HLA typing using genome wide data reveals susceptibility types for infections in a psychiatric disease enriched sample

Abstract

Background:

The infections Toxoplasma gondii (T. gondii), cytomegalovirus, and Herpes Simplex Virus Type 1 (HSV1) are common persistent infections that have been associated with schizophrenia and bipolar disorder. The major histocompatibility complex (MHC, termed HLA in humans) region has been implicated in these infections and these mental illnesses. The interplay of MHC genetics, mental illness, and infection has not been systematically examined in previous research.

Methods:

In a cohort of 1636 individuals, we used genome-wide association data to impute 7 HLA types (A, B, C, DRB1, DQA1, DQB1, DPB1), and combined this data with serology data for these infections. We used regression analysis to assess the association between HLA alleles, infections (individually and collectively), and mental disorder status (schizophrenia, bipolar disorder, controls).

Results:

After Bonferroni correction for multiple comparisons, HLA C*07:01 was associated with increased HSV1 infection among mentally healthy controls (OR 3.4, p = 0.0007) but not in the schizophrenia or bipolar groups (P > 0.05). For the multiple infection outcome, HLA B* 38:01 and HLA C*12:03 were protective in the healthy controls (OR ≈ 0.4) but did not have a statistically-significant effect in the schizophrenia or bipolar groups. T. gondii had several nominally-significant positive associations, including the haplotypes HLA D RB*03:01 ~ HLA DQA*05:01 ~ HLA DQB*02:01 and HLA B*08:01 ~ HLA C*07:01.

Conclusions:

We identified HLA types that showed strong and significant associations with neurotropic infections. Since some of these associations depended on mental illness status, the engagement of HLA-related pathways may be altered in schizophrenia due to immunogenetic differences or exposure history.

1. Introduction

Multiple genetic, environmental, and immunologic factors have been implicated in schizophrenia as well as the pathogenesis of bipolar disorder, but no definitive neurobiological mechanisms have been established. For schizophrenia, multiple genome wide association studies (GWAS) have consistently implicated polymorphisms across the broad major histocompatibility complex (MHC) region (Corvin and Morris, 2014; Schizophrenia Working Group of the Psychiatric Genomics, 2014).

In humans, the MHC region (termed the HLA system) encompasses roughly 4.5 million base pairs containing more than 200 genes, many of them involved in the immune system through coding for HLA proteins (Brucato et al., 2015; Delves, 2014). The Class I MHC genes (HLA A,B,C) code for transmembrane glycoproteins present on the surface of nucleated cells (Delves, 2014). These molecules present antigens from inside of cells to CD8+ T cells, which signals the CD8 cytotoxic T cells that a cell is infected and a target of cytolysis (Delves, 2014). Class II MHC molecules (HLA DP, DQ, DR, DM) are typically only present on specific cells (B cells, macrophages, dendritic cells) and present antigens to T-Helper cells (CD4) to stimulate their division and maturation (Delves, 2014).

The HLA region is extremely genetically diverse, providing the immune system with greater flexibility to respond to a wide array of co-evolving pathogens (Corvin and Morris, 2014). Additionally, much of the region is known to be in linkage disequilibrium (LD) (Corvin and Morris, 2014). Due to LD there exists common HLA haplotypes in certain groups – some have been identified in the Ashkenazi population primarily for the HLA loci A ~ B ~ DRB (Bonné-Tamir et al., 1978; Klitz et al., 2010).

GWAS studies implicating the MHC region in schizophrenia built on several decades of prior work indicating that schizophrenia is associated with certain HLA class I and class II alleles, which provided one explanation for the link between several autoimmune diseases and schizophrenia (Wright et al., 2001). Considering that the primary role of immune molecules in MHC is to defend the host against pathogens, HLA polymorphisms might also provide a mechanistic nexus for the role of various infections in the etiology of schizophrenia (Pearce, 2003). However, there are few studies in humans examining HLA I and II alleles in relation to specific neurotrophic infections that have been implicated in schizophrenia. There is also some evidence for a genetic association with bipolar disorder and MHC genotypes, but these findings are less consistent and compelling than for the schizophrenia association (Corvin and Morris, 2014; Figueiredo and de Oliveira, 2012; Williams et al., 2011).

In the current study we examine HLA alleles in relation to several neurotropic infections that are pertinent to schizophrenia and bipolar disorder. Toxoplasma gondii (T. gondii) is a neurotropic parasite. T. gondii resides in the brain as tissue cysts, which likely persist for the life of the host and can become reactivated (e.g. upon immunosuppression) to replicate and cause neurological disease. Even in the absence of overt reactivation, there is evidence that T. gondii can cause subtle behavioral alterations in humans (Tedford and McConkey, 2017).

Numerous studies and subsequent reviews have described a significant association between T. gondii infection (often assessed as presence of anti-T. gondii antibodies) and schizophrenia (Torrey et al., 2007; Torrey and Yolken, 2003). More recent studies have found that T. gondii is significantly associated with bipolar disorder and suicide risk (Pearce et al., 2012; Postolache and Cook, 2013). Given that bipolar disorder and schizophrenia likely share genetic underpinnings and mechanisms, this strengthens the case for a common or overlapping immunogenetic contribution to both disorders (Corvin and Morris, 2014; Lichtenstein et al., 2009).

Despite excellent studies of T. gondii peptide interactions with MHC at the molecular level, there is scant data in humans on the role of specific HLA alleles in susceptibility to T. gondii infection. A recent study found that HLA alleles associated with accelerated progression from HIV infection to AIDS were associated with resistance to toxoplasmic encephalitis (Rodrigues et al., 2016). A few small studies have found specific HLA alleles to be associated with either T. gondii infection or worse outcomes of that infection (Mack et al., 1999; Meenken et al., 1995; Shimokawa et al., 2016). Altogether, there has not been a large systematic study of this relationship even in healthy populations. Our current study fills this gap.

Cytomegalovirus (CMV) has a high prevalence in most populations but is typically asymptomatic in immune-competent individuals (Cannon et al., 2010). A few studies have found that the seroprevalence of CMV is higher in schizophrenia and bipolar disorder than in controls (Leweke et al., 2004; Tedla et al., 2011; Torrey et al., 2012). These data have not been entirely consistent but anti-psychotic treatment lowers the presence of CMV antibodies, which could make infection more difficult to detect (Leweke et al., 2004). Previous research has also implicated genetic variants outside of the MHC locus as interacting with CMV infection to increase schizophrenia risk (Avramopoulos et al., 2015; Børglum et al., 2014; Carter, 2008; Grove et al., 2014).

Herpes Simplex Virus Type 1 (HSV1) has a reported seroprevalence of 54% in the United States (Bradley et al., 2013). As with the previously discussed pathogens, HSV-1 infections are frequently subclinical. The literature remains unclear on HSV1’s relationship with schizophrenia and psychotic disorders generally. In most studies, the prevalence of HSV-1 is not higher in individuals with psychiatric disorders as compared to controls (Leweke et al., 2004; Tedla et al., 2011; Yolken, 2004). However, seropositivity has been associated with a lower level of cognitive functioning and neuroimaging abnormalities in several populations of individuals with psychiatric disorders (Dickerson et al., 2003; Prasad et al., 2011).

Previous research on the association of HSV1 with HLA has not been consistent, and many studies focused on HSV2 rather than HSV1. One study identified a significant association between HSV1 seropositivity and a few specific MHC Class I alleles (B*35, C*15) (Moraru et al., 2012). Some studies have suggested associations between HSV1 and non-HLA genes implicated in schizophrenia risk (Carter, 2011; Yolken and Torrey, 2008).

In the current study we examine the interplay between genetics, mental illness, and infection in a single analysis by including data on HLA types, schizophrenia, bipolar disorder, and several infections of interest in one cohort. The study design is further strengthened by analysis of multiple infections simultaneously. Since our cohort is exclusively of the Ashkenazi Jewish (AJ) origin, we also overcome the difficulties of confounding between ancestral origin and HLA type.

2. Methods and materials

2.1. Subjects

The Johns Hopkins institutional review board approved the recruitment methods, protocols, and informed consent documents and all volunteers provided written informed consent (Avramopoulos et al., 2015). The original cohort utilized in this study included research participants with a diagnosis of schizophrenia (including schizoaffective disorder) or bipolar disorder recruited over a 15-year period (1996–2011) through advertisements, talks, letters to leaders and service providers of the Jewish community, and a web site (Avramopoulos et al., 2015). These Ashkenazi Jewish (AJ) participants with mental illness were compared to unaffected AJ controls (termed “controls” here) who were at Jewish community professional meetings, community centers, and synagogues and were screened for psychiatric disorders (Avramopoulos et al., 2015). In addition to the controls described above, we analyzed parents of the mentally-ill cohort members as described previously (Avramopoulos et al., 2015). The frequency of a positive diagnosis in the parents was under 10%. It is therefore likely that some of the parents are also patients, yet the impact of this small group to overall statistical differences would be very small. All participants reported four grandparents of Ashkenazi ethnicity. Examiners were blinded to the subject’s diagnosis. Most of the subjects were seen in their homes. Blood for DNA and plasma from case’s parents and controls was collected and kept frozen at −80 °C (Avramopoulos et al., 2015).

For the current study we possessed serology data on 2660 subjects and genotype data on 2783 subjects; of these 1643 subjects had both serology and genotype data. This led after data cleaning to a final study size of 1636 subjects (Table 1).

Table 1.

Sample characteristics.

N = 1636	N (%) or Mean (Std)
Demographics
Age at plasma	53.1 (15.2)
Sex (Female)	777 (47.49%)
Mental illness
Sz/Sza*	502 (30.68%)
Bipolar**	420 (25.67%)
Parents	471 (28.79%)
Controls	234 (14.85%)
Infections
T. gondii positive	295 (18.03%)
Cytomegalovirus positive	589 (36.00%)
Herpes simplex virus 1 positive	638 (39.00%)
Multi-infection index
3 Infections	89 (5.44%)
2 Infections	331 (20.23%)
1 Infection	593 (36.25%)

^*Schizophrenia and schizoaffective disorder were classified together for this analysis.

^**There were no subjects with both Sz/Sza and bipolar disorder.

2.2. Immunoassay and genotyping measurements and cleaning

We used existing serology data derived from assays of plasma IgG class anti-HSV1, anti-CMV, and anti-T. gondii as described previously (Avramopoulos et al., 2015). In this previous study, we identified a bimodal distribution for these three variables of interest and determined a cut-point for each to convert these to dichotomous variables; additionally, these values had been previously adjusted for the assay plate and storage years of each sample (Avramopoulos et al., 2015). The GWAS data for this study has been described previously (Avramopoulos et al., 2015). Briefly, DNA from blood was extracted with the Gentra Puregene Kit or the QIAGEN DNeasy Blood and Tissue Kit. Genotyping was performed with the Affymetrix Human Genome-Wide SNP Array 6.0 at Emory University as described (Mulle et al., 2010). Genotypes were called using the corrected robust linear mixture model (CRLMM), an algorithm for preprocessing and genotype calling of Affymetrix SNP array data (Scharpf et al., 2011). Genotype data cleaning was performed using the software package PLINK using previously described methods (Avramopoulos et al., 2015; Purcell et al., 2007).

2.3. HLA imputation

Every subject with available genotype data had 7 HLA types imputed (A, B, C, DRB1, DQA1, DQB1, DPB1) utilizing the HIBAG software package for R (Zheng et al., 2014). This software package allows for imputation without the use of large training datasets by using published parameter estimates for given populations, which for this study were published estimates for subjects of European descent run on Affymetrix Human Genome-Wide SNP Array 6.0 (Zheng et al., 2014). The program utilizes the concept of attribute bagging, an ensemble classifier method, with haplotype inference for SNPs and HLA types. Attribute bagging is a technique which improves the accuracy and stability of classifier ensembles deduced using bootstrap aggregating and random variable selection, further information on the computational methods are available (Zheng et al., 2014). By utilizing this program in conjunction with the existing genotype data we were able to impute both alleles at 7 HLA loci by utilizing SNPs within the MHC region of chromosome 6. Through this method we obtained two imputed 4 digit alleles (as XX:XX) for each of the seven HLA types for each subject.

HIBAG provides not only alleles but also Bayesian posterior probabilities as output. The original authors of the program discuss the possibility of exclusion of observations with low probability values but also caution that while doing so can possibly increase prediction accuracy it can also drastically reduce call rates (Zheng et al., 2014). The vast majority of published literature using this software does not establish such exclusions (Abraham et al., 2015; Chang et al., 2015; Khankhanian et al., 2015; Nunes et al., 2016; Ollila et al., 2015; Parham et al., 2016). Given this and the high average posterior probabilities of our data (Appendix – Supplementary Table S1) we utilized the full breadth of available data.

2.4. Statistical analyses

The newly imputed HLA types were combined with the preexisting serology data to form a combined set of 1643 subjects. Data cleaning and all additional analyses were conducted using SAS version 9.4. Data cleaning revealed 7 controls with missing data for age who were excluded leaving a final count of 1636 subjects. Additional variables were generated for analysis purposes, these notably included a multi-infection index (0–3) indicating how many of the infections of interest yielded a seropositive result for a given subject. Table 1 describes the sample and relevant variables. Utilizing SAS proc freq the imputed alleles at each HLA locus were assessed and the most common in the sample were identified. (Tables 2 and 3). A cutoff of 3.75% prevalence in the full sample was utilized to determine a “common” allele. For each of these common alleles we additionally examined their prevalence in the seropositive portions of the sample for each of the three infections of interest. Since the parents group was uniquely overlapping with the rest of the samples we also examined whether they had a similar allele frequency as the rest of the sample population (Supplementary S2 and S3).

Table 2.

Common alleles, MHC class I.

	Percent total sample	TOXO+	CMV+	HSV1+
HLA A
01:01	12.68	13.73	13.41	14.26
11:01	5.20	4.92	5.09	5.88
02:01	16.53	13.56	15.70	15.67
02:05	3.91	4.75	4.33	3.76
24:02	9.14	7.63	8.83	8.78
26:01	15.04	15.25	13.92	15.44
03:01	8.95	9.15	8.06	8.07
33:01	3.76	3.56	3.23	3.61
Other	24.79	27.46	27.42	24.53
HLA B
14:02	11.80	11.19	11.88	10.97
35:01	5.35	5.93	5.43	5.49
35:02	8.71	7.97	9.34	8.54
35:03	3.88	3.90	4.67	3.29
38:01	17.85	17.29	15.62	16.77
52:01	5.07	5.25	4.58	5.02
57:01	4.46	6.44	5.43	5.09
08:01	4.10	6.10	3.74	5.09
Other	38.78	35.93	39.30	39.73
HLA C
12:02	5.07	5.25	4.67	5.02
12:03	21.00	20.17	19.35	19.20
04:01	19.59	18.98	20.88	19.04
06:02	11.25	12.88	12.48	11.68
07:01	8.16	11.69	7.89	10.19
08:02	11.86	11.53	11.88	11.05
Other	23.07	19.49	22.84	23.82

Table 3.

Common alleles, MHC class II.

	Percent total sample	TOXO+	CMV+	HSV1+
HLA DRB
01:02	10.06	9.66	9.93	9.64
11:01	12.93	14.07	13.24	13.32
11:04	5.23	4.41	5.52	4.86
13:01	7.21	7.46	6.79	6.82
13:02	4.03	3.90	2.97	4.70
15:02	3.79	3.73	3.57	3.45
03:01	5.75	7.12	5.86	7.12
04:02	15.89	16.44	16.04	15.28
07:01	14.33	14.58	15.28	15.13
Other	20.78	18.64	20.80	19.67
HLA DQA
01:01	11.74	10.68	11.54	10.89
01:02	8.44	6.95	7.05	9.09
01:03	11.28	11.53	10.61	10.89
02:01	14.43	15.08	15.20	15.20
03:01	17.57	18.47	17.40	17.16
05:01	5.78	7.29	5.77	7.05
05:05	22.80	22.88	23.60	22.34
Other	7.98	7.12	8.83	7.37
HLA DQB
02:01	5.75	7.29	5.69	6.97
02:02	12.50	12.71	12.99	12.85
03:01	24.39	23.90	25.89	24.53
03:02	17.60	18.31	17.23	16.61
05:01	13.20	12.54	13.67	12.46
06:01	3.76	3.73	3.57	3.37
06:03	7.37	7.80	7.05	7.29
Other	15.43	13.73	13.92	15.91
HLA DPB
104:01	5.17	4.58	5.94	5.56
02:01	25.31	25.25	24.70	25.78
04:01	43.15	44.75	42.19	42.32
04:02	9.17	9.32	10.02	9.33
Other	17.21	16.10	17.15	17.01

All analyses were conducted in a stratified fashion due to the presence of genetically related individuals in the cohort. In this dataset many subjects without mental illness were parents of cohort members with mental illness so comparison of these individuals in standard statistical models would be problematic as their genotypes (the predictors of interest) are not independent. However, by stratifying on mental illness status into four groups (schizophrenia, bipolar disorder, parents, and controls) we were not only able to solve the above independence issue, but also determine if genetic associations are modified by mental illness status.

We generated dichotomous variables for each identified common HLA allele indicating if an individual possessed one or more copies; this variable did not differentiate between heterozygotes and homozygotes. Each of these dichotomous common allele variables was used as a predictor in a logistic model using SAS 9.4 proc logistic for each of the three infections of interest as well as the multi-infection index, controlling for age and sex. The models for the multi-infection index utilized this variable as an ordinal outcome and were run as proportional odds models. These models were additionally run stratified on neuropsychiatric disease status. There were four strata, schizophrenia (including schizoaffective disorder), bipolar disorder, controls, and parents.

Since each analysis was run using a separate model there was potential for multiple comparison issues. To counteract this a Bonferroni corrected alpha level was used, and results that remained significant are noted with an asterisk in the following tables. For the common allele analyses this correction was 0.05/48 resulting yielding an alpha of 0.001 with 48 being the number of alleles tested.

After the analysis was complete additional models containing an interaction term for the alleles and schizophrenia status (yes or no) utilizing only the schizophrenia group and non-related controls with the multi-infection index outcome were run for those alleles that had significant results after Bonferroni correction in order to confirm that a differential effect of genetics by mental illness status was being observed. Graphs were produced using SPSS (Version 23, Armonk, NY) and Prism Graph Pad (Version 6, La Jolla, CA).

3. Results

3.1. Sample characteristics

The characteristics of the sample are displayed in Table 1. Sample characteristics by strata are available in appendix Table S4. Tables of all the alleles imputed are in Tables S5 through S11. The average age of the sample at the time plasma was taken was 53.1 years old (standard deviation of 15.2 years), and 47.49% of the sample was female. In this dataset those diagnosed with schizophrenia and schizoaffective disorder were grouped and analyzed together, there were a total of 502 of these individuals in the sample. This grouped condition is denoted as Sz/Sza on all tables and figures. There were 420 individuals diagnosed with bipolar disorder in the sample, and no individuals who were diagnosed with both bipolar and Sz/Sza. Parents of Sz/Sza/BP cases made up 28.79% of the sample, the remaining 14.85% were unrelated AJ controls. The prevalence of the infections of interest in this sample varied from 18.03% for T. gondii. to 39.00% for HSV1. As expected, infection prevalence increased with age. The percent of individuals who are seropositive for each infection in each diagnostic group in relation to age is displayed in Supplementary Fig. 1 (Fig. S1). Notably, serological evidence of coinfection was common, with 25.67% of the sample having 2 or 3 of these infections and 5.44% having all 3.

3.2. Common alleles

Alleles with a prevalence of at least 3.75% in the full sample were selected for further analysis. Tables 2 and 3 display the most common alleles for each HLA locus in MHC class I and class II, respectively, in the total sample as well the subset who were seropositive for each infection of interest.

Some alleles were particularly prevalent in the total sample such as HLA C*12:03, HLA DQA*05:05, HLA DQB*03:01, and HLA DPB*04:01 which each had over 20% prevalence in the sample. HLA DPB*04:01 in particular had a prevalence of 43.15% in the sample. Due to bias concerns we examined the prevalence of alleles in the parent group to observe whether this group had a significantly different observed allele prevalence than the total sample and therefore if their inclusion had altered our selection of common alleles. This did not appear to be the case (Tables S2 and S3) as the parents group had similar allele prevalence to the total sample.

Results from regression analyses testing the associations of HLA type predictor variables with infection status outcomes are shown in Tables 4–7, which show those alleles for which at least one nominally significant (p < .05) result was observed in stratified logistic models.

Table 4.

Results of stratified logistic models for statistically significant alleles predicting dichotomous T. gondii outcome controlling for age and sex.

	Sz/Sza				Bipolar				Controls				Parents
	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU
HLA B
08:01	0.0149	3.267	1.259	8.476	0.7076	1.213	0.441	3.335	0.0104	3.423	1.335	8.779	0.2783	1.477	0.73	2.988
HLA C
04:01	0.0162	0.443	0.228	0.86	0.2944	0.689	0.343	1.383	0.0044	2.628	1.353	5.107	0.4806	0.86	0.565	1.308
07:01	0.0268	2.177	1.093	4.337	0.1027	1.926	0.876	4.231	0.0751	2.109	0.927	4.796	0.3264	1.289	0.776	2.142
HLA DRB
03:01	0.0401	2.319	1.039	5.176	0.896	0.936	0.345	2.536	0.0362	2.526	1.061	6.014	0.6678	1.15	0.607	2.182
HLA DQA
01:01	0.6382	1.165	0.616	2.206	0.0451	2.025	1.016	4.038	0.5493	0.774	0.335	1.789	0.0324	0.54	0.307	0.949
05:01	0.0401	2.319	1.039	5.176	0.8619	0.915	0.338	2.478	0.0135	2.923	1.248	6.842	0.5672	1.207	0.634	2.296
HLA DQB
02:01	0.0401	2.319	1.039	5.176	0.8619	0.915	0.338	2.478	0.0135	2.923	1.248	6.842	0.4985	1.25	0.655	2.386
HLA DPB
04:02	0.3444	1.389	0.703	2.748	0.0455	2.087	1.015	4.292	0.4108	1.41	0.622	3.198	0.0065	0.413	0.218	0.781

Table 4 shows those alleles with statistically significant results for T. gondii. In the schizophrenia group there were many nominally significant alleles (p < 0.05) but none maintained their significance after Bonferroni corrections. For T. gondii infection, HLA B*08:01, HLA C*07:01, HLA DRB*03:01, HLA DQA*05:01, and HLA DQB*02:01 were associated (p < 0.05) with increased odds of T. gondii seropositive status and HLA C*04:01 was associated with decreased odds of T. gondii seropositivity. In the bipolar disorder group only one allele was nominally significant; HLA DQA*01:01 was associated with increased odds of T. gondii seropositivity. In the controls there were several nominally significant alleles, but none survived Bonferroni correction. At p < 0.05, HLA B*08:01, HLA C*04:01, HLA DRB*03:01, HLA DQA*05:01, and HLA DQB*02:01 were all associated with increased odds of T. gondii seropositivity. For the parents of cases group there were two nominally significant alleles; HLA DQA*01:01 and HLA DPB*04:02 were both associated with decreased odds of seropositivity in the parents group.

Table 5 shows the nominally-significant HLA alleles with results for cytomegalovirus (CMV). In the schizophrenia group HLA B*57:01 was associated with increased odds of CMV seropositivity. In the bipolar disorder group there were no nominally significant alleles. In the controls, HLA DRB*13:02 and HLA DQA*01:02 were associated with decreased odds of CMV seropositivity. For the parents of cases group, HLA B*38:01 and HLA DRB*13:02 were both associated with decreased odds of seropositivity.

Table 5.

Results of stratified logistic models for statistically significant alleles predicting dichotomous CMV outcome controlling for age and sex.

	Sz/Sza				Bipolar				Controls				Parents
	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU
HLA B
38:01	0.5503	0.863	0.533	1.399	0.4957	0.85	0.532	1.357	0.1377	0.628	0.34	1.161	0.0135	0.611	0.414	0.903
57:01	0.0413	2.057	1.029	4.114	0.3702	1.371	0.687	2.735	0.711	0.812	0.27	2.441	0.3799	1.361	0.684	2.711
HLA DRB
13:02	0.8339	1.092	0.481	2.479	0.8216	0.908	0.394	2.094	0.0138	0.151	0.034	0.68	0.0394	0.477	0.236	0.965
HLA DQA
01:02	0.2879	1.374	0.765	2.466	0.4419	0.786	0.425	1.453	0.0083	0.309	0.129	0.739	0.2468	0.74	0.444	1.232

Table 6 shows those alleles with statistically significant results for Herpes Simplex Virus 1. In the controls there were many nominally significant alleles and one significant after Bonferroni correction; HLA C*07:01 was significantly associated with increased odds (OR 3.4, 95% CI 1.7–6.9) of HSV1 seropositivity. Among controls, HLA B*08:01, HLA DRB*03:01, HLA DQA*05:01, and HLA DQB*02:01 were nominally associated with increased odds of HSV1 seropositivity, and HLA B*38:01 and HLA C*12:03 were associated with decreased odds of HSV1 seropositivity. In the parents of cases group HLA A*01:01 and HLA C*07:01 were nominally associated with increased odds of HSV1 seropositivity. HLA C*12:03 (p = 0.001), which was significant after Bonferroni correction, and HLA B*38:01 (p = 0.03), were associated with decreased odds of HSV1 seropositivity. In the schizophrenia and bipolar disorder groups there were no nominally significant alleles.

Table 6.

Results of stratified logistic models for statistically significant alleles predicting dichotomous HSV1 outcome controlling for age and

	Sz/Sza				Bipolar				Controls				Parents
	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU
HLA A
01:01	0.4146	0.817	0.503	1.327	0.3879	1.226	0.772	1.947	0.1186	1.654	0.879	3.113	0.008	1.81	1.168	2.806
HLA B
38:01	0.1317	1.379	0.908	2.095	0.9754	0.993	0.63	1.564	0.03	0.517	0.285	0.938	0.0225	0.641	0.437	0.939
08:01	0.3843	1.423	0.642	3.154	0.9521	0.979	0.491	1.953	0.0079	3.323	1.369	8.064	0.1223	1.755	0.86	3.58
HLA C
12:03	0.5533	1.131	0.753	1.697	0.91	1.026	0.662	1.588	0.005	0.443	0.251	0.783	0.001*	0.529	0.363	0.772
07:01	0.5847	1.167	0.671	2.031	0.6037	1.165	0.655	2.07	0.0007*	3.387	1.668	6.878	0.0486	1.651	1.003	2.717
HLA DRB
03:01	0.1655	1.551	0.834	2.882	0.7656	1.099	0.591	2.044	0.0146	2.626	1.21	5.698	0.1004	1.683	0.904	3.131
HLA DQA
05:01	0.1655	1.551	0.834	2.882	0.837	1.067	0.576	1.977	0.0238	2.403	1.123	5.138	0.1174	1.648	0.882	3.078
HLA DQB
02:01	0.1655	1.551	0.834	2.882	0.837	1.067	0.576	1.977	0.0238	2.403	1.123	5.138	0.1561	1.576	0.841	2.954

^*Significant after Bonferroni correction for multiple comparisons.

Table 7 displays those alleles which had significant effects for models where the dependent variable was the multi-infection index. These models were run as ordinal logistic regressions using the proportional odds assumption, and therefore the displayed odds ratios indicate the ratio of odds for number of infections being one higher between a person with the allele and without it. In the controls there were many significant alleles. After Bonferroni corrections, HLA B*38:01 and HLA C*12:03 were significantly associated with lower odds of additional infections among the controls. HLA B*08:01, HLA C*07:01, HLA DRB*03:01, HLA DQA*05:01, and HLA DQB*02:01 were nominally associated with increased odds of additional infections among the controls. The parents of cases group had several nominally significant alleles but none survived Bonferroni correction. HLA A*01:01, HLA B*57:01, HLA C*07:01 were nominally associated with increased odds of additional infections among the parents group. HLA B*38:01 and HLA C*12:03 were nominally associated with decreased odds of additional infections among the parents group. In the schizophrenia group there were no nominally significant alleles. In the bipolar disorder group there was one nominally significant allele; HLA DPB*04:01 was associated with decreased odds of additional infections.

Table 7.

Results of Stratified Logistic models for statistically significant alleles predicting ordinal multi-infection outcome controlling for age and sex.

	Sz/Sza				Bipolar				Controls				Parents
	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU	p-value	OR	CIL	CIU
HLA A
01:01	0.4306	0.845	0.555	1.285	0.2488	1.274	0.844	1.922	0.4735	1.235	0.693	2.202	0.0042	1.762	1.195	2.599
HLA B
38:01	0.0867	1.385	0.954	2.01	0.3493	0.825	0.552	1.234	0.0003*	0.371	0.217	0.634	0.0028	0.587	0.414	0.832
57:01	0.0649	1.743	0.966	3.145	0.2094	1.48	0.802	2.731	0.7135	1.187	0.476	2.957	0.0319	1.927	1.059	3.508
08:01	0.5287	1.264	0.61	2.617	0.3299	0.734	0.394	1.367	0.0037	3.235	1.465	7.142	0.0876	1.716	0.924	3.188
HLA C
12:03	0.179	1.279	0.893	1.83	0.2425	0.793	0.538	1.17	0.0006*	0.416	0.252	0.688	0.0016	0.576	0.409	0.811
07:01	0.4223	1.223	0.748	1.999	0.5654	0.859	0.511	1.443	0.0053	2.469	1.309	4.658	0.015	1.728	1.112	2.686
HLA DRB
03:01	0.3143	1.339	0.758	2.364	0.8838	1.042	0.6	1.811	0.0016	3.122	1.54	6.332	0.1551	1.486	0.861	2.563
HLA DQA
05:01	0.3143	1.339	0.758	2.364	0.9943	1.002	0.579	1.735	0.0016	3.069	1.528	6.165	0.1553	1.491	0.859	2.587
HLA DQB
02:01	0.3143	1.339	0.758	2.364	0.9943	1.002	0.579	1.735	0.0016	3.069	1.528	6.165	0.1985	1.44	0.826	2.51
HLA DPB
04:01	0.9136	1.021	0.702	1.486	0.0389	0.668	0.456	0.98	0.6341	0.885	0.535	1.463	0.3353	0.84	0.589	1.198

^*Significant after Bonferroni correction for multiple comparisons.

For the multiple infection outcome, the effect of HLA B*38:01 and HLA C*12:03 was in the opposite direction between the Sz/Sza and control groups. These alleles (HLA B*38:01 and HLA C*12:03) were significant after Bonferroni correction, and thus we ran additional models containing an interaction term for the alleles and schizophrenia status (yes or no) utilizing only the schizophrenia group and non-related controls with the multi-infection index outcome (also controlling for sex and age). In both cases there was a significant interaction term (p < .0001) indicating that the risk for greater number of infections differed by mental disorder status. This differential association of these two alleles by diagnostic group is displayed in Fig. 1.

Fig. 1.

Summary of differences by psychiatric affected group in the associations of HLA alleles with infection serology. Figure shows odds ratio with 95% confidence intervals (error bars) for seropositivity for multi-infection variable.

3.3. Haplotypes

We consider our results in the context of haplotypes discussed in the literature. As displayed in Table 8 there were several common haplotypes present in the sample. HLA DRB*03:01, HLA DQA*05:01, and HLA DQB*02:01 were always present as a consistent haplotype in the schizophrenia group and with a few individual exceptions in all the other groups as well. This produced nearly identical results in the common allele analysis and may indicate that this is a common haplotype in our population that could also be in linkage disequilibrium given the proximity of these alleles. The alleles HLA B*08:01 ~ HLA C*07:01 and HLA B*38:01 ~ HLA C*12:03 also co-occurred to a high degree and had very similar effects to a slightly lesser extent. The haplotypes HLA DR*03 ~ H LA DQ*02 and HLA B*8 ~ HLA C*7 have both been previously been identified as commonly occurring in a south Tunisian population but not in an Ashkenazi population (Mahfoudh et al., 2013).

Table 8.

Haplotype frequencies across strata.

	N (%)
HLA DRB*03:01 ~ HLA DQA*05:01 ~ HLA DQB*02:01
Total Sample (N = 1636)	181 (11.06%)
Sz/Sza (N = 502)	51 (10.16%)
Bipolar (N = 420)	52 (12.38%)
Controls (N = 243)	31 (12.76%)
Family (N = 471)	47 (9.98%)
HLA B*08:01 ~ HLA C*07:01
Total Sample (N = 1636)	131 (8.01%)
Sz/Sza (N = 502)	30 (5.98%)
Bipolar (N = 420)	41 (9.76%)
Controls (N = 243)	23 (9.47%)
Family (N = 471)	37 (7.86%)
HLA B*38:01 ~ HLA C*12:03
Total Sample (N = 1636)	520 (31.78%)
Sz/Sza (N = 502)	155 (30.88%)
Bipolar (N = 420)	124 (29.52%)
Controls (N = 243)	74 (30.45%)
Family (N = 471)	167 (35.46%)

4. Discussion

In this study of >1,600 individuals, we report HLA alleles associated with serology for three infections (HSV1, CMV and T. gondii) in relation to schizophrenia and bipolar diagnosis. To our knowledge this is the only study of HLA alleles and seropositivity to multiple infections in individuals diagnosed with mental disorders. Indeed, for HSV1 and T. gondii this is the largest study of HLA-types in serologically-confirmed infection in any population.

4.1. HLA alleles associated with HSV infection

HLA C*07:01 was significantly associated with HSV1 seropositivity (OR = 3.4) in the mentally healthy controls even after Bonferroni correction for multiple comparisons. This contrasted with a null effect of this allele on HSV1 in the schizophrenia and BP groups. Moreover, HLA B*38:01 and HLA C*12:03 both displayed protective effects in the mentally healthy controls (OR = 0.5 and 0.4, respectively) and null effects in the schizophrenia and bipolar disorder groups (OR = 0.99–1.38).

Because our study was performed using a large relatively homogenous Ashkenazi sample, our findings on HSV1 may have broad implications beyond psychiatry. There have been several prior studies of HLA types in HSV infections, but most of these focused on symptoms or specific sequelae (e.g. ocular infection), or were performed prior to the development of reliable HSV1-specific serology tests, or examined heterogeneous populations (reviewed in Moraru) (Moraru et al., 2012). A large study, which tested broad HLA tissue antigen types, found a negative association of HSV1 with Bw16 (Blackwelder et al., 1982). The Bw16 HLA type underwent several refinements in nomenclature and now includes HLA B*38. We found HLA B*38:01 to be negatively correlated with HSV1 in controls (OR 0.5, 95%CI 0.0.29 to 0.93; p = 0.03), but not in the schizophrenia or BP groups (P > 0.05). Hence, our results are consistent with the Blackwelder et al. study for the controls but not the mentally-ill groups. This suggest that the protection afforded against HSV1 by carrying these HLA types in the wider population is abrogated by other factors in the mentally ill patients. Studies reported subsequent to Blackwelder et. al. identified several HLA Class I alleles that were either positively or negatively correlated with HSV prevalence (Moraru et al., 2012; Samandary et al., 2014). In our study, most of these alleles were either not prevalent in our population, or did not reach a statistically significant association with HSV1. However, Jaaber et al. found HLA A*01 to be positively associated with HSV1, which is consistent with our data (Jabbar et al., 1990).

Our finding that HLA C*07:01 increases risk of HSV1 infection could be pointing to a common haplotype in our sample (HLA B* 08:01 ~ HLA C*07:01) or possibly include some influence of the broader haplotype, A*01:01 ~ B*08:01 ~ C*07:01 ~ DRB1*03:01 ~ DQB1*02:01, which is common in the United States. Among the parents in our cohort tested for HSV1, the p-value for HLA C*12:03 passed correction for multiple comparisons (OR = 0.5), and this relationship could reflect protection via the haplotype, B *38:01 ~ C*12:03, which is found in approximately 2.5% of European Caucasians in the US.

4.2. HLA alleles associated with CMV infection

In the analyses of CMV, no individual alleles were associated with infection after correction for multiple comparisons. The strongest association was a protective effect of HLA DRB*13:02 in the mentally healthy controls (OR = 0.15) but not in the schizophrenia or bipolar groups (OR = 1.1 and 0.91, respectively). Haider et. al found the DRB1*13 was more prevalent in schizophrenia than mentally healthy controls. We cannot, however, extrapolate this to a specific mechanism by which CMV may be related to schizophrenia pathogenesis because in our current dataset CMV was not more prevalent in the Sz/Sza group than in controls (Avramopoulos et al., 2015; Grove et al., 2014). HLA DQA*01:02 showed a protective effect on CMV in both control and the bipolar disorder groups (in control group, OR = 0.31, p = 0.008), though not in the schizophrenic group.

4.3. HLA alleles associated with T. gondii infection

In the analyses of Toxoplasma gondii, no individual HLA alleles were associated with infection after correction for multiple comparisons. In the common allele analysis, the alleles HLA B*08:01, HLA C*07:01, HLA DRB*03:01, HLA DQA*05:01 and HLA DQB 02:01 were all identified as having a risk-increasing effect in the schizophrenia and control group. A prior study found HLA DRB*03 may confer resistance against toxoplasmic encephalitis (though not T. gondii seroprevalence) among AIDS patients. (Rodrigues et al., 2016). The DRB*03:01 allele is protective against schizophrenia, and this has been proposed to explain the association of schizophrenia with some autoimmune disorders (Crespi and Thiselton, 2011). Our nominally-significant findings indicated that the HLA C*04:01 allele was protective in schizophrenia (OR 0.4, 95% CI 0.2–0.9) but increased susceptibility in controls (OR 2.6, 95% CI 1.4–5.1).

HLA C*04:01 may be relevant to T. gondii infection susceptibility beyond its role in antigen presentation to CD8-T cells. Specifically, NK cells are important in controlling T. gondii, and NK cells are regulated by expression of killer-cell immunoglobulin-like receptors (KIR). Some of these receptors (KIR2DL1 and KIR2DS1) are ligands for HLA C*04:01, and genotyping studies indicate they are associated with ocular toxoplasmosis (Ayo et al., 2016). Considering the inconsistent findings on NK cell number and function in schizophrenia (Wisniewski et al., 2014) (Schindler et al., 1986; Yovel et al., 2000), specific interactions between HLA C*04:01 and NK KIRs in the context of T. gondii infection should be further explored.

4.4. HLA alleles associated with combined multiple infections (infection burden)

One of the difficulties with the argument for an infectious etiology of schizophrenia is that multiple different types of infections have been implicated, and few studies have considered specific host genetic factors (Børglum et al., 2014; Pearce, 2003). To our knowledge there has not been a systematic study of HLA types in relation to multiple infections that compares schizophrenia patients to controls. The current study achieves this, and included a group with bipolar disorder. For the multiple infection outcome, among the controls there was a statistically-significant inverse association for HLA B*38:01 (OR = 0.37, p = 0.0003), which may be driven by the haplotype B*38:01 ~ C*12:03. The p-value for HLA C*12:03 also passed correction for multiple comparisons, supporting the role of this haplotype in multiple infection susceptibility. The haplotype HLA DRB*03:01 ~ HLA DQA*05:01 ~ HLA DQB*02:01 was associated with increased odds of infection in the control group.

Of note, for the multi-infection outcome, the odds ratios corresponding to the association of B*38:01 and C*12:03 were in the opposite direction for the group with schizophrenia (OR ≈ 1.3) as for the rest of the cohort, including the controls (OR ≈ 0.4). Thus, carrying the B*38:01 ~ C*12:03 haplotype appears to have differential effects on infection risk among people with schizophrenia compared to controls. This haplotype appears to be protective against simultaneous multiple infections (as measured) in the mentally healthy control group but not in the schizophrenia group. Likewise for the HSV outcome, HLA C*12:03 was protective in the controls but not the schizophrenia group. Conversely, HLA C*07:01 is associated with increased risk for HSV and multiple infections in the controls but has a null effect in the schizophrenia patients.

4.5. Possible biological mechanisms

This disjunction in the effect of these immunogenetic variables between schizophrenia patients and mentally healthy controls warrants further study. The current study is important because it suggests a specific functional immune consequence of MHC variation in schizophrenia. Nevertheless, the implication of this finding for an infectious etiology of schizophrenia (or bipolar disorder) should be inerpreted conservatively. Although various studies indicate that serology results for some infections differ between healthy subjects and people with schizophrenia, this does not imply a causal relationship between the infection and the psychiatric illness. The immune response (including antibody production) to many infections is influenced by HLA variation, but is ultimately shaped by the exposure history and complex molecular cascades that likely involve variants elsewhere in the genome (Ovsyannikova et al., 2006). In the current study we used density plots of regression residuals to clearly differentiate seropositive from seronegative individuals, and found no statistically-significant differences in the prevalence of these infections when comparing the psychiatric illness groups with controls (Avramopoulos et al., 2015). Thus, in the current study we found differences between the psychiatric illness group and controls in the HLA alleles associated with these infections despite the similar age-adjusted prevalence of these infections between these groups. This implies, for example, that patients with schizophrenia may engage subtly different immune pathways for controlling certain infections compared with controls because of underlying immunogenetic differences between these groups. The differential effect of HLA alleles on infection based on mental illness status could be a marker of these immunogenetic differences. In some populations this may be manifested as differences between schizophrenia patients and healthy controls in the seroprevalence of these infections, while in other populations factors related to infection exposure may dominate. This might help explain why some studies find a positive association of certain infections with schizophrenia while other studies do not (Pearce, 2003).

It is possible that the repertoire of peptide antigens displayed by the same HLA variant differs between patients with schizophrenia and controls. This mechanism is supported by gene expression and SNP studies in schizophrenia that have implicated genes involved in the molecular pathway for the processing, transport or loading of peptide antigens onto MHC (Fellerhoff and Wank, 2009; Wu et al., 2016). The ubiquitin system is also involved in antigen processing in herpes virus infections (Loureiro and Ploegh, 2006), and hence the dysfunction of protein ubiquitination observed in schizophrenia might also extend to antigen presentation (Rubio et al., 2013). These hypotheses could be tested by examining the interaction between genetic variants in antigen processing pathways, HLA alleles, and the effect of infections on schizophrenia risk.

4.5.1. Complement proteins

The role of the complement protein system in neuroimmune mechanisms of schizophrenia has garnered renewed attention given its role in synaptic pruning, and association with neuropsychiatric illness, including schizophrenia (Mayilyan et al., 2008b; Sekar et al., 2016; Stevens et al., 2007). Individual differences in the expression of complement C4A/C4B genes are determined to a significant extent by complex genetic structural features and copy number variants (Mayilyan et al., 2008a; Sekar et al., 2016). Blood levels of C4 are associated with risk for schizophrenia (Mayilyan et al., 2008a). A large-scale GWAS of schizophrenia honed in on the C4 locus as the key contributor to schizophrenia risk conferred by variants in the MHC region (Sekar et al., 2016).

This is relevant to our study because for T. gondii, HSV, and the multi-infection index, the positive associations with HLA loci B*08:01, DRB*03:01 and DQB*02:01 may reflect LD with the complement 4 (C4A) alleles that are associated with schizophrenia risk (Sekar et al., 2016). The presence of these HLA alleles (or related HLA haplotypes) is predicted to be in LD with the complement “C4A BS” structural allele, which is the most protective C4 allele for schizophrenia risk (Sekar et al., 2016). This C4A allele also results in low expression of C4A RNA (Sekar et al., 2016). Hence we would expect that the individuals in our cohort who were carrying HLA loci B*08:01, DRB*03:01 and DQB*02:01 would have diminished C4A expression compared to the other members of the cohort.

Our findings on the association of these complement relevant HLA alleles for the psychiatric diagnostic groups are summarized in Table 9. For HSV these alleles (B*08:01, DRB*03:01 and DQB*02:01) were linked to a statistically significant increase in seropositivity in the control group (OR = 2.4–3.3) but not in the schizophrenia group. HSV-1 reactivation has been linked to low serum C4A protein levels (Seppänen et al., 2001). This pattern of enhanced susceptibility only among the controls also appeared for the multi-infection index (Table 9).

Table 9.

Summary of alleles relevant to complement C4 genes in relation to infection prevalence.

	Sz/Sza	Bipolar	Controls	Parents
HLA B*08:01
T. gondii	↑	n.s.	↑	n.s.
HSV1	n.s.	n.s.	↑	n.s.
Multi-infection	n.s.	n.s.	↑	n.s.
HLA DRB*03:01
T. gondii	↑	n.s.	↑	n.s.
HSV1	n.s.	n.s.	↑	n.s.
Multi-infection	n.s.	n.s.	↑	n.s.
HLA DQB*02:01
T. gondii	↑	n.s.	↑	n.s.
HSV1	n.s.	n.s.	↑	n.s.
Multi-infection	n.s.	n.s.	↑	n.s.

^↑-increased prevalence with allele; n.s. non-significant (p ≥ 0.05).

Since the C4-BS allele is relatively protective against schizophrenia, we might expect this allele to be less common among our schizophrenia patients. We did not directly test C4A/C4B variants (there were too few SNPs passing QC in this region to draw conclusions), but since this locus is within the MHC class III region (positioned between MHC class 1 and Class II regions), and has been strongly linked to schizophrenia risk, this possibility is worth consideration. Indeed the DRB*03:01 ~ DQB*02:01 haplotype, and the haplotype with HLA B*08:01 (see above), were approximately 20–40% less common in the schizophrenia group compared to the controls. This is consistent with the putative role of the C4-BS allele as conferring resistance to schizophrenia (Sekar et al., 2016).

Thus, if the presence of these HLA alleles (B*08:01, DRB*03:01 and DQB*02:01) are reflecting LD with the C4 structural variants dictating low C4 expression (i.e. C4-BS), then our data implies that the lower expression of C4 increases the risk of HSV and multiple infections in the non-schizophrenic individuals but not in those with schizophrenia. An extension of this idea is that only a minor subset of people with schizophrenia carry these HLA alleles (and have low C4 expression), but among this subset, their low C4 expression fails to protect them against schizophrenia (i.e. because low C4 is normally associated with lower schizophrenia risk). However, this subset of schizophrenia patients may have an advantage over healthy controls in that the linked HLA alleles (e.g. B*08:01, DRB*03:01 and DQB*02:01) do not place them at higher risk for some neurotropic infections (e.g. HSV). In contrast, they are not similarly protected from infection with T. gondii.

4.5.2. Autoimmune mechanisms

There is also a long-standing association of schizophrenia with autoimmune disease, raising the possibility that tolerance to self-antigens may be broken in schizophrenia due to aberrations in MHC antigen display (Benros et al., 2014; Carter, 2011; Pearce, 2003). Pouget et al. found evidence that HLA B*08:01 and HLA DQB1*02:01 were associated with both schizophrenia and autoimmunity (Pouget et al., 2016). As mentioned above, we found nominally significant (p = 0.002 to p = 0.04) associations with these alleles and T. gondii, HSV1 and the multi-infection index, which further supports the possibility of infection-triggered autoimmune mechanism in schizophrenia.

In any case, there are likely multiple pathogenic pathways that lead to the behavioral phenotypes of schizophrenia and bipolar disorder. We argue these include neuroimmune pathways, as reflected in part by differences between the schizophrenia and control groups in HLA-infection interactions. Because many such interactions entail differences in amino acid sequences in the peptide-binding region of HLA, our study represents an example where the measurable genetic variation between those with schizophrenia and mentally healthy controls is manifested as a functional immune consequence.

4.6. Limitations

This analysis had several limitations, the most notable of which were the use of several stratifications reducing sample size in each model as well as the large number of models run leaving vulnerabilities to multiple comparison issues. The effects of medications were not considered, and thus future studies in first episode patients is an important future direction. We imputed HLA antigen types from genetic data. However, our AJ sample corresponded well with the expected HLA allele frequencies from the AJ bone marrow registries in Israel (e.g. HLA A*02:01 16.6% vs 15.4%; HLA B*38:01 17.8% vs 16.7%, HLA DRB1*04:02 15.8% vs 11.5%, in our study and the registry, respectively) (Manor et al., 2016) as downloaded from http://www.allelefrequencies.net. We detected many significant allele effects in the mental illness free groups despite the c the schizophrenia group.

4.7. Conclusions

As the initial systematic epidemiological analysis to consider the effects of HLA alleles on T. gondii, CMV, and HSV1 collectively, this study should have broad interest in immunology, including transplant immunology given the importance of these infections in solid organ transplant complications (Fishman, 2007).

This study discovered several HLA alleles that exhibited a differential effect on infection in mentally ill individuals versus non-mentally ill individuals. Our data suggest that immune pathways involved in controlling certain neurotropic infections differ between schizophrenia patients and healthy controls in the way they engage some HLA-related mechanisms. For example, those individuals who become schizophrenic despite having the “protective” C4 genotype may have unusual antigen presentation characteristics that protect them from some infections (e.g HSV1) but increase their vulnerability to schizophrenia via other immune mechanisms such as the production of autoantibodies against brain antigens.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.bbi.2018.03.001.