Automatic Fact Checking Using an Interpretable Bert-Based Architecture on COVID-19 Claims

We present a neural network architecture focused on verifying facts against evidence found in a knowledge base. The architecture can perform relevance evaluation and claim verification, parts of a well-known three-stage method of fact-checking. We fine-tuned BERT to codify claims and pieces of evidence separately. An attention layer between the claim and evidence representation computes alignment scores to identify relevant terms between both. Finally, a classification layer receives the vector representation of claims and evidence and performs the relevance and verification classification. Our model allows a more straightforward interpretation of the predictions than other state-of-the-art models. We use the scores computed within the attention layer to show which evidence spans are more relevant to classify a claim as supported or refuted. Our classification models achieve results compared to the state-of-the-art models in terms of classification of relevance evaluation and claim verification accuracy on the FEVER dataset.

Memory B cell responses to Omicron subvariants after SARS-CoV-2 mRNA breakthrough infection (preprint)

Individuals that receive a 3rd mRNA vaccine dose show enhanced protection against severe COVID19 but little is known about the impact of breakthrough infections on memory responses. Here, we examine the memory antibodies that develop after a 3rd or 4th antigenic exposure by Delta or Omicron BA.1 infection, respectively. A 3rd exposure to antigen by Delta breakthrough increases the number of memory B cells that produce antibodies with comparable potency and breadth to a 3rd mRNA vaccine dose. A 4th antigenic exposure with Omicron BA.1 infection increased variant specific plasma antibody and memory B cell responses. However, the 4th exposure did not increase the overall frequency of memory B cells or their general potency or breadth compared to a 3rd mRNA vaccine dose. In conclusion, a 3rd antigenic exposure by Delta infection elicits strain-specific memory responses and increases in the overall potency and breadth of the memory B cells. In contrast, the effects of a 4th antigenic exposure with Omicron BA.1 is limited to increased strain specific memory with little effect on the potency or breadth of memory B cell antibodies. The results suggest that the effect of strain-specific boosting on memory B cell compartment may be limited.

Antibody feedback regulation of memory B cell development in SARS-CoV-2 mRNA vaccination (preprint)

Feedback inhibition of humoral immunity by antibodies was initially documented in guinea pigs by Theobald Smith in 1909, who showed that passive administration of excess anti-Diphtheria toxin inhibited immune responses1. Subsequent work documented that antibodies can enhance or inhibit immune responses depending on antibody isotype, affinity, the physical nature of the antigen, and engagement of immunoglobulin (Fc) and complement (C) receptors2, 3. However, little is known about how pre-existing antibodies might influence the subsequent development of memory B cells. Here we examined the memory B cell response in individuals who received two high-affinity IgG1 anti-SARS-CoV-2 receptor binding domain (RBD)-specific monoclonal antibodies, C144-LS and C135-LS, and subsequently two doses of a SARS-CoV-2 mRNA vaccine. The two antibodies target Class 2 and 3 epitopes that dominate the initial immune response to SARS-CoV-2 infection and mRNA vaccination4-8. Antibody responses to the vaccine in C144-LS and C135-LS recipients produced plasma antigen binding and neutralizing titers that were fractionally lower but not statistically different to controls. In contrast, memory B cells enumerated by flow cytometry after the second vaccine dose were present in higher numbers than in controls. However, the memory B cells that developed in antibody recipients differed from controls in that they were not enriched in VH3-53, VH1-46 and VH3-66 genes and predominantly expressed low-affinity IgM antibodies that carried small numbers of somatic mutations. These antibodies showed altered RBD target specificity consistent with epitope masking, and only 1 out of 77 anti-RBD memory antibodies tested neutralized the virus. The results indicate that pre-existing high-affinity antibodies bias memory B cell selection and have a profound effect on the development of immunological memory in humans that may in part explain the shifting target profile of memory antibodies elicited by the 3rd mRNA vaccine dose.

Humoral immunity to SARS-CoV-2 elicited by combination COVID-19 vaccination regimens (preprint)

The SARS-CoV-2 pandemic prompted a global vaccination effort and the development of numerous COVID-19 vaccines at an unprecedented scale and pace. As a result, current COVID-19 vaccination regimens comprise diverse vaccine modalities, immunogen combinations and dosing intervals. Here, we compare vaccine-specific antibody and memory B cell responses following two-dose mRNA, single-dose Ad26.COV2.S and two-dose ChAdOx1 or combination ChAdOx1/mRNA vaccination. Plasma neutralizing activity as well as the magnitude, clonal composition and antibody maturation of the RBD-specific memory B cell compartment showed substantial differences between the vaccination regimens. While individual monoclonal antibodies derived from memory B cells exhibited similar binding affinities and neutralizing potency against Wuhan-Hu-1 SARS-CoV-2, there were significant differences in epitope specificity and neutralizing breadth against viral variants of concern. Although the ChAdOx1 vaccine was inferior to mRNA and Ad26.COV2.S in several respects, biochemical and structural analyses revealed enrichment in a subgroup of memory B cell neutralizing antibodies with distinct RBD-binding properties resulting in remarkable potency and breadth.

Antibody evolution to SARS-CoV-2 after single-dose Ad26.COV2.S vaccine (preprint)

The single dose Ad.26.COV.2 (Janssen) vaccine elicits lower levels of neutralizing antibodies and shows more limited efficacy in protection against infection than either of the available mRNA vaccines. In addition, the Ad.26.COV.2 has been less effective in protection against severe disease during the Omicron surge. Here, we examined the memory B cell response to single dose Ad.26.COV.2 vaccination. Compared to mRNA vaccines, Ad.26.COV.2 recipients had significantly lower numbers of RBD-specific memory B cells 1.5 or 6 months after vaccination. Memory antibodies elicited by both vaccine types show comparable neutralizing potency against SARS-CoV-2 and Delta. However, the number of memory cells producing Omicron neutralizing antibodies was somewhat lower after Ad.26.COV.2 than mRNA vaccination. The data help explain why boosting Ad.26.COV.2 vaccine recipients with mRNA vaccines is effective, and why the Janssen vaccine appears to have been less protective against severe disease during the Omicron surge than the mRNA vaccine.

Increased Potency and Breadth of SARS-CoV-2 Neutralizing Antibodies After a Third mRNA Vaccine Dose (preprint)

The omicron variant of SARS-CoV-2 infected very large numbers of SARS-CoV-2 vaccinated and convalescent individuals. The penetrance of this variant in the antigen experienced human population can be explained in part by the relatively low levels of plasma neutralizing activity against Omicron in people who were infected or vaccinated with the original Wuhan-Hu-1 strain. The 3rd mRNA vaccine dose produces an initial increase in circulating anti-Omicron neutralizing antibodies, but titers remain 10-20-fold lower than against Wuhan-Hu-1 and are, in many cases, insufficient to prevent infection. Despite the reduced protection from infection, individuals that received 3 doses of an mRNA vaccine were highly protected from the more serious consequences of infection. Here we examine the memory B cell repertoire in a longitudinal cohort of individuals receiving 3 mRNA vaccine doses. We find that the 3rd dose is accompanied by an increase in, and evolution of, anti-receptor binding domain specific memory B cells. The increase is due to expansion of memory B cell clones that were present after the 2nd vaccine dose as well as the emergence of new clones. The antibodies encoded by these cells showed significantly increased potency and breadth when compared to antibodies obtained after the 2nd vaccine dose. Notably, the increase in potency was especially evident among newly developing clones of memory cells that differed from the persisting clones in targeting more conserved regions of the RBD. Overall, more than 50% of the analyzed neutralizing antibodies in the memory compartment obtained from individuals receiving a 3rd mRNA vaccine dose neutralized Omicron. Thus, individuals receiving 3 doses of an mRNA vaccine encoding Wuhan-Hu-1, have a diverse memory B cell repertoire that can respond rapidly and produce antibodies capable of clearing even diversified variants such as Omicron. These data help explain why a 3rd dose of an mRNA vaccine that was not specifically designed to protect against variants is effective against variant-induced serious disease.

Conserved Neutralizing Epitopes on the N-Terminal Domain of Variant SARS-CoV-2 Spike Proteins (preprint)

Summary SARS-CoV-2 infection or vaccination produces neutralizing antibody responses that contribute to better clinical outcomes. The receptor binding domain (RBD) and the N-terminal domain (NTD) of the spike trimer (S) constitute the two major neutralizing targets for the antibody system. Neutralizing antibodies targeting the RBD bind to several different sites on this domain. In contrast, most neutralizing antibodies to NTD characterized to date bind to a single supersite, however these antibodies were obtained by methods that were not NTD specific. Here we use NTD specific probes to focus on anti-NTD memory B cells in a cohort of pre-omicron infected individuals some of which were also vaccinated. Of 275 NTD binding antibodies tested 103 neutralized at least one of three tested strains: Wuhan-Hu-1, Gamma, or PMS20, a synthetic variant which is extensively mutated in the NTD supersite. Among the 43 neutralizing antibodies that were further characterized, we found 6 complementation groups based on competition binding experiments. 58% targeted epitopes outside the NTD supersite, and 58% neutralized either Gamma or Omicron, but only 14% were broad neutralizers. Three of the broad neutralizers were characterized structurally. C1520 and C1791 recognize epitopes on opposite faces of the NTD with a distinct binding pose relative to previously described antibodies allowing for greater potency and cross-reactivity with 7 different variants including Beta, Delta, Gamma and Omicron. Antibody C1717 represents a previously uncharacterized class of NTD-directed antibodies that recognizes the viral membrane proximal side of the NTD and SD2 domain, leading to cross-neutralization of Beta, Gamma and Omicron. We conclude SARS-CoV-2 infection and/or Wuhan-Hu-1 mRNA vaccination produces a diverse collection of memory B cells that produce anti-NTD antibodies some of which can neutralize variants of concern. Rapid recruitment of these cells into the antibody secreting plasma cell compartment upon re-infection likely contributes to the relatively benign course of subsequent infections with SARS-CoV-2 variants including omicron.

Antibody Evolution after SARS-CoV-2 mRNA Vaccination (preprint)

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection produces B-cell responses that continue to evolve for at least one year. During that time, memory B cells express increasingly broad and potent antibodies that are resistant to mutations found in variants of concern. As a result, vaccination of coronavirus disease 2019 (COVID-19) convalescent individuals with currently available mRNA vaccines produces high levels of plasma neutralizing activity against all variants tested. Here, we examine memory B cell evolution 5 months after vaccination with either Moderna (mRNA-1273) or Pfizer-BioNTech (BNT162b2) mRNA vaccines in a cohort of SARS-CoV-2 naive individuals. Between prime and boost, memory B cells produce antibodies that evolve increased neutralizing activity, but there is no further increase in potency or breadth thereafter. Instead, memory B cells that emerge 5 months after vaccination of naive individuals express antibodies that are equivalent to those that dominate the initial response. We conclude that memory antibodies selected over time by natural infection have greater potency and breadth than antibodies elicited by vaccination. These results suggest that boosting vaccinated individuals with currently available mRNA vaccines would produce a quantitative increase in plasma neutralizing activity but not the qualitative advantage against variants obtained by vaccinating convalescent individuals.

Naturally enhanced neutralizing breadth to SARS-CoV-2 after one year (preprint)

Over one year after its inception, the coronavirus disease-2019 (COVID-19) pandemic caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) remains difficult to control despite the availability of several excellent vaccines. Progress in controlling the pandemic is slowed by the emergence of variants that appear to be more transmissible and more resistant to antibodies 1,2 . Here we report on a cohort of 63 COVID-19-convalescent individuals assessed at 1.3, 6.2 and 12 months after infection, 41% of whom also received mRNA vaccines 3,4 . In the absence of vaccination antibody reactivity to the receptor binding domain (RBD) of SARS-CoV-2, neutralizing activity and the number of RBD-specific memory B cells remain relatively stable from 6 to 12 months. Vaccination increases all components of the humoral response, and as expected, results in serum neutralizing activities against variants of concern that are comparable to or greater than neutralizing activity against the original Wuhan Hu-1 achieved by vaccination of naïve individuals 2,5-8 . The mechanism underlying these broad-based responses involves ongoing antibody somatic mutation, memory B cell clonal turnover, and development of monoclonal antibodies that are exceptionally resistant to SARS-CoV-2 RBD mutations, including those found in variants of concern 4,9 . In addition, B cell clones expressing broad and potent antibodies are selectively retained in the repertoire over time and expand dramatically after vaccination. The data suggest that immunity in convalescent individuals will be very long lasting and that convalescent individuals who receive available mRNA vaccines will produce antibodies and memory B cells that should be protective against circulating SARS-CoV-2 variants.

Resolving the Functional Significance of BRCA1 RING Domain Missense Substitutions (preprint)

Part 1: Development and calibration of suitably accurate functional assays for BRCA1 RING domain and BRCT domain missense substitutions could dramatically accelerate clinical classification of rare missense substitutions observed in that gene. Leveraging data from 68,000 full sequence tests of BRCA1 and BRCA2, plus data from the limited number of already classified BRCA1 RING domain missense substitutions, we used logistic regression and related techniques to evaluate three BRCA1 RING domain assays. These were recently described high throughput yeast 2-hybrid and E3 ubiquitin ligase assays, plus a newly developed mammalian 2- hybrid assay. While there were concerns about the accuracy of the yeast 2-hybrid assay and the indirect nature of the ubiquitin ligase assay, the mammalian 2-hybrid assay had excellent correlation with existing missense substitution classifications. After calibration, this assay contributed to classification of one newly reported BRCA1 missense substitution. In principal, the mammalian 2-hybrid assay could be converted to a high-throughput format that would likely retain suitable accuracy. Part 2: How does one achieve clinically applicable classification of the vast majority of all possible sequence variants in disease susceptibility genes? BRCA1 is a high-risk susceptibility gene for breast and ovarian cancer. Pathogenic protein truncating variants are scattered across the open reading frame, but all known missense substitutions that are pathogenic because of missense dysfunction are located in either the amino-terminal RING domain or the carboxy-terminal BRCT domain. Heterodimerization of the BRCA1 and BARD1 RING domains is a molecularly defined obligate activity. Hence, we tested every BRCA1 RING domain missense substitution that can be created by a single nucleotide change for heterodimerization with BARD1 in a Mammalian 2-hybrid (M2H) assay. Downstream of the M2H laboratory assay, we addressed three additional challenges: assay calibration, validation thereof, and integration of the calibrated results with other available data such as computational evidence and patient/population observational data to achieve clinically applicable classification. Overall, we found that about 20% of BRCA1 RING domain missense substitutions are pathogenic. Using a Bayesian point system for data integration and variant classification, we achieved clinical classification of about 89% of observed missense substitutions. Moreover, among missense substitutions not present in the human observational data used here, we find an additional 47 with concordant computational and functional assay evidence in favor of pathogenicity; these are particularly likely to be classified as Likely Pathogenic once human observational data become available.

Development of potency, breadth and resilience to viral escape mutations in SARS-CoV-2 neutralizing antibodies (preprint)

Antibodies elicited in response to infection undergo somatic mutation in germinal centers that can result in higher affinity for the cognate antigen. To determine the effects of somatic mutation on the properties of SARS-CoV-2 spike receptor-binding domain (RBD)-specific antibodies, we analyzed six independent antibody lineages. As well as increased neutralization potency, antibody evolution changed pathways for acquisition of resistance and, in some cases, restricted the range of neutralization escape options. For some antibodies, maturation apparently imposed a requirement for multiple spike mutations to enable escape. For certain antibody lineages, maturation enabled neutralization of circulating SARS-CoV-2 variants of concern and heterologous sarbecoviruses. Antibody-antigen structures revealed that these properties resulted from substitutions that allowed additional variability at the interface with the RBD. These findings suggest that increasing antibody diversity through prolonged or repeated antigen exposure may improve protection against diversifying SARS-CoV-2 populations, and perhaps against other pandemic threat coronaviruses.

Rapid protection from COVID-19 in nonhuman primates vaccinated intramuscularly but not intranasally with a single dose of a recombinant vaccine (preprint)

The ongoing pandemic of Coronavirus disease 2019 (COVID-19) continues to exert a significant burden on health care systems worldwide. With limited treatments available, vaccination remains an effective strategy to counter transmission of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Recent discussions concerning vaccination strategies have focused on identifying vaccine platforms, number of doses, route of administration, and time to reach peak immunity against SARS-CoV-2. Here, we generated a single dose, fast-acting vesicular stomatitis virus-based vaccine derived from the licensed Ebola virus (EBOV) vaccine rVSV-ZEBOV, expressing the SARS-CoV-2 spike protein and the EBOV glycoprotein (VSV-SARS2-EBOV). Rhesus macaques vaccinated intramuscularly (IM) with a single dose of VSV-SARS2-EBOV were protected within 10 days and did not show signs of COVID-19 pneumonia. In contrast, IN vaccination resulted in limited immunogenicity and enhanced COVID-19 pneumonia compared to control animals. While IM and IN vaccination both induced neutralizing antibody titers, only IM vaccination resulted in a significant cellular immune response. RNA sequencing data bolstered these results by revealing robust activation of the innate and adaptive immune transcriptional signatures in the lungs of IM-vaccinated animals only. Overall, the data demonstrates that VSV-SARS2-EBOV is a potent single-dose COVID-19 vaccine candidate that offers rapid protection based on the protective efficacy observed in our study.

mRNA vaccine-elicited antibodies to SARS-CoV-2 and circulating variants (preprint)

To date severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) has infected nearly 100 million individuals resulting in over two million deaths. Many vaccines are being deployed to prevent coronavirus disease-2019 (COVID-19) including two novel mRNA-based vaccines. These vaccines elicit neutralizing antibodies and appear to be safe and effective, but the precise nature of the elicited antibodies is not known. Here we report on the antibody and memory B cell responses in a cohort of 20 volunteers who received either the Moderna (mRNA-1273) or Pfizer-BioNTech (BNT162b2) vaccines. Consistent with prior reports, 8 weeks after the second vaccine injection volunteers showed high levels of IgM, and IgG anti-SARS-CoV-2 spike protein (S), receptor binding domain (RBD) binding titers. Moreover, the plasma neutralizing activity, and the relative numbers of RBD-specific memory B cells were equivalent to individuals who recovered from natural infection. However, activity against SARS-CoV-2 variants encoding E484K or N501Y or the K417N:E484K:N501Y combination was reduced by a small but significant margin. Consistent with these findings, vaccine-elicited monoclonal antibodies (mAbs) potently neutralize SARS-CoV-2, targeting a number of different RBD epitopes epitopes in common with mAbs isolated from infected donors. Structural analyses of mAbs complexed with S trimer suggest that vaccine- and virus-encoded S adopts similar conformations to induce equivalent anti-RBD antibodies. However, neutralization by 14 of the 17 most potent mAbs tested was reduced or abolished by either K417N, or E484K, or N501Y mutations. Notably, the same mutations were selected when recombinant vesicular stomatitis virus (rVSV)/SARS-CoV-2 S was cultured in the presence of the vaccine elicited mAbs. Taken together the results suggest that the monoclonal antibodies in clinical use should be tested against newly arising variants, and that mRNA vaccines may need to be updated periodically to avoid potential loss of clinical efficacy.

Evolution of Antibody Immunity to SARS-CoV-2 (preprint)

Severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) has infected 78 million individuals and is responsible for over 1.7 million deaths to date. Infection is associated with development of variable levels of antibodies with neutralizing activity that can protect against infection in animal models. Antibody levels decrease with time, but the nature and quality of the memory B cells that would be called upon to produce antibodies upon re-infection has not been examined. Here we report on the humoral memory response in a cohort of 87 individuals assessed at 1.3 and 6.2 months after infection. We find that IgM, and IgG anti-SARS-CoV-2 spike protein receptor binding domain (RBD) antibody titers decrease significantly with IgA being less affected. Concurrently, neutralizing activity in plasma decreases by five-fold in pseudotype virus assays. In contrast, the number of RBD-specific memory B cells is unchanged. Memory B cells display clonal turnover after 6.2 months, and the antibodies they express have greater somatic hypermutation, increased potency and resistance to RBD mutations, indicative of continued evolution of the humoral response. Analysis of intestinal biopsies obtained from asymptomatic individuals 4 months after coronavirus disease-2019 (COVID-19) onset, using immunofluorescence, or polymerase chain reaction, revealed persistence of SARS-CoV-2 nucleic acids and immunoreactivity in the small bowel of 7 out of 14 volunteers. We conclude that the memory B cell response to SARS-CoV-2 evolves between 1.3 and 6.2 months after infection in a manner that is consistent with antigen persistence.

Enhanced SARS-CoV-2 Neutralization by Secretory IgA in vitro (preprint)

SARS-CoV-2 primarily infects cells at mucosal surfaces. Serum neutralizing antibody responses are variable and generally low in individuals that suffer mild forms of the illness. Although potent IgG antibodies can neutralize the virus, less is known about secretory antibodies such as IgA that might impact the initial viral spread and transmissibility from the mucosa. Here we characterize the IgA response to SARS-CoV-2 in a cohort of 149 individuals. IgA responses in plasma generally correlate with IgG responses and clones of IgM, IgG and IgA producing B cells that are derived from common progenitors are evident. Plasma IgA monomers are 2-fold less potent than IgG equivalents. However, IgA dimers, the primary form in the nasopharynx, are on average 15 times more potent than IgA monomers. Thus, secretory IgA responses may be particularly valuable for protection against SARS-CoV-2 and for vaccine efficacy.

Convergent Antibody Responses to SARS-CoV-2 Infection in Convalescent Individuals (preprint)

During the COVID-19 pandemic, SARS-CoV-2 infected millions of people and claimed hundreds of thousands of lives. Virus entry into cells depends on the receptor binding domain (RBD) of the SARS-CoV-2 spike protein (S). Although there is no vaccine, it is likely that antibodies will be essential for protection. However, little is known about the human antibody response to SARS-CoV-21-5. Here we report on 149 COVID-19 convalescent individuals. Plasmas collected an average of 39 days after the onset of symptoms had variable half-maximal neutralizing titers ranging from undetectable in 33% to below 1:1000 in 79%, while only 1% showed titers >1:5000. Antibody cloning revealed expanded clones of RBD-specific memory B cells expressing closely related antibodies in different individuals. Despite low plasma titers, antibodies to three distinct epitopes on RBD neutralized at half-maximal inhibitory concentrations (IC50s) as low as single digit ng/mL. Thus, most convalescent plasmas obtained from individuals who recover from COVID-19 do not contain high levels of neutralizing activity. Nevertheless, rare but recurring RBD-specific antibodies with potent antiviral activity were found in all individuals tested, suggesting that a vaccine designed to elicit such antibodies could be broadly effective.