Schizophrenia risk from complex variation of complement component 4

Abstract

Schizophrenia is a heritable brain illness with unknown pathogenic mechanisms. Schizophrenia's strongest genetic association at a population level involves variation in the major histocompatibility complex (MHC) locus, but the genes and molecular mechanisms accounting for this have been challenging to identify. Here we show that this association arises in part from many structurally diverse alleles of the complement component 4 (C4) genes. We found that these alleles generated widely varying levels of C4A and C4B expression in the brain, with each common C4 allele associating with schizophrenia in proportion to its tendency to generate greater expression of C4A. Human C4 protein localized to neuronal synapses, dendrites, axons, and cell bodies. In mice, C4 mediated synapse elimination during postnatal development. These results implicate excessive complement activity in the development of schizophrenia and may help explain the reduced numbers of synapses in the brains of individuals with schizophrenia.

Extended Data Figure 1

Association of schizophrenia to common variants in the MHC locus in individual case-control cohorts, and schematic of the repeat module containing C4. (a–f) Data for several schizophrenia case-control cohorts that were genome-scanned before we began this work (a–d) exhibits peaks of association near chr6:32Mb (blue vertical line) on the human genome reference sequence (GRCh37/hg19). Note that association patterns vary from cohort to cohort, reflecting statistical sampling fluctuations and potentially fluctuations in allele frequencies of the (unknown) causal variants in different cohorts. Cohorts such as in (b), (e) and (f) suggest the existence of effects at multiple loci within the MHC region. Even in the cohorts with simpler peaks (a, c, d), the pattern of association across the individual SNPs at chr6:32 Mb does not correspond to the linkage disequilibrium (LD) around any known variant. This motivated the focus in the current work on cryptic genetic influences in this region that could cause unconventional association signals that do not resemble the LD patterns of individual variants. (g) A complex form of genome structural variation resides near chr6:32 Mb. Shown here are three of the known alternative structural forms of this genomic region. The most prominent feature of this structural variation is the tandem duplication of a genomic segment that contains a C4 gene, 3’ fragments of the STK19 and TNXB genes, and a pseudogenized copy of the CYP21A2 gene. (This cassette is present in 1–3 copies on the three alleles depicted above; the boundaries below each haplotype demarcate the sequence that is duplicated.) Haplotypes with multiple copies of this module (middle and bottom) contain multiple functional copies of C4, whereas the additional gene fragments or copies denoted STK19P, CYP21A2P, and TNXA are typically pseudogenized. (Rare haplotypes with a gain or loss of intact CYP21A2 have also been observed.) Note that although C4A and C4B contain multiple sequence variants, they are defined based on the differences encoded by exon 26, which determine the relative affinities of C4A and C4B for distinct molecular targets^, (Fig. 1). Many additional forms of this locus appear to have arisen by non-allelic homologous recombination and gene conversion (ref and Fig. 1).

Extended Data Figure 2

Schematic of strategy for identifying the segregating structural forms of the C4 locus. (a) Molecular assays for measuring copy number of the key, variable C4 structural features – the length polymorphism (HERV insertion) that distinguishes the long (L) from the short (S) genomic form of C4, and the C4A/C4B isotypic difference. Each primer-probe-primer assay is represented with the combination of arrows (primers) and asterisk (probe) in its approximate genomic location (though not to scale). (b) Measurement of copy number of C4 gene types in the genomes of 162 individuals (from HapMap CEU sample). The absolute, integer copy number of each C4 gene type in each genome is precisely inferred from the resulting data. To ensure high accuracy, the data are further evaluated for a checksum relationship (A + B = L + S) and for concordance with earlier data from Southern blotting of 89 of the same HapMap individuals. (c) To measure the copy number of compound structural forms of C4 (involving combinations of L/S and A/B), we perform long-range PCR followed by quantitative measurement of the A/B isotype-distinguishing sequences in droplets. (d) Analysis of transmissions in father-mother-offspring trios enables inference of the C4 gene contents of individual copies (alleles) of chromosome 6. Three example trios are shown in this schematic. (e) Examples of the inferred structural forms of the C4 locus (more shown in Fig. 1c). For the common C4 structures (AL-BL, AL-BS, AL-AL, and BS), genomic order of the C4 gene copies is known from earlier assemblies of sequence contigs in individuals homozygous for MHC haplotypes due to consanguinity and other molecular analyses of the C4 locus. For the rarer C4 structures, genomic order of C4 gene copies is hypothesized or provisional.

Extended Data Figure 3

Linkage disequilibrium relationships (r²) of MHC SNPs to forms of C4 structural variation. Correlations of SNPs in the MHC locus with (a) copy number of C4 gene types and (b) larger-scale structural forms (haplotypes) of the C4 locus. Dashed, vertical lines indicate the genomic location of the C4 locus. Note that C4 structural forms show only partial correlation (r²) to the allelic states of nearby SNPs, reflecting the relationship shown in Fig. 2, in which a structural form of the C4 locus often segregates on multiple different SNP haplotypes.

Extended Data Figure 4

RNA expression of C4A and C4B in relation to copy number of C4A, C4B, and the C4-HERV (long form of C4), in eight panels of post mortem brain tissue. Copy number of C4 structural features was measured by ddPCR; RNA expression levels were measured by RT-ddPCR. Panels a–e show data for tissues from the Stanley Medical Research Institute (SMRI) Array Consortium and consist of (a) anterior cingulate cortex, (b) cerebellum, (c) corpus callosum, (d) orbital frontal cortex, and (e) parietal cortex. Panel f shows data for the frontal cortex samples from the NHGRI Genes and Tissues Expression (GTEx) Project. Panels g and h show data for tissues from the SMRI Neuropathology Consortium (anterior cingulate cortex and cerebellum, respectively). These data were then used to inform (by linear regression) the derivation of a linear model for predicting each individual’s RNA expression of C4A and C4B as a function of the numbers of copies of AL, BL, AS, and BS. The derivation of this model, and the regression coefficients induced, are described in Methods. In the rightmost plot of each panel, expression of C4A (per genomic copy) is normalized to expression of C4B (per genomic copy) to more specifically visualize the effect of the C4-HERV by controlling for genomic copy number and for any trans-acting influences shared by C4A and C4B; the inferred regression coefficients (Methods) suggest that the observed effect is mostly due to increased expression of C4A.

Extended Data Figure 5

Detailed analysis of the association of schizophrenia to genetic variation at and around C4, in data from 28,799 schizophrenia cases and 35,986 controls (Psychiatric Genomics Consortium, ref ). SCZ, schizophrenia; β, estimated effect size per copy of the genomic feature or allele indicated; SE, standard error. Detailed association analyses of HLA alleles are in Extended Data Fig. 6–7. (*) We specifically tested C4B-null status because a 1985 study reported an analysis of 165 schizophrenia patients and 330 controls in which rare C4B-null status associated with elevated risk of schizophrenia, though two subsequent studies^, found no association of schizophrenia to C4B-null genotype. We sought to evaluate this using the large data set in this study, finding no association to C4B-null status. (**) Total copy number of C4 is also strongly correlated to copy number of the CYP21A2P pseudogene, which is present on duplicated copies of the sequence shown in Extended Data Fig. 1g.

Extended Data Figure 6

Evaluation of the association of schizophrenia with HLA alleles and coding-sequence polymorphisms. (a–e) Associations to HLA alleles and coding-sequence polymorphisms are shown in black; to provide the context of levels of association to nearby SNPs, associations to other SNPs are shown in gray. The series of conditional analyses shown (b–e) parallels the analyses in Fig. 4. Further detail on the most strongly associating HLA alleles (including conditional association analysis) is provided in Extended Data Fig. 7.

Extended Data Figure 7

Detailed association analysis for the most strongly associating classical HLA alleles. The most strongly associating HLA loci were HLA-B (in primary analyses, Fig. 4a, Extended Data Fig. 6a) and HLA-DRB1 and -DQB1 (in analyses controlling for the signal defined by rs13194504, Fig. 4c, Extended Data Fig. 6b). At these loci, the most strongly associating classical HLA alleles were HLA-B*0801, HLA-DRB1*0301, and HLA-DQB*02, respectively. These HLA alleles are all in strong but partial LD with C4 BS, the most protective of the C4 alleles; they are also in partial LD with the low-risk allele at rs13194505, representing the distinct signal several megabases to the left (Fig. 4). In joint analyses with each of these HLA alleles, genetically predicted C4A expression and rs13194505 continued to associate strongly with schizophrenia, while the HLA alleles did not. In further joint analyses with rs13194504 and genetically predicted C4A expression, 0 of 2,514 tested HLA SNP, amino-acid and classical-allele polymorphisms (from ref , including all variants with MAF > 0.005) associated to schizophrenia as strongly as rs13194504 or predicted C4A expression did.

Extended Data Figure 8

Expression of C4A RNA in brain tissue (five brain regions) from 35 schizophrenia cases and 70 non-schizophrenia controls, from the Stanley Medical Research Institute Array Consortium. C4A RNA expression levels were measured by ddPCR.

Extended Data Figure 9

Secretion of C4, and specificity of the monoclonal anti-C4 antibody for C4 protein in human brain tissue and cultured primary cortical neurons. (a) Brain tissue (from an individual affected with schizophrenia) was stained with a fluorescent secondary antibody, C4 antibody, or C4 antibody that was pre-adsorbed with purified C4 protein. Confocal images demonstrate the loss of immunoreactivity in the secondary-only and pre-adsorbed conditions. (b) Primary human neurons were stained with a fluorescent secondary antibody, C4 antibody, or C4 antibody that was pre-adsorbed with purified C4 protein. Confocal images demonstrate the loss of immunoreactivity in the secondary-only and pre-adsorbed conditions. Scale bar for all images = 25 µm. (c) Secretion of C4 protein by cultured primary neurons. Western blot for C4 protein analysis. (+) Purified human C4 protein. (−) Unconditioned medium, a negative control. (HN-conditioned) shows the same medium after conditioning by cultured human neurons at days 7 (d7) and 30 (d30). Details of Western blot protocol, antibody catalog numbers and concentrations used are in Methods. C4 molecular weight ~210 kDa.

Extended Data Figure 10

Mouse C4 genes and additional analyses of the dLGN eye segregation phenotype in C4 mutant mice and wild-type and heterozygous littermate controls. (a) The functional specialization of C4 into C4A and C4B in humans does not have an analogy in mice. Although the mouse genome contains both a C4 gene and a C4-like gene (classically called Slp), and these genes are also present as a tandem duplication within the mouse MHC locus, analysis of the encoded protein sequences indicates a distinct specialization, as illustrated by the protein phylogenetic tree. Above, mouse Slp is indicated in gray to reflect its potential pseudogenization: Slp is already known to have mutations at a C1s cleavage site, which are thought to abrogate activation of the protein through the classical complement pathway; and the M. musculus reference genome sequence (mm10) at Slp shows a 1-bp deletion (relative to C4) within the coding region at chr17:34815158, which would be predicted to cause a premature termination of the encoded protein. (In some genome data resources, mouse Slp and C4 have been annotated respectively as “C4a” (e.g. NM_011413.2) and “C4b” (e.g. NM_009780.2) based on synteny with the human C4A and C4B genes, but the above sequence analysis indicates that they are not paralogous to C4A and C4B.) (b) Sequence differences between C4A and C4B – which are otherwise 99.5% identical at an amino acid level – are concentrated at the “isotypic site” where they shape each isotype's relative affinity for different molecular targets^,. At the isotypic site, mouse C4 contains a combination of the residues present in human C4A and C4B. (c) Expression of mouse C4 mRNA in whole retina and lateral geniculate nucleus (LGN) from P5 animals and in purified retinal ganglion cells (RGCs) from P5 and P15 animals. These time points were chosen as P5 is a time of more robust synaptic refinement in the retinogeniculate system compared to P15. The same assays detected no C4 RNA in control RNA isolated from C4−/− mice (not shown). N = 3 samples for p5 retina, LGN, and P15 RGCs, N = 4 samples for P5 RGCs; * p < 0.05 by ANOVA with post hoc Tukey-Kramer multiple-comparisons test. (d) Representative images of dLGN innervation by contralateral projections (red in bottom image), ipsilateral projections (green in bottom image), and their overlap (yellow in bottom image). Scale bar = 100 µm (e) Quantification of the percentage of total dLGN area receiving both contralateral and ipsilateral projections shows a significant increase in C4−/− compared to WT littermates (ANOVA, N = 5 mice/group, p < 0.01). These data are consistent with results using R-value analysis as shown in Fig. 7. (f) Quantification of total dLGN area showed no significant difference between WT and C4−/− mice (ANOVA, N = 5 per group, p > 0.05). (g) Quantification of dLGN area receiving ipsilateral innervation showed a significant increase in ipsilateral territory in the C4−/− mice compared to WT littermates (AVOVA, N = 5 mice/group, p > 0.01). This result is consistent with defects in eye specific segregation. Scale bar = 100 µm (h) The number of RGCs in the retina was estimated by counting the number of Brn3a+ cells in WT and C4−/− mice. No differences were observed between WT and C4−/− (t-test, N = 4 mice/group, p > 0.05). Scale bar = 100 µm.

Figure 1. Structural variation of the complement component 4 (C4) gene

(a) Location of the C4 genes within the Major Histocompatibility Complex (MHC) locus on human chromosome 6. (b) Human C4 exists as two paralogous genes (isotypes), C4A and C4B; the encoded proteins are distinguished at a key site that determines which molecular targets they bind^,. Both C4A and C4B also exist in both long (L) and short (S) forms distinguished by an endogenous retroviral (C4-HERV) sequence in intron 9. (c) Structural forms of the C4 locus and their frequencies among a European-ancestry population sample (222 chromosomes from 111 genetically unrelated individuals, HapMap CEU), inferred as described in Extended Data Fig. 2. Asterisks indicate allele frequencies too low to be well-estimated.

Figure 2. Haplotypes formed by C4 structures and SNPs

SNP haplotype(s) on which common C4 structures were present. Each thin horizontal line represents the series of SNP alleles (haplotype) along a 250-kilobase chromosomal segment. Each column represents a SNP; gray and black indicate which allele is present on each haplotype. The SNP haplotypes are grouped into 13 sets of haplotypes associating with each of the four most common C4 structures. Three C4 structures (AL-BS, AL-BL, and AL-AL) each segregated on multiple SNP haplotypes (numbered at right).

Figure 3. Brain RNA expression of C4A and C4B in relation to copy numbers of C4A, C4B, and the C4-HERV

mRNA expression of C4A (a) and C4B (b) was measured (by ddPCR) in brain tissue from 244 individuals. Copy number of C4A, C4B, and the C4-HERV were measured (by ddPCR analysis of genomic DNA) in the brain donors. The results were consistent across 8 panels of brain tissue representing 5 brain regions and 3 distinct sets of donors (one set shown here, with data from 101 individuals; all panels in Extended Data Fig. 4; a few outlier points are beyond the range of these plots but are shown in Extended Data Fig. 4.) P-values were obtained by a Spearman rank correlation test. In panel c, expression of C4A (per genomic copy) is normalized to expression of C4B (per genomic copy) to control for trans-acting influences shared by C4A and C4B.

Figure 4. Association of schizophrenia to C4 and the extended MHC locus

Association of schizophrenia to 7,751 SNPs across the MHC locus and to genetically predicted expression levels of C4A and C4B in the brain (represented in the genomic location of the C4 gene). The data shown are based on analysis of 28,799 schizophrenia cases and 35,986 controls of European ancestry from the Psychiatric Genomics Consortium. The height of each point represents the statistical strength (−log₁₀(p)) of association with schizophrenia. (a, b) Association of schizophrenia to SNPs in the MHC locus and to genetically predicted expression of C4A and C4B. In (b), genetic variants are colored by their levels of correlation to rs13194504 (upper panel) or by their levels of correlation to genetically predicted brain C4A expression levels (lower panel). (c–f) Conditional association analysis. The red dashed line indicates the statistical threshold for genome-wide significance (p = 5×10⁻⁸). See also Extended Data Fig. 5–7 for detailed association analyses involving C4 locus structures and HLA alleles.

Figure 5. C4 structures, C4A expression, and schizophrenia risk

(a) Schizophrenia risk associated with four common structural forms of C4 in analysis of 28,799 schizophrenia cases and 35,986 controls. (b) Brain C4A RNA expression levels associated with four common structural forms of C4. β was calculated from fitting C4A RNA expression (in brain tissue) to the number of chromosomes (0, 1, or 2) carrying each C4 structure (across 120 individuals sampled). (c) Schizophrenia risk associated with 13 combinations of C4 structural allele and MHC SNP haplotype. The numbers on the y-axis adjacent to the C4 structures indicate the “haplogroup”, the MHC SNP haplotype background on which the C4 structure segregates, and correspond to Fig. 2. Statistical tests of heterogeneity yielded p = 0.55 for AL-AL alleles; p = 0.93 for AL-BL alleles; p = 0.06 for AL-BS alleles; and p = 5.7 × 10⁻⁵ across the overall allelic series. (d) Expression levels of C4A RNA were directly measured (by RT-ddPCR) in post mortem brain samples from 35 schizophrenia patients and 70 individuals not affected with schizophrenia. Measurements for all five brain regions analyzed exhibited the same relationship (Extended Data Fig. 8. Horizontal lines show the median value for each group. P-values were derived by a (non-parametric) one-sided Mann-Whitney test. Error bars shown in a–c represent 95% confidence intervals around the effect size estimate.

Figure 6. C4 protein at neuronal cell bodies, processes and synapses

(a) C4 protein localization in human brain tissue. Two representative confocal images (drawn from immunohistochemistry performed on samples from five individuals with schizophrenia and two unaffected individuals) within the hippocampal formation demonstrate localization of C4 in a subset of NeuN⁺ neurons. (b) High-resolution structured illumination microscopy (SIM) imaging of tissue in the hippocampal formation reveals colocalization of C4 with the presynaptic terminal marker Vglut1/2 and the postsynaptic parker PSD95. (c) Confocal images of primary human cortical neurons show colocalization of C4, MAP2, and neurofilament along neuronal processes. (d) Confocal image of primary cortical neurons stained for C4, presynaptic marker synaptotagmin, and postsynaptic marker PSD95. Scale bar for a, c, and d = 25 µm; b = 5 µm; b (inset)= 1 µm. Extended Data Fig. 9 contains additional data on antibody specificity

Figure 7. C4 in retinogeniculate synaptic refinement

(a) Representative confocal images of immunohistochemistry for C3 in the P5 dLGN showed reduced C3 deposition in the dLGN of C4−/− mice compared to WT littermates. (b) Quantification confirmed reduced C3 immunoreactivity in the dLGN (N = 3 mice/group, p < 0.05, t-test; y-axis: mean fluorescence intensity, normalized to WT). (c) Co-localization analysis revealed a reduction in the fraction of VGLUT2+ puncta that were C3+ in C4-deficient mice relative to their WT littermates (N = 3 mice/group, p = 0.0011, two-sided t-test). (d) Synaptic refinement in mice with 0, 1, or 2 copies of C4. These images represent the segregation of ipsilateral and contralateral RGC projections to the dLGN; two analysis methods were used. (Top) Projections from the ipsilateral (green) and contralateral (red) eyes show minimal overlap (yellow) in WT mice. The overlapping area is significantly increased in C4−/− mice (N = 6 mice/group, p < 0.01, ANOVA with Bonferroni post tests). (Bottom) Threshold-independent analysis using the R-value(R = log₁₀[F_ipsi/F_contra]). Pixels are pseudocolored with an R-value heat map (red indicates areas having only contralateral inputs; purple, only ipsilateral inputs). Compared to their WT littermates, C4-deficient mice exhibited lower R-value variance, indicating defects in synaptic refinement (N = 6 mice/group, p < 0.001, ANOVA with Bonferroni post tests). Control experiments analyzing total dLGN size, dLGN area receiving ipsilateral input, and number of RGCs are shown in Extended Data Fig. 10f–h, respectively. Error bars in (b), (c), and (d) represent S.E.M.