COVID-19 Susceptibility By Race, Blood Type

3 DISCUSSION
HLA genes are recognised to be important factors in the host response to foreign pathogens. In this regional study, composed of participants of European descent drawn from a relatively small catchment area, we have identified HLA genes which appear to be associated with the severity of COVID-19 infection.

HLA genes encode for glycoproteins which bind peptides in ‘peptide binding grooves’, present on the cell membrane. HLA class I molecules, which are present on all nucleated cells, present endogenous peptides to cytotoxic CD8+ T cells. HLA class II molecules are expressed on the surface of antigen-presenting cells (including macrophages, B cells and dendritic cells) and are essential in the presentation of peptides to T-helper CD4+ cells. The variance of individual amino acids within HLA molecules determine the three dimensional structure of the peptide binding groove. The structure of the binding grooves determines which peptides (which can be foreign or auto antigens) are presented at the cell's surface.32

Validated software allows virtual construction of peptide binding grooves encoded by an individual's HLA genotype.33 This allows for the calculation of the binding affinity between a particular HLA encoded peptide binding groove and a multitude of naturally occurring peptides. A number of publications have investigated the link between HLA variation and COVID severity using this in silico approach. Some of these studies were combined with epidemiological data34-36 or results derived from small case control series.37 There appears to be no clear consensus in the conclusions drawn so far.

The results of the current study indicate that HLA alleles interact with age, sex and BMI to determine clinical outcomes following COVID exposure. Caution is needed in the interpretation of HLA results; HLA genes are, by their very nature, highly polymorphic. Due to the number of alleles which are under investigation, the risk of identifying a ‘significant’ allele through chance is high38 and none of the tests withstood correction for multiple testing. However, the most robust finding appears to be the protection conferred by the presence of DRB1*04:01. Approximately 30% of Europeans carry DRB1*04, an allele with more than 50 subtypes. Of all the DRB1*04 alleles, *04:01 is the most frequent allele occurring with a frequency of 57% as compared to *04:02 which occurs at a frequency of 6.3%.39

The various statistical analyses presented in the current article indicate that while DRB1*04:01 may be protective, other DRB1*04 alleles such as DRB1*04:02 and DRB1*04:05 may be associated with an increase in disease severity. On the face of it, this may seem unusual, yet the profound impact of amino acid substitutions and MHC-restricted T-cell recognition has long been understood.40 For example, DRB1*04:02 and *04:01 molecules differ in the peptide-binding region by only three amino acids yet they have diametrically opposed clinical outcomes: protection from versus susceptibility to rheumatoid disease.41 Certain HLA alleles are superior in providing protection by presenting multiple epitopes for the activation of T cells. The evolutionary trade-off is that most autoimmune diseases have been associated with the presence of certain HLA class II molecules. This appears to be the case with HLA-DRB1*04:01, which is associated with rheumatoid disease.42 Animal studies comparing 04:01 mice with 04:02 show that, following vaccination, even though both strains of mice are protected from influenza, only *04:01 mice can generate cross-protective immunity. There is also evidence that MHC peptide loading occurs in different cell compartments between the closely related HLA genes DRB1*04:01 and DRB1*04:02.17, 43 This fundamental difference between the two molecules may lead to different peptides being loaded and presented despite the same pathogen exposure. Furthermore, in response to superantigens, *04:01 mice have been shown to produce different and varying amounts of cytokines compared to 04:02 strains.44

That the strongest predictor of COVID severity was found in the class II HLA-DR region was not surprising given recently published work.45, 46 Kachuri et al3 conducted a comprehensive study including genome-wide and transcriptome-wide association analyses to identify genetic loci associated with IgG antibody response to 16 viruses using serological data from 7924 European ancestry participants in the UK Biobank cohort. Signals in the HLA class II region dominated the antibody response to viruses, with 40 independent loci and 14 independent classical alleles implicated. The strongest associations with seroreactivity were identified within the DRβ1 locus. These results were substantiated by Hammer et al47 who carried out a cross-pathogen, genome-wide investigation of the role of host genetics in modulating the individual IgG response to common viral antigens. Hammer et al found that individuals carrying DRB1*15:01 were more likely to have detectable levels of anti-influenza A IgG, whereas the presence of DRB1*01:01 was associated with seronegativity.

Consistent with data from the biobank studies, we found an increased frequency of DRB1*15:01 in both the asymptomatic and severe groups when compared to the background controls. This finding did not achieve significance however in our study. We believe this may have been due to our study protocol, which we designed primarily to identify differences between severe and asymptomatic groups. However, in a study of HLA 99 Italian patients admitted for COVID-19 infection,48 Novelli et al did observe significantly increased frequencies of DRB1*15:01 and DQB1*06:02 (alleles which are recognised to be in strong linkage disequilibrium). The authors noted that their findings conflicted with a larger study by Ellinghaus et al,37 who conducted a genome wide association study involving 1980 patients with COVID-19 and severe disease (defined as respiratory failure) at seven hospitals in Italy and Spain. Ellinghaus et al found no SNP association signals at the HLA complex that met even the significance threshold of suggestive association with either COVID-19 or disease severity. Ellinghaus et al did, however, identify a 3p21.31 gene cluster as a susceptibility locus and confirmed a potential involvement of the ABO blood-group system.

Another large GWAS study from the United Kingdom also failed to identify an HLA link.49 What could explain the difference between our results and those of the GWAS studies? The difficulties of using GWAS based methods to unpick HLA associations is recognised.50 However, in their study, Ellinghaus et al also employed next generation sequencing in a sub set of patients. It could be that the demographics of the study cohorts played a part in the difference in findings. In the Ellinghaus study, the median ages of the patients recruited from seven centres varied from 64 to 69 years with interquartile values ranging from 54 to 79 years. These patients were significantly older and frequently exhibited other medical comorbidities and the hospitals were separated by over 1500 km. In the current study, the median age of the admitted patients was 59 years and all hospitals and the catchment for the control group, were separated by a maximum distance of 80 km. The GENOMICC GWAS identified a signal in the HLA complex which was not confirmed in their validation cohort which was not matched for ethnicity or latitude.49 It is noteworthy that a GWAS is challenged where different alleles at the same locus have contrasting effects on phenotype.

The haplotype DQA1*01:01-DQB1*05:01-DRB1*01:01 was found at a significantly lower frequency in the asymptomatic group when compared to the background population. The lower frequency of this haplotype in the severe patients suggests that either the response to COVID in these individuals may not be primarily dependent upon IgG antibody production (which would be consistent with biobank data), or that this haplotype is associated with an increased likelihood of developing symptoms. There appeared to be an interaction with patient sex, with the frequency of this haplotype at 12.5% in the severe female group compared to 3.4% in the males. Interestingly, in our ongoing investigation into responses to orthopaedic implants, DRB1*01:01 was significantly associated with a lower probability of T and B cell infiltration into periprosthetic tissues in female patients, and this was one of the factors which prompted the current investigation.

3.1 The relationship between geographic location, allelic frequencies and COVID mortality rates
A host of environmental and socioeconomic factors have been shown to influence the course of previous viral outbreaks.3, 51, 52 In light of this, De Larochelambert et al performed an extensive analysis in an attempt to unpick the interaction of health, demographic, environment, and economic parameters on the impact of COVID-19 in countries across the globe.31 They identified several factors associated with COVID mortality rates experienced by a country, which included UV exposure, temperature, humidity, life expectancy and rates of obesity and cancer. The authors substantiated the previously identified positive association between latitude and COVID mortality,29 and went on to highlight the inverse relationship between COVID mortality and longitude. We have presented a far simpler statistical analysis in the current article, but the opposing effects of latitude and longitude on COVID mortality held true using a data set which has had another 6 months to accumulate.

Latitude and longitude also correlate to the frequencies of several HLA genotypes.28 In the current study, the alleles most strongly related to COVID severity (DRB1*01:01 and DRB1*04:01) were the only alleles to show significant, positive correlations to latitude and inverse correlations to longitude. If the results of this study are confirmed, then populations based in North Western Europe have the greatest concentrations alleles associated with both asymptomatic carriage and increased severity.

Discussion
One half (50.8%) of adult COVID-19 patients in this analysis had obesity, compared with 43.1% in the total PHD-SR sample and 42.4% nationally (5), suggesting that adults with COVID-19–associated illness and obesity might commonly receive acute care in EDs or hospitals. The findings in this report are similar to those from previous studies that indicate an increased risk for severe COVID-19–associated illness among persons with excess weight and provide additional information about a dose-response relationship between higher BMI and risk for hospitalization, ICU admission, invasive mechanical ventilation, and death (1,2). The finding that risk for severe COVID-19–associated illness increases with higher BMI suggests that progressively intensive management of COVID-19 might be needed for patients with more severe obesity. This finding also supports the hypothesis that inflammation from excess adiposity might be a factor in the severity of COVID-19–associated illness (3,8). The positive association found between underweight and hospitalization risk could be explained by uncaptured underlying medical conditions or impairments in essential nutrient availability and immune response (9).

Consistent with previous studies, the dose-response relationship between risk for hospitalization or death and higher BMI was particularly pronounced among patients aged <65 years (1,2). However, in contrast to previous studies that demonstrated little or no association between obesity and COVID-19 severity among older patients (1,2), the results in this report indicate that overweight and obesity are risk factors for invasive mechanical ventilation and that obesity or severe obesity are risk factors for hospitalization, ICU admission, and death among patients aged ≥65 years. A sensitivity analysis adjusting for other underlying medical conditions found weaker associations between BMI and severe COVID-19–associated illness, which might be partially attributable to indirect effects of obesity on COVID-19 or overadjustment by including intermediate variables on the causal pathway from exposure (i.e., BMI) to outcome.

BMI is continuous in nature, and the analyses in this report describe a J-shaped association between BMI and severe COVID-19, with the lowest risk at BMIs near the threshold between healthy weight and overweight in most instances. Risk for invasive mechanical ventilation increased over the full range of BMIs, possibly because of impaired lung function associated with higher BMI (4). These results highlight the need to promote and support a healthy BMI, which might be especially important for populations disproportionately affected by obesity, particularly Hispanic or Latino and non-Hispanic Black adults and persons from low-income households, which are populations who have a higher prevalence of obesity and are more likely to have worse outcomes from COVID-19 compared with other populations.¶¶¶

The findings in this study are subject to at least five limitations. First, risk estimates for severe COVID-19–associated illness (including hospitalization) were measured only among adults who received care at a hospital; therefore, these estimates might differ from the risk among all adults with COVID-19. Second, hospitalization risk estimates might have been affected by bias introduced by hospital admission factors other than COVID-19 severity, such as a health care professional’s anticipation of future severity. Third, only patients with reported height and weight information were included; among 238 hospitals, 28% of patients were missing height information, weight information, or both. However, results of a sensitivity analysis using multiple imputation for missing BMIs were consistent with the primary findings. Fourth, the BMI of some older adults might have been misclassified because of complex interactions between height loss and sarcopenia, a condition characterized by loss of skeletal muscle mass and function (10). Finally, although this analysis includes one of the largest samples of patients with available heights and weights to be assessed to date, the results are not representative of the entire U.S. patient population.

The findings in this report highlight a dose-response relationship between higher BMI and severe COVID-19–associated illness and underscore the need for progressively intensive illness management as obesity severity increases. Continued strategies are needed to ensure community access to nutrition and physical activity opportunities that promote and support a healthy BMI. Preventing COVID-19 in adults with higher BMIs and their close contacts remains important and includes multifaceted protection measures such as masking, as well as continued vaccine prioritization (6) and outreach for this population.