In this large-scale ecological study, HLA genotypes were integrated with SARS-CoV-2 testing results across over 319,000 patients.
In the baseline analyses, blood group O was found to have a protective effect against SARS-CoV-2 infection in the EUR ancestry group, consistent with observations reported in multiple studies [33]; however, this protective effect was not replicated in other ancestry groups. This could be because of smaller sample sizes in other ancestry groups, as suggested by the wide error bars for non-EUR groups in Fig. 3. Alternatively, this could be because prior studies did not stratify analysis by ancestry or may have had a predominance of subjects of EUR ancestry in the cohorts analyzed.
Although some studies have reported that blood group A confers susceptibility to SARS-CoV-2 infection (though with less confidence) [33], no such association in any ancestry group was identified in the current study. This discrepancy could be due to differences in study populations, outcome variables, statistical methodology, sample sizes, or other factors.
No association between Rh status and SARS-CoV-2 infection was found, despite previous studies that reported a protective effect of Rh-negativity [34, 35]. Prior reports that male sex is a risk factor for SARS-CoV-2 infection were confirmed, but only in the EUR ancestry group. Age was not found to predict SARS-CoV-2 infection, even though age had been proposed as a risk factor for SARS-CoV-2 infection in prior studies [5, 6].
The non-HLA variables with the largest effects in the baseline model were estimated SES (NDI) and SARS-CoV-2 exposure risk (zip3 positivity), which were associated with increased susceptibility to SARS-CoV-2 infection across ancestries. NDI showed a stronger effect size than exposure risk and may be identifying additional elements of exposure. Associations between lower SES and increased risk of SARS-CoV-2 infection have been reported previously and may be related to a variety of factors including housing conditions, occupations, and reliance on public transportation [36,37,38,39].
Eight HLA alleles showed significant association with risk of infection after controlling for exposure and SES. Allele effect sizes were variable and, in several cases, an allele associated with increased infection risk in one ancestry was associated with a decreased (yet non-significant) risk in other ancestries (or vice-versa). It should be noted that HLA alleles are not evenly distributed across ancestry groups, so allele-specific effects may not be identified across all groups.
In relation to existing literature, the current study found that the HLA–C*01:02 allele may be associated with an increased susceptibility to infection; this aligns (at the allele family level, HLA-C*01) with one prior study [40]. Additionally, DRB1*08:02 was associated with an increased susceptibility in the HIS population; this is consistent with a 2021 study by Schindler et al. which identified this allele as a potential risk factor for symptomatic SARS-CoV-2 infection in a small group of Hispanic patients [41]. The allele family HLA-DRB1*08 was also associated with increased infection risk in a case-control study from Italy by Amoroso et al. [16]. Based on in silico studies, these authors postulate that –DRB1*08 alleles have poor binding affinity with SARS-CoV-2 peptides, which could explain the increased risk for infection [16, 41]. It should be noted, however, that the current study identified an allele in the same –DRB1*08 family (-DRB1*08:03) as having a protective effect in the API population. This could be because commonly observed—DRB1*08 alleles [42] (Supplementary Table 4) may differ at key peptide-binding residue positions, and so not all alleles may behave uniformly with respect to peptide presentation or infection risk.
The remaining HLA findings in the current study did not replicate any allele associations with SARS-CoV-2/COVID-19 susceptibility reported elsewhere. This is not surprising, given the general lack of agreement on susceptibility loci identified across previous studies. It should also be noted that, with rare exception [39], most studies that explored HLA genetics in the context of existing SARS-CoV-2 infection have not controlled for exposure risk and socioeconomic status of individuals, which was found here to be strong predictors of infection. Other potential reasons for the disagreements among HLA association studies include study size, disease phenotype classification, ancestries evaluated, alleles investigated, and HLA genotyping methods.
The lack of consistent allele effects across ancestry groups, coupled with the fact that the largest cohort (EUR) failed to identify many robust associations, may lead one to question whether the alleles identified as significant could be sampling artifacts. However, given that alleles are not evenly represented across ancestry groups, the lack of reproducible findings could be related to data sparseness in certain populations. The differences in infection risk by ancestry could also be due to non-HLA genetic or environmental factors that were not explored in this study. Though functional analysis was beyond the scope of this study, future research could explore HLA gene expression by alleles of interest in the context of SARS-CoV-2 infection.
The lack of consistent findings among susceptibility studies may also be explained by the biological role HLA molecules play in immune response. HLA molecules are primarily involved in the T-cell response, which typically occurs within seven days of symptom onset and peaks at ~14 days [43, 44]. Therefore, HLA variation may have a stronger association with infection outcome (mediated, in part, by T-cells) than infection susceptibility (mediated, initially, by antibodies). Data regarding patient symptoms and outcomes were not available for analysis in this study population; therefore, this hypothesis cannot be explored using the current dataset.
In the paired analysis, two allele pairs appeared to strengthen the effects seen in the single allele model. The first pair involved HLA-B*38:01, which showed a protective effect as a single allele in EUR (OR = 0.87); when paired with HLA-DQB1*03:02, the protective effect was augmented (OR = 0.78). Though these two alleles were not found to be in LD (D’ = 0.17) in this dataset, they represent a common haplotype seen in the Ashkenazi Jewish population [45, 46], thus it is possible that some hidden population substructure was detected, rather than a true association. The second pair involved HLA-DRB1*08:03, which showed a protective effect as a single allele in the API group (OR 0.66), and HLA-DQB1*06:01, observed in strong LD with HLA-DRB1*08:03 (D’ = 0.90), which displayed a slightly enhanced protective effect (OR = 0.63).
During the study period (December 30, 2019–June 23, 2021), the SARS-CoV-2 virus evolved, and variants emerged. The predominant viral strains present in the US during this time were the original strain, and starting in early 2021, the alpha variant (B.1.1.7) [47]. Though these viral variants have the potential to interact differently with HLA alleles due to changes in the spike protein or viral peptides, prior research has shown that the immunodominant T-cell epitopes of interest appear to be highly conserved over time [48]. Therefore, though an effect cannot be ruled out, the fluctuation of variants throughout the study period is unlikely to have an impact on results. As PCR testing for COVID-19 was reported in a binary fashion (positive/negative), it is not possible to determine which viral strain was present in the patients with positive test results, or to assess the impact of variants on the study findings.
In addition to the emergence of viral variants during the study period, COVID-19 vaccines became available. The U.S. Food & Drug Administration granted Emergency Use Authorization (EUA) for the Pfizer-BioNTech and Moderna vaccines in mid-December 2020, and the Johnson & Johnson vaccine in late February 2021. It is estimated that adult vaccination rates increased from ~6% in January 2021 to ~65% in June 2021 [49]. As vaccination status was not collected as part of the study data, it is not possible to determine the impact of vaccination status on infection risk in this study. It should be noted that an evaluation of logistic regression models for the baseline variables of patients tested prior to 2021 versus those tested in 2021 were broadly similar cross the two time periods, so the potential impact of vaccination is expected to be minimal in this dataset.
Another limitation of the study is the use of zip3-based imputation to estimate individual exposure risk. Zip3 regions typically span a range of urban, suburban and rural environments with highly heterogeneous exposure risks. The fact that this coarse-grained exposure proxy measure was significant in all ancestries suggests that it was capturing a real aspect of individual exposure risk, but it also suggests that a more fine-grained exposure metric could explain even more of the variability in infection risk. A subject’s SES was also estimated, but at the more fine-grained census tract level using NDI scores. It should be noted that NDI may also be acting as a proxy for exposure. The inclusion of zip3 positivity and NDI in the models reduced the number of significant HLA alleles; it is possible that better exposure data could result in loss of significance for some (or all) of the remaining alleles.
In summary, the largest real-world study to date exploring the relationship between HLA genetics and SARS-CoV-2 infection emphasizes the crucial importance of controlling for SARS-CoV-2 exposure risk and socioeconomic status. Controlling for these variables using coarse-grained imputations reduced the number of significant candidate risk-modifying alleles from 41 to eight; it is possible that better exposure data could reduce this number further. These results suggest that most common HLA variants do not strongly affect SARS-CoV-2 infection risk.
