one of those flip-flopping claims. It made it through the first round, was then overturned by the refs in S2 but ultimately reinstated by the organizers. I guess because it would be too embarrassing to not validate this obvious falsehood.

Claim #6 (Skipped animal trials)

Target quote: [00:23:16] Bret: […] Now, Heather and I on this program when we heard how the vaccines worked and that they had been accelerated through this process our sense was, uh, you’re gonna deliver a brand new technology with respect to deploying it in humans, and it’s going to interact with the immune system, and you’re gonna speed it through this process. We did not know that it had skipped the animal trials that might have alerted us to something really dangerous, um, [crosstalk] there.

[00:23:50] Steve: Yeah, they didn’t even- they didn’t even test the vaccine in itself in the animal trials.

Well, at least this ridiculously wrong claim was validated.

Claim #8 (WHO airborne)

Target quote: [00:14:16] Kory: So-so right now, the CDC finally came around and admits that it’s airborne transmitted, the WHO still does not.

This one was an obvious slam dunk, as at the time of the podcast, the WHO recognized airborne spread of Covid for almost two months.

Claim #10 (ZBW IVM)

Target quote: [01:35:01] Kory: […]. South Africa and Zimbabwe, especially Zimbabwe, when they were getting hurt earlier in this year, they basically eradicated COVID with the widespread adoption of Ivermectin. They were doing dealing with the South African variant.

This is the second flip-flopping claim which was overturned by refs in S2 but resurrected by the organizers. Ivermectin use was obviously far from widespread in Zimbabwe, and at the time of the podcast Zimbabwe was in the midst of its third wave with skyrocketing infection numbers.

Validated but then overturned

Now let’s look at what claims the refs initially agreed were false or unsupported but then changed their mind.

Claim #3 (Spike toxicity):

Target quote: [00:13:22] Bret: And the other- the other problem and this is your area of expertise, is that what the- what these vaccines do is they encode spike protein, alone, so that the immune system will learn to recognize by protein and will catch it quickly when one is confronted with COVID. But the spike protein itself, we now know is very dangerous and cytotoxic. Is that a fair description?

What Bret was claiming there is that the vaccines encode a lone spike protein that is known to be “very dangerous and cytotoxic” by itself, i.e. not as part of a virion. It is an obvious falsehood based on an incorrect interpretation of the “Salk study” which the study’s own author refuted on numerous occasions.

The refs initially recognized this but then for some reason decided to cite the very same Salk study as a justification for overturning the claim, completely omitting the lone part of the spike being cytotoxic, and choosing to ignore several counts of evidence that proved the lone spike is not toxic in concentrations many orders of magnitude greater than what is seen in humans after vaccination:

Yuri Deigin on Twitter: "No, there is no evidence that the spike protein alone is "very dangerous and cytotoxic" in concentrations even orders of magnitude higher than what is ever observed in vaccinated people. https://t.co/6j6MJiyVJL / Twitter"

No, there is no evidence that the spike protein alone is "very dangerous and cytotoxic" in concentrations even orders of magnitude higher than what is ever observed in vaccinated people. https://t.co/6j6MJiyVJL

https://twitter.com/ydeigin/status/1417213855177101324

I was quite disappointed to see the BS refs waver on this obvious falsehood as it formed the basis for a lot of Bret’s antivaxx fearmondering and should have been vocally refuted.

Claims #4 & 5 (vaccines and ovaries):

Target quote: [00:13:53] Steve: And-and to be fair, the FDA did not think that, um, the S1 subunit and the spike protein was toxic. And so, they knew, the FDA knew about the biodistribution. And-and one of the scary things is that the biodistribution like peaks in your ovaries.

Now, this claim is blatantly false. It has been shown time and time again that Kirsch blindly used a misleading graph that omitted tissues with concentrations 2–15x higher than that of the ovaries:

Based on his misunderstanding, Kirsch then proceeded to wax poetic on it with some illogical scaremongering that betrayed a lack of biological understanding:

Eventually, the group that made the original graph that Kirsch used subsequently corrected it by putting other tissues with higher concentrations:

But for some reason they still omitted the injection site, which dwarfs all other sites when included:

Moreover, the study in question that Kirsch got his graph from was done in rats, so “peaks in your ovaries” is doubly misleading. How in the world the BS project failed to validate this obvious falsehood is beyond me. Their explanation is as illogical as it is juvenile:

Conclusion: Ineffective. The first challenge focused on the technicality of the word “peak”, and missed an opportunity to discuss whether the concentration found in the ovaries is scary or significant, especially in the light that subsequent repeat-dose studies didn’t find elevated levels in the ovaries. That said, the word “peak” is clearly a source of ambiguity, and we feel that specifying that the quote could have clarified that the study was done in rats, especially in light of the word “your” in front of “ovaries”.

Technicality of the word ‘peak’, eh? This just might rival the infamous Clinton quote: “It depends on what the meaning of the word ‘is’ is”.

Claim #7 (Remdesivir cost):

Target quote: [02:00:17] Robert: MBAs controlling the practice of medicine and, uh, the use of alternative drugs. The other disincentive they have, and this is- this is really, you know, coming out of the federal treasury. This is coming out of your tax dollars. Okay? Is that, um, they don’t make the same compensation if-if they’re- if they prescribe remdesivir, you know, $6,000 or $8,000 pop, um, for treatment course, versus they prescribe ivermectin. Okay?

The cost of a treatment course of remdesivir is $3,120, which the BS project directly acknowledges. But then instead of validating the obviously false claim, the project comes up with the lamest excuse for not doing so:

After reading the above, it becomes crystal clear the entire project is just an exercise in exonerating Bret and Heather, and watching the (il)logical pirouettes the organizers are capable of performing to accomplish their goal could make anyone dizzy.

Claim #9 (ZBW eradicated Covid):

Target quote: [01:35:01] Kory: […]. South Africa and Zimbabwe, especially Zimbabwe, when they were getting hurt earlier in this year, they basically eradicated COVID with the widespread adoption of Ivermectin.

When the podcast in question took place, Zimbabwe was well into its third wave of the virus:

This should have been one of the easiest claims to validate. And yet, the Better Contortionists managed to wiggle their way out of doing so with yet another exercise in rationalization:

I don’t even…

Claim #11 (FLCCC 0.2 mg)

Target quote: [00:42:01] Heather: Uh, and the results were so staggeringly promising with regard to the ivermectin and carrageenan, so then they did the-the bigger study, which had 1195 healthcare workers in Argentina, uh, in which 788, about two-thirds of them were put on this combination of ivermectin and carrageenan. Uh, difference being in the main study in 1195 people study, uh, the ivermectin was, um, a-a weekly dose. It’s actually exactly the, uh, same protocol that the FLCCC has-has put up as — What is it called the [unintelligible 00:43:03] Math+ protocol I think. It’s like two mgs per kilogram bodyweight — . [00:43:08] Bret: Two milligrams. .2 milligrams per kilogram body weight.

Ok, if there is any claim I am content with being invalidated or overturned, this one’s it. Heather clearly confused the names of the FLCCC protocols, and the prophylactic one did indeed have 0.2 mg/kg at the time.

Claim #12 (IVM teratogenic):

Target quote: [01:17:01] Heather: […] Uh, to the second paper, uh, which they cite for the teratogenicity of the drug, um, the people actually find the opposite.

Here again the refs pull the Bill Clinton defense by debating what is actually meant by “find the opposite”. The opposite of labeling the drug as teratogenic is establishing that the drug is NOT teratogenic, i.e. that there have been no observed teratogenic effects in human or animal studies. This stands in stark contrast with the study we cite which reported a 15% increased risk of stillbirths and spontaneous abortions, and a 69% increased risk of congenital abnormalities.

Calling this “finding the opposite” of teratogenicity is clearly false. And yet, the BS project managed to overturn this claim.

Claim #13 (IVM fertility effects):

Target quote: [01:17:01] Heather: […] So to these particular claims, uh, we just say that the paper that they cite for testicular- testicular dysfunction, um, A, it’s in rats, it’s not in humans, and they- and the-the paper itself says, “Ivermectin and acaricide and anthelmin-helminthic drug of the family of Avermectins produced by streptomyces, avermitilis cultures, is a well-tolerated drug with no side effects in mammals in pharmacological doses.” Now, this is in the introduction of the paper which Berlinski and Deigin are claiming as finding statistically dysfunction. And when you actually read the paper, you find that there are tiny effects maybe but it’s in combination with this other drug, Verapamil, in which fertility meiotic effects become an issue. So that’s cheating. That’s not the result that they were claiming.

Here the BSers used the an interesting new tactic to overturn the claim they initially validated: “Only the abstract of the study is available for free”. I guess the fact that I posted the relevant table from the actual paper in my original submission somehow escaped their attention:

That table clearly shows that ivermectin alone does have negative effects, which is why the claim was a slam dunk to begin with, and should have stayed that way.

Rejected outright

Now let’s look as some clearly valid claims that did not even make it past the first round.

Submitted Claims #29 and 45 (vaccines not safe for women at all)

[02:20:57] Steve: Yeah. And-and-and-and it’s the-the cognitive dissonance between what the CDC says is that this is safe for pregnant women.

[02:21:05] Bret: Right. It’s cra- at some level, it’s not safe for women at all.

This one is really weird. Steve and Bret make an obviously invalid claim, which two BS refs agree should be validated as ‘unsupported’ but Paul misunderstands it and decides to reject it completely by giving it a 0 (I guess because even a 1 would have given the claim the necessary 9 points to be validated):

I then provided extra context to clear Paul’s misunderstanding, and Jocelynn and Max again agreed the claim should be upheld but Paul again gave it a 0:

Thus a clearly valid claim that got a grade of 5 from Jocelynn and a 3 from Max did not even make it through the first round.

Claims #12, 13, 14 (ivermectin can eradicate Covid and end the pandemic in a month):

1: https://www.betterskeptics.com/transcript-how-to-save-the-world-in-three-easy-steps/

[00:04:45] Steve: One-one could argue that if everybody just took Ivermectin for a month, worldwide — we would end the pandemic.

2: https://www.betterskeptics.com/transcript-jre-1671-bret-weinstein-dr-pierre-kory/

[00:57:47] Bret: All right, a couple of things. Um, one, Jamie, can you put the abstract to that paper back up? Because there’s this thing so the world is very focused on using ivermectin to treat COVID, which I understand, but we miss this other thing. So scroll down a little bit. Therapeutic advances, there it is, sentence in the middle. Okay, this might be one of the most important sentences written this century. Low certainty evidence found that ivermectin prophylactic — prophylaxis reduced COVID-19 infection by an average of 86%, 95% confidence interval between 79% and 91%.

So, that sentence actually, is a hallelujah sentence. Because what it means is, even if ivermectin were completely ineffective at treating people who have COVID, that number is high enough, because it is over the number, uh, that we understand herd immunity to be for this disease. Uh, any number that has been proposed, uh, as far as I know, because that number is so high, what it means is that ivermectin alone, if properly utilized, is capable of driving this pathogen to extinction.

3: https://www.betterskeptics.com/transcript-covid-ivermectin-and-the-crime-of-the-century-darkhorse-podcast-with-pierre-kory-bret-weinstein/

[00:58:22] Dr. Kory: You could eradicate, period.

[00:58:22] Dr. Weinstein: — is clearly enough to end the pandemic, and the benefit, very few people understand really what we’re up against because they don’t think about this in evolutionary terms, the longer we leave that bug out there, the longer it is transmitting between people, the greater the chance that it evolves into something we can’t eradicate like seasonal influenza.

[00:58:42] Dr. Kory: There’s no question. Thank you for bringing that.

On at least three separate occasions Bret or his guests claim that ivermectin alone can eradicate Covid and end the pandemic. This is an obviously ridiculous claim, and should have been easy work for the BS project. Jocelynn saw that clearly, Max was splitting hairs about what ‘is’ is, and Paul just rejected the submissions outright:

Shame.

Claims #42 and 43 (leukemias and lymphomas):

[02:21:25] Bret: — preferentially. Yes, the lipid is ending up there. And then the other place that we have a signal, which I think suggests something we need to worry farther down the road. You tell me, Robert, if I’m on the right track here or not. But the fact that it shows up concentrating in bone marrow.

[02:21:39] Steve: Bad news.

[02:21:40] Bret: Actually that suggests that you could end up with — And I’m not saying this is gonna happen, but I’m saying we need to look for something like leukemias showing up here, because of their creation in the bone marrow. I know from other work, that it also seems to show up preferentially in-in, uh, lymph nodes, which raises the question of whether or not, uh, lymphomas might be created. In any case, these are possible long-term effects that we have no way of knowing, but don’t arise because these things have not been injected into people for more than a year.

Leukemia is a cancer of the early blood-forming cells. Lymphomas are a group of blood cancers caused by an oncotransformation of lymphocytes, a subclass of immune system cells like T- and B-cells. For a cell to turn cancerous, driver mutations are necessary. Such mutations are usually endogenous but could also be caused by external factors: radiation or chronic exposure to external carcinogens, e.g. asbestos or cigarette smoke. Mutations by definition happen in the DNA of cells, and DNA is located in the nucleus.

Unlike DNA vaccines, mRNA vaccines deliver inside cells an already transcribed RNA molecule which never goes into the nucleus. This was the inspiration behind the mRNA platform, as it eliminates oncogenic risk, since mRNA cargo cannot integrate into cell genome, unlike the DNA one. Moreover, mRNA cargo delivered in mRNA vaccines disappears from the body within 1–2 weeks.

In fact, Moderna’s founder Derrick Rossi is on record saying that his inspiration for the mRNA platform was trying to deliver Yamanaka genes into cells without the risk of oncotransformation:

Moreover, all vaccines “show up in the lymph nodes”, as that is where immune system training happens. In all of the history of vaccines, no cases of vaccine-caused lymphomas have been observed. Thus, there is no plausible mechanism by which mRNA vaccines could cause leukemia or lymphomas, and Bret’s two scaremongering claims are completely unjustified.

These two obviously unsupported claims got a 4 from Jocelynn and a 3 from Paul but then Max killed them with 0 and 1 grades based on very weak reasoning:

Claims #26 and 28 (variants do not escape ivermectin)

[01:22:14] Dr. Kory: It’s no question. Then so what the phrase that we like is that we consider it a bridge to and a safety net for. The other reason why it’s a safety net is because of these escape variants. The idea that these vaccines are going to have long-term efficacy is not clear. They seem to be working quite well as far as on the efficacy side, the safety, I’ll leave alone for now, but we don’t know what it’s going to look long-term. Ivermectin, on the other hand, is doing incredibly against all the variants. We know that in the areas of Brazil that did a concerted use of it, it was crushing the P.1 variant, especially if you use it early.

We knew it for doctors on the ground. There are cities in Brazil that used it. That totally, it is completely effective against the deadly P.1 in Brazil. The South African variant, which gave them fits a few months ago, that went to Zimbabwe. 60% to 90% of the cases in Zimbabwe, whether it’s South African variant, they literally eradicated COVID using ivermectin in Zimbabwe.

We know it did well against South Africa, did well against the P.1, the UK variant, which went to Eastern Europe, Slovakia, and Czech Republic, they adopted international guidelines, cases and death counts plummeted. All of the variants, it has so many mechanisms of action. We still don’t know all the mechanisms, and we’re still not positive on what exactly the mechanisms are, but the ones that we think are, there’s multiple, and they’re not going to deal with — the variants are not going to escape this. It has many tentacles.

[01:24:39] Bret: Now, you could imagine, and in fact, I wondered, whether or not ivermectin was going to suffer the same fate that effectively we were going to get resistance because, like an antibiotic, it is having some action, and to the extent that there are variants that aren’t susceptible, they have an advantage but what you’re telling me is that. actually, empirically speaking, if nothing else, whether we know the mechanism or not empirically speaking, the variants do not appear to escape ivermectin. They do appear to escape vaccines.

These are absolutely unsupported musings by Bret and Kory with zero evidence to back them up. As such, these claims should have quickly been validated by the refs. Jocelynn gave them a 4, Max 3 and 2, but then Paul, while agreeing with the logic of his fellow refs, gave them 0s with an extremely surprising rationalization:

I agree with your reasoning and I think this reasoning is bogus, but “the podcast correctly states the evidence, reasons from it in a way they make clear, and there’s an error in their reasoning” isn’t nearly strong enough to back the claim that Dark Horse propagates falsehoods.

Wait, what?

Claim #32 (grossly underreported VAERS deaths)

[01:02:16] Steve: Which means 5,000 reports translates into 500,000 deaths. And I don’t think it’s that high. Okay?

[01:02:31] Steve: Right. It’s not that high, but-but it’s higher. It’s — I guarantee you it’s higher.

Even the 5000 VAERS death reports at the time of the podcast were yet unconfirmed to be caused by vaccines — in fact, most were ruled not to be. And the 100x estimate for underreported events cannot be extrapolated to underreported deaths, as the original estimate comes from a 2011 report assessing all possible adverse events after vaccinations, the overwhelming majority of which are benign: headache, fever, pain at injection site, etc.

Moreover, the lack of correlation between vaccination rates and reported VAERS deaths clearly refutes the causal link between vaccines and VAERS deaths:

Ironically, at some point Marinos personally tried to challenge my analysis above only to subsequently validate it:

Alexandros Marinos on Twitter: "CORRECTION. Off by one error. Should teach me to post charts without waiting a bit to cross-check. Lesson learned, I hope.First picture is deleted incorrect tweet Second picture is corrected data. pic.twitter.com/bPywloCUIX / Twitter"

CORRECTION. Off by one error. Should teach me to post charts without waiting a bit to cross-check. Lesson learned, I hope.First picture is deleted incorrect tweet Second picture is corrected data. pic.twitter.com/bPywloCUIX

https://twitter.com/alexandrosM/status/1424379421033590785

As an aside, this wasn’t the only time Marinos opined on the VAERS data. When he leapt to Bret’s defense against Sam Harris’ criticisms, he claimed that the bad design of the VAERS website might be responsible for a 10x underreporting of VAERS deaths:

Alexandros Marinos on Twitter: "...anyone, but due to the incredible amount of confusion that has been allowed to grow due to the terrible UI/UX of VAERS, the fear is that there is an order of magnitude undercounting in the numbers. Of course this isn't causation, but at those levels, you run out of options. / Twitter"

anyone, but due to the incredible amount of confusion that has been allowed to grow due to the terrible UI/UX of VAERS, the fear is that there is an order of magnitude undercounting in the numbers. Of course this isn't causation, but at those levels, you run out of options.

https://twitter.com/alexandrosM/status/1418862092975702016

Yup, he said that. That bad VAERS website design would cause 90% of the people who come to it to report a potentially vaccine-related death to turn away. This is ridiculous for several reasons: most potential adverse events are reported by medical professionals who are required by law to report them, and they are well familiar with the VAERS system. And any laypeople who might also want to submit something, would also be able to navigate through the 3 pages of a VAERS submission — it is not flashy but pretty straighforward.

Anyways, coming back to the original VAERS claim at hand — Jocelynn again gave this claim a passing grade but the other two refs did not, even despite agreeing that Kirsch’s claim is unsubstantiated in the comments:

Sigh.

Claims 30 and 44 (Quillette authors don’t know what prophylaxis means):

[00:39:34] Heather: the authors of the Quillette article don’t seem to know what prophylaxis means. …

[00:40:48] Bret: Yeah, so they absolutely screw up. And somehow it got through the editing process.

This claim refers to our Quillette article, so I am obviously an interested party here, but the claim is still clearly false and should have been upheld as such. Heather simply misunderstood what the paragraph she quoted was saying.

Here is a screenshot of the full quote from our Quillette article:

Note the italicized any in “any significant benefit”. Any there stands for either prophylactic or treatment benefit, and the reference to the JAMA study is meant to highlight that even as treatment (i.e. not as crazy a suggestion as promoting ivermectin over vaccines for prophylaxis), ivermectin has no proven benefit when the clinical trial is done properly. It is very clear from our JAMA quote that the study we cite talks about treatment (“patients with mild COVID-19”).

With lack of evidence for ivermectin’s efficacy even for treatment out of the way, our next paragraph then dives into the prophylactic comparison of ivermectin to vaccines, explaining why it’s not even close. Just from that paragraph alone it should be clear we know what prophylaxis means.

Jocelynn clearly agreed that the claim should be validated, as she gave it a score of 5 but Max and Paul simply refused to give it any consideration. Max gave it a 0 with the following eyebrow-raising explanation:

This counterclaim is trying to do too many things. You can’t falsify a claim in which someone says “seems,” because that just makes it subjective. It seems to her that X is unfalsifiable, because the seem means it’s true for her, even if she were in error. Let’s get beyond this kind of petty stuff as the referees’ time is valuable.

Even if Heather’s “seems” should have then given her a free pass to say any falsehood (by Max’s logic), Bret’s subsequent “Yeah, they absolutely screw up” agreement should have nullified the indulgence given to Heather by Pope Maximums. Paul was also unwilling to validate the claim based on highly dubious logic:

Even after I resubmitted it with clarification, the refs refused to budge. Paul’s comment is the most telling: “Still not allowing this one. The Quillette article played stupid games and won stupid prizes.”

So much for fact-based analysis.

Claim #19 (soil bacteria):

[00:17:30] Bret: the fact that the drug in question, Ivermectin, comes from, uh, soil bacteria, it’s not a completely synthetic molecule, means that that it is likely to be similar to the things that one’s ancestors have encountered before and there’s, therefore, a good chance that the body has a reasonably elegant way of dealing with it rather than uh, using some mechanism that’s-that’s not so great.

This is a clearly unsupported claim by Bret. There is no evidence that there is “a good chance that the body has a reasonably elegant way of dealing with” a molecule extracted from soil bacteria. There are plenty of counterexamples of toxic molecules found in soil bacteria:

• https://en.wikipedia.org/wiki/Anthrax
• https://en.wikipedia.org/wiki/Tetanus
• https://en.wikipedia.org/wiki/Botulinum_toxin

The refs agreed the claim was unsupported but for some reason not strongly enough:

Conclusion

Even if the referees or organizers of the BS project might not agree, thanks to their efforts, it is now crystal clear that just four of the podcasts they analyzed were rife with falsehoods and mortally dangerous medical misinformation.

As I said, since Bret and Heather have weekly podcasts where they have been repeating the same points over and over, they just might have been the biggest source of antivaxx and ivermectin misinformation in the world. They reached tens if not hundreds of millions of viewers, and almost certainly convinced many people to forgo vaccinations in favor of “prophylactic ivermectin”.

Even though they have now greatly subdued their antivaxx/ivermectin crusade and renounced some of their crazier claims, they are under the impression that the BS project has somehow provided them with a “clean bill of health”. They have repeated this on a number of occasions, and Marinos himself seems to promote this spin on the results of the project.

I hope that after reading this analysis you will be convinced that nothing could be further from the truth. Even ivermectin couldn’t rescue Bret and Heather’s suffocating antivaxx narrative, which upon close examination is revealed to be total BS.

References

Here is the full BS project spreadsheet with all of the submitted claims and referee comments:

Ground Truth Challenge

PS: If you found this article informative, you might also enjoy this one:

Why Bret Weinstein is dangerous

His name was Leslie Lawrenson. He seemed to have bought into Bret’s misinformation:

He then got Covid and died:

Leah McElrath 🏳️‍🌈 on Twitter: "Before being found dead at home, Leslie Lawrenson made this video about being glad he got COVID because he didn't trust vaccines.Family said he was "brainwashed by the stuff that he was seeing on YouTube and social media."More: https://t.co/tw6SNtZqHppic.twitter.com/Q7zo5ZCGM3 / Twitter"

Before being found dead at home, Leslie Lawrenson made this video about being glad he got COVID because he didn't trust vaccines.Family said he was "brainwashed by the stuff that he was seeing on YouTube and social media."More: https://t.co/tw6SNtZqHppic.twitter.com/Q7zo5ZCGM3

I ask two favorite Russian questions: Who is to blame? What should be done?

As some pundits have already pointed out, we are living in two pandemics: the Covid pandemic and the misinformation pandemic. The second one makes the first one deadlier than it has to be.

Besides Leslie Lawrenson, there have been other highly publicized cases of Covid killing people who bought into various Covid misinformation. Here are just a few recent ones:

• Phil Valentine
• Joe Manning
• Marc Bernier
• Caleb Wallace
• Marcus Birks
• Dick Farrel
• Tod Tucker
• Jimmy DeYoung Sr.
• Stephen Harmon
• Mikel Lowe

Between May and September 2021 Bret Weinstein has publicized dozens of false or unsubstantiated biomedical claims on his podcasts, collectively heard by millions of people. In fact, a project started by Bret’s superfan Alexandros Marinos, which was dedicated to examining Bret’s claims in just four podcasts, has collected 44 such potential claims, 36 of which were graded as false or unsupported by the only referee in that project with relevant biomedical experience.

The goal of this essay is to provide further, detailed evidence that Bret does not possess requisite competence to make biomedical claims.

A bit of context

Bret has been a friend ever since he invited me on his podcast to talk about my Medium article on the lab leak hypothesis (1, 2). We were communicating a lot in the second half of 2020 but then as Bret’s attention switched to his Unity 2020 idea, our communication gradually subsided. I would still catch some of Bret’s tweets once in a while but nothing really piqued my interest until he put out a podcast on vaccines with Geert Vanden Bossche.

That podcast surprised me because what Geert was saying should have raised a whole lot of red flags to anyone with a solid grasp on biomedicine. So that was a canary in the coalmine moment for me that something is not right with Bret’s biomedical understanding. I tried to push back on some of the claims right away:

Yuri Deigin on Twitter: "Ok, @BretWeinstein, this is seriously awful and illogical advice from @GVDBossche. "Best to be seronegative"??? Didn't help much the original seronegative people in Wuhan or the 3M dead from COVID. Where were Geert's wonderful natural antibodies there? 🤦‍♂️ pic.twitter.com/2E4AFvSaTP / Twitter"

Ok, @BretWeinstein, this is seriously awful and illogical advice from @GVDBossche. "Best to be seronegative"??? Didn't help much the original seronegative people in Wuhan or the 3M dead from COVID. Where were Geert's wonderful natural antibodies there? 🤦‍♂️ pic.twitter.com/2E4AFvSaTP

At about the same time, I also challenged Bret’s claims about ivermectin but he just kept dodging my inquiries:

Yuri Deigin on Twitter: "Interesting. So we know from clinical trials that vaccines reduce risk of getting Covid by 10-20x. By how much do you estimate IVM reduces that risk?Then if you do get Covid, your risk of dying on IVM is still 1.4-3.9%. What would you estimate your risk of dying from vaccine? pic.twitter.com/bK47rMVPhs / Twitter"

Interesting. So we know from clinical trials that vaccines reduce risk of getting Covid by 10-20x. By how much do you estimate IVM reduces that risk?Then if you do get Covid, your risk of dying on IVM is still 1.4-3.9%. What would you estimate your risk of dying from vaccine?

I should say that I was fighting antivaxxer misinformation even before Bret has turned into one. My May 2021 Medium article addressed some of the most popular antivaxx claims of the day:

Get vaccinated! It could help you, like, not die…

Thus I was quite shocked to then see Bret go full antivaxxer just weeks after my article in his “How to save the world” podcast with Malone and Kirsch:

How to save the world, in three easy steps.

As a friend should, at first I tried to challenge Bret’s increasingly vocal antivaxx messaging privately, but to no avail. So we then published a written critique of Bret’s misinformation with Claire Berlinski, but it only prompted Bret and Heather to double down on their dangerous views on Dark Horse podcast #87:

I addressed Bret and Heather’s erroneous responses to our Quillette article first on Twitter and then on Rebel Wisdom’s podcast, where I also went over 18 of Bret’s key falsehoods:

In a nutshell, my 3 key points are as follows:

• Vaccines are safe and effective
• Ivermectin is ineffective and dangerous
• Bret Weinstein is unqualified to make biomedical claims

I have addressed the first two points on numerous occasions. If you are interested in detailed rebuttals of Bret’s antivaxx or IVM misinformation, the following spreadsheet links to individual Twitter threads containing evidence for each of the 44 false or unsupported claims submitted to BetterSkeptics:

Ground Truth Challenge

Now, let me expand on my third point above.

Why Bret is Unqualified

A great example illustrating Bret’s lack of biomedical understanding is his claim that approved drugs are unsafe because they have been tested in lab mice, which, as Bret claims to have “happened upon” in his PhD work, have telomeres that are longer than wild mice. Here is the quote from Dark Horse podcast #87 in which Bret explains that because of his 2002 hypothesis, their household is highly skeptical of pills:

[00:11:10] Bret: In our family, we actually have a kind of concern about pharmaceuticals very generally and this arises from the fact that in my graduate work … I happened on to what I believe is a flaw in the drug safety system. I published this flaw. And the flaw basically amounts to the mice that are often used for things like drug safety testing and other experiments having been accidentally evolutionarily modified by the breeding protocol that is used to produce them so that their telomeres which are these repetitive sequences at the ends of chromosomes have been elongated tremendously and [this] has potentially very large impacts. Effectively, these animals have a capacity to repair their tissues so that if you poison them, but you don’t outright kill them, they actually have an extremely good capacity to fix themselves whereas we have a limited capacity, so they’re bad models.

Now, that has caused us over the many years [since] that work was published in 2002 [to] have a kind of uneasy relationship when we opened the medicine cabinet because we know that virtually every drug in there was tested against mice that are very, very capable of dealing with toxicity, and so we don’t know. And there are lots of instances which drugs have shown themselves to be toxic after they were proven- or proven safe in safety testing. So, Vioxx, Fen-Phen, Gleevec, Seldane, erythromycin, all of the NSAIDs, all of these drugs have been shown to be hazardous at one level or another. So, ordinarily, in this household, we are very pro-vaccine and we are a bit skeptical of pills.

[00:12:56] Heather: A lot skeptical.

[00:12:57] Bret: A lot skeptical of pills because we know that even if you know that they’re hazardous, you don’t know which pills it applies to and in what way.

There is plenty of bad science and even bad logic in that short quote. Let me address both.

Bad logic

First, I will elucidate the ‘bad logic’ part. Even if Bret’s claim that lab mice have longer telomeres was true, his inference that approved drugs are unsafe because of it is simply illogical. Drug safety is validated in several rounds of human trials before any drug is approved, so even if lab mice somehow masked some dangers of a given drug candidate for humans, those dangers would still be caught during human trials.

Or, conversely, if those dangers were so rare that they slipped even through human trials and were only caught by pharmacovigilance years after drug approval, as was the case in the examples Bret mentions, then no animal trials would have been able to catch a side-effect so rare that it only manifests itself at a rate of 1/100000 or rarer, as each animal study only consists of dozens of animals.

Also, even before a drug is allowed to be tested in humans, its safety must first be validated in several animal species, including non-rodent species. So even if lab mice were somehow more resistant to drug toxicity, other animals would not be. But then again, subsequent human testing makes the point about potential animal testing issues moot.

Bad science

Now, let me address the ‘bad science’ part. First of all, if you are really interested in all of the details of this “stolen Nobel Prize” story as it’s come to be known, watch this podcast entirely devoted to it. In a nutshell, the story from Bret’s perspective is as follows: he came up with a theoretical prediction that lab mice might have much longer telomeres than wild mice due to peculiarities of lab breeding protocols. He then told it to Carol Greider, and she “stole it” — first confirmed it in her lab, and then published this confirmation without acknowledging Bret as the source of the hypothesis.

There are several problems with Bret’s claim. First of all, “could lab mice have longer telomeres?” is not an earth-shattering insight worthy of a Nobel Prize. Carol Greider got hers for completely different reasons.

Second, Carol published the paper which Bret claims “stole” his idea in 2000. But Bret published his ‘prediction’ after, in 2002. This is already a bit of a mismatch. But the biggest mismatch is that Carol has actually published the same observation in 1995:

So there’s that.

Ironically, Bret’s original prediction and Carol’s observation have actually been disproved by Jackson Labs, the premier research animals breeding organization:

So there’s that as well.

Finally, telomeres were quite fashionable in early 2000s but have since largely proven to be a red herring. They’ve been disproved as the cause of cell senescence — there could be many other pathways to cause a cell to become senescent that just telomere shortening. They’ve also been disproved as a cause of aging. Here is a great illustration showing that:

The red box shows that for a ~10 kb specie telomere length, there are species with wildly different lifespans — from 2 to 210 years, and the blue box shows that for a maximum lifespan of 5–20 years, there are species with wildly different telomere lengths.

Oh, and lab mice don’t have “bizarrely long telomeres” as Bret has put it. In fact, some other species have them even longer (last column, in kb):

Squirrels, shrews, rabbits, and even tigers have longer telomeres than mice. Moreover, rats have telomeres about as long as mice. So there really isn’t anything special about telomeres in mice — lab or otherwise.

In light of all of the above, I can’t help but revisit something else Bret said in the “stolen Nobel Prize” podcast. In it Bret claims that after reviewing his 2002 paper Carol produced “pages and pages” of criticisms, which Bret implies were made in bad faith — as an attempt to somehow torpedo his paper (although the first time she made them was in private to Bret). However, after looking at his 2002 paper, I cannot help but wonder that maybe Carol was honestly trying to help. Because as it stands, Bret’s paper is, well, not very good. For example, their opening hypothesis is just wrong:

We hypothesize that (1) in vertebrates, a telomeric fail-safe inhibits tumor formation by limiting cellular proliferation. (2) The same system results in the progressive degradation of tissue function with age.

Even in 2002 it would have been shortsighted to assume that age-related tissue-degradation is driven by telomere shortening. We have over 200 different tissue types, each with its own aging dynamic. Neurons don’t divide after birth and survive for hundreds of years (in whales), while hematopoietic stem cells divide constantly and actually have active telomerase to elongate telomeres on demand.

Also, the following quote alone is enough to conclude the authors are not well-versed in the highly variable lifespans between different species, even between mammals where some rats live for less than a year and whales live for over 200:

If a simple modification of telomere-system parameters would extend life without significant costs, it should already have spread due to selection.

That quote reminds me of the following joke:

An economist and his friend are walking down the street when the friend sees a $100 dollar bill on the sidewalk.

“Look,” he says, “a $100 dollar bill!”

“Nonsense,” says the economist. “If that was a $100 dollar bill, someone would have picked it up by now.”

Bret’s telomere hypothesis, while possibly the most illustrative, is not the only evidence why he should not make biomedical claims. There are a couple of other, more recent statements that also raise eyebrows of competent biologists.

Ivermectin and soil bacteria

In the same Dark Horse podcast #87, Bret muses that ivermectin is somehow superior because it was originally extracted from soil bacteria:

The fact that the drug in question, ivermectin, comes from soil bacteria, it’s not a completely synthetic molecule, means that that it is likely to be similar to the things that one’s ancestors have encountered before, and there’s, therefore, a good chance that the body has a reasonably elegant way of dealing with it rather than using some mechanism that’s-that’s not so great.

There is absolutely no evidence that a molecule extracted from soil bacteria has any safety of efficacy advantage. There are plenty of counterexamples of toxic molecules found in soil bacteria: anthrax, botox, Clostridium tetani (cause of tetanus), etc.

Anti-GMO and anti-sunscreen

Also in podcast #87, Heather read an excerpt from their joint book which revealed that Bret and Heather espouse anti-GMO and anti-sunscreen beliefs. Heather also repeated the “if it were that easy, selection would almost certainly found a way” trope, to which I would only ask why selection hasn’t provided humans with endogenous antibiotics or vaccines, and we had to first invent science to fix selection’s many shortcomings.

Bret and Heather’s anti-sunscreen stance is quite revealing of their anti-science beliefs. Heather actually said that sunscreen “probably caused more problems that it solved”:

Heather E Heying on Twitter: "Yet another modern fix that probably caused more problems than it solved: Sunlight has benefits we don't yet understand, so eradicating the (delicious) feel of sunlight on your skin is not healthy-it's hazardous. An evolutionary take saw this coming, too.https://t.co/5fTFdsu5TN / Twitter"

Yet another modern fix that probably caused more problems than it solved: Sunlight has benefits we don't yet understand, so eradicating the (delicious) feel of sunlight on your skin is not healthy-it's hazardous. An evolutionary take saw this coming, too.https://t.co/5fTFdsu5TN

She then revealed that she has been withholding sunscreen from her own children:

Heather E Heying on Twitter: "Terrible, & not surprising. It's not difficult to raise children without sunscreen, either. @BretWeinstein and I rarely put the stuff on our boys when they were younger. We taught them to avoid burning via behavior: every 20 min you're outside, take 5 in the shade. It works. / Twitter"

Terrible, & not surprising. It's not difficult to raise children without sunscreen, either. @BretWeinstein and I rarely put the stuff on our boys when they were younger. We taught them to avoid burning via behavior: every 20 min you're outside, take 5 in the shade. It works.

Bret also weighed in on this topic a while back:

As usual, Bret tried to tie the scientific findings they were discussing on their podcast to his telomere work and his “lifetime capacity for repair” hypothesis. Let me just quickly dispel that hypothesis: animals do not have some sort of limited capacity for repair that they use up and thus die. If that were the case, athletes would keel over in their thirties, and mice undergoing caloric restriction would live shorter, rather than 50% longer lives. On the contrary, it seems various insults seem to prolong lifespan and/or make animals fitter — that’s the concept of hormesis.

In the clip above, Bret also seems to make a distinction between getting sunburns and getting “sun-damaged skin”, which reveals a misunderstanding of the biological process behind UV’s effect on skin. Both sunburn and suntan are skin’s responses to detecting UV-caused DNA damage. (I’ve actually dug into this topic quite deeply, albeit in Russian). And sun-damaged skin is the result of many repeated instances of such UV-caused DNA damage: when DNA damage is beyond repair, most cells become senescent (growth-arrested, and it’s not due to telomere shortening!), while a small proportion of such cells might actually turn cancerous.

Published papers

Some of Bret and Heather’s listeners hold a misconception that the two are highly credentialed scientists. They are not: they have not done any research in years, and any research they have done years earlier is minimal. In all of his academic career, Bret has published only two biology papers (in 2002 and 2005), while Heather — three (2001, 2006, 2007):

Yuri Deigin on Twitter: "As we are actually comparing credentials - just to clear up any confusion - even in terms of evolutionary biology, let alone drug development, Bret and Heather are the opposite of credentialed. Bret has 2 evo bio papers to his name, Heather has 3: pic.twitter.com/b6FTY8OnmB / Twitter"

As we are actually comparing credentials - just to clear up any confusion - even in terms of evolutionary biology, let alone drug development, Bret and Heather are the opposite of credentialed. Bret has 2 evo bio papers to his name, Heather has 3: pic.twitter.com/b6FTY8OnmB

I think by now it should be abundantly clear that Bret is not competent to make biomedical claims, and should retract all of the false antivaxx and ivermectin claims he has made. He should also refrain from making further biomedical pronouncements.

Bret knows that the price for spreading Covid misinformation is measured in lost lives. He conceded as much on his podcasts. However, he felt he had a moral obligation to speak up as he thought he possessed some superior knowledge that was being suppressed. Tellingly, he didn’t seem to feel the same responsibility if his highly publicized views turned out to be wrong and responsible for people’s deaths:

Yuri Deigin on Twitter: "And the fact that you knowingly continue to propagate falsehoods about vaccine safety and IVM efficacy while at the same time saying that whoever is going to listen to you will only have themselves to blame if your misinformation ultimately kills them - that is totally immoral. pic.twitter.com/kZee8ZrLBp / Twitter"

And the fact that you knowingly continue to propagate falsehoods about vaccine safety and IVM efficacy while at the same time saying that whoever is going to listen to you will only have themselves to blame if your misinformation ultimately kills them - that is totally immoral. pic.twitter.com/kZee8ZrLBp

Unfortunately, as in the past, Bret’s superior knowledge is simply a delusion. He is wrong. Moreover, when shown to be wrong, he is unwilling to admit it but would rather see himself as a kind of a martyr who is being suppressed by some DISC forces.

That’s not what good faith actors do. Good faith actors acknowledge their mistakes and correct any previously wrong claims they have made.

Bret, if you are truly a good faith actor, please publicly retract your erroneous claims and tell your followers to get vaccinated ASAP. I still hold out hope that deep down you are a good man.

PS: Leslie Lawrenson has left behind his partner of 14+ years and their two sons. You can donate to his family here:

Support for Amanda & The Boys, organized by Carla Hodges

Ok, I’ve just about had it with all the ignorance and misinformation surrounding Covid vaccines. For some reason even smart and knowledgeable people have a blind spot when it comes to them. Maybe vaccines just got a bad rep from the days of antivaxx activists creating “vaccines cause autism” memes a decade ago, and it is subconsciously coming out now at a time of great fear, uncertainly and doubt that Covid has unleashed on our planet.

Let me be clear — our kids these days get dozens of shots and those vaccines are VERY safe. But new vaccines, like any new medical treatment, definitely need to have their safety and efficacy proven before being allowed to be used in millions of people. And yes, ideally, it would be nice to have a few years to monitor them before that. But in case you haven’t noticed, we are in the middle of the worst pandemic the world has seen in a century, and unless we stop Covid, it just might take the crown for the worst pandemic ever.

Vaccines are our only hope of stopping it and thanks to science and technology, humanity is in a great position to do so — we can create novel vaccines in silico in a matter of hours, start testing them in vitro in days, in animals in a couple of months, and then test their short- and mid-term safety and efficacy in humans in another few months.

This provides us with a great advantage over the SARS2 virus, as, on average, it acquires only about 25 new mutations per year (in a 30k genome), which means that the chances are very high that we can stamp it out before it has enough time (and seronegative people!) to develop enough mutations to avoid the antibody repertoire that our vaccines create in every vaccinated person.

Vaccines 101

What is the point of a vaccine? To help your body not let the virus get into your cells and start replicating. And even if some of it still manages to get into cells, to speed up the process of your immune system killing those cells. Yes, those infected poor little cells of yours are goners, c’est la vie.

All right, so how can we help the body prevent the virus from entering cells? With neutralizing antibodies. Normally those are only generated after your immune system acquaints itself with a pathogen, and they take many days to get produced in large enough numbers. That’s what vaccines are meant to shortcut: they present the immune system with the very part of the virus that it uses to get into cells — the spike protein — so that our bodies generate neutralizing antibodies in advance. This way, if and when the virus does find its way into our body, those neutralizing antibodies will bind to its spike and not let it get into any meaningful number of cells. Which means that not only we won’t feel sick but also that we won’t produce and ‘shed’ (i.e. spread) copies of the virus that could infect others and potentially kill grandma even if you, oh brave man, are willing to take your chances playing the Covid roulette.

So do you see how vaccines are not only good for you, but good for everyone, as they very effectively can help stop the spread of a new virus? Come on, you don’t need a PhD to understand this.

But what if the vaccine kills me?!

Ok, seriously, your chances of dying from Covid itself are much higher than your chances of dying from a vaccine. There have been no vaccine-related deaths reported in clinical trials, and even after 250 MILLION doses have been administered in the US, only ~4500 deaths were reported in VAERS as potentially having any sort of a connection:

Even if all of those deaths were indeed caused by a vaccine, that is still a meager 1 in 55556 chance of dying (or ~0.002%). And even that is a huge overestimation, as CDC reports just a few of those deaths to be causally related to vaccinations.

Note that 1.5M Americans die from heart attacks, stroke or cancer every year. Which amounts to about 4K people per day. So having 4.5K deaths in a cohort of ~150M people getting 250M vaccine shots in a period of 5 months is actually significantly lower than expected background mortality.

The Covid roulette

So what are your chances of catching Covid and then dying from it? Obviously, they greatly depend on your age, as well as where you live, and a few other factors. But let’s just focus on US numbers and do some quick estimations — enough to give us the right orders of magnitude.

We have these nice CDC data from the time when there were just 15M cases in the US (after about 6 months of the pandemic). Now it’s 33M, in case you missed it.

We also have these recent data for deaths:

So using the above three sources, I have put together a quick spreadsheet to estimate the risks of catching Covid, dying from it once caught, and the cumulative annualized risk of catching Covid and dying from it as of early 2021 (when total Covid cases were ~28M). Here it is:

Nothing terribly surprising there: the older you are, the much higher your chances of dying from Covid. But even if you are aged 30–39, your risk of dying is not negligible: 0.14% is nothing to sneeze at, pardon the pun. And if you are aged 50–64, that risk is tenfold greater: 1.5%! Given that all adults had about the same 10% probability of catching Covid in the past year, the overall risk of doing so and dying from it is definitely worrying: 0.15% for the 50–64 group or a whopping 1% for the 75–84 group.

Or at least those were the numbers in the past year, and they would probably remain so if nobody was vaccinating. Fortunately, thanks to all people who got a jab, rates of new infections are rapidly dropping in the US. However, the chance to catch Covid and die is still quite sizable. Probably not 0.15% or 1% anymore for 50–64 or 75–84 groups, but even if it is 10x lower now thanks to vaccinations, it is still 0.015% or 0.1% for the unvaccinated.

Which means that even in worst case vaccine death scenario and best case reduced Covid transmission rates scenario, today your risk of getting Covid and dying is still 10–50x times higher (0.015% or 0.1%) than the risk of dying from a vaccine (<0.002%).

But what about Covid vaccines’ long-term safety?

As I said, it would have been great to wait a few years to study the long-term effects of Covid vaccines, but we just don’t have the luxury. Allowing Covid to ravage the planet and potentially develop novel variants while we wait for any potential vaccine long-term effects to manifest themselves is just dumb. This will leave millions more dead (ok, I concede, no unknown long-term issues there) and hundreds of millions Covid survivors susceptible to potential Covid long-term issues. And given a choice between potential long-term issues from Covid or the vaccine, I think the vaccine is a much safer choice.

Especially because the vaccine is essentially a much smaller subset of the virus, which, unlike the virus, does not replicate and gets fully eliminated from your body in a matter of days. Only its memories remain, in the form of antibodies and, well, memory B-cells and T-cells — and, again, all those are a subset of what usually remains after a Covid infection. So if you are willing to brave Covid for the sake of getting that “natural” immunity, you really shouldn’t worry about braving a vaccine.

What is even more encouraging is the fact that the clinical effects of a vaccine shot are obviously waaay less severe than the effects of catching a live virus. This has been documented in numerous studies, so the fact that the short- and mid-term safety of vaccines is much higher than that of Covid, makes a pretty compelling case that long-term safety of vaccines is similarly much better.

But the vaccine epitope repertoire sucks! I want MOARE protection, just give me Covid!

Wait, let me get this straight. You prefer to risk dying from Covid for protection against getting it a second time rather than get a vaccine and not risk dying from Covid? That makes zero sense.

And just because the epitope repertoire from a “natural” Covid infection is broader than that of a spike-specific vaccine, doesn’t automatically make it better for protection from subsequent infection. In fact, it might even make it worse. Read on.

What exactly is an epitope repertoire? It’s the collection of various short fragments of viral proteins that the immune system is trained to recognize and attack after a natural infection or a vaccine. When our immune cells encounter the natural virion, they will slice up its various proteins into these short fragments and will eventually raise armies of B-cells and T-cells, meant to produce antibodies against those fragments or kill infected cells displaying those fragments.

However, as I already mentioned, not all antibodies are created equal. To protect from future infections, the most useful antibodies are neutralizing antibodies — ones that will prevent the virus from entering our cells. Unfortunately, our immune system does not know which antibodies will be neutralizing, and which ones won’t, as it will blindly pump out antibodies against the epitopes it encounters. But we, humans, armed with our brains, know that only antibodies that stick to the spike protein can be neutralizing, as the spike protein is the only thing the virus has sticking out of it, and all other proteins are hidden inside. So for the purposes of a vaccine, we can just present the immune system with the spike and not divert its resources on producing useless antibodies, like against the N (nucleocapsid) protein, for example. Especially since higher N antibody titers were observed to be associated with worse disease progression:

What about ADE? Won’t vaccines cause ADE?!

In case you don’t even know what ADE is, it stands for Antibody-dependent enhancement, which is basically when your antibodies turn from friend to foe and instead of preventing the virus from entering your cells, actually help it get in. ADE has been observed in Dengue and HIV and potentially in first SARS, so scientists were rightfully concerned that it might be an issue with SARS2 as well.

Thankfully, no ADE has been observed with SARS2 in any of the human vaccine trials. And ADE is not exclusively a vaccine thing, in fact, “natural” immunity might actually have a higher chance of ADE for future reinfections, as it’s got a broader antibody repertoire than the vaccine-generated one, so more chances for some particular viral epitope to turn out to be enhancing.

But but but Geert said vaccine antibodies will mess up my natural antibodies and prevent them from neutralizing the virus!

Yeah, ok, that is just crazy talk. But yes, that is what Geert Vanden Bossche said:

Not only would people lose vaccine-mediated protection but also their precious, variant-nonspecific (!), innate immunity will be gone (this is because vaccinal antibodies outcompete natural antibodies for binding to Covid-19, even when their affinity for the viral variant is relatively low).

I was going to write my own rebuttal of Geert’s claims but when I googled him, I came across the following detailed deconstructions of his conjectures: one by Edward Nirenberg, and another one by Snopes. Here is an excellent response to Geert’s quoted claim by Nirenberg:

This is absolute, unvarnished nonsense. Bossche is referencing the production of natural IgM, which is generated by B1-B cells as a stopgap measure against infections until more potent responses can be initiated; these antibodies are polyreactive, nonspecific, and critically: constitutively produced. They are always present for as long as B1-B cells generating them live. IgM is pentameric and thus even though it has lower affinity than antibodies that have had the opportunity to evolve superior binding affinity, it can compensate with the fact that it has 10 binding sites instead of 2. However, IgG antibodies bear many of the same effector functions (actually, they tend to be better at many of them, as Table 10.27 shows) and they can diffuse into extravascular sites unlike IgM. Principally, antibodies against SARS-CoV-2 could be of value if they are neutralizing. Bossche presents no evidence to support that natural IgM is neutralizing (rather than just binding) SARS-CoV-2.

Bottom line: Geert is wrong. Don’t listen to Geert.

I heard SARS2 can get into my DNA! Isn’t this a reason to fear vaccines?

Umm, no, this could be a reason to fear SARS2, and want to get vaccinated to not get SARS2, but in no way can this be a reason to fear vaccines.

The question about SARS2 integrating into our cells is still an open one. Yes, there’s been this study that showed that in vitro (in a Petri dish) they can induce LINE-1 reverse transcriptase (which is a part of endogenous human retroelements which make up almost half of our genome) to integrate parts of SARS2 genes into human DNA. Is that a big deal? No, because those SARS2 snippets will probably not be used by our cells to make new proteins and even if they were, those proteins won’t be able to assemble full virons, and moreover those proteins, even if indeed expressed, would likely quickly trigger an immune response and death of said cell.

After their in vitro success, the authors then tried to detect whether there were similar insertions in Covid patient tissues and potentially detected some genome sequencing reads that could have been legit, or could have been due to contamination:

Chimeric reads generally accounted for 0.004–0.14% of the total SARS-CoV-2 reads in the samples. A majority of the chimeric junctions mapped to the sequence of the SARS-CoV-2 NC gene (SI Appendix, Fig. S6 C and D). This is consistent with the finding that NC RNA is the most abundant SARS-CoV-2 subgenomic RNA (56), making it the most likely target for reverse transcription and integration. However, recent data showed that up to 1% of RNA-seq reads from SARS-CoV-2–infected cells can be artifactually chimeric as a result of RT switching between RNA templates, which can occur during the cDNA synthesis step in the preparation of a RNA-seq library (57). Thus, because there is a mixture of host mRNAs and positive-strand viral mRNAs in infected cells, the identification of genuine chimeric viral–cellular RNA transcripts is compromised by the generation of artifactual chimeras in the assays.

In any case, those reads contained only the N (nucleocapsid) gene, which is just a structural protein used to pack viral RNA in the virion. No big deal. Oh, and if you think this can somehow lead to false positive PCR patient swabs, don’t: most PCR patient tests use the spike gene to establish a positive Covid diagnosis, so no, this potential integration of SARS2 N gene into DNA inside cells can in no way cause false positives in swabs from your nose.

Finally, again, if you are worried about SARS2 integrating into you genome, this is all the more reason to get vaccinated. mRNA vaccines don’t even contain the N gene (just the spike) and won’t integrate into your genome.

But don’t vaccines breed new variants?

No! On the contrary, vaccines prevent new variants from appearing. Most of the current novel variants appeared before mass vaccinations have started. So the virus clearly can create novel strains in the absence of vaccine-provided immunity. Moreover, all of the known strains are still neutralized by vaccine-mediated or natural immunity, albeit not as effectively as the original strain. Still, vaccines can protect against them.

Thinking that vaccines can make this pandemic worse is not just alarmist, it also shows a misunderstanding of the dynamic process that a pandemic is. Vaccines actually greatly reduce opportunities for the virus to develop new variants because they take away 2 main ingredients for that: new people to infect and the time to do so.

Recall that the virus infects vaccinated people at 10–20x lower rates. Basically the more vaccinated people there are, the fewer new people the virus has a chance to infect, thereby greatly reducing its effective reproductive number and ultimately extinguishing it from existence, as without reproduction it will, well, go extinct. This effect of vaccinations is orders of magnitude stronger than any potential additional selection pressure that vaccines could put on a virus.

But think of the children! How dare you risk their precious lives!

Well, children under 18 aren’t even authorized to be vaccinated in many countries yet, and are definitely the last who should be. Their Covid risk profile is highly favorable and indeed the biggest value from vaccinating some subset of the under 18 age group — probably the 12–18 subgroup — comes from further increasing herd immunity.

The biggest epidemiological issue with children is that many of them are potential silent superspreaders: asymptomatic carriers who interact with a lot of other people from diverse age groups:

Largest COVID-19 Tracing Study to Date Finds Children Are Key Spreaders

That said, if a high enough percentage of adults is vaccinated, we might be able to stamp out Covid even without vaccinating children under 12. Basically it depends on us getting the viral effective reproductive number low enough. In any case, debates about the merits of vaccinating children should have no bearing on the merits of vaccinating adults.

Moreover, the only reason not to vaccinate children is an over abundance of caution — essentially, the fear of unknown unknowns. As I said in the beginning, kids get dozens of vaccines, and they tolerate them very well. Their immune systems are in excellent shape, which can be seen from how they deal with Covid — most don’t even notice it. So tolerating a vaccine which is a subset of the virus causing Covid should be even less of a problem. Same logic applies to any potential long-term effects, as Covid’s unknown unknowns are probably likely worse than those of a vaccine.

Ivermectin will save me

Nope. Not by a long shot. Even if you believe the outcome of the severely statistically underpowered clinical trials which show IVM reducing Covid mortality, that mortality risk still remains extremely high: 1.4–3.9% depending on the trial:

Those numbers are crazy high when looked from a population level perspective, as with such a highly transmissible virus as SARS2, they imply millions more dead if we relied on IVM as an alternative to vaccinations.

Now, don’t get me wrong, if you want to use IVM as treatment, go right ahead. Or even if you think it could be an additional prophylactic measure to vaccines. But if you think IVM can protect you from Covid better than a vaccine — that is just nonsense.

I’ll just take my chances

Some young and healthy people think that a 0.1% risk of dying from Covid is negligible, so they shouldn’t even bother getting a vaccine. The problem with that line of reasoning is threefold. First, the anticipated loss in the case of that 0.1% scenario is infinite: you die. A rational person would do anything possible to lower the probability of that happening. Second, if there are many people who think like that, even having 0.1% of them die is unfortunate — they would be dying needlessly. Finally, even if 99.9% of them survive Covid, they could infect many more people whose risks of dying or developing long-term issues from Covid are much higher, and therefore the implied costs of such selfish thinking to society are many times greater than just the loss of life of the selfish 0.1%.

Also, even if you survive Covid, your subsequent risk of death is increased by 50%. And risk of developing various cardiovascular, neurological or other complications is also increased by 50% if you weren’t hospitalized or by several times if you were:

Another study in a younger cohort had similar findings:

Trust me, Covid is no picnic even when it’s mild.

But LNPs will rot my brain!

Right, those pesky brain-eating LNPs. Well, first of all, if you are really so afraid of the LNP/mRNA technology, you could just get an old-fashioned vaccine. But really, even with LNPs I don’t really see any cause for concern. First of all, they have been shown to remain well-localized at the site of an intramuscular injection:

From the above you can see that most of the injected mRNAs stay in the injected muscle and their next most popular destination is the nearest lymph node where the immune system magic happens.

Second, even though some LNPs do go on a rollercoaster ride around the body, the proportion of those that do so is very small. For example, only 1/10000 of injected mRNAs make it to the brain of mice. By the way, that was a HUGE dose those mice were injected with — 300 mcg/kg, which is way higher than the doses used in Covid mRNA vaccines: 30 or 100 mcg in Pfizer and Moderna shots, respectively, for a whole 50–100 kg human, i.e. a 150–1000x lower dose per kg than poor mice got. Which I am sure translates to an even lower percentage of LNPs going for a ride in our bodies but for the sake of an upper bound, let’s try to estimate how many brain cells we can expect to lose to zombie LNPs.

First, let’s see how many mRNA molecules are in a single vaccine shot. The vaccines encode the full viral spike which is a ~1300 amino acid protein, so an mRNA for it is ~3900 nucleotides (each aa is coded by 3 nts). The average molecular weight of a nucleotide in a long RNA chain is about 300 Da. With that, we can do the following calculation:

~1300 aa spike * 3 nt/aa * 300 Da/nt = 1.5 MDa = 1.5e6 g/mol

6e23 molecules in 1 mole, so in grams 1 copy of our mRNA weighs
(1.5e6 g/mol) / (6e23 copies/mol) = 2.5e–18 grams

Thus, 100 mcg of S mRNA (Moderna) should contain
1e–4 g / 2.5e–18 g/copy = 4e13 (40 trillion) copies, while

30 mcg (Pfizer) contains about 12 trillion mRNAs.

How many mRNAs fit inside a single LNP? This one actually turned out to be a tough question to answer, but I found this paper that used a human EPO mRNA for their LNP characterization studies and have this wonderful figure:

EPO is about 2.5 times smaller than the viral spike, so we can divide the above number by 2.5 for our mRNA vaccines. Most vaccine LNPs are reported to be about 100 nm in size, so let’s say we get 30 mRNAs per 1 Covid vaccine LNP.

This means that in a single vaccine dose we get ~0.4–1.2 trillion LNPs (between Pfizer and Moderna). Let’s just continue with 1 trillion LNPs for ease of estimation. How many LNPs do we need to get in a cell to cause significant damage? Well, a typical mammalian cell contains 360,000 mRNA molecules. Let’s say getting ~10% of that could probably overwhelm a cell and cause cell death. ~30,000 vaccine mRNAs means 1000 LNPs per cell, which means that 1T LNPs could probably kill 1B cells. Now, before you freak out, know that there are 40T cells in a human body, so 1B is just 0.0025%. Also, many LNPs probably will get intercepted by tissue-resident immune cells before they have a chance to enter muscle cells in your shoulder.

And what about brain cells? If we use the mouse data observation that 1/10000 of injected LNPs get to the brain, this would put an upper bound of ~100M LNPs that could potentially get there after an injection, and thus potentially kill 100K brain cells. Again, if that sounds like a lot, recall that we have ~100B neurons, so 100K is just 1 part per million of our total. I think I probably lost more brain cells playing beer pong.

Finally, this was just an educated guess of what the vaccine might do, but the ultimate proof in the pudding comes from clinical trials and observations of what actually happens. And no rotting brains or concerning neurological symptoms have been observed in any of the vaccine trials, in sharp contrast to Covid infections where a large percent of people develop loss of smell/taste, brain fog, anxiety and other neurological symptoms, which often persist long after the infection has been cleared.

So again, when it comes to messing up your brain, vaccines do not even come close to Covid.

Vaccines reduce severity of infections even in breakthrough cases

While vaccines can reduce your chances of getting Covid by 10–20 times, they don’t offer perfect protection. Also, you don’t get full protection right away — you actually need two shots for the full Monty, and while a single shot does provide protection, it doesn’t kick in until 2 weeks after the jab. This is why you should still play it safe and not go partying a day after the jab thinking you’re now immune to Covid.

The good news is that even a single jab provides some level of long-lasting protection, even against new variants. Here are the neutralizing antibody titers 1 year after the jab (from this preprint):

In fact, it seems that a single vaccine dose provides a similar level of neutralizing antibodies as a natural infection:

Also, as you can see from the same graph above, getting a vaccine after a prior infection greatly increases the level of neutralizing antibodies.

More good news is that even if people do get Covid after vaccinating, it seems that their average viral load is much lower (4x lower according to this preprint), and the severity of their clinical manifestations might be decreased:

Around 20% of those who were admitted to hospital for treatment had been vaccinated, but their cases weren’t serious, the Ministry of Health said Monday. Almost none of the critical and severe cases requiring intensive care had been vaccinated, and no one who has been vaccinated has died of Covid-19.

Radegonde said Thursday that only two people in the country are in intensive care.

“The conclusion is that the vaccines are protecting the people. Those who have been vaccinated are not developing any complications,” Radegonde said. “We remain confident that the vaccines — both of them — have helped the country. Things would have been worse.”

So which vaccine is best?

I like mRNA vaccines — not only do I think that mRNA LNP technology is more ‘pure’ than viral delivery vectors, the mRNA vaccines seem to have the highest safety and efficacy profiles.

For a much more detailed analysis, here is a recent Nature paper on the topic. I’ll just post this pretty straightforward graphical summary:

Basically, according to that paper, Moderna (mRNA-1273), Novavax (NVX-CoV2373), and Pfizer (BNT162b2) take the podium, while Sputnik (rAd26-S+rAd5-S) gets an honorable mention.

But I already had Covid. Should I still get a shot?

This one is a bit tricky, at least at first glance. Obviously, if you just had Covid, you should have plenty of protection against reinfection. The issue is that neutralizing antibodies don’t stick around for too long — in a matter of months their levels greatly decline. So if you catch the virus after that, it will enter your cells, and it will then be up to your T cell immunity to take the brunt of the load of eradicating the virus from your body for the second time.

Not a huge deal, but risks still exist in the case of reinfection, although reinfections do seem rare and milder than the original infections. But you could still infect others from your reinfection, continuing the virus’ march across our planet and giving it new opportunities to mutate. So getting a vaccine a few months after the original infection makes the most sense to me.

In fact, this is what I myself have done. I had Covid back in June 2020, and got a Pfizer shot in May 2021.

Oh and there’s some new data about new vairants that very nicely illustrates why you might still want to get a jab even if you had a prior infection: because a prior infection by itself (or a single vaccine dose) has very low neutralizing titers against some new variants but a jab after a prior infection has a much higher neutralizing titer:

I would also expect to see similar neutralizing titers after the second vaccine dose in the absence of prior infection.

Vaccinations should be voluntary and with full informed consent

Even though I am a strong vaccine proponent, I am absolutely against any mandatory vaccinations. That just goes against my core principles. I am also not a fan of coercing people into vaccinating. I think it should be done completely voluntarily after all the pros and cons have been examined. At the same time, I think that any reasonable person, after examining all available data, and weighing all those pros and cons, will choose to vaccinate. Because not only does that lower most adults’ chances of dying from Covid by at least 10–20x, if not 100x, but also because vaccines are the key to eradicating Covid altogether.

So. What are you waiting for?

Funding sources and Conflict of interest statement

This article was sponsored by grant FU0042069420 from the Vaccine Manufacturers Guild, grant FU200133701337 from the EcoHealth Pangolin Malayance, and grant DOS655360B from the Gates 5G Depopulation Foundation.

The author declares no conflict of interest.

Why am I such a huge fan of epigenetic clocks? Well, because in addition to being the most accurate predictors of our remaining lifespan — better than our passports or any other biomarker — to me they also provide evidence for a hypothesis I consider critical to defeating aging. The hypothesis that aging is an epigenetic program.

These days, most aging researchers agree that epigenetics play an important role in aging, and many, like David Sinclair, go so far as to recognize the role of epigenetics as critical. However, the way David formulates his Information Theory of Aging, the epigenetic changes observed during aging are seen as consequences, i.e. responses to various random or environmental factors (DNA damage, etc.):

Youth → broken DNA → genome instability → disruption of DNA packaging and gene regulation (the epigenome) → loss of cell identity → cellular senescence → disease → death (p.41).

I call that the Weak Epigenetic Theory of Aging. In contrast, I believe that, while there definitely are some epigenetic changes occurring in response to the environment, when it comes to aging, practically all of it is driven by epigenetic changes pre-programmed in our DNA — just as embryogenesis, childhood, or puberty are driven by their own pre-programmed epigenetic changes. In other words, aging is programmed, and “wear and tear” barely makes a dent in how long a given species can live. That’s my Strong Epigenetic Theory of Aging.

I see a lot of evidence in support of it. Some of the strongest evidence comes from the existence of epigenetic DNA methylation clocks in many species. What is surprising about those clocks is that with age in mammals, DNA methylation is not just stochastically diluted (or “washed out”), as one might expect if it were a random process, but more often, on the contrary, methylation increases, which usually correlates with a decrease in the expression of various necessary (or “good”) genes. Why “good” or necessary? Well, because if those genes weren’t needed in adulthood, they would have been turned off abruptly after puberty, rather than being gradually turned down as we age.

There are plenty of observations in various animals of such “good” genes having their expression levels gradually decreased with age, and new examples are constantly emerging. A bit later I will return to this topic and cite several great papers that have shown such a decrease in different species.

At the same time, an opposite process is also observed: some “bad” genes that are rather detrimental, such as harmful retrotransposons, are gradually being released from their epigenetic prison as we age. Transposons are highly repressed until adulthood but then gradually they start being more and more active, bombarding our genes with their copies. This eventually may lead to hyperactivation of innate immunity, that is, to the notorious “inflammaging” — sterile age-related inflammation, which is almost universally observed in elderly people and animals.

Why does our body allow transposons to be activated after a certain age? After all, as examples of other genes demonstrate, our organism is perfectly capable of maintaining or even increasing DNA methylation levels with age. And for precious germ cells, which are forced to reset their methylation after fertilization, the body has an additional defense mechanism against them — piRNA. So why, at the end of reproductive age, are transposons set free and allowed to wreak havoc? Is that just indifference of our genetic overlords towards their “disposable soma” which has outlived its reproductive purpose, or is that deliberate phenoptosis, i.e. are our genes trying to kill us?

It seems to me that it’s the latter. Let me explain why. In nature there is a very illustrative example of social animals: bees, ants, termites. Those animals have at least two castes: workers and breeders, and despite sharing identical DNA, breeders outlive workers by 10–100 times. Why are the “good” genes that help breeders live so long silenced in their worker twins?

Mirroring the observations of piRNA being active in our germ cells but silenced in “worker” cells, a similar observation was made in social animals: the queens enjoyed the protection of piRNA for many years, while the worker caste had its protection decreased in a matter of months:

Strikingly, genes from the PIWI-interacting RNA (piRNA) pathway, which are generally known to silence [transposable elements] in the germline of multicellular animals, were down-regulated only in old major workers but not in reproductives.

To me it seems pretty obvious that the genes just don’t allow the workers to live very long. Basically, the genes kill their workers 10–100x sooner than they do their reproductives. But this line of thinking seems blasphemous to the authors of the above observations, and they are trying to find an excuse why genes might be perpetrating this to the workers:

“It would, after all, probably be too energetically costly to maintain an active piRNA signalling pathway permanently in all the cells of an organism — or in all the individuals of a colony.”

“Too energetically costly”? I am not convinced. It’s not too costly for a queen which simultaneously lays her weight in eggs every day (20,000 eggs!), but it is costly in a worker? And what kind of energy are we talking about, food? Or was that a reference to the “trade-off” hypothesis of aging, which the queen is somehow excluded from?

An even more egregious example of genetic voluntarism in choosing how long to let their creatures live is the Monarch butterfly. In the summer, these butterflies live only a few weeks, but in the fall they migrate south for wintering, and those migrants can live up to 9 months. Moreover, these very same migrants not only have to fly for thousands of miles back and forth, but in the spring, the females that fly back do so while pregnant. Why else can’t their summer brethren live at least as long if it’s not for the genes deliberately killing them?

The Great Gladyshev Group Paper

Let us put murderous genes and transposons aside for now, and return to methylation clocks. In particular, I would like to analyze in detail an excellent work from Vadim Gladyshev’s group, in which they thoroughly analyzed what happens to methylation as we get older (or as mice do, to be precise). The authors have collected mouse methylation profiles at a whopping 16 time points — i.e. practically every 2 months of a mouse’s short life. Moreover, the profiles they collected spanned not just a few hundred methylation sites, but as many as 800,000 of which just over 20% changed significantly with age.

By the way, it is worth noting that for their paper the authors used data on age-related changes in murine methylome, obtained by their colleagues from the same lab in an earlier work. In that original work, the authors had created a novel methylation clock for mice and, more importantly, have shown that this clock is delayed by various interventions that are know to extend lifespan (caloric restriction, rapamycin, etc.). By the way, that novel Gladyshev methylation clock was praised by Steve Horvath himself, the most famous epigenetic clockmaker.

So what was so captivating for me in the Gladyshev group paper? First, the following observations:

We found 102 significantly enriched pathways (adjusted p-value < .05) associated with the loss of methylation and 1,162 pathways associated with its gain in promoters during aging. Among the pathways with decreased methylation were those related to DNA repair, immune processes, and inflammation. In the increasing group, the most overrepresented pathways were related to developmental processes. There were also significantly enriched pathways related to aging and lifespan-extending interventions, such as the response to growth factors, insulin-like growth factor and TGFβ, MAPK cascade, WNT and Notch signaling pathway, regulation of stem cells, estradiol response, and fatty acid metabolism- and transcription regulation-related pathways.

We also investigated the enrichment for genes (Table S6) and found 39 significant pathways that lost methylation and 987 pathways that gained it during aging. We observed similar patterns in gene bodies compared to promoters. Pathways with decreasing DNA methylation included DNA repair, immune function, and inflammation-related pathways, and those with increasing methylation included various developmental pathways. There were aging-related enriched pathways in the increasing group, including regulation of cell aging and senescence, growth factor response such as the response to TGFβ stimulus, stem cell proliferation and differentiation, MAPK cascade, WNT, Notch signaling, and fatty acid metabolism-related pathways (Figure S9). In addition, a pathway involving DNA methylation itself was detected, including the gene DNMT1.

That is, what do we see? That there are 10x (!) more gene pathways where methylation increases with age, which usually correlates with a decrease in gene expression, than pathways where the opposite happens. Moreover, it seems that methylation also increases in some “gears” of the methylation mechanism itself, such as DNMT1, one of the functions of which is to duplicate methyl labels during DNA replication from the original strand to the one being built. Could this be why some harmful genes like transposons gradually escape from their epigenetic prisons as the efficiency of the systems for maintaining their methylation decreases? This “escape of the transposons” seems all the more odd when considered against the background of the body having enough resources and “resolve” to increase methylation (which takes effort and energy to do) of many previously active genes. Well, at least it seems odd to those who do not believe that genes can deliberately try to slowly kill their hosts. In the framework of programmed aging it actually makes perfect sense.

But you know what’s funny? That the authors of the Gladyshev paper are staunch “stochastics”, i.e. opponents of programmed aging who instead believe that aging is caused by random accumulation of damage. Here is how one of the co-authors, Alexander Tyshkovsky, interprets what he observed (translated from Russian):

Concerning [methylation] clocks and their determinism. When we analyzed the specific changes in the methylation profile with age, we saw 2 main patterns.

The first is an increase in entropy (explains the change in methylation status with age for more than half of the sites). Sites that are fully methylated begin to demethylate with age, and vice versa. Are these changes deterministic? Yes, absolutely. Are they programmed? No, because they are explained by pure stochastics: the methylation level of fully methylated sites cannot get any higher, it can only decrease (which is what happens), the opposite is true for fully demethylated sites. In other words, their methylation status moves towards the mean from different directions as a consequence of falling control over the methylation status.

The second pattern is a drop in methylation (and an increase in activation) of genes associated with maintaining the body and repairing damage (for example, genes for DNA repair and immune response). Is this a program? Yes. But not an aging program, but a program for repairing damage that increases with age. When the car is out of order, the owner is “programmed” to take it to the mechanic. But this does not mean that there is a car breakdown program.

Not Entropy Again

Oh, this entropy. Every now and then comes up in conversations about aging. Well, yes, yes, as we age, entropy increases — and for some reason at very different rates in different species — but not because some physical law compels it to increase, but because genes allow it to and even help at some point. When genes need to reduce entropy of an organism, they do so marvelously — during embryogenesis, for example, or during puberty. Some species can even regrow severed limbs or even a head, as planaria do.

To me it is pretty clear that the reason for aging cannot be “entropy” or, more accurately, the Second Law of Thermodynamics, which states that in isolated systems entropy cannot decrease. By the way, this “cannot decrease” for some reason is often transformed into “must increase”. Moreover, the time frames in which this entropy is obliged to increase in not usually specified. In any case, there is no point in trying to discuss shades of gray when we are obviously looking at green: living organisms are not isolated systems. Quite the opposite, all they do is absorb energy from the outside and use it for various needs, including for reducing their entropy.

But let’s get back to Alexander’s thesis on entropy. On the basis of what data did he draw his conclusions? Judging by the graphs in their paper, based on these data for mice:

…and these data for humans:

Personally, I do not see a notable increase in entropy here, especially in humans. On the contrary, average entropy of methylation sites is surprisingly stable with age. Moreover, for those human sites where methylation decreases with age, the entropy actually slightly drops (bottom graph, blue dots).

By the way, the age-related decrease in the entropy of human methylation sites shown in the graphs above is in line with a very curious phenomenon: after age 75–80, the spread in epigenetics between different people, instead of monotonically increasing, begins to actually decrease. The researchers saw this in twins at first:

There is experimental evidence suggesting that epigenetic fluctuations may stop diverging in very old individuals. In twin studies, variation in global genome methylation increased gradually until 75 years of age but showed a decreasing trend in the oldest twin group (76–88 years) [17]. Likewise, methylation patterns of distant CpGs (1–5 kb) become more similar in older individuals.

And later as a general phenomenon:

Age-dependent epigenetic divergence, paradoxically, changes to convergence in the later stages of life.
…
We find that brain tissues of older individuals (>75 years) become more similar to each other, both epigenetically and transcriptionally, compared with younger individuals.

Let me go out on a limb here for a second — I think we might observe something similar in real life. The spread in health levels of people in the 50–60 age cohort is quite large: at this age there are still many people in perfect health but also many people with debilitating disease, while at age 90, almost everyone is quite decrepit. Perhaps this is what the dynamics of methylation reflect. And this is very sad, because it shows that no matter how hard we try to keep ourselves young by leading a healthy lifestyle, it seems that by a certain age we will still catch up with our less disciplined peers in terms of decrepitude. That is, our aging trajectories will inevitably converge.

By the way, it’s funny that the convergence of aging trajectories can also be traced between the rich and the poor — money can significantly increase your chances of living to 75, but not 100:

Okay, back to our mice and their entropy. The last thing I wanted to say on this topic is that if you look closely, you can see that during the overwhelming majority of murine life, the level of entropy of their methylation remains approximately unchanged — almost up to 2 years of life (and the average lifespan of a mouse is 2.5 years):

If we ignore for a second the final months of murine short life (after 22–24 months), then for the first two years entropy is practically unchanged.

Methylation with Age

Ok, let’s put entropy aside for now. Let’s take a closer look at methylation levels themselves. How accurate is the thesis that methylation changes “are explained by pure stochastics: the methylation level of fully methylated sites cannot get any higher, it can only decrease (which is what happens), the opposite is true for fully demethylated sites”? Here is a very revealing graph of methylation changes of all hundreds of thousands of sites that the authors looked at:

On the left are all sites in general, and on the right are only those that change significantly with age. What do we see? That the global methylation pattern changes very little with age: in the left graph, the blue curves (young mice) are practically indistinguishable from the red ones (old mice). And even if you look only at those sites that change significantly with age, the picture does not differ dramatically. Moreover, the main changes take place at a fairly mature age (after 20 months).

This is seen even more clearly if we look at biologically significant sites rather than all of them. For example, sites at the initial regions of genes (where methylation levels are usually inversely correlated with gene expression):

The age when the curves start to turn red is 20–22 months. As can be seen from the graph above, up to this age methylation levels at the start of genes remain practically unchanged. And afterwards it does not grow as much as in other places.

In other biologically significant zones, more pronounced dynamics are visible. For example, methylation levels of long non-coding RNAs (lncRNAs), which are an important mechanism for epigenetic “fine-tuning” of various cellular processes. Their methylation begins to increase even at the “blue” age, and varies much more noticeably with age:

In the case of harmful transposons, the dynamics are reversed; their methylation levels decrease with age:

Moreover, their methylation levels decrease at a fairly high speed and with an early, “blue” start. That is, these ancient jumping genes wake up in mice long before old age, and with each subsequent month their activity increases.

And formally — yes, methylation often falls or rises in the direction of the average value of 0.5. But! Even if a decrease in methylation can be reasonably explained by passive “dilution” during DNA copying (say, because DNMT1 could not reproduce all methyl labels with 100% accuracy), an increase in mentylation by some random process is difficult to explain. That’s because DNA methylation is an active process that requires an expenditure of energy and effort: methyl tags will not attach themselves to CpG cytosines of their own volition. To paraphrase Mayakovsky, if the tags are attached, there is someone who needs it.

Moreover, we see that in biologically significant regions — for example, in promoters or in first exons, an increase in methylation occurs much more often than a decrease. Now is a good time to recall the initial quote from the authors of this work which I cited above, that there were 10 times more increasing methylation sites than decreasing ones. This graph shows the dynamics of methylation in other functionally significant regions of the genome:

As we can see, the number of functional areas where methylation increases (regression slope > 0) overshadows areas where it decreases (slope < 0). I would like to highlight those regions, the increase in methylation of which usually correlates with a decrease in the expression of their genes — promoters, 5'UTRs, and first exons. In the graph above, their methylation increases at the fastest rate: their regression slope is 0.03. Judging by other graphs in the article, the units of this slope are the percentage of change in the methylation levels over 1 month of a mouse’s life. That is, the methylation level of each genomic region is normalized from 0% (not methylated at all) to 100% (completely methylated), and thus a value of 0.03 for the slope of the regression line means that, in 1 month, the methylation level of these elements, on average, increases by 3%, and in 30 months — by 90% (that is, growth without compounding). A negative value means methylation decreases with age.

A reasonable question is how the methylation level of CpG islands (methylation islands) can grow at a rate of 4% per month — after all, then, in 30 months of an average mouse’s life, the methylation of these areas should exceed the theoretical limit of 100%. Perhaps this is an artifact of linear regression for nonlinear data — I will assume that the methylation of these areas greatly increases in the first 20 months of mice life, and in the remaining months it practically does not change.

Although to see the rate of increase in methylation of the first exons at 3.5% per month is also very strange, because in the general graph of the initial regions of genes, we see that over the entire life of a mouse, the level of their methylation increases by no more than 20% — from 10% to 30% (or even only up to 25%, if you look closely):

In other graphs, the difference in methylation levels between 3- and 35-month-old mice rarely exceeds 30%. Therefore, I am still unsure how the authors have ended up with such high slopes for their regressions.

What happens to the sites in methylation clocks?

An analysis of 800,000 methylation sites is definitely interesting, but maybe in such a large amount of information we are missing some processes critical to aging? Let us put aside all methylation sites and for a moment focus just on those that are included in methylation clocks, that is, sites whose age-related dynamics are conserved between different individuals of the same age. After all, methylation clocks have just dozens or hundreds of sites rather than 800,000, and those sites are special because in some clocks they change alike for both rapidly dividing cells and for cells that do not divide at all with age or do so slowly.

This to me is a great property of methylation clocks: the most accurate of them (for example, the Horvath clock) show practically the same “time” in very different cells types: e.g. neurons, lymphocytes, and in dozens of other tissues:

It is rather surprising, as very different genes are active in a neuron and a T-cell; moreover, these types of cells go through a very different number of cell divisions during a lifetime of their host. And the fact that in both of them some of the same epigenetic processes are proceeding at the same speed — that’s a big surprise! Moreover, to think that these same epigenetic processes are synchronized across many tissues for a hundred years just boggles the mind. Take that, Second Law!

Do you know what is even crazier? That the human methylation clock works pretty well for chimpanzees (adjusted for the rate of aging, of course). Steve Horvath was so inspired by this that he is now working to create a universal methylation clock for all mammals. If he succeeds, then such a clock may well bring us closer to answering the favorite question of stochastics: “So where are your pro-aging genes?” Because the current answer “in the same place as the genes responsible for timing of embryogenesis and puberty”, although logically correct, from the point of view of fighting aging, is rather useless.

Ok, so what does happen to methylation clock sites as we age? Do the changes in them correspond to Alexander’s thesis about stochastic drift towards the mean? Let’s take the Horvath clock and see! Fortunately, he kindly made public not only the regression file, but also included statistics on the median methylation levels of young (under 35) and elderly cohorts (over 55). In the graph below, I plotted these curves by sorting all 353 Horvath CpG methylation sites by median methylation level in the young cohort. That is, the blue curve in the graph below is the median methylation level in young people, and the red curve is the median methylation level of the same sites in the elderly:

What do we see? First, we see a clear confirmation that the age-related change in methylation in Horvat’s clock is multidirectional: in about half of the sites, methylation increases, and in the other half, it decreases (oscillation of the red curve around the blue). Second, we see that many age-related changes are directed not towards the average methylation value (0.5), but away from it. And thirdly, we see that in the vast majority of sites, the difference in methylation levels between young and old values ​​is very small. To better see the last thesis, here is a more visual graph of the absolute difference in median methylation levels between the cohorts of young and elderly (sorted in ascending order of this difference):

As you can see, for 90% of sites, the absolute increase or decrease in methylation does not exceed 10%. To me, this graph just doesn’t fit with stochastic, random changes at all.

What genes in methylation clocks change the most with age?

By the way, I was curious, what are the genes in which the remaining 10% of sites are located — that is, those where methylation varies greatly between the young and the old? Here are the top 20 sites on each side:

Alas, this does not tell me much — well, yes, methylation decreases with age in some activator of lymphocytes, increases in some ribosomal protein, but, in general, everything is vague. Fortunately, there are much smarter people who were also curious about the answer to the same question and who did a more thorough analysis. These people are Morgan Levine and her co-authors, the creators of the PhenoAge methylation clock. In her clock, Morgan saw the following functional changes with age:

Among those with positive aging associations (overexpression among epigenetically older individuals), we observed enrichment for a number of pro-inflammatory signaling pathways.

These pathways included, but are not limited to: multiple toll-like receptor signaling pathways (7,9,3,2), regulation of inflammatory response, JAK-STAT cascade, response to lipopolysaccharide, tumor necrosis factor-mediated signaling pathway, and positive regulation of NF-kappaB transcription factor activity. Additionally, positively associated genes were also enriched for a number anti-viral response pathways — type I interferon signaling, defense response to virus, interferon-gamma-mediated signaling pathway, cellular response to interferon-alpha, etc. Other interesting GO terms enriched among positively associated genes included: response to nutrient, JAK-STAT cascade involved in growth hormone signaling pathway, multicellular organism growth, and regulation of DNA methylation.

When testing for enrichment among genes that were negatively associated with DNAm PhenoAgeAccel (decreased expression among epigenetically older persons) we observed that many were implicated in processes involving transcriptional and translational machinery, as well as damage recognition and repair.

These included: translational initiation; regulation of translational initiation; ribosomal large subunit assembly; ribosomal small subunit assembly; translational elongation; transcription initiation from RNA polymerase I promoter; transcription-coupled nucleotide-excision repair; nucleotide-excision repair, DNA incision, 5'-to lesion; nucleotide-excision repair, DNA damage recognition; DNA damage response, detection of DNA damage; and regulation of DNA damage checkpoint.

Something caught my eye here: the fact that chronic inflammation grows with age has been known for a while, but to see another confirmation of the global activation of type 1 interferon response was very interesting. After all, this ancient response is tailored to the fight against viruses (“Alert, DNA in the cytoplasm detected!”), and there is evidence (1, 2, 3) that we are talking not so much about external viruses, but about internal ones — retroviruses-transposons living in our genome, which are activated somewhere after age 45–50. Andrey Gudkov even coined a term for this group — “retrobiome”, and together with his colleagues has shown that if the activity of retroelements is suppressed with the help of reverse transcriptase inhibitors (which creates DNA copies of these retroelements in the cytoplasm), then interferon activity is significantly reduced. But transposons are a separate topic that deserves its own article.

Not all clocks are created equal

I should mention here that more than a dozen different human methylation clocks already exist, and they are rather heterogeneous. Morgan Levine et al. have produced an excellent analysis of various clocks. As you can see from their table below, some of the clocks were initially trained for correlation with passport age, others with mortality. In some, there are hundreds of methylation sites, in others — very few:

Some show a high level of synchronization between tissues (the original Horvath clock is still in the lead here), others — a low one:

It seems to me that this reflects various age-related epigenetic processes: for sure, there are some tissue-specific epigenetic aging processes, but there are also global, systemic ones. Different clocks reveal different aspects of these processes. And, of course, we are still in the early stages of our journey to unravel the biological meanings of these processes. But I am sure that eventually we will figure them out.

Turning off “good” genes with age

Returning to functional age-related changes, in the second part of the above quote from Morgan Levine, I was intrigued to once again see evidence that the activity of DNA damage recognition and repair pathways epigenetically decreases with age in humans. Incidentally, this contrasts with data in mice, where Gladyshev’s group saw the opposite trend.

But what mice and humans agree on is that with age, the activity of DNA transcription and translation mechanisms, including the construction of ribosomes themselves, epigenetically decreases. The MSU Gladyshev group had a great poster about age-related decline in ribosomal function in the 2018 conference in Kazan. Here is a quote from their abstract from that conference:

We observed an age-related decrease in the expression levels of many genes associated with translation, including genes involved in ribosomal biogenesis and mRNA recruitment. It was unexpected to see that [in the results of ribosomal profiling of cells after fixation] with age, the number of ribosomes on start codons decreased, but increased on stop codons, which corresponded to the systemic degradation of proteostasis with aging.

Corroborating the observations of the Gladyshev group, there are works of other groups showing that the activity of building ribosomes and gene translation, in general, decreases with age. Here is, for example, a work in mice, which showed an age-related decrease in the expression of not only ribosomal, but also mitochondrial genes (highlighted in red below) in mouse cardiomyocytes:

More recently, a very beautiful work on rats was published, where the authors traced age-related changes in gene expression in 4 types of tissues: liver, kidneys, muscles (gastrocnemius) and brain (hippocampus). This work also observed similar themes: activation of innate immunity genes (block 1 in the graph below) and a decrease in mitochondrial function genes (block 7):

It is interesting to observe, however, that gene expression in the hippocampus did not change that much compared to the other three tissue types. It looks like the brain, or at least the hippocampus, is aging slower than other organs. But in muscles and kidneys, apoptosis genes significantly increase their activity in old age, which can explain both age-related sarcopenia and renal failure. Here’s how the authors summarize their findings:

Most strikingly, pathways that were related to up-regulation of inflammation were the dominant theme that we observed; for example, pathways related to the innate immune response, inflammation, and cytokine signaling were up-regulated with age in the liver, gastrocnemius, and kidney (Figures 4A and 4B). Age-related up-regulation of these pathways was also seen in the hippocampus, albeit with a less-dramatic increase. Pathways linked to allograft rejection and interferon alpha and gamma responses increased strongly with age in kidney, liver, and gastrocnemius muscle and were also increased in the hippocampus (Figures 4A and 4B). The complement pathway was also up-regulated with age in all four tissues. In addition to inflammation, other pathways of interest were perturbed; the apoptosis pathway was overall up-regulated in aging liver, gastrocnemius, and kidney and to a lesser extent in hippocampus, suggesting an overall increase in cell death occurring in tissues with age.
…
The most prominent pathway down-regulated with aging was mitochondrial; oxidative phosphorylation, respiratory electron transport, and biological oxidation were all gradually down-regulated with age in the liver and kidney. Prominent down-regulation of oxidative phosphorylation and respiratory electron transport was seen in the gastrocnemius muscle. Age-related regulation of oxidative phosphorylation, respiratory electron transport, and biological oxidation pathways was not conspicuous in the hippocampus (Figures 4A and 4B). These changes were consistent with the idea that the mitochondria become less competent with age, depriving cells of critical supplies of ATP, in addition to the multitude other signals, which are mitochondrial in origin.

There is other evidence that “good” genes are turned off with age. Here is another great paper, in which the authors analyzed the methylome and transcriptome of leukocytes, and were able to find in these cells those genes whose expression correlated with changes in their methylation. As has been shown elsewhere, many of the gene cascades that change significantly with age are associated with the immune system:

To analyse the processes associated with the genes that displayed a correlation between expression and methylation levels, we performed GO term analysis and IPA for the nonagenarians. We identified 20 GO process terms (Bonferroni-corrected p-value <0.05), of which 6 (30%) were immune system associated. Numerous immune system pathways were also identified when considering GO process terms that were more loosely associated with these genes (Benjamini-Hochberg-corrected p-value <0.05), where 39 of 121 (32%) statistically significant GO process terms were immune system associated (Additional file 12). Only one GO function term (GO:0005515 Protein binding) was associated with the correlated CpGs. In addition to the immune system, pathways related to the reaction to the environment were affected. Ingenuity canonical pathway analysis revealed 15 canonical pathways (Benjamini-Hochberg-corrected p-value <0.05) (Table 6), the majority of which were directly immune system associated (Crosstalk between Dendritic Cells and Natural Killer Cells, Antigen Presentation Pathway, Fcγ Receptor-mediated Phagocytosis in Macrophages and Monocytes, T Helper Cell Differentiation) or associated with cytoskeleton remodelling and endocytosis (Integrin Signalling, Actin Cytoskeleton Signalling, Tec Kinase Signalling, Paxillin Signalling, Caveolar-mediated Endocytosis Signalling).

But most of all I liked the conclusion of the authors in bold, because it completely coincides with my understanding:

We find that certain ageing-associated immune-system impairments may be mediated via changes in DNA methylation. The results also imply that ageing-associated hypo- and hypermethylation are distinct processes: hypermethylation could be caused by programmed changes, whereas hypomethylation could be the result of environmental and stochastic processes.

The authors of this very detailed work saw something similar:

Notably, we observed an increased DNA methylation level at the BACH2 locus and decreased BACH2 expression in CD4+ T cells of the middle-aged and long-lived groups compared with the newborn group. BACH2 plays a crucial role in T-cell-mediated immune responses.
 …
This finding suggests that BACH2 down-regulation may be associated with T cell senescence, which may be controlled by DNA hypermethylation at the BACH2 locus in addition to histone modifications.

By the way, since we are talking about the role of epigenetics in hematopoiesis, I cannot but mention my other favorite paper. In it, the authors turned off the DNMT3A gene in hematopoietic stem cells of mice, that is, the gene of one of the methylases — enzymes that perform DNA methylation. And these cell lines became essentially immortal:

Here we show that Dnmt3a-null HSCs can regenerate over at least 12 transplant generations in mice, far exceeding the lifespan of normal HSCs. Molecular characterization reveals that this in vivo immortalization is associated with gradual and focal losses of DNA methylation at key regulatory regions associated with self-renewal genes.

That is, after suppressing the activity of one of the methylases, in the cells immortalized in this way, the authors saw a decrease in methylation in the regulatory regions of the genes responsible for the renewal of these very cells. A reasonable question creeps in: why, in ordinary mice, these important genes are methylated with age? These observations suggest that the degradation of hematopoietic stem cells that we observe with aging may well be caused not by external factors (wear and tear, fatigue from tough life), but by internal ones — the age-related epigenetic program.

What about other tissues? There are also interesting observations there. Here, for example, a group of German researchers saw the following epigenetic changes in intestinal tissues of mice (cells of intestinal crypts and intestinal stem cells):

We discovered hypermethylation both on genes important for intestinal homeostasis (e.g. Wnt3a, Cbx6, Pak3, Nr5a2) and on genes often epigenetically silenced in cancer (e.g. Cdk, Dkk, Sfrp gene family). These findings point out that aging-dependent CpG island hypermethylation could have a key role in both promoting intestinal loss of functionality (by silencing factors necessary to maintain intestinal structure and differentiation ability) and contributing to cancer initiation.

In summary, I believe that a rather large body of data has already been accumulated to support the idea that epigenetic changes are a driver of many (most) age-related pathologies.

Not by methylation alone

DNA methylation is a regulatory feature of gene networks in mammals and other vertebrates. But there are species in which DNA methylation either does not exist at all, or it does not play a regulatory role in epigenetics. However, these species do have histone epigenetic regulation — as, of course, do we, and ours is usually synchronized with our DNA methylation (repressed DNA methylation configurations coincide with repressed histone configurations, and vice versa):

By the way, to me the fact that all eukaryotes (including unicellular yeast) have histone regulation, but not all of them have regulatory DNA methylation, suggests that DNA methylation is a much younger evolutionary mechanism of epigenetic control.

But I digress. What I wanted to show is that even in much more “primitive” species than us or mice, there is an epigenetic switch-off of “good” genes with age. One of my favorite examples is the playfully titled nematode paper Shockingly Early: Chromatin-Mediated Loss of the Heat Shock Response. In it, the authors showed that the heat shock response, one of the main mechanisms of protein quality control, is epigenetically disabled in nematodes. Moreover, it turns off suddenly and at a fairly early age: only within 4 hours after puberty:

The heat shock response (HSR) is essential for proteostasis and cellular health. In metazoans, aging is associated with a decline in quality control, thus increasing the risk for protein conformational disease. Here, we show that in C. elegans, the HSR declines precipitously over a 4 hr period in early adulthood coincident with the onset of reproductive maturity.

Repression of the HSR occurs due to an increase in H3K27me3 marks at stress gene loci, the timing of which is determined by reduced expression of the H3K27 demethylase jmjd-3.1. This results in a repressed chromatin state that interferes with HSF-1 binding and suppresses transcription initiation in response to stress.
…
We found that the inducibility of the HSR is reduced by 60%–70% between 8 and 12 hr into the first day of adulthood and coincides with the onset of egg laying (Figures 2F and S2E). This suggests that repression of the HSR represents an actively controlled transcriptional re-tuning that marks the onset of reproductive maturity and could represent an early molecular event in the aging process. Furthermore, rapid repression of the HSR correlates with a dramatic impairment in the ability of animals to recover from acute heat shock (Figure 2G), suggesting that the transcriptional switch mid-way through the first day of adulthood has profound consequences for the organism.
…
Our data suggest that stress response pathways are not gradually and progressively dysregulated as a consequence of stochastic damage with age but are instead rapidly and precisely repressed as animals commit to reproduction.

Isn’t that obvious phenoptosis? By the way, a very similar epigenetic shutdown of proteostasis systems (maintaining the quality of proteins), only in rotifers, was observed by the authors of this work:

A decline in protein homeostasis is considered one of the hallmarks of aging across taxa [1], and our results suggest such a decline in late life in B. manjavacas. Expression of nearly all proteasome-related genes increases steadily over life until the transition to reproductive senescence, when expression of 31 out of 38 annotated structural proteasome subunit and catalytic co-factor genes decreased by up to 2.4-fold (Fig. 8); GSEA results capture the late-life decline in proteasome expression (Fig. 5).
 …
Together, these results suggest that large shifts in epigenetic markers may be partly or even largely responsible for driving the changes in gene expression over lifespan, an idea increasingly supported by results from other model systems.

Incidentally, this Fig. 5 shows very clearly how many key organismal systems decrease their activity at the end of the reproductive period:

At the same time, after the end of reproduction age, rotifers live another 20–30% of their lifespan, and nematodes may live a whopping 80% — the fertile period in nematodes lasts only 3–4 days, after which the worms remain alive for another 2–3 weeks. Why do genes begin to turn off not only important homeostasis systems, but also the reproductive system so early? After all, theoretically, from the point of view of individual selection, prolonging the ability to reproduce would give them an evolutionary advantage. Apparently, there is still selection pressure at higher fractal levels than the level of an individual, which reinforces both phenoptosis and early sterilization in such species.

Could Alzheimer’s also be epigenetically controlled?

Continuing the topic of epigenetically determined age-related degradation, I cannot fail to mention a paper, which even for such a staunch “programmer” as myself came as a surprise. In it, the authors showed that Alzheimer’s disease may be epigenetically reversed. To be honest, I thought that Alzheimer’s was caused by irreversible neuronal death, and that it can only be prevented, not reversed. Fortuitously, it seems I have been wrong.

The authors of the above work noticed that the activity of glutamate receptors is epigenetically suppressed in Alzheimer’s patients. Glutamate is the main neurotransmitter, and its receptors play an important role in memory processes. So, the authors found that this suppression occurs through one of the histone mechanisms of gene repression. The next logical step was to try to block this repressive mechanism and see what happens.

What they observed was close to a miracle: after being treated with inhibitors of the above repressive epigenetic mechanism, mice with significant memory deficits had their memory restored almost to healthy levels. Their cognitive test scores skyrocketed after the injection of a gene repression inhibitor:

Noticeably, the effect of this therapy was rather short-lived: 7 days after the administration of the drugs, the memory of the mice deteriorated back to its initial levels, which suggests that this is exactly how much time it takes for that histone repressive mechanism to restore the “bad” epigenetic profile of the activity of glutamate receptor genes. But most importantly, we saw the first signs of an absolutely startling prospect: Alzheimer’s could be epigenetically reversible.

Epigenetic rollback (rejuvenation)

Ok, I think I’ve provided enough examples of just how many bad things that happen as we age are due to epigenetic mechanisms. So let me remind that there is a silver lining in this — epigenetic changes are reversible! The most striking example of this is the epigenetic rollback accomplished via OSKM reprogramming and the accompanying functional rejuvenation of cells. Indeed, Yamanaka factors are able to rejuvenate cells even from centenary donors: for example, to restore their mitochondrial function.

Moreover, we are very fortunate that Yamanaka factors reduce epigenetic age gradually. This preprint demonstrated this beautifully for human fibroblasts — namely, that the decrease in their epigenetic age (blue curve in the graph below) is gradual, as is the decrease in the expression of functional genes that are responsible for fibroblasts remaining fibroblasts:

The discovery of this gradual property of epigenetic reprogramming allows us to hope that we will be able to find a safe therapeutic window (for example, marked by a blue rectangle in the graph above) — that is, a time period of safe epigenetic rollback, when the cell is already epigenetically rejuvenated, but has not yet lost its functional characteristics ( i.e. the skin cell remains a skin cell and does not dedifferentiate into a pluripotent cell).

By the way, I was not the only one who fell in love with the above smooth rollback graph at first sight. It is also featured in presentations of such gurus as Steve Horvath or Mike West. And yes, it’s nice to see that since I got inspired by epigenetic rejuvenation by partial reprogramming back in 2017 and founded Youthereum, several more companies have also committed to translating this paradigm into a therapy. These are Turn Biotechnologies from Stanford; AgeX spin-off REVERSE Bioengineering; as well as the famous David Sinclair with his Iduna.

Could epigenetic clocks just be a reflection of DNA damage?

Having mentioned David Sinclair, I would like to circle back to the beginning of this article and elaborate on the difference between the Weak and the Strong Epigenetic Theories of Aging.

While I share David Sinclair’s optimism about the value of partial reprogramming for epigenetic rejuvenation, I have differing understanding of the actual mechanisms of age-related epigenetic changes. In his talks and a recent preprint (co-authored with Vadim Gladyshev, by the way), David speaks of aging as a loss of epigenetic information due to the accumulation of epigenetic noise, and reprogramming as the restoration of the “correct” epigenetic settings from some kind of informational backup.

Within this paradigm, David and his supporters believe that the epigenetic clock only reflects stochastic breaks in our DNA. They believe that the methylation sites on which the epigenetic clock is built are nothing more than sites of repairs of double-strand breaks that randomly appear with age.

I disagree with this thesis for at least two reasons. First, methylation levels of CpG sites in various methylation clocks change in opposite directions with age. For example, in the Horvath clock, about half the sites increase, while the other half decrease. Which of these processes reflects DNA repair in the above thesis? Secondly, as the clock sites are the same for all humans, and their levels are highly similar at the same age, it seems naive to think that with age, DNA breaks would occur in all humans in the same spots and with about the same frequency. Moreover, in identical ways in different tissues: in the brain, skin, blood, etc. After all, as I already mentioned, CpG sites in the Horvath methylation clock are universal for 51 types of cells: both rapidly dividing and not dividing at all after birth.

In support of their thesis, David Sinclair and co-authors have recently released another preprint with a telling title “Epigenetic drift caused by DNA breaks as a cause of aging in mammals”. In this work, they created transgenic mice in which DNA breaks could be induced at authors’ choosing, and tried to show that excess DNA breaks increase epigenetic age.

In order to cause such breaks not randomly, but on demand, the authors introduced a special gene cassette with “gene scissors” into the genome of mice, which was not active by default, but was triggered if the mice were given an activator molecule. Moreover, when activated, the “scissors” cut DNA not just anywhere, but in strictly defined 20 sites (red dots in the picture below):

It seems that the authors based their key thesis that it is double-strand breaks that cause aging solely on the fact that the methylation clock in such mice showed a later time after DNA breaks (and even then, only in cell culture). The authors did not provide any data on the effect of such DNA breaks on lifespan of their mice. On the contrary, they cited other works that showed that mice with increased mutagenesis live as long as normal ones.

I was also intrigued by the following observation: although the authors said that after inducing DNA breaks in mice their epigenetic age was 1.5 times higher, upon closer examination it turns out that 1.5 times was observed according to some previously unknown methylation clock (left graph below), while when the authors used the previously validated Gladyshev’s clock (right graph), the difference in methylation age was only around 10%. At the same time, the methylation age of the control group differed between the two clocks by 2.5x (!):

Finally, further undermining their own thesis, the authors themselves admit that the sites of their induced double-strand breaks do not at all coincide with the sites of methylation of their methylation clock. This is clearly seen in the previous graph above where methylation sites (blue) and DNA break sites (red) are mapped on murine chromosomes.

In light of these facts, the proposed mechanism of the effect of DNA breaks on methylation put forward by the authors looks unjustified to me. The authors hypothesize that when a DNA break occurs, repressive complexes are irretrievably moved from methylation sites to breaks to repair them. But this does not fit well with the fact that, as I already mentioned, methylation levels of clock CpG sites changes in opposite directions with age. Moreover, the decrease in methylation with age can be explained in a simpler way: just by stochastic erosion during cell division. But how does the above DNA breaks hypothesis explain methylation increases? Recall that addition of methyl groups to CpG sites is an energy-consuming process that is unlikely to occur by chance at the same sites in all individuals of the same age.

To me, another explanation of the nature of methylation clocks and the epigenetic rejuvenation by reprogramming is much more plausible. I believe that Yamanaka factors roll back epigenetic settings to a younger state not because young settings are secretly recorded somewhere in the backup, but because the rollback of the epigenetic aging program as the initial stage of dedifferentiation is a mechanism inherently programmed in our genes. Perhaps the same mechanism is used to rejuvenate the egg after fertilization. In general, the immortal germ line has a couple more tricks up its sleeve for its rejuvenation, thanks to which an old egg cell has been turning into a new individual for millions of years without any problems — I hope that someday I will turn my two-hour lecture on this topic into a full-fledged article.

But I got distracted again. Returning to epigenetic rejuvenation by partial reprogramming, the very fact that it is possible, and occurs smoothly, tells me that the decrease in the expression of “good” genes with age is also epigenetically programmed, and does not occur randomly due to some kind of “noise”. After all, if it were a random process, we would not have any epigenetic clocks, since the variation between people would be too large, and would only increase with age. There would also be no epigenetic rollback: after all, reprogramming is an obviously smooth process, ending with a complete loss of differentiation and the transformation of a cell into a pluripotent one. Moreover, the epigenetic settings of the final pluripotent (essentially embryonic) cell are clearly different from the settings of a “young” differentiated cell of the type that it was before it was reprogrammed. That is, an old fibroblast is much closer to a young fibroblast in its epigenetics than to the pluripotent cell. So which of these “young” epigenetic settings would be recorded in a backup? And which of them are considered “young”? Before puberty? Right after? Clearly, there are more questions than answers stemming from the DNA breaks epigenetic hypothesis.

On the basis of the above, I strongly believe that aging is an epigenetic program. Luckily, the differences between the Weak and the Strong Epigenetic Theories of Aging might be more academic than practical. In practicality, many if not most aging researchers agree that, regardless of how it works, we should continue studying the wonderful phenomenon of epigenetic rejuvenation by partial rejuvenation with a goal of creating therapies that could provide humanity with significant life extension. I am positive that one day we will tame this approach to help us slow down and even reverse our epigenetic clocks. Let’s work together to help this day come sooner. Because death does not become a truly intelligent species.

References and Further Reading

In addition to the references linked to throughout the text, here are other important works that have shaped my understanding of aging mechanisms:

1. Mitteldorf, J, 2016, An Epigenetic Clock Controls Aging, Biogerontology, 17:257–265
2. Mitteldorf, J, 2017, Aging is a Group-Selected Adaptation. Theory, Evidence, and Medical Implications
3. Adiv Johnson et al, Rejuv Res 2012, The Role of DNA Methylation in Aging, Rejuvenation, and Age-Related Disease
4. Rando & Chang, Cell 2012, Aging, rejuvenation, and epigenetic reprogramming: resetting the aging clock
5. Horvath, S, and K. Raj, 2018, DNA Methylation-based Biomarkers and the Epigenetic Clock Theory of Ageing, Nature Reviews, 19:371–384
6. Galkin, F, 2020, Biohorology and biomarkers of aging: Current state-of-the-art, challenges and opportunities, Aging Research Reviews 60:1–13

If you hear anyone claim “we know the virus didn’t come from a lab”, don’t buy it — it may well have. Labs around the globe have been creating synthetic viruses like CoV2 for years. And no, its genome would not necessarily contain hallmarks of human manipulation: modern genetic engineering tools permit cutting and pasting genomic fragments without leaving a trace. It can be done quickly, too: it took a Swiss team less than a month to create a synthetic clone of CoV2.

How I Learned to Start Worrying

Oh, come on. Lab-made? Nonsense! Back in January, that was my knee-jerk reaction when ideas that Covid-19 is caused by a laboratory leak had just surfaced. Bioweapon? Well, that is just Flat Earth crazies territory. Thus, whenever I kept hearing anything about non-natural origins of SARS-CoV-2, I brushed it aside under similar sentiments. So what if there is a virology institute in Wuhan? Who knows how many of those are sprinkled throughout China.

At some point, it became necessary to brush such theories aside in a substantiated manner, as their proponents began to back up their theses about the possible artificial nature of the virus with arguments from molecular biology, and when engaging them in debate, I wanted to smash their conspiracy theories with cold, hard scientific facts. Just like that Nature paper (or so I thought).

So it was then, in pursuit of arguments against the virus’s lab-madeness, that I got infected by the virus of doubt. What was the source of my doubts? The fact that the deeper you dive into the research activities of coronavirologists over the past 15–20 years, the more you realize that creating chimeras like CoV2 was commonplace in their labs. And CoV2 is an obvious chimera (though not nesessarily a lab-made one), which is based on the ancestral bat strain RaTG13, in which the receptor binding motif (RBM) in its spike protein is replaced by the RBM from a pangolin strain, and in addition, a small but very special stretch of 4 amino acids is inserted, which creates a furin cleavage site that, as virologists have previously established, significantly expands the “repertoire” of the virus in terms of whose cells it can penetrate. Most likely, it was thanks to this new furin site that the new mutant managed to jump species from its original host to humans.

Indeed, virologists, including the leader of coronavirus research at the Wuhan Institute of Virology, Shi Zhengli, have done many similar things in the past — both replacing the RBM in one type of virus by an RBM from another, or adding a new furin site that can provide a species-specific coronavirus with an ability to start using the same receptor (e.g. ACE2) in other species. In fact, Shi Zhengli’s group was creating chimeric constructs as far back as 2007 and as recently as 2017, when they created a whole of 8 new chimeric coronaviruses with various RBMs. In 2019 such work was in full swing, as WIV was part of a $3.7 million NIH grant titled Understanding the Risk of Bat Coronavirus Emergence. Under its auspices, Shi Zhengli co-authored a 2019 paper that called for continued research into synthetic viruses and testing them in vitro and in vivo:

Currently, no clinical treatments or prevention strategies are available for any human coronavirus. Given the conserved RBDs of SARS-CoV and bat SARSr-CoVs, some anti-SARS-CoV strategies in development, such as anti-RBD antibodies or RBD-based vaccines, should be tested against bat SARSr-CoVs. Recent studies demonstrated that anti-SARS-CoV strategies worked against only WIV1 and not SHC014. In addition, little information is available on HKU3-related strains that have much wider geographical distribution and bear truncations in their RBD. Similarly, anti-S antibodies against MERS-CoV could not protect from infection with a pseudovirus bearing the bat MERSr-CoV S. Furthermore, little is known about the replication and pathogenesis of these bat viruses. Thus, future work should be focused on the biological properties of these viruses using virus isolation, reverse genetics and in vitro and in vivo infection assays. The resulting data would help the prevention and control of emerging SARS-like or MERS-like diseases in the future.

If the above quote might seem vague as to what exactly “using reverse genetics” might mean, the NIH grant itself spells it out:

Aim 3. In vitro and in vivo characterization of SARSr-CoV spillover risk, coupled with spatial and phylogenetic analyses to identify the regions and viruses of public health concern. We will use S protein sequence data, infectious clone technology, in vitro and in vivo infection experiments and analysis of receptor binding to test the hypothesis that % divergence thresholds in S protein sequences predict spillover potential.

“Infectious clone technology” stands for creating live synthetic viral clones. Considering the heights of user friendliness and automation that genetic engineering tools have attained, creating a synthetic CoV2 via the above methodology would be in reach of even a grad student.

But before delving into CoV2 origins, let’s first take a quick dive into its biology.

Biology

Ok, let’s start from the basics. What’s a furin site, an RBM, or a spike protein? Bear with me: once you wade through the jungle of terminology, conceptually, everything is pretty straightforward. For example, spike proteins are those red things sticking out of a virus particle — the very reason for which these viruses got “crowned”:

It is with the help of these proteins that the virion clings to the receptor of the victim cell (ACE2 in our case) to then penetrate inside. So it is a vitally important part of the virus, as without getting into a cell viruses cannot replicate. The spike protein also determines which animals the virus can or cannot infect, as ACE2 receptors (or other targets for other viruses) in different species can differ in structure. At the same time, out of the entire 30 kilobase genome (quite huge by viral standards), the gene of this protein makes up only 12–13%. So the spike protein is only about 1300 amino acids long. Below is how the spike (S) protein is structured in CoV2 and close relatives:

As can be seen from the figure above, the S protein consists of two subunits: S1 and S2. It is S1 that interacts with the ACE2 receptor, and the place where S1 does so is called Receptor Binding Domain (RBD), while the area of direct contact, the holy of holies, is called Receptor Binding Motif (RBM). Here is a beautiful illustration from an equally beautiful work:

When the CoV2 genome was just sequenced and made publicly available on January 10, 2020, it was a riddle, as no closely related strains were known. But quite quickly, on January 23, Shi Zhengli released a paper indicating that CoV2 is 96% identical to RaTG13, a strain which her laboratory had previously isolated from Yunnan bats in 2013. However, outside of her lab, no one knew about that strain until January 2020.

It was immediately clear that RaTG13 is special. Take a look at the figure below:

This is a genome similarity graph between CoV2 and other known strains. The higher the curve, the higher the percentage of matching nucleotides. As you can see, in the spike protein (S) gene region (between nucleotides 22k and 25k), only RaTG13 is more or less close to CoV2, while all other strains take a deep dive around this spot — both strains from other bats and the first SARS-CoV (red curve). This in itself is far from suspicious — who knows how many unknown SARS-like strains lurk in the bat caves of Yunnan? Ok, maybe it is not very clear how exactly the virus could get from there to Wuhan, but hey, with those wet markets you never know.

Pangolins

Next, pangolins appeared on the scene: in February, another group of Chinese scientists discovered a peculiar strain of pangolin coronavirus in their possession, which, while generally being only 90% similar to CoV2, in the RBM region was almost identical to it, with only a single amino acid difference (see the upper two sequences, dots indicate a match with the top sequence):

Surprisingly, in the first quarter of the S protein, the pangolin strain is highly dissimilar from CoV2, but after the RBM all three strains (CoV2, Pangolin, RaTG13) exhibit a shared high degree of similarity. Most strikingly, RaTG13’s RBM itself is quite different than that of CoV2, which can be seen from the steep dive of the green RaTG13 graph compared to the red CoV2 graph in the RBM region (pink strip) in the following graph:

This observation is confirmed by the phylogenetic analysis of the three areas highlighted in the graph above — in the RBM, the pangolin strain is closer to CoV2 than is RaTG13, but it is RaTG13 that is closer to CoV2 to the left and right of RBM. So there is obvious recombination, as the authors (and other papers) conclude.

How did the researchers obtain those pangolins? This is how:

They were confiscated from smugglers by Chinese customs and transferred to an animal rehab center in Guangdong, where they died while exhibiting severe coronavirus symptoms. This, of course, must have gotten the attention of local virologists, who took several samples:

Pangolins used in the study were confiscated by Customs and Department of Forestry of Guangdong Province in March-December 2019. They include four Chinese pangolins (Manis pentadactyla) and 25 Malayan pangolins (Manis javanica). These animals were sent to the wildlife rescue center, and were mostly inactive and sobbing, and eventually died in custody despite exhausting rescue efforts. Tissue samples were taken from the lung, lymph nodes, liver, spleen, muscle, kidney, and other tissues from pangolins that had just died for histopathological and virological examinations.

Those pangolins attracted the attention of other virologists too. For example, a team in Hong Kong also received samples of confiscated pangolins and in February 2020 they also released a paper that noted clear signs of recombination in the CoV2 spike protein:

We received frozen tissue (lungs, intestine, blood) samples that were collected from 18 Malayan pangolins (Manis javanica) during August 2017-January 2018. These pangolins were obtained during the anti-smuggling operations by Guangxi Customs. Strikingly, high-throughput sequencing of their RNA revealed the presence of coronaviruses in six (two lung, two intestine, one lung-intestine mix, one blood) of 43 samples. With the sequence read data, and by filling gaps with amplicon sequencing, we were able to obtain six full or nearly full genome sequences — denoted GX/P1E, GX/P2V, GX/P3B, GX/P4L, GX/P5E and GX/P5L — that fall into the 2019-CoV2 lineage (within the genus Betacoronavirus) in a phylogenetic analysis (Figure 1a).
 …
More notable, however, was the observation of putative recombination signals between the pangolins coronaviruses, bat coronaviruses RaTG13, and human 2019-CoV2 (Figure 1c, d). In particular, 2019-CoV2 exhibits very high sequence similarity to the Guangdong pangolin coronaviruses in the receptor-binding domain (RBD; 97.4% amino acid similarity; indicated by red arrow in Figure 1c and Figure 2a), even though it is most closely related to bat coronavirus RaTG13 in the remainder of the viral genome. Bat CoV RaTG and the human 2019-CoV2 have only 89.2% amino acid similarity in RBD. Indeed, the Guangdong pangolin coronaviruses and 2019-CoV2 possess identical amino acids at the five critical residues of the RBD, whereas RaTG13 only shares one amino acid with 2019-CoV2 (residue 442, human SARS-CoV numbering).

By the way, the authors of this article also highlighted the high phylogenetic mosaicity of the CoV2 spike protein:

Interestingly, a phylogenetic analysis of synonymous sites alone in the RBD revealed that the phylogenetic position of the Guangdong pangolin is consistent with that in the remainder of the viral genome, rather than being the closest relative of 2019-CoV2 (Figure 2b). Hence, it is possible that the amino acid similarity between the RBD of the Guangdong pangolin coronaviruses and 2019-CoV2 is due to selectively-mediated convergent evolution rather than recombination, although it is difficult to choose between these scenarios on current data.

Translated from science-speak, what this means is that if we analyze the entire RBD of the three strains, ignoring the obvious differences (i.e. non-synonymous substitutions) among them, which are mainly found in the RBM (which, recall, is identical between CoV2 and Pangolin), and construct a phylogenetic tree for synonymous substitutions, CoV2 is still closer to RaTG13 than to the pangolin strain. Which is rather strange in light of the fact that the pangolin strain and CoV2 have identical RBMs (which are segments inside RBD).

The authors go on to put forth a conjecture that this may be the result of convergent evolution, in other words, that CoV2 and the pangolin strain came to possess identical RBMs each in their own way, rather than through recombination between common ancestors. Because it would have required a rather unique recombination event — as if someone cut out a precise RBM segment from a pangolin strain and used it to replace the RBM in RaTG13. Talk about Intelligent Design!

Royal Genealogy

In order to better understand CoV2 origins, let’s take a look at spike protein sequences of our Unholy Trinity: CoV2, RaTG13 and MP789 (pangolin-2019). Let’s compare the pairwise differences between them (identical amino acids are marked with dots, red letters denote differences, and dashes indicate deleted/inserted amino acids):

The comparisons illustrate what previously quoted papers have noted: that in the first quarter of the sequence, the pangolin strain is far from CoV2 and RaTG1, and if it weren’t for the RBM region (red rectangle), RaTG13 would have been very close to CoV2. But, as I already said, the RBM in CoV2 is closest to that of the pangolin strain.

What about other pangolin strains? So far we’ve only analyzed the MP789 strain isolated from pangolins confiscated by customs in 2019. But there was another batch of pangolins confiscated in 2017, and they also had a similar coronavirus strain isolated. Let’s compare it to RaTG13 and MP789:

In the first quarter of the S protein, the 2017 pangolin strains are closer to RaTG13 (and CoV2) than their 2019 pangolin counterpart (MP789). At the same time, all three have a clear recent common ancestor in the areas marked by green rectangles, and in these areas RaTG13 and pangolin-2019 (MP789) are closer to each other than to pangolin-2017, since they have several common mutations (marked by red and blue ellipses), which are absent from pangolin-2017. But the RBM for all three is different, and different in approximately the same proportion, and in similar places.

Maybe after ancestors of RaTG13 and MP789 diverged, the MP789 ancestor had the first quarter of its protein replaced (which did not occur in RaTG13 or pangolin-2017), and the rest of the protein remained common for all three strains. Later the paths of the RaTG13 and MP789 gene pools crossed again and produced CoV2. It is also possible that the ancestor of RaTG13 arose as a result of recombination of ancestral pangolin strains.

It is also interesting to see a rather unique identical mutation (QTQTNS) in RaTG13 and pangolin-2019 right in front of the spot where CoV2 has a new furin cleavage site. That furin site, as I mentioned, arose via an insertion of 4 new amino acids (PRRA). If we look at the nucleotide sequence around this insertion, we can see that RaTG13 and CoV2 are closer to each other in that area than to pangolin-2019, since they possess several common mutations (highlighted in blue):

By the way, Orf1ab is also a phylogenetic mess in CoV2: 1a is closer to RaTG13, but 1b is closer to pangolin-2019:

Does this mean that the ancestor of CoV2 crossed with the common ancestor of pangolin-19 at least twice? First, when it (along with a common ancestor of RaTG13) inherited Orf1ab and the second half of the spike protein with the QTQTNS mutation, and second time when it acquired 1b and RBM, which differ from RaTG13. All of this is certainly possible in nature — after all, these viruses mutate and recombine constantly. Another question is where exactly bat and pangolin viruses are most likely to encounter one another for such orgies — in mountain caves, “wet markets”, shelters for confiscated animals, or even in laboratories. But let’s put those questions aside for now. First, let discuss what is arguably the most eye-catching aspect of the new virus — a 4-amino acid insertion that turned it into a natural-born killer.

A Killer Intro

It is impossible to ignore the introduction of a PRRA insert between S1 and S2: it sticks out like a splinter. This insert creates the furin cleavage site, which I mentioned at the very beginning. Let me explain what a furin site is. Remember the structure of our spike protein? Here is a detailed diagram:

The protein consists of two parts, S1 and S2, of which S1 is responsible for primary contact with the receptor (recall Receptor Binding Domain / Motif), and S2 is responsible for fusion with the cell membrane and penetration into the cell. The fusion process is started by the fusion peptide marked in yellow, but in order for it to engage in its dirty deed, someone must cut the S protein at one of the sites marked by diamonds in the diagram above. The virus does not have its own such “cutters”, so it relies on various proteases of its victims. There are several types of such proteases, as can be deduced from the abundance of colors of those diamonds. But not all proteases are equal, and not all types of cells have proteases needed by the virus. Furin is one of the most effective, and it is found not only on the surface of cells, but also inside. Most clearly, the danger of the new furin site is demonstrated by the difference between CoV2 and its grandpa, SARS-CoV:

As can be seen from the diagram, in the case of CoV2, thanks to the furin site, it is not two, but three classes of proteases (three colored PacMans) that can cut its S protein outside the cell. But perhaps the most important difference is that furin is also present inside the cell, so it can cut the S protein immediately after virion assembly, thereby providing new virions with the ability to merge with new cells right off the bat (no pun intended).

The importance of the new furin site in CoV2’s virulence was recently demonstrated by a study in hamsters where the disappearance of the furin site (due to a mutation) greatly decreased mutant CoV2’s pathogenicity and replication ability:

Infection of hamsters shows that one of the variants (Del-mut-1) which carries deletion of 10 amino acids (30 bp) does not cause the body weight loss or more severe pathological changes in the lungs that is associated with wild type virus infection.

The good news is that there already exist various furin and other protease inhibitors, and some of them (like camostat and its analogs) are already being clinically tested against CoV2.

By the way, it is possible that the new furin site could also be largely responsible for the pronounced age-dependent morbidity and mortality of CoV2:

Patients with hypertension, diabetes, coronary heart disease, cerebrovascular illness, chronic obstructive pulmonary disease, and kidney dysfunction have worse clinical outcomes when infected with SARS-CoV-2, for unknown reasons. The purpose of this review is to summarize the evidence for the existence of elevated plasmin(ogen) in COVID-19 patients with these comorbid conditions. Plasmin, and other proteases, may cleave a newly inserted furin site in the S protein of SARS-CoV-2, extracellularly, which increases its infectivity and virulence.

Furin cuts proteins in strictly defined places, namely after an RxxR sequence (that is, Arg-X-X-Arg, where X can be any amino acid). Moreover, if arginine is also in the second or third place (that is, RRxR or RxRR), then the cleavage efficiency is significantly increased.

Therefore, the appearance of a new furin cleavage site was noticed immediately, as none of the closest or even distant relatives of Cov2 have such a site — those coronaviruses that do, share only 40% of their genome with Cov2:

It was found that all Spike with a SARS-CoV-2 Spike sequence homology greater than 40% did not have a furin cleavage site (Figure 1, Table 1), including Bat-CoV RaTG13 and SARS-CoV (with sequence identity as 97.4% and 78.6%, respectively). The furin cleavage site “RRAR” in SARS-CoV-2 is unique in its family, rendering by its unique insert of “PRRA”. The furin cleavage site of SARS-CoV-2 is unlikely to have evolved from MERS, HCoV-HKU1, and so on. From the currently available sequences in databases, it is difficult for us to find the source. Perhaps there are still many evolutionary intermediate sequences waiting to be discovered.

Here is a great illustration from the source article of the quote above. Coronaviruses with a furin site are marked in pink, 3 different strains of Cov2 are shown at 10 o’clock:

The closest relative with a furin site is the HKU5 strain, isolated by the Shi Zhengli team in 2014 in Guangzhou from bats of the genus Pipistrellus (added to GenBank in 2018). But it is a very distant relative — their spike proteins share only 36%.

So the virologists are puzzled. Where did this 12 nucleotide insert come from? Could it be lab-made? Well, virologists have studied furin sites in coronaviruses for decades, and have introduced many artificial ones in a lab. For example, an American team had inserted RRSRR into the spike protein of the first SARS-CoV back in 2006:

To investigate whether proteolytic cleavage at the basic amino acid residues, were it to occur, might facilitate cell–cell fusion activity, we mutated the wild-type SARS-CoV glycoprotein to construct a prototypic furin recognition site (RRSRR) at either position.

And the Japanese have inserted a similar site (RRKR) into the SARS-CoV protein in 2008, though a bit downstream than in CoV2:

In the same year 2008, their Dutch colleagues also studied these protease sites of SARS-CoV and compared them to the murine coronavirus MHV, which also has such a site (SRRAHR | SV), one that is quite similar to the site of CoV2 (SPRRAR | SV):

In 2009, another American group also worked on “improving” SARS-CoV and, continuing the American tradition of not penny-pinching on arginines, they inserted as many as 4 of them (RRSRR):

To examine the potential use of the SARS-CoV S1–S2 and S2′ positions as sites for proteolytic cleavage, we first introduced furin cleavage recognition sites at these locations by making the following mutations 664-SLLRSTSQSI — SLLRRSRRSI-671 (S1–S2) and 792-LKPTKRSF — LKRTKRSF-799 (S2′).

Beijing 2019

But the most recent work of this kind that I came across was an October 2019 paper from several Beijing labs, where the new furin site RRKR was inserted into not just some pseudovirus, but into an actual live chicken coronavirus, infectious bronchitis virus (IBV):

An interesting side note is that, as the authors point out, the addition of a furin site allows the mutant virus to infect nerve cells. Perhaps the CoV2 furin site is the reason why some patients with CoV2 exhibit neurological symptoms, including loss of smell:

Mutation of the S2' site of QX genotype (QX-type) spike protein (S) in a recombinant virus background results in higher pathogenicity, pronounced neural symptoms and neurotropism when compared with conditions in wild-type IBV (WT-IBV) infected chickens. In this study, we present evidence suggesting that recombinant IBV with a mutant S2' site (furin-S2' site) leads to higher mortality. Infection with mutant IBV induces severe encephalitis and breaks the blood–brain barrier.
 …
In summary, our results demonstrate that the furin cleavage site upstream of the FP in S protein is an important site for CoV, modulating entry, cell–virus fusion, adaptation to its host cell, cell tropism and pathogenicity, but not antigenicity.

To be clear, many coronaviruses have naturally occurring furin sites, and they are very diverse. Obviously, they can appear as a result of random mutations. This is what happened in the case of MERS, as was pointed out in 2015 by an international team of authors, including Shi Zhengli and Ralph Baric, two stars of synthetic coronavirusology. We will come back to them many times, but for now, a few words about that article. In it the authors have shown that just two mutations allowed MERS to jump from bats to humans, and one of these mutations created a furin site. Though it was not an insertion of new amino acids, but a mutation of an existing one (marked in red on the left below):

The authors did not just show this, but actually introduced these mutations back into the original bat strain: they created the same furin site and showed that it enables the bat strain to infect human cells:

To evaluate the potential genetic changes required for HKU4 to infect human cells, we reengineered HKU4 spike, aiming to build its capacity to mediate viral entry into human cells. To this end, we introduced two single mutations, S746R and N762A, into HKU4 spike. The S746R mutation was expected to restore the hPPC motif in HKU4 spike, whereas the N762A mutation likely disrupted the potential N-linked glycosylation site in the hECP motif in HKU4 spike.
 …
We examined the capability of the mutant HKU4 spike to mediate viral entry into three types of human cells (Fig. 3A for HEK293T cells; data not shown for Huh-7 and MRC-5 cells), using a pseudovirus entry assay as previously described (14). In the absence of exogenous protease trypsin, HKU4 pseudoviruses bearing either the reengineered hPPC motif or the reengineered hECP motif were able to enter human cells, whereas HKU4 pseudoviruses bearing both of the reengineered human protease motifs entered human cells as efficiently as when activated by exogenous trypsin (Fig. 3A). In contrast, wild-type HKU4 pseudoviruses failed to enter human cells. Therefore, the reengineered hPPC and hECP motifs enabled HKU4 spike to be activated by human endogenous proteases and thereby allowed HKU4 pseudoviruses to bypass the need for exogenous proteases to enter human cells. These results reveal that HKU4 spike needs only two single mutations at the S1/S2 boundary to gain the full capacity to mediate viral entry into human cells.

By the way, how they did it might frighten those who aren’t familiar with modern biotechnology — because the authors inserted this coronavirus spike-like protein into inactivated HIV:

Briefly, MERS-CoV-spike-pseudotyped retroviruses expressing a luciferase reporter gene were prepared by cotransfecting HEK293T cells with a plasmid carrying Env-defective, luciferase-expressing HIV-1 genome (pNL4–3.luc.R-E-) and a plasmid encoding MERS-CoV spike protein.

Perhaps this is what prompted Indian researchers to look for sequences similar to HIV in the CoV2 genome (but their preprint was quickly criticized for bad methodology and erroneous conclusions). In fact, experts use such pseudoviruses regularly, and in general, one should not be scared of retroviruses as a class — their subspecies lentiviruses have been used for gene therapy for many years.

Where Did RaTG13 Come From?

RaTG13 is a very unusual strain. Odd to see that Shi Zhengli’s group was silent about it for all these years. After all, it is very different from its SARS-like siblings, especially in the spike protein, which is precisely what determines which types of cells (and in which animals) this virus can infect. Here is a genome similarity graph of CoV2 compared to other bat coronaviruses (panel B):

The red curve represents RaTG13 while the blue curve is for the strains closest to RaTG13 (ZXC21 and ZC45). These strains were isolated from Chinese horseshoe bats (Rhinolophus sinicus) in Zhoushan in 2015 (ZXC21) and 2017 (ZC45). As can be seen from the above graph, even they differ in their S proteins from RaTG13. A direct sequence comparison illustrates this difference best:

As we can see, the spike proteins of ZXC21 and ZC45 are not only 23–24 amino acid residues shorter than the RaTG13 protein, but they are shorter in the most important place — in the RBM (note the deletions in the red box marked with red dashes).

So where did RaTG13 come from? As I already mentioned, in 2020 Shi Zhengli reported that she isolated it in 2013 from Yunnan horseshoe bats (from Rhinolophus affinis, not the usual suspects R. sinicus). But until January 2020, this strain’s existence was not known, and here is how Shi Zhengli’s group described their discovery about RaTG13’s similarity to CoV2:

We then found that a short region of RNA-dependent RNA polymerase (RdRp) from a bat coronavirus (BatCoV RaTG13) — which was previously detected in Rhinolophus affinis from Yunnan province — showed high sequence identity to 2019-CoV2. We carried out full-length sequencing on this RNA sample (GISAID accession number EPI_ISL_402131). Simplot analysis showed that 2019-CoV2 was highly similar throughout the genome to RaTG13 (Fig. 1c), with an overall genome sequence identity of 96.2%.

Not much detail: previously detected, and that is that. Moreover, the quote seems to imply that until 2020, they only sequenced a part of its genome, the RdRp gene (which is part of Orf1b that precedes the spike protein gene). Ok, but where exactly in Yunnan was it obtained? The paper doesn’t mention it, and neither does GenBank. However, the GISAID entry seems to have a bit more info: collected in Pu’er City from a male bat’s fecal swab:

This rang a bell, as in my wanderings around Pubmed, I had already encountered an expedition to Pu’er in the summer of 2013:

Bats were captured from various locations in five counties of four prefectures of Yunnan Province, China, from May to July 2013.

Researchers did not report anything particularly interesting for us from that expedition, but maybe it was then that Shi Zhengli or someone from her group obtained the RaTG13 sample? Which they sequenced only partially, and for some reason decided not to publish, although it was very different from everything known before.

By the way, Shi Zhengli could well have personally participated in that expedition, as she expressed great fondness when describing them — for example, in her TED-like talk in 2018, where she showed personal photos from such expeditions:

Moreover, it was a series of exactly such expeditions that brought Shi Zhengli worldwide fame and a “Batwoman” moniker: in a 2013 Nature paper, her group triumphantly announced that in Yunnan caves they had discovered carrier bats of the RsSHC014 and Rs3367 strains that coincided with the first SARS-CoV by 85% and 96%, respectively.

It is quite a coincidence that around the same time in Yunnan, Shi Zhengli’s group also discovered RaTG13, the closest strain to CoV2, and the two also share 96% of their genomes.

UPD: Is RaTG13 the same as RaBtCoV/4991?

[UPDATED] After I had published this post, I was pointed to this preprint that alleges that RaTG13 is, in fact, RaBtCoV/4991 (KP876546), which Shi Zhengli had previously reported discovering in an abandoned mineshaft in Yunnan in 2013. There indeed are several reasons to think so. First and foremost, the only published sequence for RaBtCoV/4991 is 100% identical to that of RaTG13 at the nucleotide level, albeit being just a 370-bp stretch of the RdRp gene:

Second, the collection details of the two strains are nearly identical: both were collected in July 2013 from a fecal swab of R. affinis bats:

RaBtCoV/4991 was collected in a mineshaft located in the Mojiang county, which is under the jurisdiction of Pu’er City:

Mojiang Hani Autonomous County is an autonomous county under the jurisdiction of Pu’er City, in the south of Yunnan Province, China.
Wikipedia

And Pu’er City is listed as the collection location of RaTG13 at the GISAID database, which could well be an approximation for the Mojiang mineshaft.

It is odd that in her 2020 paper on RaTG13 Shi Zhengli fails to mention RaBtCoV/4991 or cite her 2016 paper about its discovery, for which she is listed as the one who “designed and coordinated the study”. It is not like RaBtCoV/4991 was forgotten by her group, as it is mentioned in their 2019 paper, where it is included in a phylogenetic tree of other coronaviruses:

I doubt that RaBtCoV/4991’s place in that tree was determined based solely on a 370-bp fragment, so I would think that by early 2019, Shi Zhengli’s group would have sequenced its full genome.

Intriguingly, both pangolin-2017 and pangolin-2019 genomes are also very close in this stretch of the RdRp gene, and CoV2 and pangolin-2019 share a few common mutations not found in RaTG13:

But let’s put this topic aside for now and get back to the story of Shi Zhengli’s famous 2013 Nature paper.

“Wuhan-1”

In that paper, Shi Zhengli’s group also reported that by culturing the isolated samples in monkey Vero cells, they managed to isolate a live virus that was almost identical to the Rs3367 strain. The authors named their creation WIV1 (where WIV stands for Wuhan Institute of Virology):

Most importantly, we report the first recorded isolation of a live SL-CoV (bat SL-CoV-WIV1) from bat faecal samples in Vero E6 cells, which has typical coronavirus morphology, 99.9% sequence identity to Rs3367 and uses ACE2 from humans, civets and Chinese horseshoe bats for cell entry. Preliminary in vitro testing indicates that WIV1 also has a broad species tropism.

Let’s compare RaTG13 with Rs3367 and RsSHC014:

As we can see, the spike proteins of these strains are not only 13 amino acids shorter than that of RaTG13, but they also differ in the first quarter of the protein quite substantially. By the way, it is curious that the spike proteins of Rs3367 (aka WIV1) and RsSCH014 are almost identical, and differ only in the RBD region (right sequence below). Almost like CoV2 and RaTG13 (not counting the furin insert):

Could researchers, having received coronavirus samples from pangolins that were intercepted by customs in March 2019, then want to check whether the RBM in pangolin strains can bind to the human ACE2 receptor? And could such researchers also decide to throw an extra furin site in the mix?

Theoretically, of course, they could. From a technical standpoint, it is almost routine for virologists to conduct such experiments. A reasonable question might be: why use RaTG13 as a backbone, and not, say, the tried and true WIV1? Well, it doesn’t have to be either-or: maybe a chimera with WIV1 was also tested. But in parallel, they might have decided to simulate recombination of the pangolin virus with the bat strain closest to it — after all, RaTG13 is much closer to the pangolin strains than WIV1: its spike protein is closer to them both phylogenetically and structurally — it even matches them in length, while the proteins of WIV1/Rs3367 and RsSHC014 are 13 amino acids shorter. Also, the QTQTNS mutation common to RaTG13 and pangolin-2019 (MP789) just before the protease site could not have gone unnoticed by coronavirus experts.

Other Yunnan Strains

In 2011, other researchers had also found samples of coronaviruses from the Yunnan Rhinolophus affinis. The strain LYRa11 seemed to me the most interesting:

But it is also quite distant from RaTG13, and much closer to Rs3367 (that’s the strain that shares 96% with the first SARS-CoV):

But RaTG13, isolated from the same Rhinolophus affinis bats as LYRa11, looks the least like it (left sequence comparison).

Finally, another Yunnan strain (ingenuously named Yunnan2011), isolated in 2011 from another subspecies of horseshoe bats, Rhinolophus pusillus, is even less similar to RaTG13 than LYRa11:

Between themselves, Yunnan2011 and LYRa11 (the right sequence above) are not particularly similar, apart from the highly conserved S2 region. By the way, what’s up with the differing naming conventions for these strains? Sometimes they fully spell out the year, sometimes partially, yet other times not at all (Rs3367). The carrier species sometimes leads (RaTG13), sometimes follows (LYRa11). And what do TG, LY or SHC stand for? Initials of the person sequencing the genome?

Anyways, let’s move on from viral archeology to viral engineering, namely transplanting key areas of the spike protein between species and other gain-of-function (GOF) experiments.

1999: First Chimeric Coronavirus

If you think that all of the gain-of-function coronavirus research into what exactly allows coronaviruses to jump from one species to another began in response to the first SARS outbreak in 2002, you’d be mistaken. Virologists experimented with chimeric coronaviruses long before that. Here, for example, is a 1999 paper from the Dutch group of Peter Rottier from Utrecht University with a revealing title Retargeting of Coronavirus by Substitution of the Spike Glycoprotein Ectodomain: Crossing the Host Cell Species Barrier:

Using targeted RNA recombination, we constructed a mutant of the coronavirus mouse hepatitis virus (MHV) in which the ectodomain of the spike glycoprotein (S) was replaced with the highly divergent ectodomain of the S protein of feline infectious peritonitis virus. The resulting chimeric virus, designated fMHV, acquired the ability to infect feline cells and simultaneously lost the ability to infect murine cells in tissue culture.

By the way, Shi Zhengli seems to have worked under the supervision of Peter Rottier in Utrecht for a time. At least in 2005, she co-authored a joint paper where Utrecht was listed as her affiliation (but her current address was listed at Shanghai Institute). [CORRECTION: The italicized text is incorrect, it was not Shi Zhengli but a different person]. That article itself is quite curious — in it the authors investigated what exactly allows viruses to expand their species tropism:

Only a relatively few mutations in its spike protein allow the murine coronavirus to switch from a murine-restricted tropism to an extended host range by being passaged in vitro. One such virus that we studied had acquired two putative heparan sulfate-binding sites while preserving another site in the furin-cleavage motif. The adaptation of the virus through the use of heparan sulfate as an attachment/entry receptor was demonstrated by increased heparin binding as well as by inhibition of infection through treatment of cells and the virus with heparinase and heparin, respectively.

It is interesting that the furin site in that virus (SRRAHR | SV) is similar to the site in CoV2 (SPRRAR | SV), although in CoV2 it is cut more efficiently due to dual arginines (this is what makes it a polybasic site, i.e. it has multiple basic amino acids in a row in the RxxR sequence):

But what is especially curious is that the mutations that allowed the virus to “expand its horizons” occurred not in animals, but in vitro. Moreover, it seems, they happened pretty quickly:

MHV/pi23, a virus obtained after 23 of the 600 passages that resulted in MHV/BHK, also contains a putative HS-binding site in the S1 domain at the same position as in MHV/BHK, albeit as a smaller insertion, while it lacks the putative HS-binding site immediately upstream of the fusion peptide. MHV/pi23 does infect nonmurine cells to some extent but much less efficiently than MHV/BHK. In addition to the multiple HS-binding sites, however, mutations found in other parts of the S protein, such as the HR1 domain and the putative fusion peptide (Fig. 1), might also contribute to the efficient entry into nonmurine cells. We are currently in the process of determining the S protein mutations that are required for the extended host range phenotype.

Skipping ahead, I’ll just mention that there were other groups that used in vitro mutagenesis to increase the virulence of coronaviruses, for example, MERS:

To better understand the species adaptability of MERS-CoV, we identified a suboptimal species-derived variant of DPP4 to study viral adaption. Passaging virus on cells expressing this DPP4 variant led to accumulation of mutations in the viral spike which increased replication.

Moreover, their mutations arose after just several passages (rounds of cell culture reproduction):

But those experiments occurred much later. In the meantime, let’s go back to 2002 — BEFORE the outbreak of the first SARS-CoV.

Ralph “Trailblazer” Baric

Ralph Baric is a legend in coronavirology. He is a trailblazer of synthetic genomic manipulation techniques. Back in 2002, he published a breakthrough work, which marked a milestone in both the study of various mechanisms of natural viruses and in gain-of-function research. In their paper, the Baric group described creating a synthetic clone of a natural murine coronavirus:

A novel method was developed to assemble a full-length infectious cDNA of the group II coronavirus mouse hepatitis virus strain A59 (MHV-A59). Seven contiguous cDNA clones that spanned the 31.5-kb MHV genome were isolated. The ends of the cDNAs were engineered with unique junctions and assembled with only the adjacent cDNA subclones, resulting in an intact MHV-A59 cDNA construct of ∼31.5 kb in length. The interconnecting restriction site junctions that are located at the ends of each cDNA are systematically removed during the assembly of the complete full-length cDNA product, allowing reassembly without the introduction of nucleotide changes… The method has the potential to be used to construct viral, microbial, or eukaryotic genomes approaching several million base pairs in length and used to insert restriction sites at any given nucleotide in a microbial genome.

In essense, the authors have “translated” the RNA virus into the language of DNA (using reverse transcriptase), which enabled them to manipulate its genome with the help of existing genetic engineering tools. Having created 7 such cDNA provirus segments, the authors then stitched them together “seamlessly” (i.e. without introducing any new, even silent mutations, including new restrictase sites), after which they transcribed their construct back into RNA, which was then translated into virus particles in other cells.

SARS-2003

Just a few weeks after the publication of the above work, the first SARS-CoV epidemic broke out. The Baric group sprang into action. By summer of 2003, they have submitted a paper on synthetically recreating SARS-CoV:

Using a panel of contiguous cDNAs that span the entire genome, we have assembled a full-length cDNA of the SARS-CoV Urbani strain, and have rescued molecularly cloned SARS viruses (infectious clone SARS-CoV) that contained the expected marker mutations inserted into the component clones. Recombinant viruses replicated as efficiently as WT virus and both were inhibited by treatment with the cysteine proteinase inhibitor… Availability of a SARS-CoV full-length cDNA provides a template for manipulation of the viral genome, allowing for the rapid and rational development and testing of candidate vaccines and therapeutics against this important human pathogen.

The speed of the Baric group illustrates how quickly a qualified team of virologists can create a synthetic clone from a natural virus, and therefore make genetic modifications to it. Moreover, that was back in 2003. Today, a qualified laboratory can repeat those steps in a matter of weeks.

In fact, two just did: the Swiss have created a synthetic clone of CoV2 in under a month, while it took the Galveston BSL4 lab less than 2 months to do so.

SARS-2006

Baric was the first, but far from the last. Genetic engineering developed by leaps and bounds, creating newer and better tools. Other groups explored alternative synthetic virology techniques. For example, in 2006, Spanish researchers followed in Baric’s footsteps, also creating a synthetic SARS clone, but using an alternative approach (bacterial artificial chromosome):

The engineering of a full-length infectious cDNA clone and a functional replicon of the severe acute respiratory syndrome coronavirus (SARS-CoV) Urbani strain as bacterial artificial chromosomes (BACs) is described in this study. In this system, the viral RNA was expressed in the cell nucleus under the control of the cytomegalovirus promoter and further amplified in the cytoplasm by the viral replicase. Both the infectious clone and the replicon were fully stable in Escherichia coli.
 …
The assembled SARS-CoV infectious cDNA clone was fully stable during its propagation in E. coli DH10B cells for more than 200 generations, considerably facilitating the genetic manipulation of the viral genome (data not shown). The detailed cloning strategy, plasmid maps, and sequences are available upon request.

True, they didn’t do it as elegantly as Baric, as their final assembly of the synthetic virus included their added restriction enzyme sites, while Baric learned to combine fragments “seamlessly”. But this is a minor point, the Spanish approach is just as robust — in 2013, with its help, the same authors had created a synthetic clone of MERS, and in 2015 their technique was included in a coronavirus textbook (chapter 13).

Wuhan 2007

Let’s get back to 2007. That is when the Shi Zhengli group joined the synthetic virology race with a study of the spike protein of human and bat coronaviruses, trying to determine what exactly is responsible for the ability to skip from one species to another:

A series of S chimeras was constructed by inserting different sequences of the SARS-CoV S into the SL-CoV S backbone.

That is, the authors inserted different segments from the human SARS-CoV spike protein into the spike protein of the bat virus. Here is their conclusion:

From these results, it was deduced that the region from aa 310 to 518 of BJ01-S was necessary and sufficient to convert Rp3-S into a huACE2-binding molecule.

At the same time, they tried to replace shorter fragments, including just the RBM:

For introduction of the RBM of SARS-CoV S into the SL-CoV S, the coding region from aa 424 to 494 of BJ01-S was used to replace the corresponding regions of Rp3-S, resulting in a chimeric S (CS) gene designated CS424–494.

Given that the above was written in 2007, I think today it will not be difficult for even a novice virologist to replace the RBM of one virus by an RBM from another.

Chimera-2015

In light of the above experiments, it is not very clear what caused the uproar that followed probably the most famous gain-of-function virology paper. I am referring to the joint 2015 work of Shi Zhengli and Ralph Baric, in which they created a synthetic chimeric virus:

Using the SARS-CoV reverse genetics system, we generated and characterized a chimeric virus expressing the spike of bat coronavirus SHC014 in a mouse-adapted SARS-CoV backbone. The results indicate that group 2b viruses encoding the SHC014 spike in a wild-type backbone can efficiently use multiple orthologs of the SARS receptor human angiotensin converting enzyme II (ACE2), replicate efficiently in primary human airway cells and achieve in vitro titers equivalent to epidemic strains of SARS-CoV. Additionally, in vivo experiments demonstrate replication of the chimeric virus in mouse lung with notable pathogenesis. Evaluation of available SARS-based immune-therapeutic and prophylactic modalities revealed poor efficacy; both monoclonal antibody and vaccine approaches failed to neutralize and protect from infection with CoVs using the novel spike protein. On the basis of these findings, we synthetically re-derived an infectious full-length SHC014 recombinant virus and demonstrate robust viral replication both in vitro and in vivo.

To me, the authors followed a familiar path: they took the spike-like protein from RsSHC014, which Shi Zhengli isolated from Yunnan bats in 2011, and inserted it into a murine-adapted variant of SARS-CoV for subsequent in vivo experiments. They also tested it in human cells, and almost as an aside created a recombinant clone of the same RsSHC014 strain:

The researchers also uncovered that it was not only the binding of spike protein to the receptor that determined the virus’s potential for transition from one animal species to another, because the SHC014-MA15 chimera was more virulent than SHC014 itself, even in human cells:

Notably, differential tropism in the lung as compared to that with SARS-MA15 and attenuation of full-length SHC014-CoV in [human epithelial airway cell] cultures relative to SARS-CoV Urbani suggest that factors beyond ACE2 binding — including spike processivity, receptor bio-availability or antagonism of the host immune responses — may contribute to emergence.

I especially want to highlight the spike processivity in the quote, because this is not the first time that virologists have mentioned that the ability of a spike protein to be cleaved by proteases (including furin) can have an impact on virulence.

That’s all I have to say about that paper. As a curiosity here is a common photo of its key authors, which was taken in Wuhan, in October 2018. Fittingly, Ralph Bariс and Shi Zhengli are front and center. I call this photo “The Wuhan Clan”. (Sorry, couldn’t resist).

Murine SARS-2007

One quick aside regarding the “murine virus MA15” from the above paper. That was not some kind of natural murine coronavirus, as one might think. It was a laboratory-modified human SARS-CoV, which back in 2007 the Baric group — possibly in competition with the Shi Zhengli group (remember their article from 2007) — turned into a real beast. To do this, they first iteratively “improved” it in mice, and when after several iterations it became maximally “effective”, they reproduced the observed mutations in a synthetic clone, and once again checked that it really does have increased virulence and lethality:

We adapted the SARS-CoV (Urbani strain) by serial passage in the respiratory tract of young BALB/c mice. Fifteen passages resulted in a virus (MA15) that is lethal for mice following intranasal inoculation. Lethality is preceded by rapid and high titer viral replication in lungs, viremia, and dissemination of virus to extrapulmonary sites accompanied by lymphopenia, neutrophilia, and pathological changes in the lungs. Abundant viral antigen is extensively distributed in bronchial epithelial cells and alveolar pneumocytes, and necrotic cellular debris is present in airways and alveoli, with only mild and focal pneumonitis. These observations suggest that mice infected with MA15 die from an overwhelming viral infection with extensive, virally mediated destruction of pneumocytes and ciliated epithelial cells. The MA15 virus has six coding mutations associated with adaptation and increased virulence; when introduced into a recombinant SARS-CoV, these mutations result in a highly virulent and lethal virus (rMA15), duplicating the phenotype of the biologically derived MA15 virus. Intranasal inoculation with MA15 reproduces many aspects of disease seen in severe human cases of SARS.

Baric-2008

Here is another example of the potential scientific rivalry between the Baric and Shi Zhengli groups. In 2008, the Baric group took the Bat-SCoV strain and replaced its RBD with an RBD from human SARS. That is, they essentially reproduced the work of Shi Zhengli’s group from 2007, except they didn’t limit themselves to pseudo-viruses, but created a real chimeric virus:

Here, we report the design, synthesis, and recovery of the largest synthetic replicating life form, a 29.7-kb bat severe acute respiratory syndrome (SARS)-like coronavirus (Bat-SCoV), a likely progenitor to the SARS-CoV epidemic.
 …
To test whether the RBDs of Bat-SCoV and SARS-CoV were interchangeable, we replaced the Bat-SCoV RBD (amino acid 323–505) with the SARS-CoV RBD (amino acid 319–518) (27, 28) (GenBank accession no. FJ211860), simulating a theoretical recombination event that might occur during mixed infection in vivo (Fig. 1B).

Baric-2016

The Baric group does seem to have its share of similar papers. For example, in 2016, they essentially repeated their collaboration with Shi Zhengli from 2015 to create a chimeric virus, only this time they inserted a spike protein segment into their mouse-adapted SARS not from RsSCH014, but from another strain Shi Zhengli found in Yunnan — its close relative Rs3367. Or, to be exact, from WIV1 — the laboratory clone of Rs3367 isolated at the Wuhan Institute of Virology in 2013:

Using the SARS-CoV infectious clone as a template (7), we designed and synthesized a full-length infectious clone of WIV1-CoV consisting of six plasmids that could be enzymatically cut, ligated together, and electroporated into cells to rescue replication competent progeny virions (Fig. S1A). In addition to the full-length clone, we also produced WIV1-CoV chimeric virus that replaced the SARS spike with the WIV1 spike within the mouse-adapted backbone (WIV1-MA15, Fig. S1B). … To confirm growth kinetics and replication, Vero cells were infected with SARS-CoV Urbani, WIV1-MA15, and WIV1-CoV.

To me, the 2016 paper looks a lot like the 2015 one. Moreover, its rationale is not very clear to me: after all, WIV1/Rs3367 already shared 96% of their genome with SARS-CoV. So I am not sure why one would want to insert a spike protein from its closest relative back into SARS-CoV. Maybe just because they could. In this light, the title of the article acquires a certain duality: SARS-like WIV1-CoV poised for human emergence.

Oh, and I am not sure how in 2015 Baric was granted a patent for the creation of “chimeric coronavirus spike proteins”, given all that he and Shi Zhengli previously disclosed in their papers long before 2015.

Baric-1990

Just so you appreciate how long Ralph Baric has been at this game — he was designing recombinant coronaviruses way before there were any DNA sequencing machines or other modern tools of genetic engineering. Here is his paper on the creation of “temperature mutants” from mouse coronavirus from 1990:

The A59 strain of mouse hepatitis virus (MHV-A59) was used throughout the course of this study. Virus was propagated and cloned three times in the continuous murine astrocytoma cell line (DBT).
 …
Various combinations of [temperature sensitive] mutants were mixed and inoculated onto cells at a multiplicity of infection of 10 each.

So Dr. Baric has been creating mutant viruses for over 30 years.

Wuhan-2017

The Shi Zhengli group has also not been idle since the famous 2015 paper. In 2017, they published a paper where they reported creating not one but 8 chimeric viruses — all made using transplanted RBDs from bat SARS-like viruses which they collected over a span of 5 years from the very cave around Kunming, Yunnan Province, where Shi Zhengli originally found Rs3367 and RsSCH014.

Using the reverse genetics technique we previously developed for WIV1 [23], we constructed a group of infectious bacterial artificial chromosome (BAC) clones with the backbone of WIV1 and variants of S genes from 8 different bat SARSr-CoVs. Only the infectious clones for Rs4231 and Rs7327 led to cytopathic effects in Vero E6 cells after transfection (S7 Fig). The other six strains with deletions in the RBD region, Rf4075, Rs4081, Rs4085, Rs4235, As6526 and Rp3 (S1 Fig) failed to be rescued, as no cytopathic effects was observed and viral replication cannot be detected by immunofluorescence assay in Vero E6 cells (S7 Fig). In contrast, when Vero E6 cells were respectively infected with the two successfully rescued chimeric SARSr-CoVs, WIV1-Rs4231S and WIV1-Rs7327S, and the newly isolated Rs4874, efficient virus replication was detected in all infections (Fig 7).

The authors then checked if their chimeras can infect human cells, and this time they used a live synthetic virus, rather than not pseudo-typed HIV constructs as before:

To assess whether the three novel SARSr-CoVs can use human ACE2 as a cellular entry receptor, we conducted virus infectivity studies using HeLa cells with or without the expression of human ACE2. All viruses replicated efficiently in the human ACE2-expressing cells. The results were further confirmed by quantification of viral RNA using real-time RT-PCR (Fig 8).

Baric-2019

Ralph Baric also showed no signs of slowing down. At the end of October 2019, his group submitted for publication another paper on the importance of spike protein protease cleavage (remember the furin site?) to crossing the “barrier to zoonotic infection” by coronaviruses:

Together, these results demonstrate that protease cleavage is also the primary barrier to infection of Vero cells with HKU5-CoV. Examining further, we compared the predicted cleavage at S1/S2 border, S2’, and the endosomal cysteine protease site across MERS, PDF2180, and HKU5 spikes (Fig. 6D) (26). For the S1/S2 site, MERS, Uganda, and HKU5 maintain the RXXR cleavage motif, although the different interior amino acids may alter efficiency. For the S2’ sequence, MERS and HKU5 also retain the RXXR motif; however, the Uganda spike lacks the first arginine (SNAR), potentially impacting cleavage.

As I recall the spirit of scientific competition between the groups of Baric and Shi Zhengli, I can’t help but wonder whether someone was conducting similar research in the Wuhan lab in 2019.

Gain-of-Function: Risky Business

Many people who first learn about the above research ask a very valid question: “But why?” Why do scientists create chimeric killer viruses? The politically correct answer is to develop preventive measures (drugs or vaccines) from possible natural chimeras and to understand the risks of their occurrence. Here, in fact, is what Baric, Shi Zhengli, and co-authors themselves wrote on this subject in their famous 2015 paper:

In addition to offering preparation against future emerging viruses, this approach must be considered in the context of the US government–mandated pause on gain-of-function (GOF) studies. On the basis of previous models of emergence (Fig. 4a,b), the creation of chimeric viruses such as SHC014-MA15 was not expected to increase pathogenicity. Although SHC014-MA15 is attenuated relative to its parental mouse-adapted SARS-CoV, similar studies examining the pathogenicity of CoVs with the wild-type Urbani spike within the MA15 backbone showed no weight loss in mice and reduced viral replication. Thus, relative to the Urbani spike–MA15 CoV, SHC014-MA15 shows a gain in pathogenesis (Fig. 1). On the basis of these findings, scientific review panels may deem similar studies building chimeric viruses based on circulating strains too risky to pursue, as increased pathogenicity in mammalian models cannot be excluded. Coupled with restrictions on mouse-adapted strains and the development of monoclonal antibodies using escape mutants, research into CoV emergence and therapeutic efficacy may be severely limited moving forward. Together, these data and restrictions represent a crossroads of GOF research concerns; the potential to prepare for and mitigate future outbreaks must be weighed against the risk of creating more dangerous pathogens. In developing policies moving forward, it is important to consider the value of the data generated by these studies and whether these types of chimeric virus studies warrant further investigation versus the inherent risks involved.

Were these words prophetic? At the end of 2014, the United States introduced a moratorium on state financing of such gain-of-function studies, but it was shortly canceled (in 2017). In China, no moratorium on such studies was introduced, on the contrary, they went full steam ahead with creating new “super labs” of the highest biosafety level (BSL-4), as in 2017 in Wuhan:

To be clear, the Wuhan lab was allowed to work with coronaviruses even before 2017, as these viruses only required a BSL-3 rating which the Wuhan Institute of Virology had. But their aspirations to obtain BSL-4 made a lot of people uneasy, including fellow researchers:

Future plans include studying the pathogen that causes SARS, which also doesn’t require a BSL-4 lab, before moving on to Ebola and the West African Lassa virus, which do. Some one million Chinese people work in Africa; the country needs to be ready for any eventuality, says Yuan. “Viruses don’t know borders.”
 …
The plan to expand into a network heightens such concerns. One BSL-4 lab in Harbin is already awaiting accreditation; the next two are expected to be in Beijing and Kunming, the latter focused on using monkey models to study disease.

Lina says that China’s size justifies this scale, and that the opportunity to combine BSL-4 research with an abundance of research monkeys — Chinese researchers face less red tape than those in the West when it comes to research on primates — could be powerful. “If you want to test vaccines or antivirals, you need a non-human primate model,” says Lina.

But Ebright is not convinced of the need for more than one BSL-4 lab in mainland China. He suspects that the expansion there is a reaction to the networks in the United States and Europe, which he says are also unwarranted. He adds that governments will assume that such excess capacity is for the potential development of bioweapons.

“These facilities are inherently dual use,” he says. The prospect of ramping up opportunities to inject monkeys with pathogens also worries, rather than excites, him: “They can run, they can scratch, they can bite.”

Trevan says China’s investment in a BSL-4 lab may, above all, be a way to prove to the world that the nation is competitive. “It is a big status symbol in biology,” he says, “whether it’s a need or not.”

Interestingly, in addition to Wuhan, the Chinese government planned to open a new BSL-4 lab in Kunming, with an eye to testing vaccines on primates. As you might recall, Kunming is not only the capital of Yunnan, but it is also where Shi Zhengli found the strains Rs3367 and RsSHC014 in nearby caves. By the way, primate testing was mentioned by Baric and Shi Zhengli as possible next steps for the development of preventive vaccines against potential future outbreaks of coronaviruses in their famous 2015 paper:

However, further testing in nonhuman primates is required to translate these finding into pathogenic potential in humans. Importantly, the failure of available therapeutics defines a critical need for further study and for the development of treatments. With this knowledge, surveillance programs, diagnostic reagents and effective treatments can be produced that are protective against the emergence of group 2b–specific CoVs, such as SHC014, and these can be applied to other CoV branches that maintain similarly heterogeneous pools.

Maybe by 2019 the creation and testing of potential vaccines against various SARS-like coronaviruses was already in full swing.

Beware of Lab

Let’s now take a look at the lab leak hypothesis. But first, I will provide some historical context, including previous confirmed lab leaks, as many of those happened before. First and foremost, lab leaks of the first SARS-CoV: initially, in the summer of 2003 in Singapore, then in December 2003 in Taiwan, and in the spring of 2004 twice in Beijing.

There were close calls in Europe and the USA, although thankfully no infections occurred there. For example, a French lab once lost vials with SARS, and an American BSL-4 laboratory in Texas, lost a vial containing Guanarito (Venezuelan hemorrhagic fever virus):

Only one scientist worked with the virus, and Reyes said the lab suspects that scientist accidentally threw the vial away in November.
 …
Galveston biolab requires the most stringent safety measures because it studies biosafetly level BSL-4 materials, or dangerous infectious diseases that have no vaccines or cures. BSL-4 materials include Guanarito, Ebola and smallpox.

History knows other, much larger-scale leaks. For example, the “resurrection” of the H1N1 flu virus in 1977, which had previously been considered extinct. Yes, this is the virus of the famous “Spanish flu”:

Human influenza H1N1 viruses appeared with the 1918 pandemic, and persisted, slowing accumulating small changes in its genome (with a major change in 1947), until the H2N2 “Asian” flu appeared in 1957, causing a worldwide pandemic. H1N1 influenza virus then apparently became extinct, and was not isolated for 20 years. In 1969 the “Hong Kong” H3N2 virus replaced the H2N2 virus, and is still circulating.

In September 1977 an H1N1 influenza virus was isolated from human infections in the Far East region of the Soviet Union, and in early 1978 the Chinese reported they had isolated H1N1 virus in May of 1977 in northeast China adjacent to the Soviet outbreak. Using the early genetic tools available at the time, the 1977 H1N1 virus was found to be closely related to H1N1 human influenza viruses circulating in 1949–1950, but not to those circulating earlier or later.
 …
Only since 2009–2010 did major papers begin to state directly the 1977 emergence of H1N1 influenza was a laboratory related release: “The most famous case of a released laboratory strain is the re-emergent H1N1 influenza A virus which was first observed in China in May of 1977 and in Russia shortly thereafter.”
 …
The speculation that the 1977 release may have been related to H1N1 vaccine research is supported by the observation that in the initial outbreaks in China, nine of the ten viral isolates expressed “temperature sensitivity” (Kung 1978). Temperature sensitivity normally an uncommon trait, but one that was in the 1970s (and still is) a fundamental trait for making live attenuated influenza vaccines. Temperature sensitivity generally occurs only after a series of substantial laboratory manipulations and selections.

Interestingly, further investigation indicated the circulating strains in 1977–78 were often comprised of mixed temperature-sensitive and normal components, and that temperature sensitivity apparently disappeared from the post-1978 H1N1 lineage rapidly. Escape of a mid-protocol population of H1N1 virus undergoing laboratory selection for temperature sensitive mutants would provide such a mixed population. In 1976–77 laboratory personnel in their late teens or early 20s would not have been exposed to pre-1957 H1N1 influenza viruses, and been susceptible to laboratory infections. The low severity of the 1977 pandemic might be in part due to the temperature sensitivity of the virus, a trait that limits virus replication in pulmonary tissues.

It seems that the creation of temperature-sensitive viral mutants to develop potentially attenuated vaccines was widespread at the end of the twentieth century. If you remember, in 1990, Ralph Baric himself also experimented with the creation of temperature-sensitive coronavirus strains.

Could something like this have caused the Covid-19 pandemic? Several options are possible — from a leak during development of a potential vaccine to fundamental research on laboratory recombination of the bat and pangolin viruses. Some particularly ambitious researcher could even decide to combine the two “fashionable research themes” — adding a furin site and transplanting RBM from a strain of one species (pangolin) to another (bats), so that later, confirming the increased virulence of the new chimeric virus, they can wax poetic about the dangers of the same recombination happening in Yunnan caves or wet markets. And if such a researcher could even pre-emptively develop a vaccine against this and other potential chimeras, all sorts of accolades could await.

Am I then saying this is what happened? Of course not, I do not claim to know what happened. Today, there is no evidence of this. For now, there is just a series of strange coincidences — for example, that the outbreak of the Yunnan coronavirus occurred thousands of kilometers from Yunnan in a wet market closest to the Wuhan Institute of Virology. Or maybe not at the wet market, as 3 of the first 4 patients had no ties to the market. Plus, there are coincidences in the structural features of the CoV2 genome, which resemble manipulations that virologists have repeatedly carried out in the lab. But coincidence is not proof.

Moreover, coincidences happen, and CoV2 could obviously have arisen naturally. It is not yet clear exactly how — for this, the bat and pangolin strains must have met in the same cell of some animal in Wuhan, since the outbreak occurred there (otherwise we would have seen other outbreaks along the path that animal would have taken to get to Wuhan). Given that bats were not sold in the Wuhan market, and generally hibernate at this time of the year, and that no other carriers of ancestral strains have yet been identified, the exact scenario of natural emergence remains a mystery.

On the opposite side of the balance, giving credence to the lab hypothesis, there are reports that in 2018, American experts were quite alarmed after their visit to the Wuhan Institute of Virology and conversation with Shi Zhengli. Their “lab tour” resulted in two diplomatic dispatches to Washington in which they noted a number of safety weaknesses:

Sources familiar with the cables said they were meant to sound an alarm about the grave safety concerns at the WIV lab, especially regarding its work with bat coronaviruses. The embassy officials were calling for more U.S. attention to this lab and more support for it, to help it fix its problems.
 …
“During interactions with scientists at the WIV laboratory, they noted the new lab has a serious shortage of appropriately trained technicians and investigators needed to safely operate this high-containment laboratory,” states the Jan. 19, 2018, cable, which was drafted by two officials from the embassy’s environment, science and health sections who met with the WIV scientists. (The State Department declined to comment on this and other details of the story.)

The Chinese researchers at WIV were receiving assistance from the Galveston National Laboratory at the University of Texas Medical Branch and other U.S. organizations, but the Chinese requested additional help. The cables argued that the United States should give the Wuhan lab further support, mainly because its research on bat coronaviruses was important but also dangerous.

It is somewhat ironic the Wuhan lab received guidance from the Texas laboratory in Galveston, which at one time had itself lost a vial with a Guanarito virus: Wuhan specialists were trained at Galveston, which was even reported in the Wuhan Institute’s own newsletter (though, that publication has been deleted from the WIV website, but it is still available at the Wayback Machine):

A couple of final touches to the family portrait of laboratory leaks: in November 2019, an outbreak of brucellosis (a bacterial infection) occurred in two research centers in Lanzhou, China, infecting over 100 researchers who worked there. American labs have also not been immune to outbreaks, although not on the same scale:

Inside America's secretive biolabs

Possible Hallmarks of Lab Origin?

Let us now turn our attention back to the virus itself. Does it have any obvious signs of lab manipulation? First, a few words about what “obvious” means. Any mutation can arise naturally, and even if the amino acid insert that had created the furin site in CoV2 was not “PRRA” but “MADEINWVHANPRRA”, there would still be a non-zero chance that it arose by accident. But for us, and for any court, I think this would be enough to prove lab origin beyond a reasonable doubt.

The main problem with such evidence is that even in a lab-made virus it simply may not exist. Basically, a good genetic engineer can create a synthetic virus that would be indistinguishable from a natural one. Moreover, often researchers deliberately introduce some synonymous mutations into their designs so that later they can discern their strain from natural ones. But if the creators choose not to reveal these markers, it is impossible to distinguish them from natural mutations.

But sometimes traces of manipulation may remain, especially if the creators do not try to hide them. First of all, I am talking about the spots in virus genome where its DNA is cut (recall that RNA virus manipulations are carried out in complementary DNA constructs). This occurs when virus creators need cut out a segment, or stitch together new segments. After all, DNA cannot be cut in arbitrary places (CRISPR aside), but only where its nucleotide sequence (usually 4–6 “letters”) forms a sequence recognized by some restriction enzyme, that is, an enzyme that can cut a nucleotide chain. However, such an analysis is complicated by the fact that there are hundreds of different types of restriction enzymes used in genetic engineering. But let’s try it anyways.

As a baseline, here is an example of the work of the Baric group from 2008, where they took Bat-SCoV and replaced its RBD by an RBD from human SARS. Here’s how they describe the creation of their chimera:

As you can see, the Baric group first created a synthetic clone of Bat-SCoV in the same pattern as they used for their synthetic clone of SARS-CoV. That is, for the bat clone, they used the same 6 segments with the same restriction enzyme sites that they had previously used for SARS-CoV, which allowed them to swap virus segments between different strains like Lego pieces. Here is their detailed description:

Viruses containing PCR-generated insertions within the viral coding sequence were produced by using the SARS-CoV assembly strategy (24, 33, 53) with the following modifications. Briefly, for Bat-F virus, full-length cDNA was constructed by ligating restriction products from SARS-CoV fragments A–E and Bat-SCoV fragment F, which required a BglI-NotI digestion. For Bat-SCoV and Bat-SRBD, Bat-SRBM, and Bat-Hinge, plasmids containing the 7 cDNA fragments of the Bat-SCoV genome were digested by using BglI for Bat-A, Bat-B, Bat-C, and Bat-D, BglI and AflII for Bat-E1 and Bat-E2, and BglI and NotI for Bat-F. Digested, gel-purified fragments were simultaneously ligated together. Transcription was driven by using a T7 mMessage mMachine kit (Ambion), and RNA was electroporated into Vero cells (24, 53).

All these three-letter abbreviations (BglI, AflII, NotI, etc.) in the sentence highlighted above are different types of restriction enzymes. Let’s see if there are any differences in the restriction enzyme sites in the spike protein sequence of the chimera compared to the genome of the original SARS-CoV:

As can be seen, the restriction enzyme sites of the chimera are almost identical to those in the original sequences in Bat-SCoV or SARS from where they were taken. The only differences are noticeable at the “stitching” sites of the inserted SARS piece. Here, for example, is the left (5’-) edge of the insert:

Here Bat-SCoV and SARS turned out to have a common identical region of nucleotides (the intersection of cyan and pink regions), and there are no new restriction enzyme sites at the stitching site of the two sequences, on the contrary, the SspI site from SARS disappeared. And here is the right (3’-) edge of the insert:

Here, on the contrary, all the original restriction enzyme sites remain at the site of ligation, and even new ones appear, for example, EcoRII. Had I not known that the chimeric genome is the result of lab manipulations, could I deduce this by looking at these 3 sequences? Not really, and even if some suspicion did creep in, it would certainly not be beyond a reasonable doubt. Perhaps it would be obvious to specialists in genetic engineering by some other signs, and, if so, I hope they speak up.

But in any case, let’s compare the RaTG13spike protein to CoV2 and pangolin-2019. Just in case something does jump out.

This is what the RBD (highlighted in light green) and RBM (yellow) look like for all three:

So, anything interesting? Well, I noticed some new restriction enzyme sites in CoV2 marked by red rectangles — they coincide with unique mutations in the amino acid sequence (also marked by red rectangles in the amino acid sequences on the far right). Just in case, I highlighted several other new sites: blue rectangles, and a green rectangle located in the region of the only amino acid that differs between RBMs of CoV2 and pangolin-2019.

Let’s now compare the stretch around the PRRA insert that created the furin site in CoV2 among those three strains:

Here, too, several new restriction sites have appeared (highlighted in blue) on both sides of the new insert. Could they have been used to create a furin site? Theoretically, yes. Alternatively, the insertion could have been made via existing sites or even using the “seamless” ligation method — i.e. by creation of segments with new restriction sites which disappear after the complementary ends are joined. You might remember that the Baric group have applied this technology in 2002 to create a synthetic clone of murine coronavirus:

The interconnecting restriction site junctions that are located at the ends of each cDNA are systematically removed during the assembly of the complete full-length cDNA product, allowing reassembly without the introduction of nucleotide changes.

In 2003 they have used this approach again for a synthetic clone of SARS-CoV:

To rapidly assemble consensus clones, we used class IIS restriction endonucleases that cut at asymmetric sites and leave asymmetric ends. These enzymes generate strand-specific unique overhangs that allow the seamless ligation of two cDNAs with the concomitant loss of the restriction site.

Today, genetic manipulation techniques are so advanced and have become so routine that the October 2019 Beijing paper which had inserted a new furin site into the chicken coronavirus, only devoted a couple of sentences to their methodology:

2.2. Generation of Recombinant Virus
Recombinant rYN-S2/RRKR virus containing an S protein with the furin-S2′ site was generated by vaccinia recombination, as described previously [20,28]. Briefly, plasmid with the furin-S2′ site was generated using the Seamless Assembly kit (Invitrogen, Carlsbad, CA, USA) and transfected into CV-1 cells infected by vaccinia virus containing the genome of YN-ΔS-GPT. Furin-S2’ site was introduced into the YN cDNA by homologous recombination using the transient dominant selection system [25].

The pace of progress in genetic engineering is astounding. Here is a description of the above Seamless Assembly kit:

The GeneArt® Seamless Cloning and Assembly Kit enables the simultaneous and directional cloning of 1 to 4 PCR fragments, consisting of any sequence, into any linearized vector, in a single 30-minute room temperature reaction. The kit contains everything required for the assembly of DNA fragments, and their transformation into E. coli for selection and growth of recombinant vectors.

• Speed and Ease — Clone up to 4 DNA fragments, with sequence of your choice, simultaneously in a single vector (up to 13 Kb); no restriction digestion, ligation or recombination sites required
• Precision and Efficiency — Designed to let you clone what you want, where you want, in the orientation you want, and achieve up to 90% correct clones with no extra sequences left behind
• Vector Flexibility — Use our linear vector or a vector of your choice
• Free Tools — Design DNA oligos and more with our free web-based interface that walks you step-by-step through your project
• Diverse Applications — Streamline many synthetic biology and molecular biology techniques through the rapid combination, addition, deletion, or exchange of DNA segments

Up to 4 DNA fragments can be joined in a desired orientation in about half an hour, without having to deal with restriction enzymes or ligation. Once you’re done, quickly “upload” your creation into E. coli to propagate the resulting design. Easy-peasy!

In summary, the restriction enzyme site analysis did not yield anything conclusive. It did, however, point out that not only CoV2 is quite unique, but so is RaTG13, and we should continue digging into the origins of both.

Codon Preferences

For these purposes, I decided to take a look at codon usage bias to check which strains look like CoV2 and RaTG13 the most. It is known that viruses tend to adapt their codon signature to the preferences of their hosts, so I expected to see RaTG13 exhibit a similar pattern to other bat viruses, and also hoped to see a difference from pangolin strains.

SARS-CoV, for example, is very similar to Rs3367 and RsSCH014, as one might expect:

Among themselves, by the way, SARS, MERS and CoV2 do differ:

RaTG13 is similar to CoV2, which is also to be expected:

But RaTG13 is actually not that close to the pangolin strains, and the pangolin strains are not exactly identical to each other:

RaTG13 also differs from ZXC21 and ZC45:

Looking at Yunnan strains, RaTG13 is quite distant from Rs3367 and RsSCH014, and closer to LYRa11, but also with noticeable differences:

In general, as before, RaTG13 and CoV2 stand out in a class of their own. I was also intrigued by the AAA codon — they use it much more often than their fellow strains:

This is probably just another coincidence, but a similar proportion between AAA and AAG is observed in E. coli. Can the cDNA codon signature change if it is being cultivated for a long time in cell culture? Maybe, but I haven’t yet dug into this topic very deeply.

[UPDATED] I also decided to check codon usage patterns between RaTG13 and other Ra strains collected from the same abandoned mineshaft in Mojiang where in 2013 Shi Zhengli’s group found strain RaBtCoV/4991 (KP876546) that shares an identical 370-bp RdRp segment with RaTG13. Unfortunately, only 816-bp segments of the RdRp gene were available for the other Ra strains (RaBtCoV/3750 and RaBtCoV/4307–2), so I extracted the corresponding 816-bp segment from RaTG13 for the purposes of codon usage comparison. RaTG13 again differed substantially, while the other two strains clustered together:

So codon analysis also did not reveal any obvious signs of lab origins, but once again confirmed the uniqueness of CoV2 and RaTG13. What does this leave us with? So far, just a number of oddities, which, as scientists like to say, taken together, do not allow us to reject the lab origin hypothesis of CoV2.

The Nature Paper vs. the Lab-Made Hypothesis

But didn’t that Nature article refute the lab-made hypothesis? No, not really. There is no irrefutable evidence against it in the paper, just a loud “we don’t believe so” based on a shaky foundation. Judge for yourself — here are the authors’ key arguments in support of their conclusions:

While the analyses above suggest that SARS-CoV-2 may bind human ACE2 with high affinity, computational analyses predict that the interaction is not ideal and that the RBD sequence is different from those shown in SARS-CoV to be optimal for receptor binding. Thus, the high-affinity binding of the SARS-CoV-2 spike protein to human ACE2 is most likely the result of natural selection on a human or human-like ACE2 that permits another optimal binding solution to arise. This is strong evidence that SARS-CoV-2 is not the product of purposeful manipulation.

In the original paper, the quoted sentences are just below the diagram showing identical RBMs between CoV2 and pangolin-2019. So I am puzzled as to what “computational analysis” has to do with anything. Obviously, the most likely scenario for the lab-made hypothesis is the transfer of RBM from one strain to another — which virologists have done many times before. Therefore, the author’s chain of arguments does not make sense: “computer says binding is not ideal, thus CoV2 must be the result of natural selection. Ergo, this is strong evidence that CoV2 is not lab-made.” Wait, just because CoV2 differs from some “optimal” virus, doesn’t mean it could not have been created in a lab. Not the lab trying to create “optimal” bioweapons, but a lab creating chimeras of naturally found strains, say, in bats and pangolins.

The authors continue to surprise:

Furthermore, if genetic manipulation had been performed, one of the several reverse-genetic systems available for betacoronaviruses would probably have been used. However, the genetic data irrefutably show that SARS-CoV-2 is not derived from any previously used virus backbone.

Again, the same questionable logic dressed in categorical adjectives: “genetic analysis irrefutably proves that CoV2 was not created on the basis of previously known strains!” Well thanks, Captain Obvious. But why couldn’t potential creators of CoV2 make a cDNA backbone from unpublished strains related to or even derived from RaTG13? Then they could easily insert the pangolin RBM into it, as well as add a furin site (or maybe the cDNA backbone already had one). Virologists have been doing things like this for 20 years, and modern genetic engineering tools make such manipulations accessible even to a grad student.

As for the chances of the furin site arising in cell culture, the authors also express strange ideas:

The acquisition of both the polybasic cleavage site and predicted O-linked glycans also argues against culture-based scenarios. New polybasic cleavage sites have been observed only after prolonged passage of low-pathogenicity avian influenza virus in vitro or in vivo. Furthermore, a hypothetical generation of SARS-CoV-2 by cell culture or animal passage would have required prior isolation of a progenitor virus with very high genetic similarity, which has not been described. Subsequent generation of a polybasic cleavage site would have then required repeated passage in cell culture or animals with ACE2 receptors similar to those of humans, but such work has also not previously been described.

First off, the authors themselves cite previous works where the furin site arose in vitro as viruses were cultured in cells. And second, what do they mean, a strain with high genetic similarity has not been described — what about RaTG13? If it had its RBM replaced by one from the pangolin strain, and then the chimeric strain was cultured in vitro, then the furin site could well have arisen in this matter. Additionally, the new strain could thus acquire other mutations that distinguish CoV2 from RaTG13 and pangolin-2019.

But in terms of the potential lab-based origin of the furin site, I am more inclined to hypothesize a specific insertion — as in the Beijing paper from October 2019 with chicken coronavirus. After that, the synthetic strain could have acquired new mutations by subsequent culturing in vitro or in vivo — like the MA15 murine strain in 2007, for example. Or maybe even using the same mouse model with humanized lung tissues and immune system that was created at UNC by Baric’s and other groups in 2018, in which they reported testing several viruses including MERS:

The human innate and adaptive immune system of BLT-L mice
We generated an in vivo model with human lung implants and an autologous human immune system by constructing BLT mice with autologous human lung implants (BLT-L humanized mice).

Finally, even if CoV2 is the product of selection rather than intelligent design, that does not rule out a lab leak either — selection can happen in the lab just as well, both natural and artificial kinds. Different strains can recombine in research animals or in vitro by design or by chance.

On the 4% Genome Difference between RaTG13 and Cov2

Some critics of the lab-made hypothesis claim that the observed ~4% genetic difference between RaTG13 and CoV2 is too high to have possibly occurred in a lab if RaTG13 itself was used as a backbone. Observed mutation rates for RNA viruses vary widely — from 10⁻⁶ to 10⁻⁴ nucleotides per replication in vitro, and in humans CoV2 seems to mutate at a rate of 25 mutations per year. Thus, the logic goes, it would take years, if not decades, for two strains to diverge by 4%. While that is a valid point, there are several issues with that line of reasoning.

First, in vitro mutation speeds (i.e. per unit of time) are much higher, as you can passage cells much more often than infect new animals. As SARS and MERS in vitro experiments showed, significant mutations might be observed after only a few passages. For example, the 2004 paper reported that only after 600 passages there already was a 2.1% difference in the genomic sequences of spike proteins between the original strain and its progeny:

Moreover, in the presence of some antiviral compounds, such as nucleoside analogs (e.g. ribavirin or remdesivir), mutation rates in RNA viruses can increase even further:

We obtained an estimate of the spontaneous mutation rate of ca. 10⁻⁴ substitutions per site or lower, a value within the typically accepted range for RNA viruses. A roughly threefold increase in mutation rate and a significant shift in mutation spectrum were observed in samples from patients undergoing 6 months of interferon plus ribavirin treatment. This result is consistent with the known in vitro mutagenic effect of ribavirin and suggests that the antiviral effect of ribavirin plus interferon treatment is at least partly exerted through lethal mutagenesis.

So if ancestral CoV2 was being lab-tested to assess how its mutagenesis might affect the efficacy of potential vaccines or antiviral drugs, it could have accumulated mutations at a much higher rate.

But possibly, the biggest problem with the 4% difference argument is that it relies on RaTG13 being exactly what WIV says it is. If we are to seriously consider the lab leak hypothesis, we must concede that it does not make sense to blindly trust the data released by the very lab suspected of the leak. If the leak did occur, as is the premise of the lab hypothesis, then the description of what RaTG13 is could be furthering the goal of covering up the leak.

Again, I am not claiming with certainty that is what is happening here. All I am saying is that this is what could have happened, and we need a lot more evidence before we can reach a definitive conclusion. One thing that could help rule out tampering with RaTG13 is having independent labs sequence the 2013 Yunnan samples that She Zhengli extracted RaTG13 from. WIV must still have them if they re-sequenced RaTG13 in 2020.

Shi Zhengli-2020

As I was writing this post, a fresh paper co-authored by Shi Zhengli came out, in which the authors tested a peptide which they have been studying for some time before against CoV2. That peptide was meant to be a pan-coronavirus inhibitor, and its designed mode of action was to block the fusion of a spike protein with a cell membrane. The authors, of course, mention the new furin site of CoV2, and suggest that it may play an important role in the much more efficient penetration of CoV2 into the cell:

In this study, we have shown that SARS-CoV-2 exhibits much higher capacity of membrane fusion than SARS-CoV, suggesting that the fusion machinery of SARS-CoV-2 is an important target for development of coronavirus fusion inhibitors.
 …
Generally, β-B coronaviruses lack the S1/S2 furin-recognition site, and their S proteins are uncleaved in the native state. For example, SARS-CoV enters into the cell mainly via the endosomal membrane fusion pathway where its S protein is cleaved by endosomal cathepsin L and activated. Inducing the S1/S2 furin-recognition site could significantly increase the capacity of SARS-CoV S protein to mediate cellular membrane surface infection.

In this context, I wonder whether the authors have previously conducted experiments on how adding a furin site can alter the effectiveness of their peptide (or other drugs or vaccines) against a given coronavirus.

Not to be outdone, Ralph Baric also joined the race to find drugs against CoV2. As I understand, he and co-authors took data on the effectiveness of their nucleoside analogue (β-D-N4-hydroxycytidine, NHC) against SARS-CoV and MERS that they already had, added some in vitro data on CoV2, and sent off the paper to print. Nucleoside analogues (such as the famous remdesivir) are a fundamentally different approach than Shi Zhengli et al. Here, the authors try to prevent viral replication by giving “defective” letters of the genetic alphabet to virus’ copying machine, while Shi Zhengli and coauthors try to prevent the virus from entering the cell altogether. Theoretically, these approaches could be combined.

This is the End, Beautiful Friend

If you made it here by reading rather than scrolling, mad props to you. Hey, even if you scrolled, that’s cool too, and I apologize for the verbosity. I just didn’t anticipate that the rabbit hole would turn out to be a whole underground cave system. I hope that you found this deep dive into the world of virology interesting and enjoyed the exploration of the lab-made CoV2 hypothesis. In my opinion, the data I have presented, taken together, do not allow us to reject this possibility.

Let me be clear: this does NOT prove that CoV2 was synthesized in the laboratory. Yes, as we have seen above, from a technical standpoint, it would not be difficult for a modern virologist to create such a strain. But there is no direct evidence that anyone did this, and strange coincidences cannot pass for circumstantial evidence. On balance, the current chances against this are still higher than for the natural origins of CoV2. Moreover, even if CoV2 was indeed an unfortunate lab leak, the scientists themselves are not to blame, as they were working within the established international laws and guidelines on such research. Now, those who might be trying to cover up that leak, that’s a different story.

The opposite point is worth repeating too: the inverse hypothesis about the exclusively natural origin of the virus does not yet have strong evidence either. Until intermediate ancestors between RaTG13, pangolin-2019 and CoV2 are found, in whom we could trace the mosaic recombination that we observe in CoV2, the question of its origins remains open. In closing, there is no one better to quote on this matter than Ralph Baric himself:

What is the reservoir species of SARS-CoV-2?
They have not identified the actual reservoir species. Reports show that pangolins are potentially the intermediate host, but pangolin viruses are 88–98% identical to SARS-CoV-2. In comparison, civet and racoon dog strains of SARS coronaviruses were 99.8% identical to SARS-CoV from 2003. In other words, we are talking about a handful of mutations between civet strains, racoon dog strains and human strains in 2003. Pangolins [strains of CoV2] have over 3000 nucleotide changes, no way they are the reservoir species. Absolutely no chance.

So there you have it. It remains possible that the mysterious virus host was a lab:

Bad pun? Sorry, last one.

How I Learned to Hate the GOF

I hope this post is not used to prematurely assign blame or propagate one-sided theories. What I do hope it highlights is the scale of dangerous gain-of-function research that has been and is going on in virology. The Covid-19 pandemic really exposed its huge risks in the face of few benefits: GOF research hasn’t protected us from this outbreak, hasn’t provided us with any effective treatments or vaccines in time to save hundreds of thousands of lives lost to CoV2, and if there is even a 0.1% chance GOF research caused the whole thing, that chance is too high.

When I first heard about the concept of mind uploading, I thought it meant creating a digital copy of one’s mind in a computer where it would then continue to live all on its own. Well, I thought, this is of little comfort to me, the original. Who cares how many of my clones are created, as if I — this particular instance of me — die, it would be the end of the only ‘me’ I have ever been a part of.

To counter that view, some say that a clone of you waking up and believing that it is you is the same as you waking up every morning — after all, you lose continuity with reality every time you go to sleep — but for me personally the above kind of mind uploading is still the least preferred. I would be willing to try it as a last resort in the case of, say, recovery after cryopreservation, but I would much rather upload my mind gradually.

Trending AI Articles:

1. Deep Learning Book Notes, Chapter 1

2. Deep Learning Book Notes, Chapter 2

3. Machines Demonstrate Self-Awareness

4. Visual Music & Machine Learning Workshop for Kids

Gradual mind uploading makes the most sense. If fact, most of us already engage in it, as we are outsourcing more and more of our memories to digital substrates — how many phone numbers do you remember today? And it’s not just memories that we already outsource. A lot of our daily mental processing is now done by our phones and computers. The only problem is that our interaction with them is limited by what Elon Musk (and many others) calls our bandwidth — the ability to input and output information between our wetware and our digital extensions — that is limited by our reading and typing (or speaking) speeds.

That is also why Elon is a big fan of the neural lace technology which is essentially an interface to greatly increase the communication bandwidth between our brain and our computers. Imagine that instead of having to type into Google and read back the Wikipedia article it finds you, you can accomplish both activities simply with the power of thought. And yes, you would have a direct link to the Internet in your brain rather than at your fingertips. Just imagine how much more Facebooking you could accomplish!

Of course, neural lace is a huge step in the direction of complete mind uploading, but it is only one of many. We would obviously need to replicate human mental and emotional processing in silico. When you don’t just store memories or perform computational tasks in your digital extension, but also analyze, think and feel in it, only then would you have accomplished full mind uploading. Note that we would not necessarily need to decode and replicate the underlying biological mechanisms behind those mental or emotional centers of our brains, just their functionality. Or maybe we could even improve upon it — avoid anxiety, reduce anger, cure addiction.

So to me, ultimate mind uploading would involve having all our memories and all our processing moved or parallelized to a machine safer than our brain. That way, even if our brain dies, we could still retain all or most of our personality. This would reduce losing our entire brain to what happens today if one suffers a minor head injury or stroke: a small region of the brain might die and one might lose some memories or the ability to drive, but they are still alive, able to enjoy life and interact with loved ones.

The next interesting question is avatars. If most of our personality would reside outside our brain, then would our original body be reduced to essentially an avatar? Could we then get more than one like in the Hardcore Henry movie? I don’t see why not. Obviously, avatars would need to be in constant (or at least frequent) communication with one’s primary consciousness, or things could get too out of sync, but I am sure that those are solvable technological issues. Growing new avatars might be a bigger challenge that keeping them in sync, and possibly more of an ethical one than a biological one. But that’s food for thought for a different post. Obviously there are many risks and ethical questions, and mind uploading has its detractors.

Among other negatives, opponents of mind uploading charge that even a gradually uploaded you would cease to be you, and while agreeing with them in principle, I disagree that this is a negative. Just like a 1-year old you is very different from a 15-, 35-, or 65-year old you, so would a digitally extended you be a very different you indeed. While others might prefer the 3-year old version of you, if you personally would rather stay the 35-year old version of yourself, then no one else should deny you this choice.

Would the difference between the superhuman you and you before uploading be greater than between the 1-year old you and the 65-year old one? I bet it would. And I bet it would be awesome.

Don’t forget to give us your 👏 !

Mind Uploading Done Right was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.