Genetic Determinants of Antibody-Mediated Immune Responses to Infectious Diseases Agents: A Genome-Wide and HLA Association Study.

BACKGROUND: Infectious diseases are causally related to a large array of noncommunicable diseases (NCDs). Identifying genetic determinants of infections and antibody-mediated immune responses may shed light on this relationship and provide therapeutic targets for drug and vaccine development.
METHODS: We used the UK biobank cohort of up to 10 000 serological measurements of infectious diseases and genome-wide genotyping. We used data on 13 pathogens to define 46 phenotypes: 15 seropositivity case-control phenotypes and 31 quantitative antibody measurement phenotypes. For each of these, we performed genome-wide association studies (GWAS) using the fastGWA linear mixed model package and human leukocyte antigen (HLA) classical allele and amino acid residue associations analyses using Lasso regression for variable selection.
RESULTS: We included a total of 8735 individuals for case-control phenotypes, and an average (range) of 4286 (276-8555) samples per quantitative analysis. Fourteen of the GWAS yielded a genome-wide significant (P < 5 ×10-8) locus at the major histocompatibility complex (MHC) on chromosome 6. Outside the MHC, we found a total of 60 loci, multiple associated with Epstein-Barr virus (EBV)-related NCDs (eg, RASA3, MED12L, and IRF4). FUT2 was also identified as an important gene for polyomaviridae. HLA analysis highlighted the importance of DRB1*09:01, DQB1*02:01, DQA1*01:02, and DQA1*03:01 in EBV serologies and of DRB1*15:01 in polyomaviridae.
CONCLUSIONS: We have identified multiple genetic variants associated with antibody immune response to 13 infections, many of which are biologically plausible therapeutic or vaccine targets. This may help prioritize future research and drug development.

Infectious agents have been implicated in the pathogenesis of many noncommunicable diseases (NCDs) [1]. They rank as the third leading cause of cancers worldwide [2] and are associated with multiple chronic conditions such as rheumatoid arthritis [3], Alzheimer’s dementia [4], and multiple sclerosis [5]. Measuring the antibody immune response to infections is a common approach to study this relationship, as seropositivity can both serve as a marker of infection exposure and provide clues about the pathophysiology between infections and NCDs (eg, through molecular mimicry [6]). This could potentially lead to important diagnostic and therapeutic advances, especially in the fields of vaccination and immunotherapy [7-10]. Further, given the recent global coronavirus disease 2019 pandemic, understanding immune responses to infection could help to identify common pathways that, when perturbed, influence susceptibility to infection and/or immunological response. However, the underlying pathophysiology tying infectious agents and NCDs often remains poorly understood, and their true causal effect also often remains unclear.

One approach to improve our understanding of these associations is through the study of genetic markers of infectious diseases susceptibility or host immune response. Genome-wide association studies (GWAS) have been used to identify the genetic determinants of a large array of diseases, improving our understanding of their pathophysiology and leading to therapeutic advances [11]. The infectious disease immune response is also intrinsically tied to the human leukocyte antigen (HLA) system [12], encoded by the major histocompatibility complex (MHC) gene complex. The MHC is a region of chromosome 6 with a high density of highly polymorphic genes, often in linkage disequilibrium, rendering any disease association study difficult to perform without adequate sample sizes and appropriate statistical methods [13]. However, few GWAS and HLA association studies have been published on genetic determinants of infectious diseases, and most were limited by small sample sizes [14] or relied on patient self-reported infectious diseases diagnosis [15]; both factors led to a decrease in statistical power and difficulties with clinical interpretation.

Therefore, in order to better understand the human immunological response to infectious diseases, we undertook GWAS and HLA association studies using the UK Biobank [16] cohort (UKB) in up to 10 000 serological measurements of 20 infectious diseases.

METHODS

Phenotype

We performed this analysis using the UKB’s serological measurements of infectious agents. The UKB recruited over half a million British adults between 2006 and 2010, among whom a subsample of 9724 participants provided serum samples for serological measurements of 20 different microorganisms. Samples were tested for total antibody levels against multiple antigens. These were measured using fluorescent bead-based multiplex serology technology at a dilution of 1:1000 using the Luminex 100 platform (Luminex Corporation, Austin, TX, USA). This method provides the median fluorescence intensity (MFI), a standardized quantification of the amount of antibody in the sample obtained by measuring the fluorescence emitted by the analyte-capture agent complex. Validation was performed using separate serum samples and a reference gold standard [17]. This method and the choice of seropositivity threshold have been previously validated for multiple infectious agents [17]. Finally, of the 20 original pathogens, we selected the ones with a seroprevalence of >15% for our GWAS [1] in order to ensure adequate statistical power to identify associated loci. Table 1 provides more details on the selected infectious agents.

Table 1.

Details of Infectious Agents Studied

Infectious Agents Groups	Infectious Agents	Antigen Targets	MFI Seropositivity Threshold Used for Quantitative GWAS	Number of Seropositive Samples (%) (for MFI Quantitative GWAS)	Seropositivity Definitions if Multiple Antigens Available (for Case–Control GWAS)
Herpesviridae	Herpes simplex virus-1	mgG-1	150	6199 (69.0)	n/a
	Herpes simplex virus-2	mgG-2	150	1382 (15.4)	n/a
	Epstein-Barr virus	VCA p18	250	8518 (94.8)	Positive for 2 or more antigens
		EBNA-1	250	7972 (88.7)	Total: 8477 (94.4%)
		ZEBRA	100	8191 (91.2)
		EA-D	100	7763 (86.4)
	Human cytomegalovirus	pp150	100	5136 (57.2)	Positive for 2 or more antigens
		pp52	150	5681 (63.2)	Total: 5045 (56.2%)
		pp28	200	5087 (56.6)
	Human herpesvirus-6	IE1A	100	6968 (77.6)	Three seropositivity definitions assessed separately:
		IE1B	100	7119 (79.2)	-Positive for any antigen, total: 8171 (91.0%)
					-Positive for IE1A, total: 6968 (77.6%)
		p101k	100	1951 (21.7)	-Positive for IE1B, total: 7119 (79.2%)
	Human herpesvirus-7	U14	100	8528 (94.9)	n/a
	Varicella zoster virus	Glycoproteins E and I	100	7595 (84.5)	n/a
Polyomaviridae	Human polyomavirus BKV	BK VP1	250	8555 (95.2)	n/a
	Human polyomavirus JCV	JC VP1	250	5118 (57.0)	n/a
	Merkel cell polyomavirus	MC VP1	250	5915 (65.8)	n/a
Bacteria	Chlamydia trachomatis	momp A	100	964 (10.7)	Seropositive if either:
		momp D	100	1371 (15.3)	-Positive for pGP3
		tarp-D F1	100	1635 (18.2)	-Negative for pGP3 but positive for 2 out of 5 remaining antigens (but momp D and momp A can only contribute once together)
		tarp-D F2	100	2074 (23.1)
		PorB	80	273 (3.04)
		pGP3	200	1784 (19.9)	Total: 1784 (19.9%)
	Helicobacter pylori	CagA	400	985 (11.0)a	Positive for 2 or more antigens, except for CagAa
		VacA	100	1571 (17.5)	Total: 2674 (29.8%)
		OMP	170	2640 (29.4)
		GroEL	80	2716 (30.2)
		Catalase	180	1558 (17.3)
		UreA	130	2251 (25.1)
Parasite	Toxoplasma gondii	p22	100	1308 (14.6)	Positive for either antigen
		sag1	160	3919 (43.6)	Total: 2449 (27.3%)

This table is adapted from the UKB documentation [17]: infectious agents selected for the GWAS analyses, the antigen targets used for antibody measurements, and the seropositivity definitions used for case–control status. All counts and ratios are restricted to the 8761 samples from white British individuals.

Abbreviations: UKB, UK Biobank; GWAS, genome-wide association study.

aCagA was not available for all participants and was therefore not used for seropositivity calculation.

Genome-Wide Association Study

For the GWAS, we used the UKB Version 3 imputed genotype data set, with genome-wide genotyping data available for 488 000 UK Biobank participants. Details on the collection of this data can be found elsewhere [16]. For each of the selected microorganisms, we performed 3 GWAS. First, we performed case–control analyses by splitting participants into seropositive and seronegative samples, based on the UKB suggested seropositivity definitions (Table 1). These GWAS aim to identify genetic variants associated with previous infections to each given pathogen. Second, we performed quantitative analyses using antibody MFI measurements. Given that all serological tests are at risk of low-level cross-binding with other nonspecific antibodies that are not representative of infection [18], we restricted the quantitative analyses to samples above the seropositivity threshold, using thresholds suggested by UKB (Table 1). Therefore, these GWAS aim to identify genetic variants responsible for varying antibody-mediated immune responses within the seropositive population. Third, given the risk of heavily skewed data leading and inflation of variance with higher MFI (in violation of linear regression assumptions), we performed the same quantitative analysis but using a logarithmic transform on antibody MFI.

For all GWAS, we restricted our analysis to White British individuals, as identified using principal component analysis [19]. This was performed to minimize bias from population stratification, which tends to confound the relationship between the genetic variants and the phenotype of interest [20]. For the 2 quantitative analysie, we standardized the data to have a mean of 0 and a standard deviation of 1 before applying the algorithm. We performed all GWAS using the fastGWA mixed model package [21] with the following covariates: sex, age, UKB assessment center visited, and the first 20 principal components. We excluded single nucleotide polymorphisms (SNPs) with minor allele frequencies <1%. A prespecified P value threshold of P < 5 ×10-8 was then used to identify genome-wide significant SNPs [22]. Lead genetic variants at each loci were identified using the plink software (276 to 8555) [23, 24] clump function with a linkage disequilibrium threshold of 0.2 and a physical distance threshold of 500 kilo-base pairs.

HLA Association Study

Next, we performed HLA association studies on the same 3 phenotypes as those defined above for our GWAS. This was also done for each of the selected pathogens. HLA classical allele imputation was performed by the UKB using the HLA*IMP:02 algorithm [16]. The following HLA class I and II genes were studied: A, B, C, DRB1, DRB3, DRB4, DRB5, DPA1, DPB1, DQA1, and DQB1. As recommended by the UKB [25], we set alleles with a posterior probability call of <0.7 to a copy number of 0. Similarly, the UKB used the 99:01 suffix to indicate that no DRB3-4–5 alleles were present and recommends setting them to 0. Given the high degree of collinearity between HLA gene alleles and the large number of statistical tests planned, we used Lasso regression for variable selection. Lasso imposes a penalty on models selecting larger variable numbers and is therefore less likely to uncover false associations than stepwise regression methods [26]. The analysis used all HLA allele copy numbers, age, sex, and the first 20 principal components as variables from which to select a sparse model. Analyses were also restricted to White British individuals. To further reduce false associations, we performed 10-fold cross-validation 100 times and only selected variables that were chosen by Lasso in at least 95 of our analyses, using the 1 standard error selection rule [26]. In cases of repeated measurements from the same individuals, we average their results (for quantitative MFI phenotypes), and an individual was called seropositive if at least 1 serology test was above the seropositivity threshold (for case–control phenotypes). Analyses were performed using the glmnet package (version 2.0.16) on R (version 3.5.0).

Amino Acid Residue Association Study

Given the highly polymorphic HLA genes, different alleles may encode similar sequences of amino acids, and studying the association between diseases, those amino acid residue sequences can be a more powerful and informative statistical analysis [27]. To do this, we first used the IMGT/HLA database [28] to translate each HLA allele copy number (as defined above) to amino acid copy numbers. In cases where the specific amino acid residue was unknown (ie, an asterisk in the IMGT/HLA database), we set it to 0. We used the same Lasso analysis pathway described above, using amino acid copy numbers, age, sex, and the first 20 principal components. Repeated serology measurements were handled as above. Note that given that some alleles and amino acid residues are not commonly found in every population, amino acid residues with 100% correlation were analyzed as the same and were reported as such in the results.

Review of Previously Reported Associations

We used the NHGRI-EBI GWAS catalog [29] and the Phenoscanner [30, 31] tool to review previously reported disease associations with the genome-wide significant SNPs found in our GWAS, and with the alleles selected using the HLA allele Lasso analyses. Note that the Phenoscanner uses SNP rsID rather than allele identifiers for the HLA region. We restricted our results to genome-wide significant associations.

Patient Consent Statement

Consent was obtained by the UKB for every enrolled participant.

RESULTS

A total of 13 pathogens were chosen from the original list of infectious disease agents (Table 1). Most microbes were viruses of the Herpesviridae family, though the 3 Polyomaviridae viruses most commonly associated with human NCDs [32] were also selected. Seroprevalence was highly variable, with Epstein-Barr virus (EBV) antibodies found in >94% of individuals and Chlamydia trachomatis in <20%. There were a total of 8735 White British individual participants (55.9% female), providing a total of 8984 individual samples. There was an average (range) of 4286 (276–8555) samples used for quantitative analyses. The median age (interquartile range) was 58 (51–64) years at the time of enrollment into the UKB.

As suspected before performing the analyses, log-transformed MFI analyses had more stable estimates than their untransformed counterparts. Visual inspection of the untransformed antibody MFI GWAS Manhattan plots showed a large amount of genome-wide significant loci, even in analyses with smaller sample sizes (results not shown). This suggests that the variance-stabilizing logarithmic transform was able to reduce most of the noise observed in the signals. Therefore, here we will only report on results from case–control and log-transformed antibody MFI analyses. Manhattan plots from these analyses are available in Supplementary Data 1.

GWAS

A total of 46 GWAS were performed: 15 case–control analyses and 31 logarithm-transformed MFI analyses. Genome-wide significant loci are shown in Table 2. Genomic inflation factors (Supplementary Data 2) were smaller than 1.04 for all GWAS, except for momp D MFI (log-transformed) at 1.24. This analyses also yielded 22 separate loci using LD clumping, the largest number of genome-wide significant peaks. However, this was one of the analysis with the smallest sample size (1371 individuals). Given the genomic inflation, an underlying population stratification and an elevated rate of false-positive associations should be suspected.

Table 2.

Lead Significant Variants per GWAS as Determined by LD Clumping

Agent	Analysis	Variant	CHR	Beta	SE	P Value	Effect Allele	Other Allele	Effect Allele Frequency	Overlapping or Nearest Gene
BKV	VP1 MFI (log)	rs492602	19	–0.09	0.015	4.3e-09	A	G	0.479	FUT2
C. trachomatis	Seropositivity case–control	rs143335233	2	–0.14	0.023	5.4e-09	C	T	0.983	NCK2
	PorB MFI (log)	rs74725117	3	–1.71	0.31	4.5e-08	T	G	0.987	TP63
		Chr4:13265941	4	–1.53	0.28	3.1e-08	CT	G	0.983	RNU6-962P
		rs201129973	6	–1.74	0.31	3.1e-08	G	A	0.981	RP11-302L19.1
		rs61957300	13	–1.28	0.23	3.6e-08	A	C	0.975	TPTE2P2
		rs140031044	18	–2.05	0.33	7.4e-10	A	G	0.989	CTD-2008L17.1
	pGP3 MFI (log)	Chr7:66874490	7	–0.32	0.059	4.2e-08	CTCTT	C	0.916	AC006480.1
CMV	pp28 MFI (log)	rs12698418	7	0.13	0.024	4.4e-08	G	A	0.248	EN2
EBV	Seropositivity case–control	rs71437272	13	0.07	0.012	2.1e-08	C	T	0.980	RASA3
	VCA MFI (log)	rs9379862	6	0.10	0.018	1.1e-08	T	C	0.740	BTN3A2
		rs9271536	6	0.19	0.020	7.3e-21	A	T	0.180	HLA-DQA1 (MHC)
	ZEBRA MFI (log)	rs34034915	6	0.10	0.017	3.3e-09	T	TG	0.660	RP1-97D16.1
		Chr6:32597087	6	–0.29	0.016	3.0e-75	CA	C	0.637	HLA-DQA1 (MHC)
	EBNA-1 MFI (log)	rs67886110	3	0.09	0.016	2.1e-08	G	T	0.599	MED12L
		rs6927022	6	0.30	0.016	1.9e-76	A	G	0.505	HLA-DQA1 (MHC)
	EA-D MFI (log)	rs2316515	6	0.09	0.016	4.2e-08	A	G	0.388	IRF4
		rs2395192	6	–0.13	0.016	3.1e-16	C	T	0.440	HLA-DRB9 (MHC)
		rs73067509	7	0.14	0.025	2.2e-08	C	G	0.892	AC004538.3
HHV6	IE1A MFI (log)	rs13079586	3	–0.10	0.018	4.6e-08	C	T	0.674	ITGA9
		rs2844606	6	–0.10	0.017	1.2e-08	A	G	0.584	RPL3P2 (MHC)
	IE1B MFI (log)	rs28752523	6	–0.12	0.021	7.9e-09	C	T	0.806	HLA-DQA1 (MHC)
HHV7	U14 MFI (log)	rs139299944	6	0.11	0.016	4.1e-12	C	CT	0.658	HLA-DQA1 (MHC)
		rs75438046	11	0.27	0.047	1.0e-08	G	A	0.973	CXCR5
		rs1808192	17	–0.09	0.016	6.8e-09	A	G	0.329	TBKBP1
H. pylori	OMP MFI (log)	rs3104361	6	–0.18	0.029	6.5e-10	T	C	0.418	HLA-DQB1 (MHC)
	UreA MFI (log)	rs71569678	6	–0.34	0.061	3.0e-08	A	C	0.940	RP11-439H9.1
HSV2	Seropositivity case–control	rs538162817	3	–0.14	0.025	2.9e-08	T	C	0.988	GRK7
		rs7503464	17	0.03	0.0054	1.2e-08	G	A	0.484	CTD-3195I5.5
	mgG-2 MFI (log)	rs144232229	3	–1.0	0.17	1.9e-09	G	A	0.989	TAMM41
		Chr5:166282898	5	0.73	0.13	4.6e-08	CT	C	0.020	CTB-63M22.1
		rs117705146	8	–0.94	0.16	3.8e-09	G	A	0.986	ADAM32
JCV	Seropositivity case–control	rs17843569	6	–0.15	0.010	1.7e-47	C	G	0.157	HLA-DQA1 (MHC)
		rs2432132	19	–0.06	0.0082	8.8e-15	C	T	0.380	FUT2
	VP1 MFI (log)	rs1610401	1	0.35	0.063	1.9e-08	G	C	0.026	LRRN2
		rs374949924	6	–0.26	0.026	4.5e-24	G	A	0.198	HLA-DRB6 (MHC)
MCV	Seropositivity case–control	rs55792153	5	0.05	0.0082	3.6e-10	A	C	0.255	TMEM173
		rs9269771	6	–0.12	0.0096	7.3e-37	T	C	0.252	HLA-DRB1 (MHC)
	VP1 MFI (log)	rs7444313	5	0.16	0.021	6.5e-15	G	A	0.273	TMEM173
		rs76148407	6	0.31	0.052	1.8e-09	C	G	0.967	HIST1H4PS1
		rs114708114	6	0.37	0.066	2.1e-08	C	T	0.980	OR2E1P
		rs75040706	6	0.30	0.054	3.4e-08	A	G	0.970	RP1-86C11.7
		rs28393149	6	–0.26	0.029	1.0e-18	C	G	0.882	HLA-DRB6 (MHC)
T. gondii	sag1 MFI (log)	rs541989586	3	–0.26	0.047	3.9e-08	A	AT	0.934	SOX2-OT
		rs148929820	5	–0.44	0.072	8.5e-10	G	A	0.975	RP11-510I6.1
		rs11881343	19	–0.47	0.086	3.9e-08	A	T	0.982	NOTCH3
VZV	Seropositivity case–control	rs1766	6	–0.04	0.0055	1.1e-11	A	G	0.530	HLA-DQB1 (MHC)
	Glycoproteins E and I MFI (log)	rs13197633	6	–0.20	0.025	1.4e-15	G	A	0.880	TOB2P1
		rs34073492	6	–0.19	0.025	5.8e-14	C	T	0.877	RP11-457M11.5
		rs56401801	6	–0.19	0.025	6.7e-14	T	A	0.883	VN1R10P
		rs13204572	6	–0.16	0.026	1.1e-09	G	C	0.892	HIST1H4D
		rs1048381	6	–0.22	0.021	6.3e-25	G	A	0.815	HLA-DQA1 (MHC)

Where applicable, variant positions are given using the GRCh37/hg19 human genome assembly. For variants within the major histocompatibility complex, only the variant with lowest P value is reported.

Abbreviations: BKV, BK polyomavirus; CMV: cytomegalovirus; CHR, chromosome; EBV: Epstein-Barr virus; HHV6, human herpes virus 6; HHV7, human herpes virus 7; HSV2, herpes simplex virus type 2; JCV, JC polyomavirus; MCV, Merkel cell polyomavirus; MFI, mean fluorescence intensity; MHC, major histocompatibility complex; SE, standard error; VZV: varicella zoster virus.

Reassuringly, 14 analyses showed a locus at the MHC (Figure 1), with EBV (P = 1.9 ×10-76 to 3.1 ×10-16), JC virus (JCV; P = 1.7 ×10-47 to 4.5 ×10-24), and Merkel cell virus (MCV; P = 7.3 ×10-37 to 1.0 ×10-18) showing the strongest associations. Interestingly, the FUT2 gene on chromosome 19 was identified in both the case–control JCV seropositivity analysis and the BK virus VP1 log-transformed MFI analysis. Otherwise, no other genetic loci were identified more than once, including within infectious agent families.

Figure 1. 

QQ plot and Manhattan plots from selected GWAS from the seropositivity case–control and antibody log-transformed MFI analyses. Each dot on a Manhattan plot (right) represents the P value (y-axis, on a logarithmic scale) associated with the association test at a genetic variant. Values above the dashed line are considered genome-wide significant. QQ plots (left) show the observed P values (y-axis) against the expected P values (x-axis). Any deviance from the red line suggests that the effect seen is not explained only by chance alone. As can be seen, the MHC is a commonly identified locus. A, EBV EA-D MFI (log). B, EBV EBNA-1 MFI (log). C, EBV ZEBRA MFI (log). D, JCV seropositivity case–control. Abbreviations: EBV, Epstein-Barr virus; JCV, JC polyomavirus; GWAS, genome-wide association studies; MHC, major histocompatibility complex; MFI, mean fluorescence intensity.

HLA Allele Association Analyses

EBV antibody MFIs (EA, EBNA-1, and ZEBRA) were associated with the most HLA alleles, with Lasso selecting 10 alleles for EBNA-1. The DQB1*02:01 and DQA1*03:01 alleles were found in 2 out of the 3 EBV antibody analyses (EA-D, EBNA-1, and ZEBRA) in Table 1. Interestingly, despite a strong MHC locus in the GWAS, Lasso did not select any alleles for VCA antibody MFI. This is most likely because the effect sizes and collinearity from any of the HLA alleles were likely too small to confidently select them in a multivariate analysis such as Lasso. Among the Polyomaviridae, JCV and MCV replicated their GWAS finding, with DRB1*15:01 and DRB5*01:01 (also identified in EBNA-1) being selected in both antibody MFI analyses. Of note, these 2 HLA alleles are in almost complete linkage disequilibrium in European populations [33, 34]. Finally, varicella zoster virus antibody MFI was associated with 3 HLA alleles, 1 of which (DQB1*02:01) was also associated with 2 EBV antibodies (EA-D and EBNA). Results from the HLA allele association studies are summarized in Table 3.

Table 3. 

HLA Allele Association Analysis Summary

Agent	Analysis	Selected Components	Univariate Effect (95% CI)	Univariate P Value	Multivariate Effect (95% CI)	Multivariate P Value
EBV	EA-D MFI (log)	B*08:01	–0.16 (–0.20 to –0.11)	6.17 ×10-12	–0.065 (–0.13 to –0.0011)	.05
		DRB1*09:01	0.40 (0.27 to 0.54)	8.67 ×10-9	0.24 (0.081 to 0.40)	.003
		DQB1*02:01	–0.17 (–0.21 to –0.12)	1.38 ×10-13	–0.092 (–0.16 to –0.028)	.005
		DQA1*03:01	0.11 (0.070 to 0.15)	2.96 ×10-8	0.072 (0.032 to 0.11)	.0004
		DQB1*03:03	0.18 (0.11 to 0.25)	7.3 ×10-7	0.10 (0.019 to 0.18)	.02
		Sex	–0.23 (–0.27 to –0.19)	<2 ×10-16	–0.24 (–0.29 to –0.20)	<2 ×10-16
		Age	0.007 (0.005 to 0.01)	1.35 ×10-7	0.064 (0.042 to 0.086)	1.04 ×10-8
		PC1	0.08 (0.06 to 0.11)	9.61 ×10-14	0.078 (0.056 to 0.10)	4.16 ×10-12
	EBNA-1 MFI (log)	DRB4*01:03	–0.17 (–0.20 to –0.13)	<2 ×10-16	–0.11 (–0.15 to –0.067)	3.75 ×10-7
		DRB3*02:02	0.21 (0.17 to 0.26)	<2 ×10-16	0.16 (0.11 to 0.21)	3.88 ×10-10
		DRB1*07:01	–0.22 (–0.27 to –0.18)	<2 ×10-16	–0.089 (–0.45 to 0.27)	.63
		DRB1*12:01	0.53 (0.40 to 0.67)	3.76 ×10-15	0.30 (0.16 to 0.43)	1.57 ×10-5
		DRB1*15:01	0.26 (0.22 to 0.31)	<2 ×10-16	0.14 (0.061 to 0.22)	.0005
		DQB1*02:01	–0.24 (–0.28 to –0.19)	<2 ×10-16	–0.25 (–0.30 to –0.20)	<2 ×10-16
		DQA1*01:02	0.23 (0.19 to 0.27)	<2 ×10-16	0.058 (–0.016 to 0.013)	.12
		DQA1*02:01	–0.22 (–0.27 to –0.18)	<2 ×10-16	–0.091 (–0.45 to 0.27)	.62
		DPB1*03:01	–0.20 (–0.25 to –0.14)	1.99 ×10-12	–0.19 (–0.25 to –0.14)	5.49 ×10-12
		DPB1*04:02	0.16 (0.11 to 0.21)	4.96 ×10-10	0.11 (0.058 to 0.16)	2.78 ×10-5
	ZEBRA MFI (log)	DRB4*01:01	0.28 (0.23 to 0.34)	<2 ×10-16	0.074 (–0.000 29 to 0.15)	.05
		DRB1*03:01	–0.20 (–0.24 to –0.16)	<2 ×10-16	–0.081 (–0.12 to –0.038)	.0002
		DRB1*04:04	0.45 (0.37 to 0.53)	<2 ×10-16	0.36 (0.27 to 0.46)	3.19 ×10-14
		DQB1*03:02	0.24 (0.19 to 0.28)	<2 ×10-16	0.045 (–0.029 to 0.12)	.23
		DQB1*04:02	0.36 (0.26 to 0.46)	1.18 ×10-11	0.45 (0.35 to 0.55)	<2 ×10-16
		DQA1*02:01	0.27 (0.23 to 0.31)	<2 ×10-16	0.29 (0.23 to 0.35)	<2 ×10-16
		DQA1*03:01	0.17 (0.13 to 0.20)	<2 ×10-16	0.13 (0.075 to 0.18)	1.29 ×10-6
		Sex	–0.27 (–0.31 to –0.23)	<2 ×10-16	–0.29 (–0.33 to –0.25)	<2 ×10-16
		Age	0.051 (0.029 to 0.072)	4.17 ×10-6	0.059 (0.039 to 0.080)	2.43 ×10-8
JCV	Seropositivity case–controla	DRB5*01:01	0.55 (0.50 to 0.60)	<2 ×10-16	1.02 (0.51 to 1.55)	.95
		DRB1*15:01	0.55 (0.50 to 0.59)	<2 ×10-16	0.63 (0.31 to 1.26)	.19
		DQB1*06:02	0.55 (0.21 to 0.60)	<2 ×10-16	0.94 (0.64 to 1.39)	.75
		DQA1*01:02	0.61 (0.57 to 0.66)	<2 ×10-16	0.89 (0.78 to 1.02)	.11
		Sex	1.25 (1.15 to 1.36)	1.87 ×10-7	1.26 (1.16 to 1.37)	8.06 ×10-8
	VP1 MFI (log)	DRB5*01:01	–0.29 (–0.35 to –0.23)	<2 ×10-16	–0.29 (–0.35 to –0.23)	<2 ×10-16
MCV	VP1 MFI (log)	A*29:02	–0.41 (–0.50 to –0.31)	<2 ×10-16	–0.41 (–0.50 to –0.32)	<2 ×10-16
		DRB1*04:04	–0.39 (–0.49 to –0.29)	1.9 ×10-14	–0.41 (–0.51 to –0.31)	7.01 ×10-16
		DRB1*15:01	–0.19 (–0.24 to –0.13)	1.67 ×10-11	–0.12 (–0.36 to 0.11)	.30
		DQB1*05:01	0.21 (0.16 to 0.27)	1.53 ×10-15	0.068 (–0.055 to 0.19)	.28
		DQB1*06:02	–0.18 (–0.24 to –0.13)	2.62 ×10-11	–0.059 (–0.29 to 0.18)	.62
		DQA1*01:01	0.21 (0.16 to 0.26)	<2 ×10-16	0.10 (–0.016 to 0.22)	.09
VZV	Glycoproteins E and I MFI (log)	A*01:01	0.17 (0.13 to 0.20)	<2 ×10-16	0.089 (0.039 to 0.14)	.0004
		B*08:01	0.20 (0.16 to 0.25)	<2 ×10-16	0.040 (–0.033 to 0.11)	.28
		DQB1*02:01	0.21 (0.17 to 0.25)	<2 ×10-16	0.14 (0.080 to 0.20)	8.67 ×10-6
		Sex	0.21 (0.16 to 0.25)	<2 ×10-16	0.21 (0.17 to 0.26)	<2 ×10-16

As explained in the text, the untransformed antibody MFI analyses are not shown. Only pathogens with Lasso-selected alleles are reported. Age is on the standardized scale. Only variables selected by Lasso are shown. PC1: first principal component (effect is on the standardized scale). Multivariate effects stand for the effects observed when all alleles selected by Lasso (for each respective infection) are included in the same regression analysis.

Abbreviations: EBV, Epstein-Barr virus; JCV, JC polyomavirus; MCV, Merkel cell polyomavirus; MFI, mean fluorescence intensity; VZV, varicella zoster virus.

aEffects for the JCV seropositivity case–control analysis are reported as odds ratios.

HLA Amino Acid Residue Association Analyses

Lasso selected amino acid residues from most selected alleles in the HLA allele association above, and also selected amino acid residues from proteins encoded by genes whose alleles were not previously selected. There were some differences, most significantly for the ZEBRA MFI analysis. This is likely to be due to the high rate of multicollinearity between amino acid residues encoded by MHC genes, which even Lasso was not able to entirely untangle. The full results can be found in Supplementary Data 3.

Of the 88 associations obtained from either GWAS or HLA allele Lasso analysis, the NHGRI-EBI GWAS catalog found previously reported associations in 17 variants. In total, there were 73 previously reported associations, 57 in the HLA regions. Using the Phenoscanner and the 52 loci obtained from our GWAS, 24 had previously reported associations (total of 453 associations), of which 14 were in non-HLA regions (total of 271 associations). With both tools, associations included a wide range of diseases, including autoimmune diseases, white blood cell counts, body mass index, or respiratory functions (Supplementary Data 4 and Supplementary Data 5).

DISCUSSION

Infectious diseases are a major contributor to the global burden of diseases and play a significant role in many NCDs. Given that multiple factors, both heritable and environmental, contribute to their transmission, their acquisition, and the human host response, studying their genetic determinants can be challenging. In this study, we have used measurements from the UKB’s serological results to perform the largest reported genetic association study of infectious diseases antibody measurements to date. While the HLA region was commonly associated with the host’s antibody-mediated immune responses, we have uncovered multiple biologically plausible genetic determinants of infectious diseases. For example, for EBV, we found loci at RASA3 [35], MED12L [36], and IRF4 [37], which are related to viral transcription or related to EBV-associated carcinogenesis. Similarly, the TMEM173-encoded STING [38] protein is associated with MCV-caused Merkel cell carcinoma [39]. Further, while our results cannot prove causality, multiple previously reported epidemiological associations support our methodology. For example, allele DRB1*15:01 was associated with both JCV analysis and EBV ZEBRA analysis. DRB1*15:01 is the main genetic risk factor for multiple sclerosis [40], a disease previously associated with EBV [41]. DRB1*15:01 was also selected by Lasso as a predictor of JCV serology, the cause of progressive multifocal leukoencephalopathy, another demyelinating disease [42] that can also be triggered when giving immunosuppression to patients with multiple sclerosis [43]. Similarly, FUT2 was associated with both BKV and JCV, 2 Polyomaviridae with high genetic homology [44], and is therefore likely to be truly associated with their pathophysiology. FUT2 is a major determinant of the blood group secretor status and has been associated with multiple other viruses [45] and NCDs, most notably kidney diseases [46, 47]. Uncovering these genetic determinants may allow for better informed drug and vaccine development, which typically is a long and expensive process. Similar genetics-informed prioritization of therapeutic targets for further drug and vaccine development has previously been used in multiple diseases with success [48].

There are few published GWAS on human infectious diseases [14, 15, 49–52], and each has employed a different methodology to enroll patients, making comparisons difficult. Our study used serology data, but others have used prospective patient enrollment [52] or electronic medical record data [50, 51], and the biggest GWAS on infectious diseases to date used self-reported history using questionnaires [15]. A similar preprint using the same data as ours is currently accessible and found similar significant associations [53]. However, the authors performed a slightly different analysis that did not account for intrinsically limited serology testing specificity or UKB HLA quality control recommendations. Nevertheless, as we found here, the HLA emerges as an important risk locus across most published GWAS.

Our study’s main strength lies in its careful methodology. Given the large number of statistical tests performed and the inherent diagnostic limitations of serological tests, we made multiple analytic choices to ensure interpretable results while lowering the risk of false associations. Most notably, we used Lasso regression to select HLA alleles and amino acid residues, and we used logarithmic transform to stabilize the variance of the antibody MFI analyses.

However, given that serological tests may have multiple possible interpretations, our results should be interpreted with caution. That is because the differential diagnosis of a negative serological test includes having never been in contact with the infectious agent, the host not being able to mount an antibody mediated response, or antibodies not being a good proxy for either contact or immune response [54]. Alternatively, a positive antibody titer may be explained by cross-reactivity with other antigens, especially if the antibody titers are low [55]. Further, antibody levels are known to vary in time due to multiple host and environmental factors [56]. To better assess antibody-mediated response in hosts with a likely exposure, we have limited our quantitative analyses to individuals above the seropositivity threshold. Nevertheless, given that the population was randomly selected from the UKB cohort, we cannot rule out that unmeasured environmental or socioeconomic confounders [57] may have affected our results.

Our study has several limitations that can help guide future efforts in studying genetic determinants of infectious diseases. First, ideally future serological studies should be performed in individuals with a clear history of exposure (or lack thereof) to the infectious agent. This would increase the serological test’s specificity and improve the chance of finding clinically significant genetic associations. Second, as the environment is a major nonheritable determinant of infectious diseases, it should be factored in the design of future genetic studies. Third, as individuals are possibly constantly re-exposed to infectious diseases, longitudinal follow-up and serial measurements of both IgM and IgG would also improve the chance of finding clinically important genetic variants. Lastly, our results are mostly hypothesis-generating, and despite the fact that many of the loci we found to be associated with infectious diseases have been associated with other diseases, our study needs to be followed by in vitro or in vivo studies to establish true causal associations. This is especially true given that most of the associations we found were in the HLA, a region of the human genome that does not lend itself well to further in silico study.

In summary, here we present the largest GWAS and HLA association analyses on infectious diseases and the resultant host antibody–mediated immune response to date. While this work is hypothesis-generating and should not be used to infer causality between infections and NCDs without further research, with careful planning, we hope that future genetic studies will lead to further advances in our understanding of the interplay between, host, environment, and disease pathophysiology.

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.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.