Genetically Determined Partial Complement C4 Deficiency States Are Not Independent Risk Factors for SLE in UK and Spanish Populations

Abstract

Systemic lupus erythematosus (SLE) is a chronic, multisystem autoimmune disease. Complete deficiency of complement component C4 confers strong genetic risk for SLE. Partial C4 deficiency states have also shown association with SLE, but despite much effort over the last 30 years, it has not been established whether this association is primarily causal or secondary to long-range linkage disequilibrium. The complement C4 locus, located in the major histocompatibility complex (MHC) class III region, exhibits copy-number variation (CNV) and C4 itself exists as two paralogs, C4A and C4B. In order to determine whether partial C4 deficiency is an independent genetic risk factor for SLE, we investigated C4 CNV in the context of HLA-DRB1 and MHC region SNP polymorphism in the largest and most comprehensive complement C4 study to date. Specifically, we genotyped 2,207 subjects of northern and southern European ancestry (1,028 SLE cases and 1,179 controls) for total C4, C4A, and C4B gene copy numbers, and the loss-of-function C4 exon 29 CT indel. We used multiple logistic regression to determine the independence of C4 CNV from known SNP and HLA-DRB1 associations. We clearly demonstrate that genetically determined partial C4 deficiency states are not independent risk factors for SLE in UK and Spanish populations. These results are further corroborated by the lack of association shown by the C4A exon 29 CT insertion in either cohort. Thus, although complete homozygous deficiency of complement C4 is one of the strongest genetic risk factors for SLE, partial C4 deficiency states do not independently predispose to the disease.

Introduction

Systemic lupus erythematosus ([SLE] also known as lupus [MIM 152700]) is a chronic, multisystem, clinically heterogeneous autoimmune disease characterized by the presence of autoantibodies directed against nuclear and cellular components, complement activation, and immune complex deposition resulting in tissue inflammation and organ damage. There is a strong and complex yet incompletely understood genetic component to disease susceptibility. SNP-based genome-wide association studies have shown that polymorphisms within the major histocompatibility complex (MHC) region, located on the short arm of chromosome 6, confer the greatest genetic risk for SLE.^1–3 The classical MHC region comprises three subregions: the telomeric class I region, the centromeric class II region, and the intervening class III region. The MHC class I and class II regions encode the human leucocyte antigen (HLA) class I and class II molecules involved in antigen presentation to T lymphocytes.⁴ Genetic variants within classical HLA class I and HLA class II molecules as well as deficiency states of complement C4, encoded in the class III region of the MHC, were among the first to show association with SLE in the early 1970s.^5–7 Investigation of MHC associations with SLE has been hampered by a number of factors, including long-range linkage disequilibrium (LD) across disease-associated haplotypes, high polymorphism, copy-number variation at associated loci, and lack of study power. The most consistent HLA associations with SLE reside within the class II alleles, HLA-DRB1^∗03:01 (DR3) and HLA-DRB1^∗15:01 (DR2) and their respective extended haplotypes in European populations.⁸

The complement C4 locus is structurally complex: C4 genes show significant copy-number variation (CNV) because of segmental duplication of the RCCX module. The module spans four genes (from which the name is derived): STK19 (also known as RP1, a serine/threonine protein kinase [MIM 604977]), complement C4, CYP21A2 (cytochrome P450 steroid 21-hydroxylase [MIM 201910]) and TNXB (extracellular matrix protein [MIM 600985]).⁹ The C4 sequence in each module is usually functional and can code for either of the two C4 paralogs, C4A (MIM 120810) or C4B (MIM 120820). C4A and C4B show 99% sequence identity over 41 exons and are differentiated by five conserved nucleotide changes in exon 26, causing four isotype-specific amino acid substitutions at positions 1101to 1106: PCPVLD for C4A and LSPVIH for C4B.¹⁰ The amino acid substitutions result in different chemical activities for C4A and C4B. C4A has a longer half life and a higher affinity for amino groups, suggesting a role in the clearance of immune complexes, whereas C4B binds more effectively to hydroxyl groups, resulting in a reduced half-life compared to that in C4A and implying a possible role in membrane attack complex formation and defense against bacterial pathogens.¹¹ It has been suggested that haplotypes harboring a single RCCX module represent the ancestral state of the locus and subsequent duplications of the C4 gene, driven by selective pressures to strengthen innate and adaptive immune responses, led to the diversification of its paralogs and the CNV seen in the modern population.

Complete or homozygous deficiency of complement C4, which is rare, is one of the strongest genetic risk factors for SLE and results in lupus-like disease in approximately 80% of the 28 known affected individuals^12–14. Complement component C4 is essential for the activation of the classical and mannose-binding lectin complement pathways. As such, C4 plays a vital role in the integrity of innate and adaptive host immune responses, including protection against bacterial pathogens through opsonisation of target antigens and cell lysis through the formation of the membrane attack complex, clearance of immune complexes and apoptotic cell debris, and negative selection of autoreactive B cells.¹⁵

Partial C4 deficiency states have also shown association with SLE, but despite much effort over the last 30 years, it has not been established whether this association is primarily causal or secondary to long-range LD across the MHC region.^15–18 Partial C4 deficiency can arise because of CNV at the C4 locus (i.e., low C4 gene copy number [GCN]) as well as mutations in the gene sequence, resulting in nonexpressed (or null) C4 alleles (C4AQ0 and C4BQ0). The most frequent loss-of-function mutation is that of a 2 bp CT insertion in exon 29 of the gene, which causes a frameshift change, resulting in a premature stop codon in exon 30.¹⁹ In healthy European populations, partial and homozygous deficiencies of C4A and C4B are frequent and occur as a consequence of haplotypes bearing monomodular RCCX cassettes, which comprise either C4A or C4B. Hence, the alternate C4 paralog will be physically absent and not expressed. Monomodular RCCX cassettes occur commonly, but not exclusively, on extended HLA-DRB1^∗03:01 haplotypes.²⁰ A monomodular RCCX cassette, coding for a single C4B copy, lacking C4A and resulting in C4A deficiency, is found on the extended A^∗01-B^∗08-(C4A^∗Q0)-C4B1-DRB1^∗03:01 (B8) haplotype. This is the most common HLA-DRB1^∗03:01 haplotype in northern European populations and shows association with SLE and many other autoimmune diseases. Because of tight LD, it has not been possible to identify the location of the causal variant(s) on this conserved haplotype. C4B deficiency has shown association with SLE in Spanish populations and frequently arises as a consequence of a monomodular RCCX cassette carrying a single C4A copy on the A^∗30-B^∗18-C4A3-(C4B^∗Q0)-DRB1^∗03:01 (B18) haplotype, which is observed at a higher frequency in the Spanish population compared with northern Europeans.

In general, C4A and C4B GCNs were not directly determined in the aforementioned studies but inferred by measuring relative plasma protein concentrations of C4A and C4B (immunophenotyping). In 2007, a European-American case-control study used Southern blot analysis to determine total C4, C4A, and C4B GCN distributions in SLE.²¹ This study demonstrated disease predisposition with low C4A GCN, in addition to a protective effect arising from high C4A GCN. No significant differences were observed in the distribution C4B GCN.

Recent family and case-control SNP studies in SLE populations of northern European ancestry have identified primary MHC class III association signals surrounding the RCCX module that are independent of the HLA-DRB1^∗03:01 class II region signal.^22,23 We have previously demonstrated that these class III region SNPs display moderate correlation with monomodular C4A-deficient haplotypes in healthy individuals from the UK and the HapMap CEU cohort.²⁴

Therefore, in this study we wanted to establish whether genetically determined partial C4 deficiency states are independent risk factors for SLE in northern and southern Europeans or whether the association observed is due to LD with MHC polymorphism elsewhere such as HLA-DRB1^∗03:01 or class III region SNPs. In order to do so, we have amalgamated C4 CNV, HLA-DRB1 (MIM 142857), and MHC region SNP data from haplotypically diverse UK and Spanish SLE populations in the largest study of C4 CNV in SLE to date. The different LD relationships between MHC polymorphisms in the two cohorts should allow determination of independent genetic risk. We have developed a high-throughput paralog ratio test (PRT) to reliably determine C4 CNV genotypes.²⁴ We have also developed a paralog-specific restriction enzyme digest variant ratio (REDVR) assay to detect the presence of the most common cause of a C4 null allele—the 2 bp exon 29 CT insertion.²⁵ One would expect enrichment of this indel in cases if partial C4 deficiency states are indeed causal in SLE. Specifically, we genotyped 2,207 subjects (1,028 SLE cases and 1,179 controls) for total C4 GCN, C4A GCN, C4B GCN, and the C4 exon 29 indel. High-density SNP data were available for all individuals, allowing us to explore SNP-C4 CNV correlations in the region. In addition, we have performed multiple logistic regression analyses in the UK and Spanish SLE cohorts to determine whether low complement C4 GCNs are independent of known MHC region SNP and HLA-DRB1 associations as well as examining the region for interaction effects between these markers.

Subjects and Methods

Ethics Statement

This study was approved by the London Research Ethics Committee, United Kingdom (Ref: 06/MRE02/9), Comité de Ética del CSIC, Granada, Spain and Clinical Research Ethics Committee of Vall d'Hebron University Hospital, Barcelona, Spain.

Study Cohorts

UK Cohort

The UK cohort comprised 501 UK SLE probands and 719 healthy, unrelated individuals from the 1958 British Birth Cohort. All individuals had been previously genotyped to high-density at the MHC with a custom Illumina panel.²³ Four digit HLA-DRB1 genotypes were available for 481/501 (96%) of the cases. Two-digit genotypes were available for 661/719 (92%) of the controls.

Spanish Cohort

The Spanish cohort comprised 527 Spanish SLE cases and 460 healthy, unrelated individuals; 2,553 ancestry informative markers (AIMs) and 4,179 SNP genotypes across the MHC region were available for all subjects. Samples were typed with a custom Illumina panel as part of the IMAGEN consortium study²⁶ or the Illumina OMNI-1 array chip; 192 SLE subjects were genotyped on both platforms with 100% concordance for all SNP genotypes. Four-digit HLA-DRB1 genotypes were available for 199/527 (38%) cases and 224/460 (48%) controls.

All SLE probands fulfilled the American College of Rheumatology criteria for the classification of SLE.²⁷ Written informed consent was obtained from all study participants.

Complement C4 Paralog Ratio Test

We used our complement C4 PRT to determine total C4 GCN, C4A GCN, and C4B GCN in the UK cohort according to methodology described previously²⁴.

For the Spanish cohort, each plate run was normalized with 16 control samples. The six UK normalization samples were used in addition to ten Spanish SLE samples with total C4, C4A, and C4B GCN previously determined by Southern blot. The 16 samples used for normalization covered copy-number genotypes of two to six. Normalized PRT and REDVR A ratios for each plate were clustered and genotype calls made for each sample, as previously described.²⁴

Complement C4 Exon 29 CT Insertion Assay

A high-throughput, paralog-specific REDVR assay was used to genotype samples for the 2 bp CT insertion located in exon 29 of C4.²⁵ The assay, combined with C4 GCN data, provided information on the copy number of C4A and C4B genes harboring the insertion in an individual.

HLA Genotyping

HLA typing was performed with Luminex One Lambda SSO.

Four-digit HLA-DRB1 typing was performed in 481 of 501 of UK SLE cases (96%). The typing was performed at the Anthony Nolan Trust, London, UK. Two-digit HLA-DRB1 data were obtained for 661 of the 719 UK controls (92%) from the 1958 British Birth Cohort.

Four-digit genotyping for HLA-B (MIM 142830), HLA-DRB1, and HLA-DQB1 (MIM 604305) was performed in the Spanish cohort at Hospital Virgen del Rocío, Seville, Spain, and Hospital Virgen de las Nieves, Granada, Spain. Four-digit HLA-DRB1 and HLA-DQB1 genotypes were available for 224/460 (48%) Spanish control samples. Four-digit HLA-DRB1 and HLA-DQB1 genotypes were available for 199/527 (38%) of the Spanish SLE cases.

C4 CNV/SNP Correlation

We tested for correlation between C4 GCN genotypes and surrounding MHC SNPs by using standard linear regression as described previously.²⁴ We analyzed individuals with zero, one, or two copies of C4A or C4B. In such individuals C4 GCN genotypes can be downcoded as SNP genotypes assuming that the two-copy individuals carry a single C4 gene on each chromosome: 0 for a two-copy, 1 for a single-copy, and 2 for a zero-copy homozygote of either C4A or C4B. This would allow investigation of the relationships between low C4 copy alleles, which have shown association with SLE and surrounding SNPs. For C4A, 533 (74%) UK controls, 393 (78%) UK cases, 322 (73%) Spanish controls, and 347 (74%) Spanish cases with C4A GCN of two or less were analyzed for C4A GCN/SNP correlation. For C4B, 621 (86%) UK controls, 379 (75%) UK SLE cases, 360 (80%) Spanish controls, and 391 (84%) Spanish SLE cases with C4B GCN of two or less were analyzed for C4B GCN/SNP correlation. HLA alleles in both cohorts were recoded to SNP genotypes and included in the correlation analyses. Correlation coefficients, r², were calculated between C4 CNV data and 1,230 MHC SNPs for the UK cohort and 4,178 MHC SNPs for the Spanish cohort.

SNP Quality-Control Filters

SNP quality-control (QC) filters for the UK cohort have been previously described.²³ For the Spanish cohort, all QC analyses except principal components analyses were performed with PLINK.²⁸ Samples and SNPs were put forward for analysis if they met the following QC filters: SNPs greater than 95% genotyping efficiency, MAF greater than 1%, samples greater than 95% genotyping efficiency, and PI-Hat scores less than 0.2 on identity-by-state analysis with AIMs in order to exclude cryptic relatedness and duplicate samples. SNPs were excluded for deviation from Hardy-Weinberg equilibrium in controls on the basis of a false discovery rate (FDR) of 0.05 (n = 61). In order to correct for population stratification, samples were excluded if they were outliers on principal components analysis with post-QC AIMs (performed with EIGENSTRAT and defined as greater than 4 standard deviations (SDs)from the mean).²⁹ The genomic inflation factor (λ_GC) was calculated with the post-QC AIMs after correction for population stratification (λ_GC = 1.04).

Statistical Analyses

The mean, SD and p value for C4 copy-number frequencies between cases and controls and between cohorts were calculated with SPSS (SPSS, Chicago, IL, USA). One-way analysis of variance (ANOVA) was performed in SPSS to test for differences in integer GCNs of total C4, C4A, and C4B across SLE cases and controls. To estimate the odds ratio (OR) for low (0 or 1) and high (3 or more) C4A and C4B GCNs relative to a C4A or C4B GCN of two, we performed logistic regression by using the glm function in the statistical package R, with copy-number class (low or high) coded as a categorical variable and GCN of two used as a reference. We performed a test for heterogeneity of odds ratios for the two cohorts by coding up cohort as a binary variable and testing for interaction between the cohort and OR while letting the intercept term and the term for the ancestry covariate vary between groups. This was performed in R and tests against the null-hypothesis of equal ORs.

The statistical power to detect an association for the C4A CT insertion for the Spanish and UK cohorts is 57% and >99%, respectively, assuming an odds ratio (OR) of 1.93 for the Spanish and an OR of 2.13 for the UK at the 5% significance level. The assumed effect sizes are equivalent to the ORs for C4A zero-copy (nonexpressed) alleles in each cohort.

Single-marker association analyses employing logistic regression (LR) and multiple logistic regression (MLR) were performed via PLINK for the five markers chosen for analysis: rs558702 (the most highly associated SNP in the single-marker analysis in both cohorts), C4A, C4B, HLA-DRB1^∗03, and HLA-DRB1^∗15. Principal component one (PC1) was used as a covariate in all analyses to correct for population substructure. Only individuals with C4A and C4B GCNs of zero, one, or two were included in these analyses. We performed MLR to determine independent effects for each of the five chosen markers at the MHC; every marker was tested for association conditional on PC1 as a covariate and each of the other four markers as a covariate, resulting in four tests for every marker used as a covariate. In order to correct for multiple testing, we permuted the data 10,000 times for each of the five sets of tests. We included genotypes for the well-established SLE risk alleles, HLA-DRB1^∗03 and HLA-DRB1^∗15 in the UK cohort and tag SNPs for HLA-DRB1^∗03 and HLA-DRB1^∗15 in the Spanish cohort (rs2187668, r² = 0.79 and rs3135391, r² = 0.82, respectively). We looked for interactions between C4A and C4B and independently associated variants across the MHC by using the glm function in R. We examined LD relationships between SNPs and HLA alleles in each cohort by calculating the correlation coefficient, r², by using the Tagger algorithm in Haploview.³⁰

Stepwise logistic regression with the Bayesian information criterion (BIC) was used to determine which of the five variables (rs558702, C4A, C4B, HLA-DRB1^∗03, and HLA-DRB1^∗15) best explained disease status. This analysis was performed in R. Along with the model return by the stepwise procedure, we also checked the BIC for models within the vicinity of the best model (by adding and subtracting variables).

Results

Complement C4 GCN Range in UK and Spanish SLE Cohorts

Using the C4 PRT assay, we determined the GCN for total C4, C4A, and C4B in 501 UK SLE cases, 719 UK controls, 527 Spanish SLE cases, and 460 Spanish controls. We have previously reported C4 copy-number ranges in UK and Spanish control populations.²⁵ These data are summarized here for comparison with the UK and Spanish SLE case statistics.

UK SLE Cases and Controls Demonstrate a Wide Range of C4 CNV Corroborating Previous European-American SLE Data

In the UK cohort, total C4 GCN varied from two to six in controls, and from two to eight in cases (Figure 1 and Table 1). Both C4A GCN and C4B GCN ranged from zero to four in controls and from zero to five in cases. Mean total C4 GCN and C4A GCN were lower in cases compared to controls, whereas mean C4B GCN was higher in cases compared to controls. These data are consistent with those previously observed in European-American SLE populations.²¹

Figure 1.

Total Complement C4, C4A, and C4B GCN Distributions in UK and Spanish SLE

From top to bottom, the histograms demonstrate total complement C4, C4A, and C4B GCN distributions in healthy controls (blue) and SLE cases (red) from the UK and Spain.

Table 1.

Total Complement C4, C4A, and C4B GCN Frequencies in UK and Spanish SLE Cohorts

	GCN Range	0	1	2	3	4	5	6	7	8	Mean (±SD)	p Valuea
C4 GCN
UK SLE (n = 501)	2–8	–	–	36 (7.18%)	158 (31.53%)	210 (41.96%)	70 (13.97%)	25 (4.99%)	1(0.19%)	1(0.19%)	3.79 ± 0.98	0.046
UK controls (n = 719)	2–6	–	–	25 (3.47%)	168 (23.36%)	385 (53.00%)	137 (19.00%)	4 (0.55%)	0	0	3.89 ± 0.76	0.046
Spanish SLE (n = 464)	2–8	–	–	28 (6.03%)	141 (30.83%)	213 (45.90%)	67 (14.40%)	11 (2.37%)	3(0.64%)	1(0.21%)	3.80 ± 0.92	<0.001
Spanish controls (n = 449)	2–8	–	–	19 (4.23%)	86 (19.15%)	210 (46.77%)	93 (20.70%)	32 (7.12%)	7(1.55%)	2(0.44%)	4.13 ± 1.02	<0.001
C4A GCN
UK SLE	0–5	26 (5.18%)	165 (32.90%)	202 (40.30%)	88 (17.50%)	18 (3.59%)	2(0.40%)	0	–	–	1.82 ± 0.93	<0.001
UK controls	0–4	9 (1.25%)	149 (20.70%)	375 (52.20%)	143 (19.90%)	43 (5.90%)	0	0	–	–	2.08 ± 0.83	<0.001
Spanish SLE	0–5	5 (1.07%)	111 (23.90%)	231 (49.80%)	99 (21.30%)	13 (2.80%)	5(1.07%)	0	–	–	2.04 ± 0.85	0.007
Spanish controls	0–6	5 (1.11%)	54 (12.02%)	263 (58.60%)	107 (23.80%)	16 (3.6%)	3(0.70%)	1(0.22%)	–	–	2.19 ± 0.78	0.007
C4B GCN
UK SLE	0–5	13 (2.60%)	125 (24.90%)	241 (48.10%)	110 (21.90%)	11 (2.20%)	1(0.20%)	–	–	–	1.96 ± 0.82	<0.001
UK controls	0–4	26 (3.60%)	185 (25.70%)	410 (57.00%)	94 (13.00%)	4 (0.60%)	0	–	–	–	1.81 ± 0.72	<0.001
Spanish SLE	0–4	14 (3.00%)	170 (36.60%)	207 (44.60%)	62 (13.40%)	11 (2.40%)	0	–	–	–	1.75 ± 0.81	0.019
Spanish controls	0–5	12 (2.70%)	117 (26.00%)	231 (51.40%)	64 (14.30%)	24 (5.30%)	1(0.20%)	–	–	–	1.94 ± 0.86	0.019

^at test p value.

Spanish Populations Display a Broader Range of C4 CNV Compared to the UK Population

Within the Spanish cohort, 11/460 controls and 63/527 SLE cases were excluded for relatedness or because they were population outliers, resulting in a post-QC cohort size of 464 cases and 449 controls.

In both Spanish controls and cases, total C4 GCN varied from two to eight, identical to that observed in UK cases but less than in UK controls (Figure 1 and Table 1). C4A and C4B GCN ranges were similar in UK and Spanish cohorts. Mean total C4, C4A, and C4B GCNs were all lower in Spanish cases compared to controls. Mean total C4, C4A, and C4B GCNs were higher in Spanish controls compared to UK controls (Table 2). There was no significant difference between total C4 GCN in UK and Spanish SLE cases. C4A and C4B GCNs demonstrated contrasting effects in SLE cases between the two cohorts with greater mean values in Spanish cases compared to UK cases for C4A and lower levels in Spanish cases compared to UK cases for C4B (Table 2).

Table 2.

Mean Total Complement C4, C4A, and C4B GCN in UK and Spanish SLE Cohorts

	GCN Range	Mean (± SD)	p Valuea
C4 GCN
UK SLE (n = 501)	2–8	3.79 ± 0.98	0.864
Spanish SLE (n = 464)	2–8	3.80 ± 0.92
UK controls (n = 719)	2–6	3.89 ± 0.76	<0.001
Spanish controls (n = 449)	2–8	4.13 ± 1.02
C4A GCN
UK SLE	0–5	1.82 ± 0.93	<0.001
Spanish SLE	0–5	2.04 ± 0.85
UK controls	0–4	2.08 ± 0.83	0.019
Spanish controls	0–6	2.19 ± 0.78
C4B GCN
UK SLE	0–5	1.96 ± 0.82	<0.001
Spanish SLE	0–4	1.75 ± 0.81
UK controls	0–4	1.81 ± 0.72	0.007
Spanish controls	0–5	1.94 ± 0.86

^at test p value.

Similar C4A but Differential C4B Copy-Number Associations in UK and Spanish SLE

We compared the total C4, C4A, and C4B integer copy-number distributions in cases and controls in both cohorts by using a one-way ANOVA. We found significant differences between the distributions of total C4 (UK: p < 0.001, Spanish: p < 0.001), C4A (UK, p < 0.001 and Spanish, p = 0.001) and C4B (UK, p < 0.001 and Spanish, p = 0.005) in SLE cases compared to controls in both the UK and Spanish cohorts.

In order to delineate the effects of C4 integer GCN on risk of disease, we estimated the risk of low (0 or 1) or high (3 or more) copy number of C4A and C4B relative to a GCN of 2 (Table 3). Within the UK cohort, we found that low C4A GCN was significantly associated with risk of SLE, compared to a GCN of 2 (OR = 2.27, 95% confidence interval [CI] 1.72–2.99, p = 1.14 × 10⁻⁹), whereas high C4A copy number demonstrated no association with disease. With regard to C4B, there was no association between low C4B GCN and UK SLE, whereas high copy numbers conferred significant risk compared to a GCN of 2 (OR = 2.07, 95% CI 1.51–2.83, p = 4.44 × 10⁻⁶).

Table 3.

High and Low C4A and C4B GCN Effects in UK and Spanish SLE Cohorts

UK	OR	95% CIa	p Valueb	Spain	OR	95% CI	p Value	OR Heterogeneity p Valuec
low C4A GCN	2.27	1.72–2.99	1.14 × 10−9	low C4A GCN	2.22	1.56–3.20	1.14 × 10−5	0.99
high C4A GCN	1.07	0.79–1.43	0.63	high C4A GCN	1.04	0.77–1.42	0.76	0.94
low C4B GCN	1.09	0.85–1.41	0.49	low C4B GCN	1.58	1.18–3.26	1.80 × 10−3	0.01
high C4B GCN	2.07	1.52–2.84	4.44 × 10−6	high C4B GCN	0.92	0.65–1.31	0.63	1.81 × 10−4

^aOdds ratio 95% CI.

^bLogistic regression p value.

^cTest of heterogeneity of odds ratios in cohorts p value (see Statistical Analyses).

In parallel with the UK data, the Spanish cohort demonstrated significant SLE risk with low C4A GCN (OR = 2.22, 95% CI 1.56–3.20, p = 1.14 × 10⁻⁵), whereas high C4A copies were not associated with disease susceptibility. However, in contrast to the UK data, the Spanish cohort demonstrated significant risk between low C4B GCN and SLE (OR = 1.58, 95% CI 1.18–3.25, p = 1.80 × 10⁻³) but no association with high C4B GCN.

We performed a test of heterogeneity on the odds ratios for C4A and C4B high- and low-GCN classes in the UK and the Spanish populations. The odds ratios for high and low C4A GCNs were not significantly different between cohorts; however, the odds ratios for high and low C4B GCN were significantly heterogeneous between cohorts (low C4B p = 0.01, high C4B p = 1.81 × 10⁻⁴).

The Loss-of-Function C4A Exon 29 CT Insertion Is Not Associated with SLE in UK and Spanish Populations

All samples were tested for the presence of a 2 bp CT insertion in exon 29, which results in a nonfunctional C4 allele (Table S1, available online, and Figure 2). We have previously shown that 30/719 (4.17%) UK controls harbored the insertion in C4A.²⁵ This is similar to the frequency of the CT insertion in UK SLE cases (21/501 [4.19%] p = 0.55). No C4B insertions were observed in UK controls, whereas one individual carried the exon 29 CT insertion in C4B in the UK SLE cases.

Figure 2.

Frequency of the Complement C4A exon 29 CT Insertion in UK and Spanish SLE Cohorts

The histogram shows the frequency of the loss-of-function complement C4A exon 29 CT insertion in healthy controls (blue) and SLE cases (red) from the UK and Spain.

In the Spanish controls, eight samples out of 449 (1.78%) carried the insertion in C4A, lower than that observed in UK controls. The frequency of C4A CT insertions in Spanish SLE cases was 1.29%, not significantly different from the frequency observed in Spanish controls but less than that observed in UK cases. A single individual in the Spanish controls carried the insertion in C4B, along with one additional functional C4B copy. No C4B insertions were observed in Spanish SLE cases.

Whereas the frequency of the C4A exon 29 CT insertion is significantly different between healthy controls from the UK and Spain (p = 0.014) and SLE cases from the UK and Spain (p = 0.011), we observed no significant differences in the frequency of the insertion between cases and controls within the two cohorts. There is no evidence of enrichment of the C4A exon 29 CT insertion in SLE cases compared to controls. Therefore, this loss-of-function mutation does not predispose to SLE in UK and Spanish populations.

Neither MHC Region SNPs nor HLA-DRB1 Alleles Provide Surrogate Markers for C4 Copy-Number Variants in UK and Spanish SLE Cohorts

We used linear regression to test for correlation between C4A and C4B GCN and surrounding MHC polymorphism in each cohort. We have previously reported SNP/C4 CNV correlations for the UK control population and SNP/C4 exon 29 CT indel correlations for the UK and Spanish controls.^24,25 The data are summarized below for comparative purposes in addition to the novel Spanish control population SNP/C4 CNV correlations, and novel UK and Spanish SLE case SNP/C4 CNV and SNP/C4 exon 29 CT indel correlations.

SNP/C4A CNV Correlation Is Greater in UK Compared to Spanish Data Sets

In both UK cases and controls, the highest SNP/C4A CNV correlation values were observed for SNPs in class III, on either side of the RCCX module (Figure S1 and Table S2). The best correlated SNPs were rs558702, (intronic C2 [MIM 613927], r² = 0.61 in controls and 0.58 in cases), the intronic SNPs, rs3131378 and rs3131379, in MSH5 (MIM 603382) (r² = 0.60 in controls and 0.56 in cases) and the intronic SNP, rs1269852, in ATF6B (MIM 600984) (r² = 0.59 in cases and controls). These four SNPs are in strong LD and have demonstrated association with SLE in a high-density SNP study across the MHC and an SLE GWAS.^23,31 The HLA-DRB1^∗03:01 allele showed moderate correlation in UK cases (r² = 0.49) and controls (r² = 0.45 for HLA-DRB1^∗03).

Lower SNP/CNV correlation coefficents were observed in the Spanish cohort compared to that of the UK, most likely reflecting the broader haplotypic diversity at the MHC seen in the former.³² Within the Spanish population, higher correlation values were observed for the same SNPs in cases compared to controls (Figure S1 and Table S2). The best correlated SNPs in the Spanish cohort were rs558702 (r² of 0.44 for cases and 0.20 for controls) and rs3131379 (r² = 0.44 in cases and 0.19 in controls). These SNPs are identical to the best correlated C4A CNV SNPs in the UK cohort and have recently shown primary association with SLE in a high-density Spanish MHC SNP study.²⁶ The HLA-DRB1^∗03:01 allele showed weaker correlation with C4A CNV compared to the UK cohort (r² = 0.12 in Spanish cases and r² = 0.03 in Spanish controls).

SNP/C4A CNV correlation coefficients observed in these two populations show that MHC region SNPs cannot be used as surrogate markers for the C4A zero-copy state in either of the two populations and confirm the results of our previous study in healthy populations.²⁴

SNP/C4B CNV Correlations Are Low in UK and Spanish Populations

The correlation values between surrounding MHC region SNPs and C4B genotypes were low for all cohorts (Figure S2 and Table S3).

In UK cases and controls, the highest correlation values were observed with the SNP, rs605203, located 5′ SLC44A4 (MIM 606107) (r² = 0.11 controls, r² = 0.12 cases) and the intronic SNP, rs659445, in EHMT2 (MIM 604599) (r² = 0.11 controls, r² = 0.12 cases) telomeric to the RCCX module.

Slightly higher SNP/C4B CNV correlation values were observed in the Spanish cohort. The most highly correlated markers were an intronic SNP in NOTCH4, rs379464 (r² = 0.19 controls and r² = 0.18 cases), and rs9267785 in RDBP (MIM 154040) (r² = 0.20 controls and r² = 0.19 cases). Correlation values for all SNPs were below 0.3 in the UK and Spanish cohorts demonstrating that haplotypes lacking C4B genes are diverse in both populations.

SNP/C4A Exon 29 CT Insertion Correlations Are Low in UK and Spanish Cohorts

In both UK cases and controls the top correlated SNPs with the C4A exon 29 insertion were located in the class III region, telomeric to the RCCX module (Figure S3 and Table S4). The SNP with the highest correlation value was rs2734331, a synonymous SNP in SKIV2L (MIM 600478) (0.43 for controls and r² = 0.40 for cases).

The top correlated SNPs for Spanish controls were also located in the MHC class III region(best SNP, rs2228088, a synonymous variant in TNF [MIM 191160]; r² = 0.25). Low correlation values were seen in the Spanish SLE cases. The top correlated SNPs were rs11752643 downstream of HLA-DQB1 (r² = 0.13) and rs9469027 intronic LST1 in class III ([MIM 109170]; r² = 0.11).

Complement C4A and C4B GCN Variants Are Not Independent Genetic Risk Factors for SLE in UK and Spanish Populations

We analyzed 501 UK SLE cases and 719 controls by using logistic regression and MLR on 1230 MHC SNPs, HLA-DRB1 alleles and C4A and C4B GCN genotypes (Table 4). The most strongly associated marker in the univariate analysis was rs558702, (intronic C2, in the class III region of the MHC, OR = 2.61, 95% CI 2.07–3.29, nominal p = 4.26 × 10⁻¹⁶) and other SNPs in strong LD that all show moderate correlation with C4A CNV and have previously demonstrated association with SLE in northern and southern European populations.²³ The C4A zero-copy allele and HLA-DRB1^∗03 were also associated with lupus in the single-marker analysis (C4A OR = 2.13, 95% CI 1.68–2.70, nominal p = 5.27 × 10⁻¹⁰, HLA-DRB1^∗03 OR = 2.55, 95% CI = 2.03–3.21, nominal p = 1.43 × 10⁻¹⁵). C4B zero-copy and HLA-DRB1^∗15 alleles were not associated with SLE in the univariate analysis.

Table 4.

Single-Marker and Multiple Logistic Regression Analysis of MHC Polymorphism in UK and Spanish SLE

	Single Marker		Conditioned on rs558702		Conditioned on C4A		Conditioned on C4B		Conditioned on DRB1∗03		Conditioned on DRB1∗15
	UK	Spain	UK	Spain	UK	Spain	UK	Spain	UK	Spaina	UK	Spaina
rs558702
MAFb	0.11	0.04
OR (95%CI)	2.61 (2.07–3.29)	2.92 (1.91–4.45)	–	–	2.46 (1.66–3.64)	2.49 (1.46–4.24)	2.84 (2.17–3.71)	3.27 (2.03–5.26)	1.58 (1.04–2.39)	2.30 (1.41–3.75)	2.73 (2.15–3.48)	3.07 (2.00–4.71)
p Valuec	1.00 × 10−4	1.00 × 10−4	–	–	1.00 × 10−4	2.90 × 10−3	1.00 × 10−4	1.00 × 10−4	0.11	2.40 × 10−3	1.00 × 10−4	1.00 × 10−4
C4A
MAF	0.16	0.10
OR (95%CI)	2.13 (1.68–2.70)	1.93 (1.36–2.72)	1.09 (0.74–1.60)	1.24 (0.81–1.90)	–	–	2.24 (1.69–2.96)	2.10 (1.38–3.19)	1.27 (0.91–1.79)	1.58 (1.10–2.28)	2.20 (1.72–2.82)	1.95 (1.38–2.76)
p Value	1.00 × 10−4	1.20 × 10−3	0.98	0.77	–	–	1.00 × 10−4	1.50 × 10−3	0.49	0.05	1.00 × 10−4	8.00 × 10−4
C4B
MAF	0.19	0.19
OR (95%CI)	1.06 (0.84–1.32)	1.55 (1.19–2.02)	1.28 (1.01–1.63)	1.74 (1.32–2.29)	1.63 (1.14–2.34)	1.59 (1.13–2.25)	–	–	1.12 (0.88–1.43)	1.40 (1.09–1.88)	1.05 (0.83–1.33)	1.61 (1.23–2.11)
p Value	0.99	4.70 × 10−3	0.18	6.00 × 10−4	0.02	0.03	–	–	0.80	0.04	0.98	2.00 × 10−3
DRB1∗03
MAF	0.13	0.12
OR (95%CI)	2.55 (2.03–3.21)	1.78a (1.36–2.33)	1.77 (1.18–2.64)	1.33a (0.97–1.82)	2.13 (1.51–3.00)	1.65a (1.20–2.26)	2.64 (2.04–3.41)	1.63a (1.21–2.20)	–	–	2.78 (2.19–3.51)	1.94a (1.47–2.55)
p Value	1.00 × 10−4	2.00 × 10-4a	0.02	0.24a	1.00 × 10−4	7.10 × 10−3a	1.00 × 10−4	4.00 × 10−3a	–	–	1.00 × 10−4	1.00 × 10−4a
DRB1∗15
MAF	0.14	0.08
OR (95%CI)	1.23 (0.98–1.55)	1.58a (1.13–2.22)	1.46 (1.15–1.86)	1.71a (1.22–2.41)	1.22 (0.94–1.59)	1.57a (1.07–2.30)	1.25 (0.96–1.65)	1.79a (1.22–2.64)	1.53 (1.20–1.95)	1.83a (1.30–2.58)	–	–
p Value	0.27	0.03a	7.40 × 10−3	8.10 × 10−3a	0.39	0.08a	0.24	0.01a	2.00 × 10−3	1.50 × 10−3a	–	–

^aTag SNPs were used as surrogate markers for the HLA-DRB1^∗03 and HLA-DRB1^∗15 alleles in the Spanish cohort because of the low proportion of samples with available HLA-DRB1 genotypes in this cohort. The SNP, rs2187668, which was used as a surrogate marker for the HLA-DRB1^∗03 allele, demonstrated a correlation coefficient, r², of 0.79 with the HLA-DRB1^∗03 allele in 34 Spanish controls of known HLA-DRB1 genotype. The SNP, rs3135391, was used as a surrogate marker for the HLA-DRB1^∗15 allele, r² = 0.82 in 400 Spanish controls.

^bMinor allele frequency.

^cp value permuted 10,000 times.

Next, we performed MLR to assess the independence of these five markers in UK SLE (Table 4). Conditioning on the SNP, rs558702, or HLA-DRB1^∗03, abrogated the effect of low C4A GCN. In contrast to the univariate analysis, significant association was observed with HLA-DRB1^∗15, whereas that of C4B GCN was unchanged. Conversely, if C4A is used as a covariate in the MLR, rs558702 and HLA-DRB1^∗03 remain significantly associated with SLE. Conditioning on C4B or HLA-DRB1^∗15 had no effect on the association of any of the four remaining markers in UK SLE.

Within the Spanish SLE cohort, 4,178 MHC region SNPs and C4A and C4B GCN genotypes were analyzed in 464 SLE cases and 449 controls. In addition, rs2187668 and rs3135391 were used as proxy SNPs for HLA-DRB1^∗03:01 and HLA-DRB1^∗15:01, respectively, in this population. As observed in the UK cohort, the most strongly associated univariate marker in the Spanish group was rs558702 (OR = 2.92, 95% CI 1.93–4.45, nominal p = 6.73 × 10⁻⁷). This SNP demonstrated moderate correlation with C4A CNV in the Spanish population. HLA-DRB1^∗03:01, HLA-DRB1^∗15:01, C4A and C4B zero-copy alleles were also associated with SLE in this population in the single-marker analysis. When rs558702 is used as a covariate, the effects of C4A GCN and HLA-DRB1^∗03:01 were not significant. However, the C4B and HLA-DRB1^∗15:01 association results remained largely unchanged. Conditioning on HLA-DRB1^∗15:01, C4A or C4B had little effect on the univariate results, whereas the use of HLA-DRB1^∗03:01 as a covariate abrogated the effect of C4A, showed a trend for reducing the effect size of rs558702, and did not influence the association signals observed for C4B or HLA-DRB1^∗15:01. These data demonstrate that the low C4A GCN association observed in UK and Spanish SLE is not primary but due to LD of the C4A zero-copy allele with polymorphism elsewhere in the MHC region tagged by rs558702 and HLA-DRB1^∗03.

We found no evidence of interaction effects between C4A, C4B, rs558702, HLA-DRB1^∗03, and HLA-DRB1^∗15 in UK and Spanish SLE.

Model selection with the BIC identified the variables HLA-DRB1^∗03+HLA-DRB1^∗15 alone as the best fit to the data in the UK cohort (BIC = 1,570.23). However, the addition of rs558702 (BIC = 1,572.59) or C4B (BIC = 1,572.75) to these two alleles were also a good fit. The addition of C4A, however, is strongly rejected (BIC = 1,577.04) in favor of HLA-DRB1^∗03+HLA-DRB1^∗15 alone (estimated Bayes factor in favor of HLA-DRB1^∗03+HLA-DRB1^∗15 alone = 33). In the Spanish data, the variables rs558702+C4B+HLA-DRB1^∗15 were the best fit (BIC = 1,250.42). The addition of HLA-DRB1^∗03 (BIC = 1,254.7) or C4A (BIC = 1,254.9) to these three alleles was also rejected in favor of rs558702+C4B+HLA-DRB1^∗15 alone (Bayes factor of 8.5 and 9.4, respectively).

Discussion

In order to determine whether partial complement C4 deficiency is an independent genetic risk factor for SLE, we have investigated C4 CNV in the context of HLA-DRB1 and MHC region SNP polymorphism in the largest and most comprehensive complement C4 study to date. For the first time, we have examined complement C4 CNV in a southern European Spanish SLE cohort in conjunction with that of a northern European UK SLE population. We have used our previously validated PRT to assess total C4, C4A, and C4B GCN variation in 965 SLE cases and 1,168 healthy individuals (post-QC). In addition, we have genotyped the most common C4 nonsense mutation—an exon 29 CT insertion—in all study subjects to assess disease association.

In this study, we have clearly shown that genetically determined partial C4 deficiency states are not independent genetic risk factors for SLE in UK and Spanish populations. The haplotypic diversity within the MHC region between the northern and southern European cohorts, consequent on ancestry and environment, has allowed delineation of such effects. These results are further corroborated by the lack of association shown by the loss-of-function C4A exon 29 CT insertion in both SLE cohorts. Hence, although complete homozygous deficiency of complement C4 is one of the strongest genetic risk factors for SLE, partial C4 deficiency states do not predispose to disease. These data parallel the reported associations between SLE and C1q deficiency. C1q is also an early complement cascade component and initiates the classical complement pathway by binding immunoglobulin. Complete homozygous deficiency of C1q is the strongest genetic risk factor for SLE known thus far, and approximately 90% of affected individuals develop lupus or lupus-like disease.¹⁵ However, subjects who are heterozygous for C1q nonsense mutations do not show a propensity for autoimmunity. In conjunction, these data suggest that complete lack of early classical complement cascade components are important in susceptibility to SLE but that these deleterious effects can be overcome by partial expression of these proteins.

The molecular mechanisms underlying susceptibility to SLE as a consequence of complete C4 deficiency have not been fully elucidated. Mice deficient in complement component C4 demonstrate lupus-like disease in a strain-dependent manner, therefore complete C4 deficiency itself is not sufficient to cause SLE—a permissive genetic background is required.³³ It has been suggested that deficiency of C4A rather than C4B is important in the development of SLE because of the greater efficiency of C4A in handling immune complexes.³⁴ However, both C4A and C4B levels vary in parallel with disease activity, suggesting that there is no selective utilization of C4A over C4B.³⁵ Moreover, published reports have described associations between complete C4A deficiency as well as complete C4B deficiency states and immune complex disease.³⁶ A number of hypotheses, which are not mutually exclusive, have been put forward to explain the association of C4 deficiency with SLE. These include failure to clear apoptotic cell debris, breakdown in peripheral tolerance mechanisms caused by dendritic cell dysfunction, and failure of central tolerance mechanisms where autoreactive B lymphocytes escape to the periphery because of defective negative selection.^33,37 None of these proposals are predicated on the differential chemical properties observed between C4A and C4B, thus the data presented herein are consistent with these views.

The reasons underlying the difficulty in demonstrating an independent genetic effect for partial C4 deficiency in SLE over the past 30 years are manifold. Partial C4 deficiency was first identified as either a lack of or a reduced serum concentration of C4A or C4B based on the assumption that each chromosome 6 encoded a single copy of C4A and C4B. Hence, the resulting immunophenotyping data that were used as surrogate markers for genetic variation at the locus failed to account for CNV at the locus. It is now clear that low serum C4 levels frequently arise as a consequence of the normal CNV observed at the C4 locus. Moreover, serum C4 is consumed in active lupus, further confounding the results of immunophenotyping studies. In order to accurately determine partial C4 deficiency status, therefore, it became necessary to directly interrogate the genome for C4 CNV. Studies using Southern blot analysis to determine C4 CNV have confirmed the association of low C4A GCN with SLE in populations of European-American ancestry but have failed to account for the effects of LD in the MHC region.²¹

The C4 CNV association results in our UK SLE cohort are similar to those previously described in European-American SLE cohorts.²¹ The distribution of total C4 and C4A GCN is skewed toward the lower range in SLE cases compared to the range in controls such that significantly lower mean total C4 and C4A GCNs were observed in cases compared to controls. A broader range of total C4, C4A, and C4B CNV was observed in the Spanish population, reflecting the greater haplotypic diversity of this southern European population.³² In common with UK SLE, mean total C4 GCN and C4A GCN were significantly lower in Spanish cases than controls. Contrary to the UK data, C4B GCNs were also skewed to the lower copy range in Spanish cases compared to the ranges in Spanish controls, and the mean C4B GCN was significantly lower in the former than in the latter. The predominance of low C4B copy-number haplotypes in the Spanish cases is likely correlated with the prevalence of the SLE risk A^∗30-B^∗18-C4A3-(C4B^∗Q0)-DRB1^∗03:01 (B18) haplotype in this population. The frequency of the HLA-DRB1^∗03 allele was 12.85% in UK controls and 10.94% in Spanish controls. Whereas HLA-B^∗18 occurred in conjunction with 7.25% of HLA-DRB1^∗03 haplotypes in the UK controls, 32% of HLA-DRB1^∗03 haplotypes in the Spanish controls harbored HLA-B^∗18 and consequently did not carry C4B.

The association results observed for C4A CNV are similar in both UK and Spanish cohorts: low C4A GCN significantly predisposes to SLE, whereas high C4A GCN is not associated with disease in either cohort. On the other hand, low C4B GCN shows no association in the UK SLE cohort, but demonstrates significant risk in the Spanish cohort. Conversely, high C4B copy numbers demonstrate significant risk the UK SLE cohort, whereas no association is observed in the Spanish cohort. Although the low C4A GCN risk effect is consistent across the cohorts and could be primary or due to the LD with causal variation, the lack of consistency of C4B GCN effects across the cohorts suggests that C4B CNV effects in SLE are secondary to LD with as yet undefined polymorphisms within the MHC. The SNP/CNV correlation analyses presented thus far have used low C4 GCN as covariates (zero, one, and two copies). We were not able to find any significant correlations specifically between high C4B copy numbers and HLA-DRB1 alleles or surrounding SNPs (data not shown). However, as high C4B integer copy number can represent a number of copy-number haplotypes, it is difficult to establish such correlations.

A previous study that investigated the role of C4 CNV in European-American SLE demonstrated disease predisposition with low C4A GCN and protection with high C4A GCN.²¹ We confirm that low C4A GCN predisposes to SLE in UK and Spanish populations. However, high C4A GCNs did not confer protection in either population when compared to a GCN of two. The protective effect attributed to high C4A GCN in the previous study was probably due to the confounding risk effects of low C4A GCN.

In summary, we have conclusively shown that partial C4 deficiency states secondary to low C4A or C4B GCN are not independent genetic risk factors for SLE. This study highlights the utility of interrogating ancestrally and thus haplotypically diverse populations in order to resolve genetic association signals through exploitation of differential LD patterns.

Web Resources

The URLs for data presented herein are as follows:

• Luminex One Lambda SSO, www.onelambda.com

• R, http://www.R-project.org