Introduction

Vasculitides represent a heterogeneous group of complex disorders characterised by chronic inflammatory lesions of the blood vessels. Although the pathogenesis of vasculitides is far from being completely understood, cumulating data clearly suggest that both the innate and adaptive responses contribute to their development and progression1. Vasculitides show a large spectrum of clinical manifestations that depend on the affected blood vessel. In this regard, the Chapel Hill Consensus Conference proposed a nomenclature system in which the vasculitides were subdivided into three main groups: small-vessel, medium-vessel, and large-vessel vasculitis (LVV). The LVV group includes giant cell arteritis (GCA) and Takayasu’s arteritis (TAK), which mainly involve arteries of large calibre such as the aorta and its major branches2. These two forms of vasculitis develop predominantly in women, with GCA generally affecting people over 50 years of age in Western countries, especially those of European origin, and TAK affecting younger patients with a higher prevalence in Turkey, Japan, India, and China3,4.

In the last years, the use of novel technologies has produced a substantial advance in the elucidation of the genetic component of LVV5. Large-scale genetic analyses have been recently published separately for both GCA and TAK using the Immunochip platform6,7. The Immunochip has been shown to be one of the most successful platforms to identify immune-related risk variants for a large spectrum of immune-mediated diseases. The use of the same platform in these studies has facilitated the identification of shared aetiopathogenic pathways amongst these disorders, supporting the hypothesis of a common genetic background underlying autoimmunity8.

To contribute to the development of better diagnostic and prognostic markers of LVV, we evaluated the genetic similarities between GCA and TAK by performing an inter-disease meta-analysis of genomic data.

Results

Analysis of the HLA region

The HLA region harboured the main disease-specific associations in our study cohort (Fig. 1). In this context, GCA was mostly associated with class II genes, with the SNP rs9405038 (located between HLA-DRA and HLA-DRB1) representing the lead signal (P = 6.65E-16, OR = 1.60). In contrast, the main associations with TAK were located within the class I subregion, with rs12524487 (located between HLA-B and MHC class I polypeptide-related sequence A; MICA) as the strongest hit (P = 1.92E-16, OR = 3.70). Neither SNP showed even suggestive P-values in the analysis of the other type of vasculitis (TAK: rs9405038, P = 0.010; GCA: rs12524487, P = 0.244). As a consequence, a high heterogeneity (Q < 0.05) was observed across the region. Consequently, a random effects model was used to meta-analyse the HLA data. Although some class I and II markers surpassed the study-wide significance threshold (e.g. class I: rs9263969, P = 3.01E-07, ORGCA = 0.77, ORTAK = 0.77; class II: rs9272105, P = 3.74E-11, ORGCA = 1.38, ORTAK = 1.57), both the number of associations and their effect size was considerably reduced in comparison with the analysis of GCA and TAK separately (Fig. 1, see Supplementary Table S1).

Figure 1
figure 1

Manhattan plot representation of the results of the HLA region in (A) giant cell arteritis, (B) Takayasu’s arteritis, and (C) the meta-analysis of both forms of vasculitis. The log10 of the P values are plotted against their physical chromosomal position. A red/green color gradient was used to represent the effect size of each analysed polymorphism (red for risk and green for protection). The red line represents the study-wide level of significance (P < 1.13E-06). HLA class I and II subregions are highlighted in grey.

Full size image

Analysis of the non-HLA region

Outside the HLA region, only one variant surpassed the study-wide significance threshold in the overall meta-analysis including both diseases (rs755374, P = 7.54E-07; ORGCA = 1.19, ORTAK = 1.50; Table 1, see Supplementary Figure S1). This SNP is located in an intergenic region at 71 kb 5′ of the interleukin 12B (IL12B) gene (see Supplementary Figure S2).

Table 1 Suggestive shared signals (P < 1E-04) between giant cell arteritis and Takayasu’s arteritis outside the HLA region.
Full size table

Other suggestive common susceptibility factors for both diseases that showed trends of association included glutamate ionotropic receptor NMDA type subunit 2 A (GRIN2A; rs1448258, P = 2.69E-06, ORGCA = 1.23, ORTAK = 1.29), G-protein signaling modulator 1 (GPSM1; rs28489139, P = 1.38E-05, ORGCA = 1.27, ORTAK = 1.98), nitric oxide synthase 2 (NOS2; rs7406657, P = 2.65E-05, ORGCA = 0.76, ORTAK = 0.88), ASH1 like histone lysine methyltransferase (ASH1L; rs7340058, P = 6.26E-05, ORGCA = 0.61, ORTAK = 0.58), REL proto-oncogene, NF-kB subunit (REL; rs79657074, P = 6.73E-05, ORGCA = 1.32, ORTAK = 1.82), SMG6, nonsense mediated mRNA decay factor (SMG6, rs10852932; P = 6.88E-05, ORGCA = 0.83, ORTAK = 0.80), protein kinase C theta (PRKCQ, rs587198; P = 7.87E-05, ORGCA = 1.20, ORTAK = 1.22), endoplasmic reticulum aminopeptidase 1 (ERAP1, rs2255637; P = 8.77E-05, ORGCA = 1.18, ORTAK = 1.27), and ubiquitin conjugating enzyme E2 E3 (UBE2E3, rs7349232; P = 9.84E-05, ORGCA = 1.24, ORTAK = 1.25).

As previously described7, a group of variants in high linkage disequilibrium (LD), located downstream of the proteasome assembly chaperone 1 (PSMG1) gene on chromosome 21q22, also showed evidence of association with TAK in the analyses of each disease separately (lead variant: rs35819975, P = 7.98E-07, OR = 0.62).

Additional analyses of the association of IL12B with large-vessel vasculitis

To further analyse the consistency of the putative shared association with the IL12B variant rs755374, we checked the signal in the remaining cohorts included in the published GCA Immunochip, which comprised 650 additional cases of GCA and 12,491 controls from UK, North America (USA/Canada), Germany, and Norway6 (see Supplementary Table S2). Significant results at the nominal level of significance were observed when these replication cohorts were tested for IL12B rs755374 (P = 4.69E-02, OR = 1.13, 95% CI = 1.01–1.27), as well as when a meta-analysis including all GCA cohorts was performed (P = 5.52E-04, OR = 1.16, 95% CI = 1.07–1.26). Finally, an overall P = 3.41E-07 was obtained after meta-analysing all the available data for this SNP (including the six GCA cohorts and the two TAK cohorts), with no heterogeneity observed amongst the different ORs (Q = 0.19).

To further understand this common association, we looked for SNPs in high LD (r2 > 0.8) with IL12B rs755374 in the European populations of the 1000 genomes project using the online annotation tool HaploReg v4.1 (http://www.broadinstitute.org/mammals/haploreg/haploreg.php)9. Three markers were identified (rs6871626, rs56167332, and rs4921492), all of them previously associated with other immune-mediated diseases (Table 2). Interestingly, different functional annotations were observed for rs4921492, including enhancer and promoter histone marks (H3K4me1 and H3K4me3, respectively) as well as DNAse hypersensitivity peaks in different immune cell types. Additionally, the associated hit of our study, rs755374, also overlapped with the H3K4me1 enhancer histone mark in primary B cells from peripheral blood. Furthermore, the “genome-wide repository of associations between SNPs and phenotypes”10 showed 589 expression quantitative trait loci (eQTL) hits for rs6871626 in normal prepouch ileum, including key genes of the immune response like CD40, IL2RA, IL6R, IL10RA, IL12RB1, and different HLA class II molecules.

Table 2 Functional annotations of the lead signal IL12B rs755374 and its proxies in the European populations of the 1000 genomes project.
Full size table

Genetic correlation between giant cell arteritis and Takayasu’s arteritis

We estimated the whole genetic overlap between these two forms of LVV using a bivariate REML analysis on the Immunochip data (Table 3). A significant correlation was suggested only outside the HLA region (rG = 0.500, SE = 0.194, P = 5.00E-03) but not inside the region (rG = 0.012, SE = 0.192, P = 0.5). Similar results were obtained when we quantified the correlation by analysing polygenic risk scores on one disease calculated with the ORs of the markers that showed suggestive P-values (P < 1.00E-04) on the other disease (Table 3). GCA cases had a significant enrichment of non-HLA risk alleles for TAK when compared to controls (PGCA = 3.53E-03) and vice-versa (PTAK = 3.60E-02), with no correlation observed within the HLA region (PGCA = 0.27 and PTAK = 0.70).

Table 3 Genetic pleiotropy between giant cell arteritis and Takayasu’s arteritis using non-HLA data, HLA data only, and all Immunochip data.
Full size table

Discussion

This cross-disease analysis of Immunochip data represents the first interrogation of the genetic overlap between GCA and TAK. Although both conditions are characterised by inflammatory damage of the wall of large arteries2, the patterns of vascular involvement differ somewhat between them. In TAK the most affected vessels correspond with the aorta and its major branches, whereas in GCA the main lesions are usually localised in more peripheral arteries (such as the branches of the external carotid artery) and GCA is sometimes associated with the development of polymyalgia rheumatica11. Despite the evident differences that these two types of LVV show in the clinical manifestations, geographic distributions, and average age of disease onset, their similar histopathological features (with presence of inflammatory infiltrates within the vessel walls and granulomatous lesions12) have raised controversy over whether or not these conditions represent different subtypes of a single disease entity3. Comparative analyses of their genetic components may definitively help to answer this question.

Our results support the existence of a shared portion of the genetic susceptibility between GCA and TAK, but only outside the HLA region. As previously described6, GCA is mostly associated with class II genes (HLA-DRB1/HLA-DQA1), although some less intense class I signals may be also involved in disease predisposition. The opposite is observed in TAK, that is, the peak HLA associations are located within class I (HLA-B/MICA), with lower but still significant signals in class II7,13. The meta-analysis of this genomic region in our study cohorts reduced considerably the statistical significance of the disease-specific associations, thus confirming that distinct HLA haplotypes define each form of LVV. In this sense, GCA can be grouped with vasculitides such as ANCA-associated vasculitis or IgA vasculitis into class II diseases associated with HLA-DRB1 alleles14,15, while TAK and Behçet’s disease would represent archetypal class I diseases7,16.

Despite the similar histological features of GCA and TAK (which may be a consequence of the activation of dendritic cells within the vessel wall3,17), the different genetic architecture between these two diseases within the HLA region may reflect distinctive effects of the initial inflammatory stimuli. In this context, whereas the infiltrates in GCA are mostly composed of CD4+ T cells and macrophages12, infiltrations of CD8+ T cells are characteristic in TAK lesions18, which is in agreement with their specific associations with the HLA class II and I loci, respectively. Indeed, early studies described an increased in vitro cytotoxicity and a direct action of CD8+ T cells on large arteries from TAK patients19.

Regarding the non-HLA region, different relevant genes for the development of autoimmunity processes were suggested as shared risk factors for LVV, including NOS2, ERAP1, REL and PRKQC, which have been associated with psoriasis, Behçet disease, ankylosing spondylitis (AS), and rheumatoid arthritis, amongst others20,21,22. In the case of NOS2, which encodes a nitric oxide (NO) synthase involved in the release of NO during the immune response, previously published genetic evidences supported a role of this gene in GCA pathogenesis23,24.

However, a SNP located 5′ of IL12B, rs755374, represented the most consistent common associated signal between GCA and TAK. IL12B is a well-established risk gene for TAK7,13,25, but this is the first time that it has been implicated in the predisposition of GCA. Although it should be noted that this genetic variant represented a suggestive signal in the original Immunochip of this disease (P = 5.52E-04, OR = 1.16)6. This gene encodes the P40 subunit that is shared between the interleukins IL-12 and IL-23. It has been described that IL-12 induces Th1 differentiation, whereas IL-23 along with IL-1β promote Th-17 differentiation and function26. Consistent with the association with IL12B reported here, previous candidate gene studies have reported genetic associations between GCA and receptors of these cytokines27. Increasing evidence points to Th-1 and Th-17 cells as pivotal players in the development of LVV12,28. Specifically, in GCA, recent studies have shown that these cell types are directly involved in the main immunopathological pathways responsible for the clinical phenotypes of this type of vasculitis29,30,31,32,33,34. Interestingly, blocking of IL-12/23 p40 with ustekinumab resulted in an improvement of symptoms in patients with refractory GCA35.

The associated IL12B SNP is in high LD (r2 > 0.9) with other IL12B variants (rs6871626, rs56167332, and rs4921492) that overlap with different regulatory marks in immune cells (Table 2). One of them, rs6871626, has been recently established as a marker for disease severity in TAK25. These proxies have been previously identified as key susceptibility factors for several immune-mediated diseases, including TAK, inflammatory bowel diseases (both Crohn’s disease and ulcerative colitis), AS, and sarcoidosis, and leprosy7,13,36,37,38,39,40,41.

In summary, through an inter-disease meta-analysis of large scale genotyping data we evaluated the extent of genetic similarities between GCA and TAK. Our results suggest that the genetic architecture of these disorders differs more than expected, especially in the HLA region, considering their similar patterns of histological disease. Nevertheless, common non-HLA associations were suggested, including IL12B. Given that these conditions are often diagnosed after periods of low-level symptoms or even no symptoms, these data may lead to both reliable disease-specific diagnostic molecular markers and more targeted therapies for each form of LVV.

Methods

Study population

In total, 1,434 patients diagnosed with LVV and 3,814 unaffected controls were analysed. The study cohort comprised the two populations of patients with TAK included in the Immunochip analysis7, one of European ancestry from North America (USA/Canada; 110 TAK cases and 558 unaffected controls) and one from Turkey (327 TAK cases and 481 unaffected controls), as well as two of the six cohorts included in the Immunochip analysis of GCA6, a cohort from Spain (759 GCA cases and 1,505 unaffected controls) and a cohort from Italy (238 GCA cases and 1,270 unaffected controls) (see Supplementary Figure S3). The reason for not including all the available datasets of the Immunochip of GCA was to maintain a balance between the sample sizes of both diseases. All cases were diagnosed following the 1990 American College of Rheumatology classification criteria for both TAK and GCA42,43. The main clinical features of the analysed patients were detailed elsewhere6,7. All participants signed a written informed consent before being included in the study, and the procedures were followed in accordance with the ethical standards of the Ethics Committees on human experimentation of Consejo Superior de Investigaciones Científicas (Spain), University of Cantabria (Spain), Hospital Clínic de Barcelona (Spain), University of Parma (Italy), University of Michigan (USA), Marmara University (Turkey), and Istanbul University (Turkey), which provided approval for the study and all experimental protocols.

Quality control and data imputation

To ensure consistency amongst datasets, different standard quality filters were applied to the Immunochip raw data of both diseases in parallel with PLINK v1.0744 prior imputation: single-nucleotide polymorphisms (SNPs) with cluster separation <0.4, call rates <98%, minor allele frequencies (MAF) <1%, and those deviating from Hardy-Weinberg equilibrium (HWE; P < 0.001) were excluded; samples with <95% successfully called SNPs, first-degree relatives (identity by descent >0.4), and those showing >4 standard deviations from the cluster centroids of each population using the first ten principal components (PC; estimated using the ancestry markers included in the Immunochip) were also removed. Sex chromosomes were not analysed.

SNP genotype imputation was performed separately for each dataset using IMPUTE v.245 and the 1000 Genome Project Phase III data as reference panel (www.1000genomes.org)46. For that, the SNP map was updated to rs# and build 37 (HG19) using PLINK. Subsequently, chunks of 50,000 Mbp were generated and imputed with a probability threshold of 0.9 for merging genotypes. SNP data were also tightly filtered in PLINK after imputation as follows: call rate <98%, MAF <1%, HWE P < 0.001. A total of 213,188 SNPs were shared amongst the different imputed studies after QC.

Statistical Analysis

All analyses were carried out with PLINK and the R-base software under GNU Public license v2. First, each case-control study was tested for association by logistic regression on the best-guess genotypes (>0.9 probability) assuming an additive model and using the ten first PCs and gender as covariates. Next, all studies were meta-analysed with the inverse variance weighted meta-analysis method under a fixed effects models, except for the HLA region that was analysed under a random effects model. Cochran’s Q test was used to measure the heterogeneity of the ORs amongst the different datasets. The threshold for statistical significance in our study was established at 1.13E-06, accordingly with the estimation by the genetic type I error calculator software, which implements a Bonferroni-based validated method to control for type I errors47.

Analysis of the Genetic Pleiotropy

The genetic pleiotropy between GCA and TAK was assessed using both a bivariate and a polygenic risk score (PRS) analysis on Immunochip data, as previously described48. In brief, the genetic correlation (rG) was estimated by GCTA bivariate restricted maximum likelihood (REML) analysis using a genetic relationship matrix, containing data of identity by descent relationship for all pair-wise sets of individuals, and the first ten PCs as covariates. The statistical significance was determined by a likelihood ratio test (LRT). The genetic overlap between both types of vasculitis was also calculated by analysing PRS in one disease predicting risk for the other disease. We obtained for each participant included in the GCA/control cohorts a weighted mean of genotype dosage using the log of the ORs of set of tag SNPs (r2 < 0.20 within 500 kb windows) showing suggestive P-values in the TAK meta-analysis (P < 1.00E-04), and vice versa. We then analysed the difference between the score distribution in case and control subjects (considering the first ten PCs, country of origin, and gender as variables) through a LRT to quantify the relationship between the computed scores and disease status.

Additional Information

How to cite this article: Carmona, F. D. et al. Analysis of the common genetic component of large-vessel vasculitides through a meta-Immunochip strategy. Sci. Rep. 7, 43953; doi: 10.1038/srep43953 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.