We’ve created calculator that aggregates deep mutational scanning data to estimate how mutations to #SARSCoV2 RBD affect recognition by human polyclonal antibodies. The calculator emphasizes extensive antigenic change in #Omicron variant. jbloomlab.github.io/RBD_escape_cal… (1/n)
Calculator uses experimental measurements of how all mutations to #SARSCoV2 RBD affect binding by 33 neutralizing antibodies (
) to calculate antigenic effects of mutating arbitrary combinations of RBD sites. (2/n)
Below is video of toy example with just 3 antibodies to explain principle. As mutations ablate individual antibodies, that reduces binding of polyclonal mix at their epitope sites. Explore this toy example at jbloomlab.github.io/SARS2_RBD_Ab_e… (3/n)
Actual calculator uses 33 antibodies. As video below shows, it calculates net effect of mutating sites. Experimental data define importance of sites & how mutating one site affects binding at others. Explore using interactive calculator at jbloomlab.github.io/SARS2_RBD_Ab_e… (4/n)
This approach is *not* a complex black-box computational algorithm: it’s just simple aggregation of direct experimental data to define polyclonal antibody binding map, and then intuitive calculation of how it’s affected by mutations. (5/n)
What about Omicron? Here’s how all 15 mutations in Omicron RBD affect antibody binding. Seems dire: all biggest peaks are gone. If you explore interactively at jbloomlab.github.io/SARS2_RBD_Ab_e… you’ll see mutations at sites 417, 446, & 484 biggest culprits, but others contribute too. (7/n)
Plot above also shows sites of possible future Omicron escape (eg, 346, 378, 444, 504). @richardneher notes there is already Omicron subvariant w R346K, including sequences from Seattle by @pavitrarc. R346K subvariant should be monitored. nextstrain.org/groups/neherla… (8/n)
Here’s total calculated RBD antibody binding to Omicron (w/o R346K) & #SARSCoV2 variants. Omicron has much less binding than other variants; is close to artificial spike polymutant PMS20 of @PaulBieniasz@theodora_nyc (9/n)
For reference, polymutant PMS20 spike (nature.com/articles/s4158…) has avg ~20 to 80-fold reduction in neutralization for various human cohorts. This is similar to ~25 to 60-fold reduction predicted for Omicron by extrapolating correlation in Tweet 6 of this thread. (10/n)
Some caveats of escape calculator: only considers RBD antibodies, assumes antibodies w data represent polyclonal sera, treats all mutations at site equivalently, & assumes people immunized w RBD similar to Wuhan-Hu-1 (currently mostly true, but becoming less so). (11/n)
Despite these caveats, calculator works pretty well on current data. More generally, #SARSCoV2 is going to continue to evolve new variants, so we need prospective approaches in addition to reactively running neutralization assays on new variants. (12/n)
When a new SARS-CoV-2 variant arises, there are three main questions: (1) How transmissible? (2) How virulent? (3) How much antigenic change? Third question important as it’s the most actionable: we can update vaccines & develop new antibodies. (13/n)
The first two questions (transmissibility and virulence) can only be answered by waiting for epidemiology and clinical data, as transmissibility and virulence are *so complicated* we can’t even begin to predict them from sequence. (14/n)
But while direct neutralization assays will always be gold standard for antigenic change, we can interpret a lot about antigenicity from sequence. Antibody neutralization has complexity, but not mind-boggling complexity like transmissibility and virulence… (15/n)
Ultimately, antibody neutralization involves binding of antibodies to spike, which we can understand. Scientific community has now run 1000s of neutralization assays on variants, solved ~100 X-ray/cryoEM structures, plus deep mutational scanning: covdb.stanford.edu/page/susceptib… (16/n)
It’s imperative to keep generating such data for new variants, but we also need to put the current data into coherent frameworks to synthesize the knowledge accumulated so far. (17/n)
Escape calculator takes step in that direction by aggregating data to intuitively visualize what is known about antigenic effects of RBD mutations to promote understanding & interpretation. So go to jbloomlab.github.io/SARS2_RBD_Ab_e… & try it out. (18/n)
While epistasis can have arbitrary form, as I note in article, due to action of natural selection it's more likely that mutations in Omicron will interact in way favorable for virus. I suspect this will be true of mutations at site 498 & 501 for ACE2 binding, for instance. (2/n)
For those following my Tweets only for public-health info about #SARSCoV2, you can stop reading here as we don't know more than 👆 right now. But as scientist interested in epistasis, I want to link above discussion of Omicron to basic research in molecular evolution. (3/n)
Here's how mutations in #SARSCoV2 Nu variant (B.1.1.529) will affect polyclonal and monoclonal antibodies targeting RBD. These assessments based on deep-mutational scanning experiments; underlying data can be explored interactively at jbloomlab.github.io/SARS2_RBD_Ab_e… (1/n)
First, Nu variant has lot of antigenic change. Below are how mutations relate to escape averaged over 36 human antibodies. Many mutations at peak escape sites, especially E484, G446, K417, & Q493. This means even in polyclonal mix, lot of RBD antibodies will be affected. (2/n)
Another way to assess polyclonal escape is how many epitope classes affected (nature.com/articles/s4146…). We do this using epitope scheme of @bjorkmanlab@cobarnes27 as adopted by @AllieGreaney. In this scheme, three potently neutralizing epitopes: class 1, 2, class 3. (3/n)
), patient zero was infected probably at least ~1 month before mid-Dec cases @MichaelWorobey is discussing, possibly substantially earlier. (2/n)
Note this misunderstanding about patient zero comes from newspaper headlines, and isn't fault of @MichaelWorobey. In fact, he's done work suggesting patient zero was infected between mid-Oct to mid-Nov (science.org/doi/10.1126/sc…), although there's still a lot of uncertainty (3/n)
@RolandBakerIII I don't think this paper suggests people were exposed to #SARSCoV2 20 years ago. Rather, it suggests that at a very low frequency some human antibody gene rearrangements will bind strongly to the #SARSCoV2 RBD even in the absence of an immune response selecting for this. (1/n)
@RolandBakerIII This is not terribly surprising. For instance, it's known that even naive humans sometimes have a bit of antibody reactivity to the #SARSCoV2 RBD (see Fig 1B of this paper by @SCOTTeHENSLEY). Indeed, this type of rare low-level naive reactivity...
Additionally, these antibodies have genes similar to IGHV3-53, which is known to naturally bind well to the RBD with minimal somatic hypermutation. (3/n)
In a new study led by @AllieGreaney, we show that infection with a #SARSCoV2 variant elicits an antibody response with somewhat shifted specificity relative to early Wuhan-Hu-1-like viruses that were circulating early in the pandemic: biorxiv.org/content/10.110… (1/n)
It's now known that #SARSCoV2 variants have mutations that reduce neutralization by antibodies elicited by early viruses, which are source of spike in current vaccines. This figure from @VirusesImmunity shows neutralization drops for common variants:
But do the antibodies elicited by infection with these variants have different specificities, such that humoral immunity from infection with variants will be differentially affected by specific mutations? (3/n)
To answer below question, most bat CoV don't bind human ACE2 strongly, but can happen incidentally in evolution. Presumably because some mutations that increase binding to bat ACE2s incidentally increase binding to human ACE2, which has substantial homology to bat ACE2s. (1/6)
More broadly, we recently did large yeast-display survey of SARS-related CoV RBDs and found that some bind human ACE2 (and some ACE2s from other species) well despite being from bats (biorxiv.org/content/10.110…). (3/6)