A New Antibody-Dependent Enhancement Hypothesis

 By Derek Lowe

Here’s another post that I will regret writing, but a great many people have asked me about a new preprint that brings up the possibility of antibody-dependent enhancement with the current vaccines and the Delta variant. To be frank, some of the people promoting this seem to be rooting for the virus, just so long as it humiliates their enemies and proves their own positions to be correct, but there are a lot of honestly worried people out there who are wondering what this paper means. So let’s look at it, with an eye to lessons for evaluating such papers in general.

The authors are building on another recent paper (Li et al.) on neutralizing and non-neutralizing antibodies against the current coronvirus. The preprint version of that one came out in February, and the final version went online in June. That work appears to be very solid, and represents a great deal of effort, so let’s discuss it for a bit before returning to the preprint above. In it, the authors isolated antibodies from human patients that target the receptor-binding domain (RBD) and others that target the N-terminal domain (NTD). They found using in vitro assays that both neutralizing and (especially) non-neutralizing antibodies that bound to the NTD showed antibody-dependent enhancement in cell infection. This took place partly through a well-known ADE pathway involving the Fc-gamma receptor, which allows for infection of macrophages, and this was indeed the main mechanism seen with ADE of the earlier (2003) SARS virus. It should be noted, though, that these seem to use different subtypes of the Fc-gamma receptor, so they’re not completely identical. And there were other cellular ADE events that were Fc-receptor-independent, though a mechanism that has not yet been worked out.

But the Li et al. paper went on to demonstrate that this does not seem to happen in animal models for these SARS-CoV-2 antibodies. Indeed, antibodies that showed ADE in the cell culture models still protected primates from viral replication when they were challenged by the actual virus. Three of the 36 monkeys in the study had increased lung inflammation compared to controls, but still had decreased viral replication, which makes it likely not ADE (which depends on viral infection being worse) but some sort of effect of antibody treatment that is not mediated by virus. None of the animals that got the highest doses of antibodies, for example, showed any of these effects.

And the authors note that if antibody treatment led to ADE in humans, then it would have been seen in the convalescent serum trials and in its clinical usage. But it was not. Convalescent serum was not very beneficial in the end, but it was definitely not harmful. The paper also points out that the vaccine trials and the use of the vaccines in clinical practice have led to no signs of ADE, either. So ADE in cells for SARS-CoV-2 does not seem to translate to animal models of infection, and nothing of the sort has so far been observed in the (huge) human population.

Perhaps that’s why I haven’t been sent so many copies of the Li et al. manuscript, but rather this new one (Yahi et al.) I think it’s the title as well, which is an eye-catcher: “Infection-enhancing anti-SARS-CoV-2 antibodies recognize both the original Wuhan/D614G strain and Delta variants. A potential risk for mass vaccination?” The authors build on the earlier paper and make comparisons to the protein sequence of the Delta variant. They believe that the antibodies identified in the Li et al. paper as causing ADE in cell assays cause the Delta variant NTD to be more tightly bound to the membrane of human cells through an interaction with cell-surface lipid rafts, and they speculate that the balance between neutralization and infection enhancement, while favorable for earlier strains of the coronavirus, is tipped the other way for the Delta variant (thus the title of the paper, and this its rather bizarre Figure 2). About that title, while we’re on it – I’m not completely sure why they reference mass vaccination risks and not risks posed by previous infection by non-Delta strains, because I would have to think that the same concerns would apply.

This one is not a very long paper, and there’s a good reason for that – it contains no experimental data. The postulated binding enhancement via lipid rafts is completely a molecular modeling result, and has not been demonstrated in actual cells. Their computations are explained in more detail in another paper, where you can see that the authors believe that it’s the kinetics of the lipid-raft interaction that may drive transmissibility of various viral strains.

I have no comment on that per se, but I do have to note that modeling-only conclusions about large protein binding events have to be confirmed by experimental data before they can be taken seriously. There are a great number of things that look plausible in such simulations that do not translate to reality. To illustrate that, here is a paper from the same three authors, again completely based on molecular modeling, that goes into a detailed mechanistic explanation of how azithromycin and hydroxychloroquine work together as an effective therapy against the coronavirus. The azithromycin binds to the RBD, you see, while the HCQ is affecting the conformation down at the lipid-raft-binding part of the NTD (this one’s also lipid-raft-centric). That latter interaction is the subject of this earlier paper. The problem is, these two drugs together (or separately) are not actually an effective therapy for coronavirus infection, a fact that the earlier papers’ citations of the work of Didier Raoult cannot overcome. It is a detailed calculated hypothesis to explain something that does not, in fact, exist.

This is a constant danger with simulations. Readers who have not encountered much molecular modeling are often (and understandably) impressed by the graphics and tables that appear with such work, but if you’ve been involved with actual experimentation you’ve seen many, many examples of such hypotheses that turned out to be built on air. Confusing the graphics with reality is a constant danger for all of us.

And in my view, the Yahi et al. paper is not aligned with reality. They do work in a line about how “although the results obtained so far have been rather reassuring. . .” with a reference to the Li et al. paper, but they should also refer to the massive amount of real-world evidence now available. We have hundreds of millions of people who have been vaccinated to produce antibodies against the non-Delta coronavirus protein domains and are who are now being exposed to the Delta variant. To reiterate, there is (to the best of my knowledge) no evidence whatsoever of ADE in this situation. In fact, we see the opposite: people who have been vaccinated are far less likely to become infected with the Delta variant, and if they become infected, they are far less likely to experience severe disease. These trends have been seen over and over in different populations, and they are the exact opposite of what you would see if ADE were operating. If the mechanism proposed by Yahi et al. were happening in the real world, then we should see higher Delta infection rates among vaccinated people, with more severe disease. We are not. We are seeing the reverse. The vaccines simply to not appear to be causing ADE, no matter how many reasons one might be able to spin for them to do so.

In short, get real.

https://blogs.sciencemag.org/pipeline/archives/2021/08/16/a-new-antibody-dependent-enhancement-hypothesis