3. Results

3.1. MHC Class II Expression in Cingulate Cortex by Exon Array in Bipolar Disorder Compared to Controls (Experiment 1)
The data analysis for the Affymetrix Human Exon 1.0 ST array used an ANOVA with diagnosis and probesets as between subject factors. There were 11,807 transcript identifiers analyzed composed of 230,659 probesets. Affymetrix array transcript identifiers that passed this filter for diagnosis by probeset interaction are listed (Supplementary Tables 2–4), and there were 30 genes (29 known genes as ENSG00000185790 did not map to a known RefSeq gene) that passed Bonferroni multiple comparisons correction for diagnosis by probeset. The analysis was restricted to male control and bipolar patients to minimize heterogeneity of expression due to sex differences, as well as there being unequal and fewer females in the study. There were no genes that passed Bonferroni corrections for diagnosis factor.
Two MHC Class II genes (HLA-DPA1 and CD74) passed Bonferroni correction (Supplementary Tables 2–4). Post hoc analysis of HLA-DPA1 showed downregulation of HLA-DPA1 mRNA in the ACC of bipolar disorder patients relative to healthy control subjects using the Affymetrix Human Exon 1.0 ST array data (GEO Accession Number: GSE78246) (Table 4) along the entire length of the gene (Figure 2A). There was a statistically significant decrease in BD in ACC on HLA-DPA1 gene expression (p = 0.013); however, the interaction effect between diagnosis and probesets was highly significant (p = 1.6 × 10−6). The individual probesets in the exon array that were significantly decreased in bipolar subjects in ACC compared to controls were the following: Affymetrix probesets: 2950331, 2950332, 2950333, 2950336, 2950341, 2950342, 2950343, 2950346, 2950348 (Figure 2A). CD74 was significantly decreased in expression in BD compared to controls by exon array (p = 0.038, Figure 2B), and showed a highly significant interaction effect between diagnosis and probesets (p = 3.3 × 10−5). For the other MHC Class II transcript, HLA-DRB1, the transcript level was not altered in BD compared to controls (p = 0.27), nor was there a significant interaction effect.
The HLA-DRB1 diagnosis by probeset was not significant (p = 0.93). A variable expression of HLA-DPA1 between subjects by exon array probesets (Figure 3) suggested that exons 2, 3, and 4 might be sites of alternative splicing, which were investigated by PCR sequencing (see below). For validation of the exon array results, we chose additional qPCR candidates based upon microarray results shown in Supplementary Table 1 of significant evidence of diagnosis by probeset. Our validations Figure 4 were consistent for microarray and exon array fold changes (r = 0.80).
The Affymetrix exon array fold changes were consistent with the result shown for exon array of anterior cingulate cortex.

3.2. qPCR Validation of Candidate MHC II Genes Identified in Exon Array Data (Experiment 2)
The main effects for all qPCR analyses are shown in Table 6. Age and pH were selected as covariates, while diagnosis, brain region, and sex were main effects. The ANCOVA for age was significant (HLA-DRB1 and HLA-DPA1 p < 0.0005, while CD74 was nominally significant p = 0.05). Including pH in the ANCOVA was not significantly related to gene expression, and sex was significant for HLA-DRB1 expression (p = 0.001).
Highly significant main effects were found for HLA-DPA1 and CD74 (p-values < 10−20), and the resulting line plots are shown in Figure 5. For diagnosis effect, CD74 (p = 3.6 × 10−5) and HLA-DPA1 (p = 1.7 × 10−7) were significant, while HLA-DRB1 did not reach significance (p = 0.078). Only CD74 was significant for region by diagnosis interaction. Regional analyses were conducted and comparison results are shown in Table 7 A,B,C. The lack of findings for HLA-DRB1 in all brain regions by qPCR is consistent with the exon microarray findings. Three neuropsychiatric disorder groups (SZ, MDD, and BD) had a significantly lower expression of HLA-DPA1 in the hippocampus. Further, CD74 and HLA-DPA1 showed concordant expression at the brain region level, suggesting broad co-expression. SZ showed reduced expression in similar regions as CD74 and HLA-DPA1. HLA-DRB1 showed a variable relationship across diagnosis and brain regions, and, due to higher cycle numbers, might not be reliably detected in brain, at least the isoform that we measured by qPCR.

3.3. Expression by Genotype Interaction of HLA-DPA1
The same cohort of subjects which had brain gene expression measured for HLA-DPA1 (Table 2) was genotyped for SNPs around the HLA-DPA1 gene and an F-ratio was calculated for association. The HLA-DPA1 expression levels were measured by qPCR for the association analysis with the genotypes of SNP rs9277341.This intronic SNP was located in intron two of HLA-DPA1. We sequenced the specific amplicon resulting from qPCR amplification reactions (Supplementary Table S1) to find the expected product and ensuring that SNPs or large exonic deletions were not within the primer region of the qPCR amplicon.
The results for HLA-DPA1 SNP rs9277341 by brain region are shown (Figure 6). The main effect for rs9277341was significant (p = 0.0002), the major ‘T’ allele showed increased expression, while the homozygous minor “C” allele has significantly lower expression. The interaction of region (five brain regions tested) by SNP rs9277341 genotype (0, 1, 2) was not significant. The overall effect of diagnosis factor (SZ, BD, MDD, C) on HLA-DPA1 expression was significant (p = 1.68 × 10−6). Post hoc comparisons between each psychiatric diagnosis and controls were significant, and decreased significantly in expression (Table 8 and Table 9).

3.4. Alternative Splicing Validation HLA-DPA1 (Experiment 3)
Gel electrophoresis of amplicons from cDNA for HLA-DPA1 exon 2–4 amplification showed the predominant expected band size of 538 bp. However, in many brain samples, distinct smaller bands were seen (Figure 7). These gel-purified bands were directly sequenced and the results confirmed alternative splicing of these smaller bands as containing parts of exons 2, 3, and 4 (Table 10). The HLA-DPA1 sequencing primers are located in exon 2 and exon 4 of NM_033554 as:
Forward5’-TGGACAAGAAGGAGACCGTCT-3’(positon307/327);
Reverse5’-TTTATGATGAGGACGGTGCC-3’(position 844/825). All the sequences were in the range from 307 to 844 of NM_033554.
The cDNA alignment to HLA-DPA1 RefSeq showed that differential splicing of exon 2–3 and 3–4 was involved, although did not involve exon skipping, tending rather toward a novel splice site extended into adjacent exons (Figure 8). The sequence of four subjects shows full-length amplicon of exons 2, 3, and 4 (512–515 bp), and truncated versions of exons 2, 3, and 4 (215–400 bp). The UCSC sequence tracks show truncation of exon 2, 3, and 4, thereby suggesting different cryptic splice sites that occur (Figure 8).
Total sequences from forward primer to reverse primer are 538 bp, and in position 307−844 of NM_033554.3. Exon Bound1aries 2 (264–453) 3 (454–735) 4 (736–902) representative samples from four different subjects using anterior cingulate cDNA. The forward primer sequences (-f.ab1 are matched to reverse primer sequence nomenclature (-r.ab1).
For more details, the next section shows the sequencing results for forward and reverse primers for HLA-DPA1_07 splicing site position from 427 to 641 in NM_033554 and the HLA-DPA1_01 splicing site at position 478 to 643 in NM_033554.
Box 1  Sequencing Result of Bands from HLA-DPA1-07 Gel.  Sequencing of cDNA HLA-DPA1_ (07-f.ab1 with Forward primer)   350AGCCTTTTCCTTTGAGGCTCAGGGCGGGCTGGCTAACATTGCTATATTGAACAACAACTTGAATACCTTGATCCAGCG427 (spliced out region) ---------------------------------------------------------------------------------------------------------------------------------------------642TTCCATTACCTGACCTTTGTGCCCTCAGCAGAGGACTTCTATGACTGCAGGGTGGAGCACTGGGGCTTGGACCAGCCGCTCCTCAAGCACTGGG735 736AGGCCCAAGAGCCAATCCAGATGCCTGAGACAACGGAGACTGTGCTCTGTGCCCTGGGCCTGGTGCTGGGCCTAGTCGGCATCATCGTGGGCACCGTCCTCATCATAAA844   Sequencing of cDNA HLA-DPA1_(07-r.ab1 with Reverse primer)   801GCACCnGGCCCAGGGCACAGAGCACAGTCTCCGTTGTCTCAGGCATCTGGATTGGCTCTTGGGCCT736 735CCCAGTGCTTGAGGAGCGGCTGGTCCAAGCCCCAGTGCTCCACCCTGCAGTCATAGA AGTCCTCTGCTGAGGGCACAAAGGTCAGGTAATGGAA642 (spliced out region) ---------------------------------------------------------------------------------------------------------------------------------------------427CGCTGGATCAAGGTATTCAAGTTGTTGTTCAATATAGCAATGTTAGCCAGCCCGCCCTGAGCCTCAAAGGAAAAGGCTTGGCCAAACTCCTCCAGATGCCAGACGGTCTCCTTCTTGTCCA307  The sequencing result for both forward and reverse primers are consistent to show a novel splice site for HLA-DPA1 located in exons 3 and 4.
Box 2  Sequencing Result of Bands from HLA-DPA1-01 Gel.  cDNA HLA-DPA1_(01-f.ab1 with Forward primer). 360TTTGAGGCTCAGGGCGGGCTGGCTAACATTGCTATATTGAACAACAACTTGAATACCTT GATCCAGCGTTCCAACCACACTCAGGCCGCCAATGATCCCCCTGAGGTGACCGTGTTTC478 (spliced out region)---------------------------------------------------------------------------------------------------------------------------------------------644CATTACCTGACCTTTGTGCCCTCAGCAGAGGACGTCTATGACTGCAGGGTGGAGCACTGGGGCTTGGACCAGCCGCTCCTCAAGCACTGGGAGGCCCAAGAGCCAATCCAGATGCCTGAGACAACGGAGACTGTGCTCTGTGCCCTGGGCCTGGTGCTGGGCCTAGTGGGCATCATCGTGGGCACCGTCCTCATCATAAA844   cDNA HLA-DPA1_(01-r.ab1 Reverse primer)   796GGCCCAGGGCACAGAGCACAGTCTCCGTTGTCTCAGGCATCTGGATTGGCTCTTGGGCCTCCCAGTGCTTGAGGAGCGGCTGGTCCAAGCCCCAGTGCTCCACCCTGCAGTCATAGACGTCCTCTGCTGAGGGCACAAAGGTCAGGTAATG644 (spliced out region)------------------------------------------------------------------------------------------------------------------------------------478 GAAACACGGTCACCTCAGGGGGATCATTGGCGGCCTGAGTGTGGTTGGAACGCTGGATCAAGGTATTCAAGTTGTTGTTCAATATAGCAATGTTAGCCAGCCCGCCCTGAGCCTCAAAGGAAAAGGCTCGGCCAAACTCCTCCAGATGCCAGACGGTCTCCTTCCTTGTCC307  The sequencing result for both the forward and reverse primer shows a consistent alternative splice site in exon 3.

3.5. Potential Biomarker Analysis (Experiment 4)
Since these MHC Class II molecules are predominantly immune in function, we explored potential biomarkers using 87 EBV-transformed LCLs (Table 3, demographics) collected from Costa Rica. The four groups were closely matched in demographics. This sample was comprised of gender-and age-matched living subjects collected for a separate genetic study of cases and controls in Costa Rica [32]. The subjects are of Spanish ancestry, and admixed [33]. The samples of LCLs were grown in similar passages and timeframes in the lab (LM), using similar protocols for each cell line and RNA extraction. An ideal biomarker would focus on whole blood from subjects, instead of derived LCLs.
We tested expression of three MHC Class II genes for differential expression by qPCR. The results showed that CD74 and HLA-DPA1 were both significantly decreased in BD, and trends for a decrease in SZ were also found (Table 11). These findings were supportive of the decreases seen in different brain regions for these MHC Class II genes for BD and SZ. Interestingly, HLA-DRB1 was not different between the three groups, consistent with a lack of differences in MDD, SZ, or BD as shown by brain expression studies.

3.6. Blood−Brain Comparison
We extracted from exon array analysis the three MHC II mRNA expression levels for HLA-DPA1, CD74, and HLA-DRB1 from 187 different samples that had been run in-house for a variety of experiments. Those samples are broadly categorized as brain (n = 29); whole blood and PBMC (n = 84); and LCLs (n = 74) (Table 12 shows details for each category). These three tissues were summarized for exon array expression (brain, transformed cell lines, and non-transformed cells) and the results are shown in Figure 9 A,B,C. As expected, both transformed and non-transformed cells show almost identical expression levels, and were significantly increased from 4–16 times higher expression compared to brain. To determine that residual blood levels were not affecting the brain levels measured by exon array, a series of three brains were measured by analogous rat exon array, with one brain cleared with saline rinse perfusion via carotid artery. The brain levels of three MHC Class II molecules were not altered by perfusion, thus indicating that brain levels, although lower than blood levels, were not the result of residual amounts of blood within the brain.
The HLA-DPA1 gene is located at the edge of the strong MHC association signal to schizophrenia [18] shown in Figure 10. A highly significant SNP associated with schizophrenia is 15 kB downstream of the eQTL for HLA-DPA1 and D’ is 0.82 between rs112790520, a proxy for rs155327711, and rs9277341. The complete absence of HLA-DPA1 gene expression in cortex from fetal cortical samples, and near absence in infant cortex by RNA-Seq [34], is contrasted to later developmental epochs (child, teen, adult, and elderly) that show high levels of HLA-DPA1 (at FWER < 5%).