Dense genotyping of immune-related disease regions identifies 14 new susceptibility loci for juvenile idiopathic arthritis.

We used the Immunochip array to analyze 2,816 individuals with juvenile idiopathic arthritis (JIA), comprising the most common subtypes (oligoarticular and rheumatoid factor-negative polyarticular JIA), and 13,056 controls. We confirmed association of 3 known JIA risk loci (the human leukocyte antigen (HLA) region, PTPN22 and PTPN2) and identified 14 loci reaching genome-wide significance (P < 5 × 10(-8)) for the first time. Eleven additional new regions showed suggestive evidence of association with JIA (P < 1 × 10(-6)). Dense mapping of loci along with bioinformatics analysis refined the associations to one gene in each of eight regions, highlighting crucial pathways, including the interleukin (IL)-2 pathway, in JIA disease pathogenesis. The entire Immunochip content, the HLA region and the top 27 loci (P < 1 × 10(-6)) explain an estimated 18, 13 and 6% of the risk of JIA, respectively. In summary, this is the largest collection of JIA cases investigated so far and provides new insight into the genetic basis of this childhood autoimmune disease.

Juvenile idiopathic arthritis (JIA) is the most common chronic rheumatic disease of childhood and describes a group of clinically heterogeneous arthritides which begin before the age of 16 years, persist for at least 6 weeks and have an unknown cause [1]. It is established that there is a strong genetic contribution to the risk of JIA, with a sibling risk ratio of ≈ 11.6 [2] and an increased risk for other autoimmune diseases for families of JIA patients [3]. Using the International League of Associations for Rheumatology (ILAR) criteria, JIA can be divided into subtypes based on clinical features [4]. A recent genome wide association study (GWAS) identified a number of JIA susceptibility regions [5, 6]. Additional loci have also been identified through candidate gene association studies and confirmed in multiple, independent studies [7-14]. However, to date only three loci reach genome-wide significance thresholds (HLA, PTPN22 and PTPN2) [5].

Many confirmed and nominally associated JIA susceptibility loci show association with other autoimmune diseases [5]. This striking overlap of autoimmune disease susceptibility loci may occur where the same variants contribute to multiple diseases or it may be that different variants in the same gene lead to different autoimmune disease. Thus, dense mapping of as many of the susceptibility loci as possible will be important in order to understand how individual variants contribute to the risk of disease. To facilitate these efforts a custom Illumina Infinium genotyping array has been designed by the ImmunoChip Consortium based on confirmed risk loci from 12 autoimmune diseases [15], not including JIA. The chip includes dense coverage of the extended HLA region and 186 non-HLA loci [15]. In this study, we report on analysis of the ImmunoChip in 2816 individuals with oligoarticular or rheumatoid factor (RF) negative polyarticular JIA and 13056 controls post quality control (). There is overlap in the samples used in this study and in previous genetic studies of JIA [5-12, 14]; further detail can be found in the online methods. Restriction to these two subtypes (~70% of JIA cases) reduces phenotypic heterogeneity. Given that JIA is a complex genetic disorder that shares risk loci with other autoimmune diseases, the ImmunoChip provides a unique opportunity to discover novel JIA-risk loci. In addition, the dense coverage for many regions allows for fine-mapping analysis to identify possible causal variants and help inform future studies into the functional role of JIA-risk loci.

After stringent data quality control () 123,003 SNPs with MAF≥1% were available for analysis. The inflation factor (λGC) (calculated using a set of SNPs included on the ImmunoChip for a study investigating the genetic basis for reading and writing ability) for this study was λGC=1.265, λGC1000=1.057. Seventeen of the 187 autoimmune regions investigated were significantly associated with oligoarticular and RF negative polyarticular JIA (p < 5 × 10-8) (). These data strengthen the associations for three established JIA susceptibility loci (HLA, PTPN22 and PTPN2) and provide evidence for an additional 14 regions which reach genome-wide significance for the first time. Among the three established associations, the most significant associations were observed within the MHC region (). Specifically, rs7775055 (MAFcontrols=2%) provided the strongest evidence of association with JIA (OR = 6.01, p = 3.14 × 10-174). In addition, stepwise logistic regression identified 14 SNPs that showed separate effects in the region (). The most significant SNP, rs7775055 tags the DRB1*0801-DQA1*0401-DQB1*0402 haplotype, which has been consistently implicated as conferring risk to JIA [16, 17], however other haplotypes have also been associated with JIA. The HLA SNP rs7775055 showed a highly significant difference in SNP allele frequencies between the two subtypes (). The association was stronger in the oligoarticular subtype compared to the RF negative polyarthritis subtype which is consistent with previous studies showing differences in HLA associations between the two subtypes [16, 17]. Further analysis at the amino acid level is necessary to fully understand this complex region in JIA and its subtypes. The most significant association outside the MHC region is with rs6679677 (OR = 1.59, p = 3.19 × 10-25) on chromosome 1p13.2, which contains the PTPN22 gene; rs6679677 is in linkage disequilibrium (LD) (r2=1) with rs2476601, the SNP previously associated with JIA [5, 7] and implicated as the PTPN22 causal variant [18]. We also confirmed association to PTPN2, with rs2847293 (OR = 1.31, p = 1.44 × 10-12) which lies in the intergenic region 3′ of the PTPN2 gene and is in LD (r2=0.94) with rs1893217, a SNP previously associated with oligoarticular and RF negative polyarticular JIA [5]. Stepwise logistic regression including the most significant SNP in the PTPN2 region as a covariate suggests that there is a uncommon variant, rs149850873, (MAFcontrols= 2%) that confers an independent secondary effect in the region ().

Of the 14 loci confirmed as novel JIA susceptibility loci in this study at the genome wide significance level (p < 5 × 10-8) (), five (STAT4, ANKRD55, IL2/IL21, IL2RA and SH2B3/ATXN2) have supportive evidence with JIA susceptibility from previous studies. The most significant SNP in the STAT4 region (rs10174238) is in high LD with a SNP (rs7574865) previously reported in JIA [5, 8, 10] and other autoimmune diseases [19]. However, stepwise logistic regression analysis suggests two additional independent effects (rs45539732 and rs13029532), which are located within the adjacent STAT1 gene (). Notably rs45539732 is an uncommon SNP (MAFcontrols = 3%).

There were 11 additional regions showing suggestive (p <1 × 10-6 and p > 5 × 10-8) evidence for association with oligoarticular and RF negative polyarticular JIA (), of which four have supportive evidence from previous studies (COG6, CCR1/CCR3, C3orf1/CD80, AFF3/LONRF2).

We imputed across the non-HLA JIA risk loci identified in this study using the 1000 Genomes Project (online methods) ( and ). We found only modest differences between the p-values of the top genotyped SNP compared to the top imputed SNP. We note two regions that are minor exceptions, the PRM1/C16orf75 and the C5orf56/IRF1 region (. For the latter region the top imputed SNP lies within the C5orf56 gene. The lack of a substantial gain of information from imputation of the regions is consistent with other reports on the performance of ImmunoChip imputation20, 21. This likely is due to the dense fine mapping of most of the regions on the ImmunoChip.

Of the top 17 regions, that reach genome-wide significance, 13 regions are densely mapped on ImmunoChip. LD patterns and functional annotation provide strong evidence that the signal localizes to a single gene in eight cases (PTPN2, IL2RA, STAT4, IL2RB and ZFP36L1 based on LD patterns and PTPN22, SH2B3/ATXN2 and TYK2 based on the most significant SNP being a non-synonymous coding variant) (Table 3, Supplementary Table 7 and Supplementary Fig. 1), however further functional analysis is required for confirmation.

Table 3

Potential causal SNPs within the JIA risk regions

Lead SNP	SNP in strong LD (r2>0.9) with the lead SNP	Chr	Position[*]	r2 with lead SNP	Location	Regulatory potential	Conservation	Functional prediction[*]	eQTL#
Genome wide significant SNPs
rs6679677	rs2476601	1	114377568	1	Exon of PTPN22	0.14	0.999	benign; tolerated
rs11265608	rsl205591	1	154298374	1	Intron of ATP8B2	0.89	0
rs1479924	rs13144509	4	123473487	0.94	Intergenic between IL2 and IL21	0.17	1
rs27290	rs27290	5	96350088	-	Intron of LNPEP	0.21	0		Yes[35-37]
rs3184504	rs3184504	12	111884608	-	Exon of SH2B3	0.29	0.005	benign; tolerated
rs12434551	rs3825568	14	69260588	0.98	5′ UTR of ZFP36L1	0.55	0.002
rs34536443	rs34536443	19	10463118	-	Exon of TYK2	0.40	0.19	probably damaging; deleterious
rs34536443	rs74956615	19	10427721	1	Intron of RAVER1	0	0.998
rs2266959	rs2266959	22	21922904	-	Intron of UBE2L3	0.47	0.003
rs2266959	rs2298428	22	21982892	1	Exon of YDJC	0.37	1	benign; tolerated
rs2266959	rs4820091	22	21940189	1	Intron of UBE2L3	0	0		Yes[35, 37]
Suggestive SNPs
rs4688013	rs17203104	3	119139575	0.92	Intergenic between CDGAP and TMEM39A	0	0.998
rs2364480	rs2364481	12	6497260	1	Intron of LTBR	0.36	0.002
rs2364480	rs2364480	12	6495275	-	Exon of LTBR	0.34	0.005		Yes[35-37]

SNPs in strong LD (r2>0.9) with the lead SNP on ImmunoChip with evidence for either strong regulatory potential (>0.35)[30] or conservation (>0.998)[29]

Coordinates are based on the NCBI37 assembly.

Functional prediction based on PolyPhen[38]

Data from three studies was considered: lymphoblastoid cell line (LCL) from HapMap3 (Stranger et al, 2012[35]), fibroblast (F), LCL and T-cell(T) from umbilical cords of 75 Geneva Gencord individuals (Dimas et al, 2009[36]) and adipose (A), LCL and skin (S) from 856 healthy female twins of the MuTHER resource (Grundberg et al 2012[37]). Yes if evidence for eQTL (p < 1 × 10-3).

All but one of the variants which reached genome-wide significance were common (>5% MAF). One variant, a non-synonymous coding variant within the TYK2 gene had a low allele frequency (MAFcontrols = 5%). In addition a couple of the secondary effects in PTPN2 and STAT4 were uncommon.

For three regions (TYK2, SH2B3/ATXN2 and LTBR) the most significant SNP (or a SNP in r2>0.9) lies within a coding region and are therefore strong candidates for the causal variant. For SH2B3/ATXN2, the same variant has also been associated with celiac disease (CeD) [22], vitiligo [23], RA [24], type 1 diabetes (T1D) [25] and multiple sclerosis (MS) [26]. The TYK2 SNP (rs34536443) is also the lead SNP in the region in RA [27], primary biliary cirrhosis [20] and psoriasis [28]. Other regions (IL6R, ZFP36L, IL2/IL21, UBE2L3, LTBR and C3orf1/CD80) contain SNPs that show evidence for high mammalian conservation (17-way vertebrate conservation) [29] or have a high regulatory potential score () calculated using alignments of seven mammalian genomes [30]. There is eQTL evidence for the associated SNPs in LTBR, UBE2L3 and LNPEP (). The SNP in LNPEP, rs27290, is also in LD (r2=0.78) with rs2248374, a SNP which lies within a splice site for ERAP2 [31]. The rs2248374-G allele results in a spliced ERAP2 mRNA which encodes a truncated protein. For JIA the rs2248374-G minor allele showed protective association (OR = 0.76, p = 1.8 × 10-7).

IL2RA, the IL2/IL21 region and IL2RB are now all considered confirmed susceptibility loci for JIA and implicate an important role for the IL-2 pathway in JIA disease pathogenesis. This pathway plays a vital role in T cell activation and development as well as a key role in maintenance of immune tolerance through the dependence of regulatory T cells on IL-2. Other confirmed JIA loci identified here are related to this pathway, SH2B3 is an adaptor protein involved in T cell activation and STAT4 is a transcription factor important in T cell differentiation.

We next considered the top non-HLA SNP associations separately for each JIA subtype (oligoarticular and RF negative polyarticular JIA). Only one region showed evidence for differential association, the C5orf56/IRF1 region, where the association was limited to the oligoarticular subtype of JIA. All other regions showed associations with similar effect sizes and direction of effect ().

As expected, many of the JIA-associated regions shown in are also associated with other autoimmune diseases () with the same SNP, or a highly correlated SNP associated in the same direction (assessed by comparing with information from the Catalogue of published GWAS and recent publications investigating the ImmunoChip in other autoimmune diseases [20-22, 27, 28, 32, 33]). We find a strong overlap with RA loci, which is not surprising due to the clinical similarities with JIA, and is consistent with previous studies [8, 10, 34]. In addition, there is notable overlap with T1D and CeD. Some regions (IL2/IL21, C5orf56/IRF1, IL2RB, ATP8B2/IL6R, Chr13q14, CCR1/CCR3, RUNX3 and C3orf1/CD80) show association with other autoimmune diseases but their top SNP is not highly correlated with our top JIA SNP. Some regions have not been previously associated by GWAS or ImmunoChip. In depth analysis of the results across all ImmunoChip studies will be of great value to understanding the contributions of the individual loci to the various diseases.

This study of 2816 JIA cases is the largest collaborative cohort study of JIA to date, and includes samples from across the United States, United Kingdom and Germany. The power derived from this cohort plus the large control sample size, combined with the comprehensive coverage for SNPs in regions implicated in autoimmune disease on the ImmunoChip has substantially increased our power to detect association. In setting the statistical threshold at stringent genome-wide significance levels (p < 5 × 10-8) we report 14 new loci. In addition, a second tier of 11 regions with suggestive evidence for association (p < 1 × 10-6) has been identified that are plausible candidates as risk factors but require validation. While this study dramatically increases the number of susceptibility loci identified for JIA, additional genetic risk factors likely remain to be discovered, which is supported by the QQ plot () that suggests there are residual associations after removing the above implicated regions. In addition we calculated that the entire ImmunoChip loci, the HLA region and the top 27 loci explain an estimated 18%, 13% and 6% of risk of JIA, respectively. This also suggests there must be other regions of the genome that harbor additional JIA-risk loci. In summary, this analysis of ImmunoChip has substantially enhanced our understanding of the genetic component of JIA, increasing the number of confirmed JIA loci from 3 to 17. The dense mapping of confirmed regions has narrowed down the regions to take forward into future functional studies. Importantly, these studies allow us to begin to understand where JIA fits in the spectrum of autoimmune diseases and identified a number of novel genes and pathways as potential targets for future therapeutic intervention.

Online Methods

Subjects

All cohorts comprised individuals from populations of European descent from the US, UK and Germany.

The post QC US cohorts comprised 1596 US oligoarthritis and RF negative polyarthritis JIA patients and 4048 US controls. Less than one half of these cases have already been included in a genome-wide association study and previously described [5-6]. Notably, 95 of these patients were from multiplex pedigrees such that for each pedigree one RF negative polyarthritis or oligoarticular JIA case was randomly selected for genotyping. Clinics enrolling the JIA patients for Cincinnati-based studies (listed in order of number contributed) were located in Cincinnati, OH; Atlanta, GA; Columbus, OH; Little Rock, AR; Long Island, NY; Chicago, IL; Dover, DE; Salt Lake City, UT; Cleveland, OH; Philadelphia, PA; Toledo, OH; Nashville, TN; Milwaukee, WI; and Charleston, SC. Additional DNA from JIA cases collected independently by investigators in Salt Lake City, UT (314 cases, where about 75% overlap with replication cohort in previous GWAS studies [5-6]) and Boston, MA (13 cases) or enrolled as part of the Trial of Early Aggressive Therapy in Juvenile Idiopathic Arthritis (TREAT) study (clinical trials identifier NCT00443430) (22 cases) were made available for genotyping in Cincinnati.

The US controls were derived from four sources: 793 healthy children without known major health conditions recruited from the geographical area served by Cincinnati Children's Hospital Medical Center (CCHMC) and 119 healthy adults collected at CCHMC. Previous JIA GWAS studies have include about 75% of only the pediatric controls, 484 healthy adult controls from Utah screened for autoimmune diseases and all were included in the replication cohort of previous GWAS studies [5-6]. 848 healthy adult controls collected at the Oklahoma Medical Research Foundation; and 1804 healthy US adult controls from the Genotype and Phenotype registry (www.gapregistry.org) and the NIDDK IBD Genetics Consortium. Healthy controls from the Oklahoma Medical Research Foundation (OMRF) were provided by the Lupus Family Registry and Repository (LFRR)[39] and the Oklahoma Immune Cohort (OIC). Each individual completed the Connective Tissue Disease Screening Questionnaire (CSQ)[40] and individuals with a “probable” systemic rheumatic disease were excluded. Each individual was enrolled into these studies after appropriate written consent and IRB approval by the OMRF and the University of Oklahoma Health Sciences Center. Healthy controls were also provided from the University of Minnesota SLE sibship collection[41] and these subjects were enrolled after appropriate written consent and IRB approval by the University of Minnesota.

The US collections and their use in genetic studies have been approved by the Institutional Review Board of CCHMC and each collaborating center.

The post QC UK cohort comprised 772 UK oligoarthritis and RF negative polyarthritis JIA patients from five sources: The British Society for Paediatric and Adolescent Rheumatology (BSPAR) National Repository of JIA; a group of UK patients with long-standing JIA, described previously [42]; a cohort collected as part of the Childhood Arthritis Prospective Study (CAPS), a prospective inception cohort study of JIA cases from 5 centers across UK[43]; a cohort of children recruited for the SPARKS-CHARM (Childhood Arthritis Response to Medication) study, who fulfill ILAR criteria for JIA and are about to start new disease-modifying medication for active arthritis[44] and an ongoing collection of UK cases, the UK JIA Genetics Consortium (UKJIAGC). There is overlap in the JIA cases used in this study and in previous UK candidate gene studies of JIA [7, 9-12]. JIA cases were classified according to ILAR criteria [4]. All UK JIA cases were recruited with ethical approval and provided informed consent [North-West Multi-Centre Research Ethics Committee (MREC 99/8/84), the University of Manchester Committee on the Ethics of Research on Human Beings and National Research Ethics Service (NRES 02/8/104)]. The 8530 UK controls comprised the shared UK 1958 Birth cohort and UK Blood Services Common Controls. The collection was established as part of the WTCCC[45].

The post QC German Cohort comprised 448 German oligoarthritis and RF negative polyarthritis JIA patients and 478 controls. These cases have already been included as a replication cohort in a genome-wide association study and previously described [5-6]. These patients were recruited from the German Center for Rheumatology in Children and Adolescents, Garmisch-Partenkirchen; the Department of Pediatrics, University of Tübingen; Children's Rheumatology Unit Sendenhorst, Germany; and the Department of Pediatrics, University of Prague, Czech Republic. JIA was determined retrospectively by chart review. German population-based control samples were prepared from cord blood obtained from healthy newborns in the Survey of Neonates in Pomerania (SNiP) consortium [46]. The respective Institutional Review Boards approved the collection of these samples and participation in this study. Demographic breakdown of the cohorts is shown in .

Genotyping and quality controls

Samples were genotyped using ImmunoChip, a custom-made Illumina Infinium array, described previously[22]. Genotyping was performed according to Illumina's protocols at labs in Hinxton, UK, Manchester, UK, Cincinnati, US, Utah, US, Charlottesville, US and New York, US. The Illumina GenomeStudio GenTrain2.0 algorithm was used to recluster all 15872 samples together.

SNPs were excluded if they had a call rate <98% and a cluster separation score of <0.4. Samples were then excluded for call rate <98% across 178203 markers or if there were inconsistencies between recorded and genotype inferred gender. Duplicates and first- or second-degree relatives were also removed. Principal component (PC) analysis was computed, using Eigensoft v4.2 (http://www.hsph.harvard.edu/faculty/alkes-price/software/) [47, 48], on the samples, merged with HapMap phase 2 individuals (CEU, YRI and CHB) as reference populations, to identify genetic outliers. PC analysis was performed on a subset of SNPs, removing SNPs in known regions of high linkage disequilibrium (LD), with MAF < 0.05 and pruned for LD between markers. To maximize genetic homogeneity within the samples the initial PC analysis was followed by five subsequent PC analyses where at each iteration individuals 5 standard deviations from the mean were removed. The PCs from the 5th iteration were used as covariates in the logistic regression analysis. A SNP was removed from the primary analysis if it exhibited significant differential missingness between cases and controls (p<0.05), had significant departure from Hardy-Weinberg equilibrium (p<0.001 in controls) or had a MAF<0.01.

Statistical analysis

To test for an association between a SNP and case/control status, a logistic regression analysis was computed using the 5 PCs as covariates. The primary inference was based on the additive genetic model, unless there was significant lack-of-fit to the additive model (P<0.05). If there was evidence of a departure from an additive model, then inference was based on the most significant of the dominant, additive and recessive genetic models. The additive and recessive models were computed only if there were at least 10 and 20 individuals homozygous for the minor allele, respectively. For analysis of the X chromosome, the data analysis was first stratified by gender followed by a meta-analysis. The genomic control inflation factor (λGC) was calculated using a set of SNPs included on the ImmunoChip for a study investigating the genetic basis for reading and writing ability (submitted by J.C.Barrett). We visually inspected the cluster plots for the most associated SNPs in the regions to confirm the genotyping quality. Additionally, concordance of genotyping data was compared with data previously generated on other platforms. A subset of cases had high resolution HLA genotyping. These data were used to investigate if the SNPs with the strongest statistical associations with JIA were in high linkage disequilibrium with classical HLA alleles/haplotypes. To investigate subtype effects the two main subtypes (Oligoarticular JIA and RF negative polyarticular JIA) were compared separately against the same controls. Disease association heterogeneity was tested by testing for significant differences in SNP allele frequencies between the two subtypes. To determine how many independent associations were within a genomic region, a manual stepwise procedure (i.e., forward selection with backward elimination, entry and exit criteria of P<0.0001) was computed [49]. Specifically, for each region which reached genome-wide significance, the top SNP was included as a covariate and the association statistics re-calculated. SNPs were allowed to enter and exit models in this stepwise fashion until no additional SNPs met a significance threshold of P<0.0001. The stepwise procedure was modified slightly in the greater MHC region to have an entry and exit criteria of P<0.00001. These statistical analyses were performed using PLINK v1.07[50] and SNPGWA version 4.0 (www.phs.wfubmc.edu).

The cumulative variance explained by common SNP variation was estimated using a variance component model and restricted maximum likelihood estimation as implemented in the program GCTA [51] and adjusting for the PCs as covariates and using Yang's correction factor (c=0 from formula 9) for imperfect LD with causal variants. Estimates are based on SNPs that had <1% missing genotypes and stringent relatedness threshold of 0.025.

We computed SNP genotype imputation across the regions of the ImmunoChip. We used the program SHAPEIT (www.shapeit.fr/) to pre-phase our ImmunoChip data and IMPUTE2 (https://mathgen.stats.ox.ac.uk/impute/impute_v2.html) with the 1000 Genomes Phase 1 integrated reference panel to impute the SNP genotypes. To account for phase uncertainty, we tested for association using SNPTEST (https://mathgen.stats.ox.ac.uk/genetics_software/snptest/snptest.html). Only genotyped SNPs of high quality were used to inform imputation. Imputed SNP quality was assessed using the information score (> 0.5) and the confidence score (> 0.9).

Regional plots of association and adjusting for the strongest SNP association was computed using Locuszoom (http://csg.sph.umich.edu/locuszoom/) [52].

Table 1

Regions reaching genome-wide significant association with oligoarticular and RF negative polyarticular juvenile idiopathic arthritis.

Gene Region	Chr	Position[*]	Most significant SNP	Minor allele	MAF controls (n=13056)	MAF cases (n=2816)	Best p-value	Model	Odds ratio	95% confidence intervals	SNP position
HLA-DQB1/HLA-DQA2	6	32657916	rs7775055	G	0.02	0.12	3.14 × 10-174	Dominant	6.01	5.3-6.81	Intergenic
PTPN22	1	114303808	rs6679677	A	0.1	0.14	3.19 × 10-25	Additive	1.59	1.45-1.73	Intergenic
STAT4	2	191973034	rs10174238	G	0.23	0.28	1.28 × 10-13	Additive	1.29	1.20-1.37	Intron
PTPN2	18	12782448	rs2847293	A	0.17	0.2	1.44 × 10-12	Additive	1.31	1.22-1.41	Intergenic
ANKRD55	5	55440730	rs71624119	A	0.25	0.2	4.40 × 10-11	Additive	0.78	0.73-0.84	Intron
		55442249	rs10213692[#]	C	0.25	0.2	2.73 × 10-11	Additive	0.79	0.74-0.8	Intron
IL2/IL21	4	123387600	rs1479924	G	0.29	0.24	6.24 × 10-11	Additive	0.79	0.74-0.85	Intergenic
TYK2	19	10463118	rs34536443	G	0.05	0.03	1 × 10-10	Additive	0.56	0.47-0.67	Coding (NS)
IL2RA	10	6089841	rs7909519	C	0.11	0.08	8 × 10-10	Additive	0.72	0.64-0.8	Intron
SH2B3/ATXN2	12	111884608	rs3184504	A	0.49	0.54	2.60 × 10-09	Additive	1.2	1.13-1.27	Coding (NS)
		111932800	rs7137828[#]	C	0.49	0.54	1.61 × 10-09	Additive	1.20	1.13-1.28	Intron
ERAP2/LNPEP	5	96350088	rs27290	G	0.44	0.47	7.5 × 10-09	Dominant	1.32	1.20-1.45	Intron
		96357178	rs27293[#]	A	0.44	0.47	7.37 × 10-09	Dominant	1.31	1.19-1.43	Intron
UBE2L3	22	21922904	rs2266959	A	0.19	0.22	6.2 × 10-09	Dominant	1.24	1.15-1.33	Intron
C5orf56/IRF1	5	131813219	rs4705862	T	0.44	0.39	1.02 × 10-08	Additive	0.84	0.79-0.89	Intergenic
		131797547	rs6894249[#]	G	0.39	0.35	9.73 × 10-10	Dominant	0.76	0.70-0.83	Intron
RUNX1	21	36715761	rs9979383	G	0.37	0.33	1.06 × 10-08	Dominant	0.78	0.72-0.85	Intergenic
		36712588	rs8129030[#]	T	0.37	0.33	5.44 × 10-09	Dominant	0.78	0.71-0.84	Intergenic
IL2RB	22	37534034	rs2284033	A	0.44	0.39	1.55 × 10-08	Additive	0.84	0.79-0.89	Intron
ATP8B2/IL6R	1	154364140	rs11265608	A	0.1	0.12	2.75 × 10-08	Dominant	1.33	1.2-1.47	Intergenic
		154379369	rs72698115[#]	C	0.1	0.12	1.26 × 10-08	Dominant	1.36	1.22-1.52	Intron
FAS	10	90762376	rs7069750	C	0.44	0.48	2.93 × 10-08	Additive	1.18	1.11-1.25	Intron
ZFP36L1	14	69253364	rs12434551	A	0.47	0.43	1.59 × 10-08	Dominant	0.77	0.71-0.85	Intergenic
		69260588	rs3825568[#]	T	0.46	0.42	1.24 × 10-08	Dominant	0.77	0.70-0.84	5′UTR

Coordinates are based on the NCBI37 assembly.

Imputed SNP results are included when they show a better p-value than the most significant directly genotyped SNP in the region.

Chr=chromosome. MAF=minor allele frequency, NS=non-synonymous.

Table 2

Regions with suggestive significant association with oligoarticular and RF negative polyarticular juvenile idiopathic arthritis (p < 1 × 10-6 and p > 5 × 10-8)

Gene region	Chr	Position[*]	Most significant SNP	Minor allele	MAF controls (n=13056)	MAF cases (n=2816)	Best p-value	Model	Odds ratio	95% confidence intervals	SNP position
LTBR	12	6495275	rs2364480	C	0.25	0.28	5.10 × 10-08	Additive	1.2	1.12-1.28	Coding (NS)
		6493351	rs10849448[#]	A	0.24	0.27	4.54 × 10-09	Additive	1.24	1.15-1.33	5′UTR
IL6	7	22798080	rs7808122	A	0.44	0.48	5.80 × 10-08	Additive	1.19	1.11-1.25	Intergenic
		22809490	rs6946509[#]	T	0.45	0.48	3.36 × 10-08	Additive	1.19	1.12-1.26	Intergenic
COG6	13	40350912	rs7993214	A	0.35	0.31	1.61 × 10-07	Additive	0.84	0.79-0.9	Intergenic
		40355913	rs9532434[#]	T	0.36	0.32	4.52 × 10-08	Additive	0.84	0.79-0.89	Intron
Chr13q14	13	43056036	rs34132030	A	0.32	0.29	1.77 × 10-07	Additive	1.18	1.11-1.26	Intergenic
CCR1/CCR3	3	46253650	rs79893749	A	0.15	0.12	1.88 × 10-07	Additive	0.78	0.72-0.86	Intergenic
PRR5L	11	36363575	rs4755450	A	0.35	0.31	3.35 × 10-07	Dominant	0.8	0.74-0.87	Intergenic
		36343693	rs7127214[#]	G	0.35	0.31	1.90 × 10-08	Dominant	0.78	0.71-0.85	Intron
PRM1/C16orf75	16	11428643	rs66718203	C	0.18	0.14	4.46 × 10-07	Additive	0.81	0.74-0.88	Intergenic
		11471414	rs11074967[#]	G	0.42	0.38	2.4 × 10-07	Additive	0.85	0.80-0.91	Intergenic
RUNX3	1	25197155	rs4648881	G	0.49	0.53	4.66 × 10-07	Additive	1.16	1.1-1.23	Intergenic
C3orf1/CD80	3	119229486	rs4688013	A	0.19	0.22	6.30 × 10-07	Additive	1.2	1.12-1.29	Intron
		119221064	rs11714843[#]	A	0.18	0.21	3.64 × 10-07	Additive	1.22	1.13-1.31	Intron
JAZF1	7	28182306	rs10280937	G	0.11	0.13	6.60 × 10-07	Additive	1.25	1.15-1.37	Intron
		28187344	rs73300638[#]	C	0.11	0.14	1.12 × 10-07	Additive	1.28	1.17-1.41	Intron
AFF3/LONRF2	2	100813499	rs6740838	A	0.39	0.43	8.83 × 10-07	Dominant	1.25	1.14-1.37	Intergenic
		100834217	rs10194635[#]	T	0.39	0.43	8.10 × 10-07	Dominant	1.24	1.14-1.36	Intergenic

Coordinates are based on the NCBI37 assembly.

Imputed SNP results are included when they show a better p-value than the most significant directly genotyped SNP in the region.

Chr=chromosome. MAF=minor allele frequency.