The Association of Human Leukocyte Antigen and COVID-19 in Southern China

Abstract

Human leukocyte antigen (HLA) polymorphism is hypothesized to be associated with diverse immune responses toward infectious diseases. Herein, by comparing against multiple subpopulation groups as control, we confirmed that HLA-B*15:27 and HLA-DRB1*04:06 were associated with coronavirus disease 2019 susceptibility in China. Both alleles were predicted to have weak binding affinities toward viral proteins.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has contributed to nearly 4 million deaths worldwide [1]. Delayed or inadequate treatment contributes to a large portion of coronavirus disease 2019 (COVID-19) mortality. Improving risk stratification and clinical management is crucial to identify the potentially vulnerable population. Several clinical features and genetic factors have been suggested to escalate risks of SARS-CoV-2 infection, the occurrence of severe conditions, and mortality [2].

Human leukocyte antigen (HLA), the most polymorphic genetic locus, plays a pivotal role in viral antigen recognition and presentation [3]. After decades of study, several HLA alleles have been identified to be responsible for individual differences in host response, such as viral clearance and disease prognosis [4–8]. Thus far, a few HLA alleles have been reported to be predisposed to COVID-19 infection [9–12]. However, these findings are difficult to replicate [13]. Some reasons for this are small sample size and HLA being population-dependent, but the leading cause is that these case–control studies are extra sensitive to population stratification. A previous study pointed out that, even within the same ethnic group, the choice of controls would drastically affect the perceived associations [9]. On top of that, there are complicated issues specific to COVID-19 studies. In the case of investigating patients recruited during an early stage of the pandemic in Southern China, it may be inappropriate to use the local population as control, as most cases were made up of people who fled from Wuhan. To better elucidate these situations, the use of multiple control groups has been suggested [14].

Herein, we performed an empirical study in 399 unrelated patients to explore the effect of HLA on SARS-CoV-2 infection. We highlighted the influence of subpopulation by comparing it against multiple controls. Our findings provide deeper insights into the pathogenetic mechanism of COVID-19 and lay valuable groundwork for future HLA research.

METHODS

Study Subjects

Patients with confirmed COVID-19 diagnosis in Quanzhou, Fujian (n = 42), during early 2020 were enrolled. Their demographic information, clinical characteristics, and peripheral blood samples were collected with informed consent. The HLA alleles were identified via next generation sequencing–based typing using NanoWES Human Exome, version 1.0 (Berrygenomics, Beijing, China), according to the manufacturer’s instructions. Class I and class II HLA alleles were estimated with OptiType, version 1.3.3 [15], and ATHLATES, version 1.0 [16], respectively. HLA allele frequencies and clinical information for patients in the other 2 centers (Shenzhen, n = 332, and Zhejiang, n = 82) were retrieved from prior studies [9, 17].

Choice of Control Populations

Aiming to make our findings more plausible, we performed analyses against multiple sets of controls. First, we chose 3 controls representing the general population in China, including (1) individuals from the China Marrow Donor Program (CMDP), recruited during 2010–2012 (n = 169 995) [18], referred to as General 1; (2) individuals from the CMDP, recruited during 2010–2015 (n = 812 211) [19], referred to as General 2, and (3) individuals from the deep sequencing MHC region study who were enrolled independent of the CMDP (n = 10 689) [20], referred to as General 3. To rule out associations caused by proportionate representation of the Southern China population, we again selected 2 controls resembling the population in Southern China, including (1) individuals from the Hong Kong Bone Marrow Donor Registry retrieved from the Allele Frequency Net Database (AFND; n = 5266), referred to as South 1, and (2) individuals from the CMDP recruited in Shenzhen (n = 4262) [21], referred to as South 2. HLA typing for HLA-DQA1 and HLA-DPB1 was limited and only available in General 3 and South 1.

Prediction of HLA Binding Affinities

As suggested by Nguyen et al. [22], binding affinities for all SARS-CoV-2 viral peptides were predicted with NetMHCpan-4.1 (for HLA class I) and NetMHCIIpan-4.0 (for HLA class II) [23]. Currently, only DRB1 alleles are available in NetMHCIIpan-4.0. The publicly available SARS-CoV-2 proteome sequence Wuhan-Hu-1 (NC_045512.2) was used as the reference. To inspect whether the binding pattern of certain viral proteins would have a special impact on viral vulnerability, we grouped peptides with open reading frames and highlighted binding capabilities of major viral proteins such as spike protein (S), nucleocapsid protein (N), membrane protein (M), and envelope protein (E). More precisely, we assessed the binding affinities from peptides derived from the receptor binding domain (RBD) region. We later labeled these bindings strong binder (IC50 <50 nM) and weak binder (IC50 <500 nM), following the recommendations of previous studies [24, 25]. The numbers of strong binders and weak binders were calculated per HLA allele.

Statistical Analysis

Before analyses, several quality control procedures were performed. Closely related individuals were excluded through kinship estimation in PLINK, version 1.9 [26]. Hardy-Weinberg equilibrium (HWE) was assessed in Arlequin, version 3.5 [27]. Cochran’s rule was applied to eliminate HLA alleles with a frequency <5. The following statistical analyses were performed in R, version 4.0.2: HLA heterogeneity among 3 centers was measured by pairwise comparisons using the Fisher exact test. The homogeneities among each control group were examined using the chi-square test. In addition, a t test was performed to compare the allele frequency between controls representing the general and Southern populations. Subsequently, to estimate the effect of HLA on disease susceptibility, we utilized the Fisher exact test to compare HLA allele frequencies between COVID-19 patients and selected controls. The corrected P value (P_c) was then calculated using the Benjamini-Hochberg method to adjust for multiple comparisons.

RESULTS

A total of 399 unrelated individuals with confirmed COVID-19 diagnoses were included in the present study. No departure from Hardy-Weinberg equilibrium was observed when multiple comparisons were considered. There were 14 HLA-A, 26 HLA-B, 13 HLA-C, 13 HLA-DPB1, 14 HLA-DQA1, 15 HLA-DQB1, and 22 HLA-DRB1 alleles that met the criteria for analyses. No significant HLA distribution differences were detected among the 3 centers.

Couples of alleles appeared to have significant heterogeneity among intragroup controls (full list presented in Supplementary Tables 1 and 2). Besides, 10 alleles showed significant allele frequency differences between the general and Southern populations, as detailed in Table 1. The full results of the between-group comparison are appended in Supplementary Table 3. These findings suggest that extra caution is needed when conducting studies involving such significant heterogeneity alleles among the population.

Table 1.

The Allele Frequency Difference Between the General Population and the Southern Population

Allele	General 1 AF	General 2 AF	General 3 AF	South 1 AF	South 2 AF	Intra-General P Value	Intra-South P Value	Between-Group P Value
A*02:06	0.052	0.052	0.057	0.041	0.037	.010	.163	.023
A*02:07	0.085	0.084	0.073	0.119	0.113	1.70E-08	.205	.006
B*15:18	0.013	0.014	0.016	0.003	0.005	.002	.106	.004
B*15:27	0.008	0.009	0.009	0.007	0.007	.022	1.000	.037
B*46:01	0.102	0.103	0.078	0.146	0.134	3.89E-30	.019	.020
B*58:01	0.059	0.061	0.049	0.088	0.080	1.60E-20	.052	.022
C*01:03	0.006	0.006	0.007	0.004	0.003	.337	.372	.034
C*15:02	0.034	0.033	0.037	0.028	0.026	.004	.385	.017
DQB1*02:01	0.049	0.050	0.042	0.067	0.061	6.03E-07	.118	.031
DQB1*06:01	0.102	0.102	0.094	0.113	0.109	1.35E-04	.430	.042

Abbreviation: AF, allele frequency.

We estimated the relationship between HLA and COVID-19 susceptibility by comparing cases against 3 controls representing the general population in China. We determined an HLA allele to have strong evidence for the association if the associations were observed to be significant against 2 or more controls. As illustrated in Table 2, 5 alleles were identified as risk alleles for COVID-19 vulnerability. These alleles are HLA-A*11:01, HLA-A*11:02, HLA-B*15:27, HLA-B*40:01, and HLA-DRB1*04:06. Next, we performed additional comparisons against 2 controls reflecting the Southern China population. Only HLA-B*15:27 and HLA-DRB1*04:06 remained significant. We noticed that HLA-B*15:27 was 1 of the alleles previously identified to vary between the general and Southern populations. Nevertheless, the frequency of HLA-B*15:27 in COVID-19 patients (2.2%) was significantly higher than in both groups. Similarly, we found HLA-DQA1*05:03 and HLA-DQA1*05:05 to have significantly different allele frequencies when compared with both the general and Southern China populations. However, replications of these significances cannot be achieved due to the limited typing data for DQA1.

Table 2.

The Association Between HLA and COVID-19 Susceptibility

Allele	No. (Case)	Frequency in Case, %	No. (Control)	Frequency in Control, %	Odds Ratio	P Value	Pc Value	Control Label
A*11:01	214	26.8	71 884	21.1	1.367	1.32E-04	.002*	General 1
			339 270	20.9	1.388	6.00E-05	8.39E-04*	General 2
			3970	18.6	1.607	2.15E-08	3.01E-07*	General 3
			3144	29.9	0.861	.071		South 1
			2216	26.0	1.043	.613		South 2
A*11:02	27	3.4	6517	1.9	1.792	6.01E-03	.084	General 1
			27 495	1.7	2.034	8.37E-04	.012*	General 2
			351	1.6	2.098	7.08E-04	.010*	General 3
			425	4.0	0.833	.399		South 1
			247	2.9	1.173	.442		South 2
B*15:27	16	2.2	2752	0.8	2.738	4.26E-04	.011*	General 1
			13 895	0.9	2.590	7.48E-04	.019*	General 2
			191	0.9	2.479	2.42E-03	.063	General 3
			77	0.7	3.034	2.93E-04	.008*	South 1
			63	0.7	3.001	3.80E-04	.010*	South 2
B*40:01	119	14.9	33 849	10.0	1.585	1.09E-05	2.83E-04*	General 1
			155 053	9.5	1.661	1.28E-06	3.34E-05*	General 2
			2051	9.6	1.651	2.71E-06	7.04E-05*	General 3
			1624	15.4	0.961	.760		South 1
			1194	14.0	1.076	.489		South 2
DQA1*05:03	16	2.5	183	0.9	2.998	2.22E-04	.003*	General 3
			114	1.1	2.366	3.27E-03	.046*	South 1
DQA1*05:05	86	10.8	1322	6.2	1.832	1.30E-06	1.82E-05*	General 3
			815	7.7	1.440	3.43E-03	.048*	South 1
DRB1*04:06	38	5.2	8531	2.5	2.127	4.21E-05	9.26E-04*	General 1
			41 741	2.6	2.076	5.82E-05	.001*	General 2
			538	2.5	2.121	7.30E-05	.002*	General 3
			241	2.3	2.338	1.20E-05	2.64E-04*	South 1
			219	2.6	2.076	1.47E-04	.003*	South 2

Abbreviations: COVID-19, coronavirus disease 2019; HLA, human leukocyte antigen.

*Significant.

Furthermore, to explore the potential clinical interpretation of our results, we investigated binding affinities between viral peptides and HLA alleles (the full summary is presented in Supplementary Table 4). As shown in Figure 1, HLA-B*15:27 only had 1 strong binder with spike protein (outside of the RBD region), 3 with nucleocapsid, 1 with membrane, and 0 with envelope. Meanwhile, HLA-DRB1*04:06 had 0 strong binders across all major viral proteins. These results suggest that the number of peptides recognized by HLA alleles may play a role in COVID-19 susceptibility.

Figure 1.

Numbers of strong binders derived from major viral proteins per human leukocyte antigen (HLA) allele. Only HLA alleles examined in this study are included. We labeled bindings from spike protein (S) in red, from nucleocapsid protein (N) in green, from membrane protein (M) in blue, and from envelope protein (E) in orange. More specifically, we examined the binding capabilities of peptides derived from the receptor binding domain region, which are marked in purple.

DISCUSSION

Our study confirmed that HLA-B*15:27 and HLA-DRB1*04:06, which have appeared uniquely in Asia, according to the AFND, are risk alleles for COVID-19 susceptibility. Despite that HLA-DRB1*04:06 was not significant after multiple comparison correction, both alleles have been mentioned as risk alleles in a prior study [17]. It is interesting that, while we suspect the association of HLA-A*11:01, HLA-A*11:02, and HLA-B*40:01 may be driven by population stratification, both HLA-A*11 and HLA-B*40 have been previously linked with COVID-19 infection [11, 12, 28]. Despite that HLA-B*40 did show weak peptide binding capability, HLA-A*11 exhibited moderate binding affinities in the bioinformatic prediction. Taken together, a comprehensive investigation of the impact of the subpopulation is much needed.

Additionally, DRB1*04:06 has been identified as a risk factor for autoimmune diseases [29, 30] and drug-induced maculopapular exanthema [31]. This finding is consistent with previous observations that autoimmune diseases correlate with COVID-19 severity and mortality [12, 32]. One popular explanation is that the 2 conditions share the same etiologies such as extensive inflammation. However, recent findings also suggest that the locus between DRB1 and DQA1 is related to circulating IL-6 levels [33], which often accompany cytokine storms. Taken together, these findings indicate that individual vulnerability to infection may go beyond HLA affinity.

The major limitations of this study include the partial availability of haplotype data and the exclusion of asymptomatic patients. Although strain information, viral loads, COVID-19 policies, and other confounding factors were not detailed in the present study, their impacts were diminished when restricting the study within the same country and similar time frames.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.