Schizophrenia, Human Leukocyte Antigen (HLA), and Herpes Viruses: Immunogenetic Associations at the Population Level

Abstract

Several factors have been implicated in schizophrenia (SZ), including human herpes viruses (HHV) and the adaptive immunity Human Leukocyte Antigen (HLA) genes. Here we investigated these issues in 2 complementary ways. In one analysis, we evaluated SZ-HLA and HHV-HLA associations at the level of a single allele by computing (a) a SZ-HLA protection/susceptibility (P/S) score based on the covariance between SZ and 127 HLA allele prevalences in 14 European countries, (b) estimating in silico HHV-HLA best binding affinities for the 9 HHV strains, and (c) evaluating the dependence of P/S score on HHV-HLA binding affinities. These analyses yielded (a) a set of 127 SZ-HLA P/S scores, varying by >200× (maximum/minimum), which could not be accounted for by chance, (b) a set of 127 alleles × 9 HHV best-estimated affinities, varying by >600×, and (c) a set of correlations between SZ-HLA P/S scores and HHV-HLA binding which indicated a prominent role of HHV1. In a subsequent analysis, we extended these findings to the individual person by taking into account the fact that every individual carries 12 HLA alleles and computed (a) the average SZ-HLA P/S scores of 12 randomly chosen alleles (2 per gene), an indicator of HLA-based SZ P/S for an individual, and (b) the average of the corresponding HHV estimated affinities for those alleles, an indicator of overall effectiveness of HHV-HLA binding. We found (a) that HLA protection for SZ was significantly more prominent than susceptibility, and (b) that protective SZ-HLA scores were associated with higher HHV-HLA binding affinities, indicating that HLA binding and subsequent elimination of several HHV strains may confer protection against schizophrenia.

Keywords: HHV1; Human leukocyte antigen; epidemiology; immunogenetics; schizophrenia.

Figure 1.

Schematic diagram to illustrate the sliding window approach for estimating exhaustively in silico the binding affinity of all consecutive 9- and 15-mer peptides. The figure refers to HHV1 for illustration but the glycoproteins of all HHV strains (Table 4) were tested, as described in the text.

Figure 2.

Frequency distributions of $\mathrm{LPR}$ [A] and $\ln (\mathrm{LPR})$ [B]. N = 113 alleles × 9 HHV strains = 1017 for each distribution.

Figure 3.

Probability-probability plots of $\mathrm{LPR}$ [A] and $\ln (\mathrm{LPR})$ [B] to illustrate the substantial departure of the $\mathrm{LPR}$ distribution from normal (diagonal) and conversion to close to normal when log-transformed.

Figure 4.

SZ-HLA PSCorr scores are plotted against SZ-PSCov scores. See text for details.

Figure 5.

Ranked SZ-HLA PSCorr are plotted against their ranks. Notice that protective scores (blue) are more numerous and stronger than susceptibility scores (red). See text for details.

Figure 6.

(A) frequency distribution of the 127 SZ-HLA PSCorr scores. (B) boxplot of the data in Figure 5. Q1, 25th percentile; Q3, 75th percentile; IQR, interquartile range.

Figure 7.

Prevalence of schizophrenia is plotted against frequency of B*27:02 (r = .743, PSCorr = −.956), B*35:01 (r = −.703, PSCorr = −.872), and DRB1 * 13:05 (r = −.675, PSCorr = −.820), as indicated. The curves are a power fits. The fits are linear between log-log transformed data, from which the correlations above are derived.

Figure 8.

Mean (±95% CI) protective preponderance of SZ-HLA scores (negative/protective – positive/susceptibility SZ-HLA score) for the groups indicated. See text for details.

Figure 9.

A, frequency distribution of 1 million bootstrap ξ* values.

Figure 10.

Boxplot of the data in Figure 9. Conventions as in Figure 6B.

Figure 11.

Frequency distribution of protective (A) and susceptibility (B) ξ* values.

Figure 12.

Average (±95% CI) HHV1-HLA affinities [ln(LPR)] for HLA Class I and II.

Figure 13.

Average (±95% CI) HHV1-HLA affinities [ln(LPR)] for genes of HLA Class I and II.

Figure 14.

SZ-HLA PSCorr scores are plotted against $\ln (\mathrm{LPR})$ for HHV1 for 9-mer (A; left panel) and 10-mer (B; right panel) analyses. See text for details.

Figure 15.

One-thousand ξ* values are plotted against the corresponding averages of 9 HHV ln(LPR) values. r = .638, P < .001. See text for details.