Abstract
Many common illnesses differentially affect men and women for unknown reasons. The autoimmune diseases lupus and Sjögren’s syndrome affect nine times more women than men1,2, whereas schizophrenia affects men more frequently and severely3–5. All three illnesses have their strongest common-genetic associations in the Major Histocompatibility Complex (MHC) locus, an association that in lupus and Sjögren’s syndrome has long been thought to arise from HLA alleles6–13. Here we show that the complement component 4 (C4) genes in the MHC locus, recently found to increase risk for schizophrenia14, generate 7-fold variation in risk for lupus (95% CI: 5.88-8.61; p < 10−117 in total) and 16-fold variation in risk for Sjögren’s syndrome (95% CI: 8.59-30.89; p < 10−23 in total), with C4A protecting more strongly than C4B in both illnesses. The same alleles that increase risk for schizophrenia, greatly reduced risk for lupus and Sjögren’s syndrome. In all three illnesses, C4 alleles acted more strongly in men than in women: common combinations of C4A and C4B generated 14-fold variation in risk for lupus and 31-fold variation in risk for Sjögren’s syndrome in men (vs. 6-fold and 15-fold among women respectively) and affected schizophrenia risk about twice as strongly in men as in women. At a protein level, both C4 and its effector (C3) were present at greater levels in men than women in cerebrospinal fluid (p < 10−5 for both C4 and C3) and plasma among adults ages 20-5015–17, corresponding to the ages of differential disease vulnerability. Sex differences in complement protein levels may help explain the larger effects of C4 alleles in men, women’s greater risk of SLE and Sjögren’s, and men’s greater vulnerability in schizophrenia. These results nominate the complement system as a source of sexual dimorphism in vulnerability to diverse illnesses.
Systemic lupus erythematosus (SLE, or “lupus”) is a systemic autoimmune disease of unknown cause. Risk of SLE is heritable (66%18), though SLE may have environmental triggers, as its onset often follows events that damage cells, such as infections and severe sunburns19. Most SLE patients produce autoantibodies against nucleic acid complexes, including ribonucleoproteins and DNA20.
In genetic studies, SLE associates most strongly with variation across the major histocompatibility complex (MHC) locus6,7,21. However, conclusive attribution of this association to specific genes and alleles has been difficult; the identities of the most likely genic and allelic culprits have been frequently revised as genetic studies have grown in size8–11. In several other autoimmune diseases, including type 1 diabetes, celiac disease, and rheumatoid arthritis, strong effects of the MHC locus arise from HLA alleles that cause the peptide binding groove of HLA proteins to present a disease-critical autoantigen22–24. In SLE, by contrast, MHC alleles associate broadly with the presence of diverse autoantibodies25.
The complement component 4 (C4A and C4B) genes are also present in the MHC locus, between the class I and class II HLA genes. Classical complement proteins help eliminate debris from dead and damaged cells, attenuating the exposure of diverse intracellular proteins to the adaptive immune system. C4A and C4B commonly vary in genomic copy number26–28 and encode complement proteins with distinct affinities for molecular targets29,30. SLE frequently presents with hypocomplementemia that worsens during flares, possibly reflecting increased active consumption of complement31. Rare cases of severe, early-onset SLE can involve complete deficiency of a complement component (C4, C2, or C1Q)32,33, and one of the strongest common-variant associations in SLE maps to ITGAM, which encodes a receptor for C3, the downstream effector of C421,34. Though total C4 gene copy number associates with SLE risk35–37, this association is thought to arise from linkage disequilibrium (LD) with nearby HLA alleles38, which have been the focus of fine-mapping analyses6–11,21.
The complex genetic variation at C4 – arising from many alleles with different numbers of C4A and C4B genes – has been challenging to analyze in large cohorts. A recently feasible approach to this problem is based on imputation: people share long haplotypes with the same combinations of SNP and C4 alleles, such that C4A and C4B gene copy numbers can be imputed from SNP data14. To analyze C4 in large cohorts, we developed a way to identify C4 alleles from whole-genome sequence (WGS) data (Fig. 1), then analyzed WGS data from 1,265 individuals (from the Genomic Psychiatry Cohort39,40) to create a large multi-ancestry panel of 2,530 reference haplotypes of MHC SNPs and C4 alleles (Extended Data Fig. 1) – ten times more than in earlier work14. We then analyzed SNP data from the largest SLE genetic association study7 (ImmunoChip 6,748 SLE cases and 11,516 controls of European ancestry) (Extended Data Fig. 2), imputing C4 alleles to estimate the SLE risk associated with common combinations of C4A and C4B gene copy numbers (Fig. 2a).
Groups with the eleven most common combinations of C4A and C4B gene copy number exhibited 7-fold variation in their risk of SLE (Fig. 2a and Extended Data Fig. 3). The relationship between SLE vulnerability and C4 gene copy number exhibited consistent, logical patterns across the 11 genotype groups. For each C4B copy number, greater C4A copy number associated with reduced SLE risk (Fig. 2a, Extended Data Fig. 3). For each C4A copy number, greater C4B copy number associated with more modestly reduced risk (Fig. 2a). Logistic-regression analysis estimated that the protection afforded by each copy of C4A (OR: 0.54; 95% CI: [0.51, 0.57]) was equivalent to that of 2.3 copies of C4B (OR: 0.77; 95% CI: [0.71,0.82]). We calculated an initial C4-derived risk score as 2.3 times the number of C4A genes, plus the number of C4B genes, in an individual’s genome. Despite clear limitations of this risk score – it is imperfectly imputed from flanking SNP haplotypes (r2 = 0.77, Extended Data Table 1) and only approximates C4-derived risk by using a simple, linear model (to avoid over-fitting the genetic data) – SNPs across the MHC locus tended to associate with SLE in proportion to their level of LD with this risk score (Fig. 2b).
Combinations of many different C4 alleles generate the observed variation in C4A and C4B gene copy number; particular C4A and C4B gene copy numbers have also arisen recurrently on multiple SNP haplotypes14 (Extended Data Fig. 1). Analysis of SLE risk in relation to each of these alleles and haplotypes reinforced the conclusion that C4A contributes strong protection, and C4B more modest protection, from SLE, and that C4 genes (rather than nearby variants) are the principal drivers of this variation in risk levels (Fig. 2c).
These results prompted us to consider whether other autoimmune disorders with similar patterns of genetic association at the MHC locus might also be driven in part by C4 variation. Sjögren’s syndrome (SjS) is a heritable (54%41) systemic autoimmune disorder of exocrine glands, characterized primarily by dry eyes and mouth with other systemic effects. At a protein level, SjS is (like SLE) characterized by diverse autoantibodies42, including antinuclear antibodies targeting ribonucleoproteins43, and hypocomplementemia44,45. The largest source of common genetic risk for SjS lies in the MHC locus46, with associations to the same haplotype(s) as in SLE12,13 and with heterogeneous HLA associations in different ancestries47. We imputed C4 alleles into existing SNP data from a European-ancestry SjS case-control cohort (673 cases and 1153 controls). As in SLE, logistic-regression analyses found both C4A copy number (OR: 0.41; 95% CI: [0.34, 0.49]) and C4B copy number (OR: 0.67; 95% CI: [0.53, 0.86]) to be protective against SjS. The risk-equivalent ratio of C4B to C4A gene copies was similar in SjS and SLE (about 2.3 to 1); also, as with SLE, nearby SNPs associated with SjS in proportion to their LD with a C4-derived risk score ((2.3)C4A+C4B) (Fig. 2d). The distribution of SjS risk across the individual C4 alleles and haplotypes revealed a pattern that (as in SLE) supported greater protective effect from C4A than C4B, and little effect of flanking SNP haplotypes (Fig. 2e).
The association of SLE and SjS with C4 gene copy number has long been attributed to the HLA-DRB1*03:01 allele. In European populations, DRB1*03:01 is in strong LD (r2 = 0.71) with the common C4-B(S) allele, which lacks any C4A gene and is the highest-risk C4 allele in our analysis (Fig. 2c); many MHC SNPs associated with SLE and SjS in proportion to their LD correlations with both C4 and DRB1*03:01 (Extended Data Fig. 4a, b). Cohorts with other ancestries can have recombinant haplotypes that disambiguate the contributions of alleles that are in LD in Europeans. Among African Americans, we found that common C4 alleles exhibited far less LD with HLA alleles; in particular, the LD between C4-B(S) and DRB1*03:01 was low (r2 = 0.10) (Extended Data Table 2). Thus, genetic data from an African American SLE cohort (1,494 cases, 5,908 controls) made it possible to distinguish between these potential genetic effects. Joint association analysis of C4A, C4B, and DRB1*0301 implicated C4A (p < 10−14) and C4B (p < 10−5) but not DRB1*0301 (p = 0.29) (Extended Data Table 3). Each C4 allele associated with effect sizes of similar magnitude on SLE risk in Europeans and African Americans (Fig. 3a). An analysis specifically of combinations of C4-B(S) and DRB1*03:01 allele dosages in African Americans showed that C4-B(S) alleles consistently increased SLE risk regardless of DRB1*03:01 status, whereas DRB1*03:01 had no consistent effect when controlling for C4-B(S) (Fig. 3b). Although C4 alleles had less LD with nearby variants on African American than on European haplotypes, SNPs associated with SLE in proportion to LD correlations with C4 in African Americans as well (Extended Data Fig. 4c).
We next sought to find other potential contributions of the MHC locus to SLE risk by accounting for contributions from C4. SNPs across the MHC locus display very different associations with SLE in Europeans and African Americans7,11, though the SNPs with European-specific associations tend to have strong LD to C4 in Europeans (Fig. 3c). To control for C4 genotypes, many of which exhibit strong LD across the MHC locus in Europeans (Extended Data Fig. 1), we adjusted the association data for C4-derived risk using a more-complete C4-derived risk score derived from the genotype-group risk measurements in Fig. 2a. Once adjusted for C4 effects, the residual association signals in the two populations became strongly correlated (Fig. 3d). Both populations also pointed to the same small haplotype of two variants as the most likely driver of an additional genetic effect independent of C4 (Fig. 3d and Supplementary Note). The two variants defining this short haplotype reside within the XL9 regulatory region48,49, a well-studied region of open chromatin that contains abundant chromatin marks characteristic of active enhancers and transcription factor binding sites (Supplementary Note). One of these variants, rs2105898, disrupts a binding site for ZNF14350, which anchors interactions of distal enhancers with gene promoters51 (Supplementary Note). Data from the GTEx Consortium 52 (v7) included 227 instances (gene/tissue pairs) in which this haplotype associated with elevated (HLA-DRB1, -DRB5, -DQA1, and -DQB1) or reduced (HLA-DRB6, -DQA2, and -DQB2) expression of an HLA class II gene with at least nominal (p < 10−4) significance. Some of the strongest associations at each gene (p < 10−8 to 10−76) were in whole blood, but expression QTLs elsewhere can also reflect the presence of blood and immune cells within those tissues.53 (Although eQTL analyses of HLA genes may be affected by read-alignment artifacts in these genes’ hyperpolymorphic domains, most such observed signals are robust after adjusting for individual HLA alleles.54)
The haplotype with elevated expression of HLA-DRB1, -DRB5, -DQA1, and -DQB1 (allele frequency 0.20 among Europeans, 0.22 among African Americans) associated with increased SLE risk (odds ratio) of 1.52 (95% CI: 1.44-1.61;p < 10−48) in Europeans and 1.49 (95% CI: 1.35-1.63; p < 10−16) in African Americans in analyses adjusting for C4 effects. The risk haplotype was in strong LD with DRB1*15:01 in Europeans and DRB1*15:03 in African Americans, which may explain earlier findings of population-specific associations with DRB1*15:01 in Europeans and DRB1*15:03 in African Americans7,11. The risk haplotype tagged by rs2105898 tended to be on low-risk C4 haplotypes in Europeans, a relationship that may have made both genetic influences harder to recognize in earlier work; controlling for either rs2105898 or C4 (Extended Data Fig. 5a) greatly increased the association of SLE with the other genetic influence (Extended Data Table 3). Controlling for the simpler (2.3)C4A+C4B model in SNP associations with SjS (as precision of estimates of individual alleles were low due to sample size) also pointed strongly to the same haplotype, with the same allele of rs2105898 associating in the same direction but larger effect (OR: 1.96; 95% CI: 1.64-2.34) as compared to SLE (Extended Data Fig. 5b).
Alleles at C4 that increase dosage of C4A, and to a lesser extent C4B, appear to protect strongly against SLE and SjS (Fig. 2a-c); by contrast, alleles that increase expression of C4A in the brain are more common among individuals with schizophrenia6. These same illnesses exhibit striking, and opposite, sex differences: SLE and SjS are nine times more common among women of childbearing age than among men of a similar age1,2, whereas in schizophrenia, women exhibit less severe symptoms, more frequent remission of symptoms, lower relapse rates, and lower overall incidence 3–5. Hence, we sought to evaluate the possibility that the effects of C4 alleles on the risk of each disease might also differ between men and women.
Analysis indicated that the effects of C4 alleles in both lupus and schizophrenia were stronger in men. When a sex-by-C4 interaction term was included in association analyses, this term was significant for both SLE (p < 0.01) and schizophrenia (p < 0.01), indicating larger C4 effects in men for both disorders. (Analysis of SjS had limited power due to the small number of men affected by SjS – 60 of the 673 cases in the cohort – but pointed to the same direction of effect at p = 0.07). For both SLE and schizophrenia, the individual C4 alleles consistently associated with stronger effects in men than women (Fig. 4a, b). These relationships explained previously reported sex biases 55 in SNP associations across the MHC locus (Fig. 4c-e).
The stronger effects of C4 alleles on male relative to female risk could arise from sex differences in C4 RNA expression, C4 protein levels, or downstream responses to C4. Analysis of RNA expression in 45 tissues, using data from GTEx52, identified no sex differences in C4 RNA expression. We then analyzed C4 protein in cerebrospinal fluid (CSF) from two panels of adult research participants (n = 589 total) in whom we had also measured C4 gene copy number by direct genotyping or imputation. CSF C4 protein levels correlated strongly with C4 gene copy number (p < 10−10, Extended Data Fig. 6a), so we normalized C4 protein measurements to the number of C4 gene copies. CSF from adult men contained on average 27% more C4 protein per C4 gene copy than CSF from women (meta-analysis p = 9.9 × 10−6, Fig. 4f). C4 acts by activating the complement component 3 (C3) protein, promoting C3 deposition onto targets in tissues. CSF levels of C3 protein were also on average 42% higher among men than women (meta-analysis p = 7.5 × 10−7, Fig. 4g).
The elevated concentrations of C3 and C4 proteins in CSF of men parallel earlier findings that, in plasma, C3 and C4 are also present at higher levels in men than women15–17. The large sample size (n > 50,000) of the plasma studies allows sex differences to be further analyzed as a function of developmental age. Both men and women undergo age-dependent elevation of C4 and C3 levels in plasma, but this occurs early in adulthood (age 20–30) in men and closer to menopause (age 40–50) in women, with the result that male– female differences in complement protein levels are observed primarily during the reproductive years (ages 20–50). We replicated these findings using measurements of C3 and (gene copy number-corrected) (Extended Data Fig. 6b) C4 protein in plasma from adults, finding (as in the earlier plasma studies and in CSF) that these differences are most pronounced during the reproductively active years of adulthood (ages 20-50) (Fig. 4h, i). We also observed that SjS patients have lower C4 serum levels than controls (p < 1×10−20, Extended Data Fig. 6c) even after correcting for C4 gene copy number (p < 1×10−8, Extended Data Fig. 6d), suggesting that hypocomplementemia in SjS is not simply due to C4 genetics but also reflects disease effects on ambient complement levels, for example due to complement consumption. The ages of pronounced sex difference in complement levels corresponded to the ages at which men and women differ in disease incidence: in schizophrenia, men outnumber women among cases incident in early adulthood, but not among cases incident after age 404,56; in SLE, women greatly outnumber men among cases incident during the child-bearing years, but not among cases incident after age 50 or during childhood57; in SjS, the large relative vulnerability of women declines in magnitude after age 50 58,59.
Our results indicate that the MHC locus shapes vulnerability in lupus and SjS – two of the three most common rheumatic autoimmune diseases – in a very different way than in type I diabetes, rheumatoid arthritis, and celiac disease. In those diseases, precise interactions between specific HLA alleles and specific autoantigens determine risk 22–24. In SLE and SjS, however, the genetic variation implicated here points instead to the continuous, chronic interaction of the immune system with very many potential autoantigens. Because complement facilitates the rapid clearance of debris from dead and injured cells, elevated levels of C4 protein likely attenuate interactions between the adaptive immune system and ribonuclear selfantigens at sites of cell injury, pre-empting the development of autoimmunity. The additional C4-independent genetic risk effect described here (associated with rs2105898) may also affect autoimmunity broadly, rather than antigen-specifically, by regulating expression of many HLA class II genes (including DRB1, DQA1, and DQB1). Mouse models of SLE indicate that once tolerance is broken for one self-antigen, autoreactive germinal centers generate B cells targeting other self-antigens60; such “epitope spreading” could lead to autoreactivity against many related autoantigens, regardless of which antigen(s) are involved in the earliest interactions with immune cells. Our genetic findings address the development of SLE and SjS rather than complications that arise in any specific organ. A few percent of SLE patients develop neurological complications that can include psychosis61; though psychosis is also a symptom of schizophrenia, neurological complications of SLE do not resemble schizophrenia more broadly, and likely have a different etiology.
The same C4 alleles that increase vulnerability to schizophrenia appeared to protect strongly against SLE and SjS. This pleiotropy will be important to consider in efforts to engage the complement system therapeutically. The complement system contributed to these pleiotropic effects more strongly in men than in women. Moreover, though the allelic series at C4 allowed human genetics to establish dose-risk relationships for C4, sexual dimorphism in the complement system also extended to complement component 3 (C3). Why and how biology has come to create this sexual dimorphism in the complement system in humans presents interesting questions for immune and evolutionary biology.
Methods
Creation of a C4 reference panel from whole-genome sequence data
We constructed a reference panel for imputation of C4 structural haplotypes using whole-genome sequencing data for 1265 individuals from the Genomic Psychiatry Cohort39. The reference panel included individuals of diverse ancestry, including 765 Europeans, 250 African Americans, and 250 people of reported Latino ancestry.
We estimated the diploid C4 copy number, and separately the diploid copy number of the contained HERV segment, using Genome STRiP64. Briefly, Genome STRiP carefully calibrates measurements of read depth across specific genomic segments of interest by estimating and normalizing away sample-specific technical effects such as the effect of GC content on read depth (estimated from the genome-wide data). To estimate C4 copy number, we genotyped the segments 6:31948358–31981050 and 6:31981096–32013904 (hg19) for total copy number, but masked the intronic HERV segments that distinguish short (S) from long (L) C4 gene isotypes. For the HERV region, we genotyped segments 6:31952461–31958829 and 6:31985199–31991567 (hg19) for total copy number. Across the 1,265 individuals, the resultant locus-specific copy-number estimates exhibited a strongly multi-modal distribution (Fig. 1a) from which individuals’ total C4 copy numbers could be readily inferred.
We then estimated the ratio of C4A to C4B genes in each individual genome. To do this, we extracted reads mapping to the paralogous sequence variants that distinguish C4A from C4B (hg19 coordinates 6:31963859–31963876 and 6:31996597–31996614) in each individual, combining reads across the two sites. We included only reads that aligned to one of these segments in its entirety. We then counted the number of reads matching the canonical active site sequences for C4A (CCC TGT CCA GTG TTA GAC) and C4B (CTC TCT CCA GTG ATA CAT). We combined these counts with the likelihood estimates of diploid C4 copy number (from Genome STRiP) to determine the maximum likelihood combination of C4A and C4B in each individual. We estimated the genotype quality of the C4A and C4B estimate from the likelihood ratio between the most likely and second most likely combinations.
To phase the C4 haplotypes, we first used the GenerateHaploidCNVGenotypes utility in Genome STRiP to estimate haplotype-specific copy-number likelihoods for C4 (total C4 gene copy number), C4A, C4B, and HERV using the diploid likelihoods from the prior step as input. Default parameters for GenerateHaploidCNVGenotypes were used, plus -genotypeLikelihoodThreshold 0.0001. The output was then processed by the GenerateCNVHaplotypes utility in Genome STRiP to combine the multiple estimates into likelihood estimates for a set of unified structural alleles. GenerateCNVHaplotypes was run with default parameters, plus -defaultLogLikelihood −50, -unknownHaplotypeLikelihood −50, and - sampleHaplotypePriorLikelihood 2.0. The resultant VCF was phased using Beagle 4.1 (beagle_4.1_27Jul16.86a) in two steps: first, performing genotype refinement from the genotype likelihoods using the Beagle gtgl= and -maxlr=1000000 parameters, and then running Beagle again on the output file using gt= to complete the phasing.
Our previous work suggested that several C4 structures segregate on different haplotypes, and probably arose by recurrent mutation on different haplotype backgrounds14. The GenerateCNVHaplotypes utility requires as input an enumerated set of structural alleles to assign to the samples in the reference cohort, including any structurally equivalent alleles, with distinct labels to mark them as independent, plus a list of samples to assign (with high likelihood) to specific labeled input alleles to disambiguate among these recurrent alleles. The selection of the set of structural alleles to be modeled, along with the labeling strategy, is important to our methodology and the performance of the reference panel. In the reference panel, each input allele represents a specific copy number structure and optionally includes a label that differentiates the allele from other independent alleles with equivalent structure. We use the notation <H_n_n_n_n_L> to identify each allele, where the four integers following the H are, respectively, the (redundant) haploid count of the total number of C4 copies, C4A copies, C4B copies and HERV copies on the haplotype. For example, <H_2_1_1_1> was used to represent the “AL-BS” haplotype. The optional final label L is used to distinguish potentially recurrent haplotypes with otherwise equivalent structures (under the model) that should be treated as independent alleles for phasing and imputation.
To build the reference panel, we experimentally evaluated a large number of potential sets of structural alleles and methods for assigning labels to potentially recurrent alleles. For each evaluation, we built a reference panel using the 1265 reference samples, and then evaluated the performance of the panel via cross-validation, leaving out 10 different samples in each trial (5 samples in the last trial) and imputing the missing samples from the remaining samples in the panel. The imputed results for all 1265 samples were then compared to the original diploid copy number estimates to evaluate the performance of each candidate reference panel (Extended Data Table 1).
Using this procedure, we selected a final panel for downstream analysis that used a set of 29 structural alleles representing 16 distinct allelic structures (as listed in the reference panel VCF file). Each allele contained from one to three copies of C4. Three allelic structures (AL-BS, AL-BL, and AL-AL) were represented as a set of independently labeled alleles with 9, 3, and 4 labels, respectively.
To identify the number of labels to use on the different alleles and the samples to “seed” the alleles, we generated “spider plots” of the C4 locus based on initial phasing experiments run without labeled alleles, and then clustered the resulting haplotypes in two dimensions based on the Y-coordinate distance between the haplotypes on the left and right sides of the spider plot. Clustering was based on visualizing the clusters (Extended Data Fig. 1) and then manually choosing both the number of clusters (labels) to assign and a set of confidently assigned haplotypes to use to “seed” the clusters in GenerateCNVHaplotypes. This procedure was iterated multiple times using cross-validation, as described above, to evaluate the imputation performance of each candidate labeling strategy.
Within the data set used to build the reference panel, there is evidence for individuals carrying seven or more diploid copies of C4, which implies the existence of (rare) alleles with four or more copies of C4. In our experiments, attempting to add additional haplotypes to model these rare four-copy alleles reduced overall imputation performance. Consequently, we conducted all downstream analyses using a reference panel that models only alleles with up to three copies of C4. In the future, larger reference panels might benefit from modeling these rare four-copy alleles.
The reference panel will be available in dbGaP (accession # pending) with broad permission for research use.
Genetic data for SLE
For analysis of systemic lupus erythematosus (SLE), collection and genotyping of the European-ancestry cohort (6,748 cases, 11,516 controls, genotyped by ImmunoChip) as previously described7. Collection and genotyping of the African-American cohort (1,494 cases, 5,908 controls, genotyped by OmniExpress) as previously described11.
Genetic data for SjS
For analysis of Sjogren’s syndrome (SjS), collection and genotyping of the European-ancestry cohort (673 cases, 1,153 controls, genotyped by Omni2.5) as previously described47 and available in dbGaP under study accession number phs000672.v1.p1.
Genetic data for schizophrenia
The schizophrenia analysis made use of genotype data from 40 cohorts of European ancestry (28,799 cases, 35,986 controls) made available by the Psychiatric Genetics Consortium (PGC) as previously described62. Genotyping chips used for each cohort are listed in Supplementary Table 3 of that study.
Imputation of C4 alleles
The reference haplotypes described above were used to extend the SLE, SjS, or schizophrenia cohort SNP genotypes by imputation. SNP data in VCF format were used as input for Beagle v4.165,66 for imputation of C4 as a multi-allelic variant. Within the Beagle pipeline, the reference panel was first converted to bref format. From the cohort SNP genotypes, we used only those SNPs from the MHC region (chr6:24-34 Mb on hg19) that were also in the haplotype reference panel. We used the conform-gt tool to perform strand-flipping and filtering of specific SNPs for which strand remained ambiguous. Beagle was run using default parameters with two key exceptions: we used the GRCh37 PLINK recombination map, and we set the output to include genotype probability (i.e., GP field in VCF) for correct downstream probabilistic estimation of C4A and C4B joint dosages.
Imputation of HLA alleles
For HLA allele imputation, sample genotypes were used as input for the R package HIBAG67. For both European ancestry and African American cohorts, publicly available multi-ethnic reference panels generated for the most appropriate genotyping chip (i.e. Immunochip for European ancestry SLE cohort, Omni 2.5 for European ancestry SjS cohort, and OmniExpress for African American SLE cohort) were used68. Default parameters were used for all settings. All class I and class II HLA genes were imputed. Output haplotype posterior probabilities were summed per allele to yield diploid dosages for each individual.
Associating single and joint C4 structural allele dosages to SLE and SjS in European ancestry individuals
The analysis described above yields dosage estimates for each of the common C4 structural haplotypes (e.g., AL-BS, AL-AL, etc.) for each genome in each cohort. In addition to performing association analysis on these structures (Fig 2c), we also performed association analysis on the dosages of each underlying C4 gene isotype (i.e. C4A, C4B, C4L, and C4S). These dosages were computed from the allelic dosage (DS) field of the imputation output VCF simply by multiplying the dosage of a C4 structural haplotype by the number of copies of each C4 isotype that haplotype contains (e.g., AL-BL contains one C4A gene and one C4B gene).
C4 isotype dosages were then tested for disease association by logistic regression, with the inclusion of four available ancestry covariates derived from genome-wide principal component analysis (PCA) as additional independent variables, PCc, where θ=E[SLE|X]. For SjS, the model instead included two available multiethnic ancestry covariates from dbGaP that correlated strongly with European-specific ancestry covariates (specifically, PC5 and PC7) and smoking status as independent variables. Coefficients for relative weighting of C4A and C4B dosages were obtained from a joint logistic regression,
The values per individual of β1C4A + β2C4B were used as a combined C4 risk term for estimating both association strength (Fig. 2b) as well as evaluating the relationship between the strength of nearby variants’ association with SLE or SjS and linkage with C4 variation (Extended Data Fig. 5a,b).
Joint dosages of C4A and C4B for each individual in the same cohort were estimated by summing across their genotype probabilities of paired structural alleles that encode for the same diploid copy numbers of both C4A and C4B (Extended Data Fig. 2a,b). For each individual/genome, this yields a joint dosage distribution of C4A and C4B gene copy number, reflecting any possible imputed haplotype-level dosages with nonzero probability. Joint dosages for C4A and C4B diploid copy numbers were tested for association with SLE in a joint model with the same ancestry covariates (Fig. 2a),
Calculation of composite C4 risk for SLE
Because SLE risk strongly associated with C4A and C4B copy numbers (Fig. 2a) in a manner that can be approximated as – but is not necessarily linear or independent – a composite C4 risk score was derived by taking the weighted sum of joint C4A and C4B dosages multiplied by the corresponding effect sizes from the aforementioned model of the joint C4A and C4B diploid copy numbers. The weights for calculating this composite C4 risk term were computed from the data from the European ancestry cohort, and then applied unchanged to analysis of the African American cohort.
Associations of variants across the MHC region to SLE and SjS
Genotypes for non-array SNPs were imputed with IMPUTE2 using the 1000 Genomes reference panel; separate analyses were performed for the European-ancestry and African American cohorts. Unless otherwise stated, all subsequent SLE analyses were performed identically for both European ancestry and African American cohorts. Dosage of each variant, vi, was tested for association with SLE or SjS in a logistic regression including available ancestry covariates (and smoking status for SjS) first alone (Fig. 2b, d), then with C4 composite risk (Extended Data Fig. 6a), and finally with C4 composite risk and rs2105898 dosage (Extended Data Fig. 6b), where θ=E[SLE|X]. For SjS, the simpler weighted (2.3)C4A+C4B model was used instead of composite risk term, as the cohort’s size gave poor precision to estimates of risk for many joint (C4A, C4B) copy numbers (Extended Data Fig. 6c, d). The Pearson correlation between the C4 composite risk term and each other variant was computed and squared (r2) to yield a measure of linkage disequilibrium between C4 composite risk and that variant in that cohort.
Association analyses for specific C4 structural alleles
The C4 structural haplotypes were tested for association with disease (Fig. 2c, 3a) in a joint logistic regression that included (i) terms for dosages of the five most common C4 structural haplotypes (AL-BS, AL-BL, AL-AL, BS, and AL), (ii) (for SLE and SjS) rs2105898 genotype, and (iii) ancestry covariates and (for SjS) smoking status, where θ=E[SLE|X]. Several of these common C4 structural alleles arose multiple times on distinct haplotypes; we term the set of haplotypes in which such a common allele appeared as “haplogroups”. The haplogroups can be further tested in a logistic regression model in which the structural allele appearing in all member haplotypes is instead encoded as dosages for each of the SNP haplotypes in which it appears. These association analyses (Fig. 2c) were performed as in (6), with structural allele dosages for ALBS, ALBL, and ALAL replaced by multiple terms for each distinct haplotype.
To delineate the relationship between C4-BS and DRB1*03:01 alleles – which are highly linked in European ancestry haplotypes – allelic dosages per individual in the African American SLE cohort were rounded to yield the most likely integer dosage for each. Although genotype dosages for each are reported by BEAGLE and HIBAG respectively, probabilities per haplotype are not linked and multiplying possible diploid dosages could yield incorrect non-zero joint dosages. Joint genotypes were tested as individual terms in a logistic regression model,
Sex-stratified associations of C4 structural alleles and other variants with SLE, SjS, and schizophrenia (Fig. 4a-d)
Determination of an effect from sex on the contribution of overall C4 variation to risk for each disorder was done by including an interaction term between sex and C4; ie. (2.3)C4A+C4B for SLE and SjS and estimated C4A expression for schizophrenia:
Each variant in the MHC region was tested for association with among European ancestry cases and cohorts in a logistic regression as in models (4)–(6) using only male cases and controls, and then separately using only female cases and controls (Fig. 4c-e). Likewise, allelic series analyses were performed as in (7), but in separate models for men and women (Fig. 4a, b).
To assess the relationship between sex bias in the risk associated with a variant and linkage to C4 composite risk (as non-negative r2), male and female log-odds were multiplied by the sign of the Pearson correlation between that variant and C4 composite risk before taking the difference.
Analyses of cerebrospinal fluid
Cerebrospinal fluid (CSF) from healthy individuals was obtained from two research panels. The first panel, consisting of 533 donors (327 male, 126 female) from hospitals around Utrecht, Netherlands, was described previously69,70. The donors were generally healthy research participants undergoing spinal anesthesia for minor elective surgery. The same donors were previously genotyped using the Illumina Omni SNP array. To estimate C4 copy numbers, we used SNPs from the MHC region (chr6:24-34 Mb on hg19) as input for C4 allele imputation with Beagle, as described above in Imputation of C4 alleles.
The second CSF panel sampled specimens from 56 donors (14 male, 42 female) from Brigham and Women’s Hospital (BWH; Boston, MA, USA) under a protocol approved by the institutional review board at BWH (IRB protocol ID no. 1999P010911). These samples were originally obtained to exclude the possibility of infection, and clinical analyses had revealed no evidence of infection. Donors ranged in age from 18 to 64 years old. Blood samples from the same individuals were used for extraction of genomic DNA, and C4 gene copy number was measured by droplet digital PCR (ddPCR) as previously described14. Samples were excluded from measurements if they lacked C4 genotypes, sex information, or contained visible blood contamination.
C4 measurements were performed by sandwich ELISA of 1:400 dilutions of the original CSF sample using goat anti-sera against human C4 as the capture antibody (Quidel, A305), FITC-conjugated polyclonal rabbit anti-human C4c as the detection antibody (Dako, F016902-2), and alkaline phosphatase–conjugated polyclonal goat anti-rabbit IgG as the secondary antibody (Abcam, ab97048). C3 measurements were performed using the human complement C3 ELISA kit (Abcam, ab108823).
Because C4 gene copy number had a large and proportional effect on C4 protein concentration in these CSF samples (Extended Data Fig. 7a), we corrected for C4 gene copy number in our analysis of relationship between sex and C4 protein concentration, by normalizing the ratio of C4 protein (in CSF) to C4 gene copies (in genome). Therefore, these analyses included only samples for which DNA was available or C4 was successfully imputed. In total, 495 (332 male, 163 female) C4 and 304 (179 male, 125 female) C3 concentrations were obtained across both cohorts. Log-concentrations of C3 (ng/mL) and C4 (ng/[mL, per C4 gene copy number]) protein were then used separately in linear regression models to estimate a sex-unbiased cohort-specific offset for each protein, to be applied to all concentrations for that protein. Estimation of average measurements by age for each sex was done by local polynomial regression smoothing (LOESS). To evaluate the significance of sex effects, we used these cohort-corrected concentrations estimates and analyzed them with the non-parametric unsigned Mann-Whitney rank–sum test comparing concentration distributions for males and females.
Analyses of blood plasma
Blood plasma was collected and immunoturbidimetric measurements of C3 and C4 protein in 1,844 individuals (182 men, 1662 women) by Sjögren’s International Collaborative Clinical Alliance (SICCA) from individuals with and without SjS as previously described71. C4 copy numbers for these individuals were previously imputed for use in logistic regression of SjS risk. As C4 copy number has an effect on measured C4 protein similar to CSF (Extended Data Fig. 7b), we normalized C4 levels to them in all following analyses. Estimation of average measurements by age for each sex was done by local polynomial regression smoothing (LOESS) on log-concentrations of C3 (mg/dL) and C4 (mg/[dL, per C4 gene copy number]) protein. To evaluate the significance of sex bias within age ranges displaying the greatest difference (informed by LOESS), we analyzed individuals in these bins with the non-parametric unsigned Mann-Whitney rank-sum test comparing concentration distributions for males and females.
Difference in C4 protein levels between individual with and without SjS was done by performing a non-parametric unsigned Mann-Whitney rank–sum test on C4 protein levels normalized to C4 genomic copy number (Extended Data Fig. 7c).
Author Contributions
S.A.M., N.K., and A.S. conceived the genetic studies. M.T.P., C.N.P., and M.B. collected and contributed whole-genome sequence data for the Genomic Psychiatry Cohort. R.E.H. and C.W.W. genotyped C4 structural variation in the Genomic Psychiatry Cohort and optimized variant selection for use as a reference panel in the imputation of C4 variation into lupus and schizophrenia cohorts (Fig. 1 and Extended Data Fig. 1). T.J.V., R.R.G., L.A.C., C.D.L., R.P.K., J.B.H., K.M.K., D.L.M., and P.T. contributed genotype data and imputation of non-C4 variation for analysis of SLE cohorts. K.E.T. and L.A.C. contributed genotype and phenotype data along with imputation of non-C4 variation for analysis of the SjS cohort. Investigators in the Schizophrenia Working Group of the Psychiatric Genomics Consortium collected and phenotyped cohorts and contributed genotype data for analysis of schizophrenia cohorts. N.K did the imputation and association analyses (Fig. 2, 3, 4a-e, and 4h, i and Extended Data Fig. 2–5, 6b-d, 7, and 8). T.J.V., R.R.G., and D.L.M. provided valuable advice on the analysis and interpretation of SLE association results. R.A.O. and L.M.O.L collected and provided CSF samples composing the group from Utrecht, Netherlands. C.E.S. collected and provided CSF samples composing the Brigham & Women’s Hospital group. H.d.R and K.T. performed the C4 and C3 immunoassay experiments on CSF samples (Fig. 4f, g and Extended Data Fig. 6a). S.A.M and N.K. wrote the manuscript with contributions from all authors.
Competing interests
The authors declare no competing interests.
Acknowledgements
This work was supported by the National Human Genome Research Institute (HG006855), the National Institute of Mental Health (MH112491, MH105641, MH105653), and the Stanley Center for Psychiatric Research. In addition, this work was supported by the Biomedical Research Centre based at Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London (D.L.M., P.T., and T.J.V.). We thank Christina Usher for contributions to the figures and manuscript text, Marta Florio for suggestions regarding figure display, and David Curtis and Chris Patil for suggestions on the manuscript.