An Immunochip-based interrogation of scleroderma susceptibility variants identifies a novel association at DNASE1L3

Systemic sclerosis (SSc) or scleroderma is a multisystem, autoimmune disorder characterised by progressive vascular, inflammatory and fibrotic dysfunction. Skin and visceral complications of cardiac, pulmonary, gastro-intestinal, muscle and renal disease can have devastating effects on quality of life and life expectancy [1].

Scleroderma has a well-established genetic component [2]-[4]. Most of the identified SSc susceptibility loci overlap with those of other autoimmune diseases, in particular the rheumatic disorders such as rheumatoid arthritis and systemic lupus erythematosus (SLE) [5]. For example, Carmona et al. recently confirmed an association between SSc and the SLE risk haplotype at the IRF5 locus [6] and Martin et al. recently performed a pan-meta-analysis of SSc and SLE to look at the susceptibility overlap between the two diseases [7]. Using 6,835 cases and 14,274 controls they identified a novel pleiotropic locus at KIAA0319L on chromosome 1 and identified two SLE loci (near PXK and JAZF1) that also contribute to SSc [7].

To identify further susceptibility loci and to explore the genetic overlap with other antibody-mediated immune diseases, we undertook an SSc association study using the Immunochip, a custom array including SNPs of interest in a wide variety of autoimmune disorders [8]. A replication study was then performed using previously published case data from a genome-wide association study (GWAS) of scleroderma [9], and control data from a GWAS of bone density variation [10].

After sample and SNP exclusions, in the Immunochip dataset genotypes were analysed at 145,921 autosomal SNPs in 486 cases and 4,458 controls, with a genomic inflation factor of 1.02 (Q-Q plot, Figure S1 in Additional file 1). Eighty-six percent of the cases were female, mean age was 60 years and mean disease duration 14 years. Twenty-five percent of cases had diffuse disease, 72% limited pattern, and 3% were intermediate; 43% were anti-centromere antibody (ACA) positive and 15% anti-Scl-70 antibody positive. Considering the replication set, 700 scleroderma cases and 1,899 controls remained after quality control. Of the replication cases, 36% had diffuse disease, 64% limited pattern; 32% were ACA positive. The genomic inflation factor for the replication set was 1.045. The Immunochip study has 80% power to detect associations at P <10^-4.5 for a variant with minor allele frequency (MAF) of 0.3 with D' = 0.8 with a SNP with heterozygote odds ratio (OR) of 1.65.

We detected suggestive associations in the Immunochip data (uncorrected P <10^–4.5) at eight loci (Table 1), five of which showed association in the replication cohort (HLA-DRB1, DNASE1L3, STAT4, TNPO3-IRF5, and VCAM1) (Table 1, Manhattan plot Figure 1). There was also evidence of association at the previously reported genome-wide significant CD247 SNP [9] (rs2056626, OR = 0.76, P = 1.1 × 10⁻⁴), but no evidence of association at the genome-wide significant TNIP1 locus [27] (rs2233287, P = 0.94). The overlap between previously reported genome-wide significant SSc loci (outside the MHC) and our data, at an unadjusted P <0.01, is shown in Table 2. Not all previously associated SSc loci could be investigated owing to the limitations of markers available on the array.

Chr	Location range (base pairs, hg18)	Best SNP	MAF cases	MAF controls	Immunochip OR	ImmunochipPvalue	Replication OR	ReplicationPvalue	Combined analysis	Gene annotation#
6	32,039,116-32,888,448	rs2857130	0.293	0.389	0.68	2.8 × 10−7	0.81	0.003	1.3 × 10−8	Intergenic (HLA region)
1	101,009,225	Novel SNP	0.024	0.008	3.31	1.8 × 10−6	2.49	0.031	1.9 × 10−7	Intergenic (VCAM1)
3	58,158,676-58,289,303	rs35677470	0.122	0.083	1.63	3.4 × 10−6	1.27	0.027	1.2 × 10−6	DNASE1L3 (coding)
3	58,158,676-58,289,303	rs35677470	0.174	0.083	2.36	2.3 × 10−10	1.74	3.0 × 10−4	8.71 × 10−13	DNASE1L3 (coding)^
7	128,372,852-128,499,110	rs34381587	0.158	0.113	1.53	1.2 × 10−5	1.38	8.7 × 10−4	5.2 × 10−8	IRF5/TNPO3 (intron)
16	73,863,956-74,046,823	rs11149824	0.469	0.391	1.35	1.4 × 10−5	1.054	0.442	3.2 × 10−4	CFDP1 (intron)
2	43,775,459-43,784,213	rs13403030	0.384	0.318	1.36	1.5 × 10−5	0.92	0.242	0.027	PLEKHH2 (intron)
2	191,608,694-191,641,499	rs13426947	0.253	0.191	1.42	1.8 × 10−5	1.25	5.4 × 10−3	6.1 × 10−7	STAT4 (intron)
1	61,883,642	rs2886326	0.249	0.195	1.41	2.0 × 10−5	0.96	0.606	7.4 × 10−3	Intergenic (TM2D1)

^#Gene annotation is based on the location of the most significant SNP; ^{^}ACA-positive cases only. SNP, single nucleotide polymorphism; SSc, systemic sclerosis; ACA, anti-centromere antibody; Chr, chromosome; MAF, minor allele frequency; OR, odds ratio.

Consistent with previous findings, DRB1*11:04 is a particularly strong risk factor for Scl70-positive SSc [28] (Table 3). However, compared to controls the frequency of this allele is also elevated in ACA-positive cases and in cases negative for both antibodies. The protective effect of DRB1*07:01 is strongest against ACA-positive disease, and there was no evidence that this allele protects against anti-Scl70 antibody-positive disease.

The most significant STAT4 locus association was at the SNP rs13426947 (Table 1). This was one of 15 associated intronic SNPs in STAT4 (P <3 × 10⁻⁴), all correlated with rs13426947 (r² > 0.55, MAF 0.19 to 0.26). These included the top-ranked SNP at this locus from a previous SSc GWAS (rs3821236, P = 2.6 × 10⁻⁵, r² = 0.97 with rs13426947) [9]. No SNPs in this region were significantly associated with SSc (P <0.01) after conditioning on rs13426947. rs13426947 also showed association in the replication set (P = 5.4 × 10⁻³), and in the combined analysis (P = 6.1 × 10⁻⁷).

At the IRF5/TNPO3 locus, there was a group of 23 associated SNPs (P <5 × 10⁻⁵) highly correlated with the top-ranked SNP rs34381587 (r² > 0.90, MAF 0.11 to 0.13), including the top-ranked SNP reported previously (rs10488631, P = 2.4 × 10⁻⁵, r² = 0.998 with rs34381587) [9]. There was some evidence of association with a rare SNP between the two genes (128,379,270 base pairs, MAF 0.028, OR 1.83, P = 6.0 × 10⁻⁴; P = 0.05 after correction for the 83 independent tests). rs34381587 also showed association in the replication set (P = 8.7 × 10⁻⁴), and near genome-wide significance in the combined analysis (P = 5.2 × 10⁻⁸).

Association was observed and replicated at chromosome 3p14, spanning DNASE1L3 to AXOX2 (including PXK; Figure 2), a region that has previously been associated with both SLE [29] and SSc [7]. While the peak association with SLE was originally identified at an intronic SNP rs6445975 in PXK (a PX domain containing serine/threonine kinase), the strongest association with SSc on the Immunochip was at a missense SNP rs35677470 (R206C) in DNASE1L3 (deoxyribonuclease I-like 3); 187 kb distal of rs6445975. Linkage disequilibrium between the two SNPs is modest (r² = 0.13), and there is weak evidence of a secondary association with rs6445975 in our data after conditioning on rs35677470 (OR = 1.19, P = 0.03). No other associations at this locus are significant after correction for multiple testing. The association with rs35677470 is confined to ACA-positive cases (estimated OR 2.36, P = 2.3 × 10⁻¹⁰), with no association in ACA-negative cases (P = 0.76). rs35677470 also showed association in the replication set (P =0.027), and in the combined analysis (P = 1.2 × 10⁻⁶). As in the Immunochip analysis, in the replication set association was much stronger in the ACA-positive group (P = 3.0 × 10⁻⁴), and overall (P = 8.71 × 10⁻¹³). Association was particularly significant in limited scleroderma (P = 3.36 × 10⁻⁹ in overall dataset) compared with diffuse scleroderma (P = 0.57), consistent with the association of ACA antibody status with limited disease. The non-synonymous DNASE1L3 variant is predicted to be deleterious to the protein product using in silico functional prediction tools including both SIFT [30] and PolyPhen [31]. Apart from the DNASE1L3 locus, no other associations showed evidence of heterogeneity by antibody status, or between limited and diffuse SSc.

A novel intergenic SNP in VCAM1 achieved suggestive association with SSc overall in Immunochip cases (OR = 3.31, P = 1.8 × 10⁻⁶). Association was seen with both limited (P = 1.8 × 10⁻⁴), and diffuse disease (8.9 × 10⁻⁵). The findings for overall and limited disease were supported in the replication dataset (P = 0.031 and 3.1 × 10⁻³ respectively) but not with diffuse disease (P = 0.7). A combined meta-analysis gave P values of 1.9 × 10⁻⁷ for all cases, 1.9 × 10⁻⁶ for limited and 4.3 × 10⁻⁴ for diffuse disease.

The major novel finding of this study is the significant association of a functional SNP (rs35677470) in DNASE1L3 with ACA-positive SSc (P = 2.3 × 10⁻¹⁰). This locus has been previously reported in a SLE GWAS [29] and meta-analysis [32], however the peak associations were reported for SNPs in the nearby gene PXK. While the PXK and DNASE1L3 associations may be independent, no SNPs on the Illumina HumanHap300 chip used in the SLE GWAS are good tags for the missense SNP in DNASE1L3 (maximum r² of 0.21). More recently, a functional variant in DNASE1L3 was implicated in a familial form of SLE [33], and a pan-meta-analysis of SLE and SSc also confirmed the locus near PXK, but in particular for ACA-positive SSc (rs2176082 [7]; P = 1.4 × 10⁻⁴ in our data) strengthening the evidence that this locus plays a role in both diseases. These findings were independently identified in a study published whilst the current manuscript was in review [34].

The associated missense DNASE1L3 variant in our data, (rs35677470 encoding R206C) affects a highly conserved residue and there is very strong evidence that this results in loss of function of the protein [35]. DNASE1L3 encodes a member of the DNase family and functions as an endonuclease capable of cleaving DNA, mediating the breakdown of DNA during apoptosis. Al-Mayouf et al.[33] hypothesised that, in the context of SLE, dysfunction of this gene may lead to impaired DNA breakdown and clearance from apoptotic cells, resulting in the formation of self-directed DNA-specific antibodies and immune complexes. Since the same kinds of DNA-driven immune complexes (such as anti-nuclear and ACA antibodies) are also characteristic of SSc, this hypothesis is also applicable.

Suggestive association was also observed between a novel SNP in VCAM1 and overall scleroderma and in limited disease cases, both of which were also associated in the replication dataset. VCAM1 has not previously been reported to be associated with scleroderma or SLE. VCAM-1 is a member of the Ig superfamily and encodes a cell surface sialoglycoprotein expressed by cytokine-activated endothelium. It mediates leukocyte-endothelial cell adhesion and signal transduction. VCAM-1 levels have previously been shown to be elevated in early, inflammatory-phase scleroderma [36] and in limited scleroderma [37].

There were no stand-out functional variants at the STAT4 and IRF5/TNPO3 loci. These associations included many highly correlated SNPs, indicating that larger sample sizes and/or functional studies will be needed to understand and dissect these associations. It would, however, be prudent to investigate any functional variation at these loci identified in related autoimmune diseases, particularly SLE. The other novel associations are merely suggestive and require confirmation in additional datasets.

There is a significant association of a functional SNP in DNASE1L3 with anti-centromere antibody-positive SSc, previously reported in SLE. There is strong evidence for a loss of function of the protein. A novel association was also observed and replicated with an intergenic SNP in VCAM1.

This study serves to highlight that, even with a small but well-characterised disease cohort, significant associations can be obtained by tools such as the Immunochip, which are targeted towards analysis of disease-relevant and occasionally functional variation.