Pleiotropy in the Genetic Predisposition to Rheumatoid Arthritis - a phenome-wide association study and inverse-variance weighted meta-analysis

Abstract

Objective:

To examine the hypothesis that a genetic predisposition toward rheumatoid arthritis (RA) increases the risk for 10 cardiometabolic and autoimmune disorders previously associated with RA in epidemiological studies and to define new genetic pleiotropy.

Methods:

We tested the hypothesis using two approaches. First, we constructed a weighted genetic risk score (wGRS) and studied its association with the 10 prespecified disorders; additionally, we performed a phenome-wide study (PheWAS) to identify potential new associations. Second, we used inverse variance weighted regression (IVWR) meta-analysis to characterize the association between genetic susceptibility to RA and the prespecified disorders.

Results:

The wGRS for RA was significantly associated with type I diabetes (T1D) (OR 1.10 [95% CI 1.04–1.16], P=9.82×10⁻⁴) and multiple sclerosis (OR 0.82 [0.77–0.88], P=1.73×10⁻⁸), but not with other cardiometabolic phenotypes. In the PheWAS, the wGRS was additionally associated with increased risk of several autoimmune phenotypes including RA, thyroiditis, and systemic sclerosis, and with decreased risk of demyelinating disorders. In the IVWR meta-analyses, RA was significantly associated with T1D (P=1.15×10⁻¹⁴) with evidence of horizontal pleiotropy (MR-Egger intercept estimate P=0.001) likely driven by rs2476601, a PTPN22 variant. The association between T1D and RA remained significant (P=9.53×10⁻⁹) after excluding rs2476601 with no evidence of horizontal pleiotropy (intercept estimate P=0.939). RA was also significantly associated with type 2 diabetes and C-reactive protein; these associations were driven by variation in the major histocompatibility complex (MHC).

Conclusion:

We found evidence suggesting pleiotropy between the genetic predisposition for RA and other autoimmune disorders including T1D.

INTRODUCTION

Rheumatoid arthritis (RA) is a chronic inflammatory disease that affects approximately 1% of the population (1). Epidemiologic and clinical studies have noted an increased prevalence of several conditions occurring in association with RA, including coronary artery disease (CAD) (2), ischemic stroke (3), atrial fibrillation (AF) (4), diabetes - particularly type 1 diabetes (T1D) (5, 6), chronic kidney disease (CKD) (7), hypertension (HTN) (8), central obesity (8), and low levels of low density lipoprotein cholesterol (LDL-C) (9), as well as with decreased prevalence of multiple sclerosis (MS) (10). Inflammation contributes to the pathogenesis of many of these conditions, raising the possibility of shared genetic pathogenesis with RA.

Numerous common single nucleotide polymorphisms (SNPs) associated with risk of RA have been identified by genome wide association studies (GWAS) (11). Such information can be harnessed to construct a genetic risk score (GRS) to predict not only RA but also conditions that may have shared pathogenesis; for example, as was used to define the inverse relationship between genetic predictors of LDL-C concentrations and risk of type 2 diabetes (T2D) (12). Moreover, in addition to examining the relationship between a GRS and a few prespecified phenotypes, phenome wide association studies (PheWAS) allow the unbiased identification of new phenotype associations (13).

Summary statistics from large GWAS studies can be used to test whether SNPs associated with one disease or risk factor (e.g. RA) are also associated with a second disease or risk factor (e.g., T1D, LDL-C) using different approaches including inverse variance weighted regression (IVWR) meta-analysis (14). Importantly, the genetics underlying many of the conditions associated with RA have been characterized by large GWAS.

To define whether a genetic predisposition to RA increases risk of other autoimmune and cardiometabolic conditions, we used two approaches: 1) we constructed a weighted GRS for RA and examined its associations with selected prespecified phenotypes reported to be associated with RA and also performed a PheWAS to identify new pleiotropic associations; 2) we used IVWR meta-analysis to test the hypothesis that genetic predisposition to RA increases the risk of selected prespecified phenotypes associated with RA.

METHODS

Data Sources

We studied the association between the genetic liability for RA and 10 autoimmune and cardiometabolic outcomes previously associated with RA: AF, ischemic stroke, CAD, T2D, obesity, HTN, CKD, hyperlipidemia, T1D, and MS using BioVU, the Vanderbilt University Medical Center (VUMC) DNA Biobank. A full description of BioVU has been published (15). Briefly, BioVU accrues DNA from blood samples obtained during routine clinical care from patients who have consented to have a DNA sample collected. DNA is extracted from samples that would otherwise be discarded, de-identified, and linked to a de-identified version of the electronic health record (EHR) at VUMC. Approval for the present study was obtained from the Vanderbilt Institutional Review Board. We used data from the eMERGE network (16) to replicate PheWAS findings. To replicate significant findings in the IVWR analysis, we used summary statistics from the UK Biobank (Neale Lab GWAS round 2 available at http://nealelab.is/uk-biobank).

For the IVWR analyses, we selected the largest genetic meta-analysis with summary-level data available for individuals of European ancestry for the selected outcomes (or proxies when the exact phenotype was not available) previously associated with RA (Supplementary Table S1). We studied the same 10 outcomes as used for the GRS association analyses and also 2 additional immunological biomarkers for which there are no good phenotype equivalents in the EHR, C-reactive protein (CRP) and interleukin 6 (IL-6) concentrations.

Genotyping

In the BioVU EHR cohort, genotyping was performed by the Vanderbilt Technologies for Advance Genomics (VANTAGE) according to standard protocols on the Illumina Infinium Multi-Ethnic Genotyping Array (MEGA^EX) platform. eMERGE subjects were genotyped on multiple platforms and underwent QC analyses and imputation as previously described. (17, 18) Quality control (QC) analyses used PLINK version 1.90β3 (19, 20) and included reconciling strand flips, verifying that allele frequencies were concordant among data sets, and identifying duplicate and related individuals (one of each pair of subjects with a pi-hat>0.05 was excluded).. Data sets were standardized using the HRC-1000G-check tool v4.2.5 (http://www.well.ox.ac.uk/~wrayner/tools/) and pre-phased using SHAPEIT (21). BioVU data was imputed using IMPUTE2 (22), in conjunction with the 10/2014 release of the 1000 Genomes cosmopolitan reference haplotypes. All other genetic data were imputed using the Michigan Imputation Server (HRC v1.1)(23). . Imputed data were filtered for a sample missingness rate <2%, a SNP missingness rate <4% and SNP deviation from Hardy-Weinberg P< 10^-6. Principal components (PCs) were calculated using the SNPRelate package (24).

Phenotype Data

For the 10 prespecified autoimmune and cardiometabolic outcomes, we extracted clinical diagnoses from the EHR identified by the International Classification of Diseases, Ninth Revision (ICD9) codes that mapped to the prespecified outcomes and transformed these ICD9 codes into phecodes, which aggregate one or more related ICD-9 codes into distinct diseases or traits (13). For each phenotype, cases were defined as individuals with 2 or more instances of the specific phecode in the EHR. Controls were defined as individuals without the phecode or any close related phecode. For the PheWAS analysis, we followed the same procedures and included 1162 clinical phenotypes with 100 or more cases in the EHR to assure statistical power. Controls were matched by the represented age of cases.

Statistical analysis

Because the risk alleles were derived largely from GWAS studies in whites of European ancestry, only individuals of European ancestry were included in the analyses. Genetically-predicted risk for RA was calculated using summary statistics for RA-related SNPs that were fed into a weighted GRS using the following equation:

$$\text{Weighted}\text{genetic}\text{risk}\text{score}(\text{wGRS})=\sum _{i=1}^{\#SN{P}_S}\left({\text{β}}_i\times {[\text{SNP}\text{genotype}]}_{\text{i}}\right)$$

where β is the effect size (log odds-ratio) of the risk allele and the genotype is the number of copies of the risk allele coded as 0,1, or 2. Only independent SNPs (r²<0.10) that passed quality control and reached genome-wide significance (P<5×10⁻⁸) were included in the wGRS (Supplementary Table S2). A multivariable logistic regression adjusting for 5 PCs, median age in the EHR, and sex was performed for the pre-specified phenotype analysis and the PheWAS using the wGRS as main predictor. For the 10 pre-specified phenotypes, a Bonferroni-corrected P-value<0.005 (0.05/10 phenotypes) was considered significant. For the global PheWAS, a false discovery rate (FDR)<0.1 was considered statistically significant. Analyses were performed using the PheWAS R package (13).

To replicate PheWAS findings in eMERGE (16) we calculated the wGRS in a set of ~26,000 individuals of European ancestry and performed a logistic regression analysis adjusting for 5 PCs, sex, and median age in the EHR for phenotypes that were significantly associated with the wGRS in the PheWAS. Phecodes in eMERGE include ICD9 and ICD10 codes. A Bonferroni-adjusted replication P-value of 0.05 divided by the number of phenotypes tested in the replication analysis was considered significant (<0.05/ #phenotypes for replication).

To study the relationships between genetic liability for RA and selected phenotypes we performed IVWR meta-analyses (25, 26). Summary statistics for the exposure (RA) and the selected outcomes (Supplementary Table S1) were extracted and the SNPs were aligned to the 1000 Genome Project phase 3. To ensure the selection of independent SNPs from the exposure and each outcome dataset, we selected SNPs that were significantly associated with RA at P<5×10⁻⁶ (11) and a linkage disequilibrium (LD)-reduced (r²<0.05) set of SNPs with a minor allele frequency>0.05 were selected as instrumental variables (IVs) to estimate the combined effect of these SNPs in the IVWR meta-analysis.

As a sensitivity analysis, we performed weighted median regression since this approach, while less powered than the IVWR approach, provides better estimates of the true effect size when less than 50% of the IVs are not valid (25, 26). In addition, we also tested for unbalanced horizontal pleiotropy using MR-Egger regression (27). The analyses were performed using the Two-Sample Mendelian Randomization R-package and a P-value<0.004 (0.05/12 outcomes) was considered significant. P-value<0.05 for the intercept estimate in the Egger regression was considered significant and indicates the presence of horizontal pleiotropy.

Because the major histocompatibility complex (MHC) region is one of the most polymorphic regions in the human genome and is characterized by a complex LD structure and long-range haplotypes, identifying causal variants is challenging (28). Thus, we performed a sensitivity analysis excluding SNPs in this region (chr.6: from 25–35 Mb, GRCh37). All associations were expressed as odds ratio (OR) and 95% confidence interval (95% CI).

To replicate significant findings in the IVWR analysis, we extracted summary statistics from the UK Biobank. Low confidence SNPs (e.g. minor allele count < 25 or MAF<0.001 in cases) were filtered out and an LD-reduced set of SNPs (as described above) were selected as IVs. A P-value<0.05 was considered significant for replication.

RESULTS

GRS analysis

Ninety five of the 102 RA-associated SNPs passed quality control and were included in the wGRS calculation. The wGRS was calculated in 15,939 (53.4%) women and 13,922 (46.6%) men of European ancestry with genotype information and clinical data available in BioVU. The median (interquartile range) for the GRS was 0.057 (0.054, 0.060) and median age was 59 years (48, 69).

Among the 10 preselected phenotypes, the wGRS for RA was positively associated with T1D (OR [95%CI] =1.10 [1.04, 1.16], P=9.82×10⁻⁴) and negatively associated with MS (0.82 [0.77, 0.88], P=1.73×10⁻⁸) (Table 1, Figure 1). None of the other preselected phenotypes were significantly associated with the RA wGRS.

Table 1:

Association of the weighted genetic risk score for RA with cardiometabolic and autoimmune phenotypes

Phenotype	phecode	# cases	# controls	OR [95%CI]	P-value
Atrial fibrillation	427.21	5195	14453	0.98 [0.94, 1.02]	0.249
Ischemic stroke	433.21	1399	23097	1.02 [0.97, 1.08]	0.392
Coronary atherosclerosis	411.4	8720	16746	0.99 [0.96, 1.02]	0.489
Type 2 diabetes	250.2	6309	19612	1.02 [0.98, 1.05]	0.252
Essential hypertension	401.1	16488	11767	0.98 [0.95, 1.01]	0.131
Chronic renal failure	585.3	3486	21257	1.01 [0.97, 1.05]	0.580
Hyperlipidemia	272.13	587.6	15039	1.00 [0.97, 1.03]	0.988
Obesity	278.1	4438	22965	0.99 [0.96, 1.02]	0.621
Type 1 diabetes	250.1	1456	19571	1.10 [1.04, 1.16]	9.82×10−4
Multiple sclerosis	335	1009	21542	0.82 [0.77, 0.88]	1.73×10−8

Figure 1: Clinical diagnoses associated with a weighted genetic risk score (wGRS) for rheumatoid arthritis (RA).

Green filled triangles and black filled dots in the volcano plot represent significant and non-significant associations at false discovery rate (FDR)< 0.1 in BioVU, respectively. Table in the right provides the clinical labels and results for significant associations in BioVU and replication in eMERGE. * Less than 100 cases were available for this phenotype.

In addition to the findings for T1D and MS, the RA wGRS was associated with increased risk of several autoimmune related disorders including RA (P=1.54×10⁻²⁷), T1D retinopathy (P=1.56×10⁻⁶), phenotypes related to hypothyroidism (P≤3.92×10⁻⁶), autoimmune thyroiditis related phenotypes (P≤7.27×10⁻⁵), and systemic sclerosis (P=4.33×10⁻⁴) in the PheWAS. In addition to the decreased risk of MS, there was also an inverse association with the risk of demyelinating disorders (0.83 [0.77, 0.91], P=4.93×10⁻⁵) (Figure 1, Supplementary Table S3). When the RA-associated SNP in the MHC region was excluded from the analysis, the inverse associations between the wGRS and MS and demyelinating disorders and the positive associations with T1D were no longer significant (Supplementary Table S4). Several other immune and non-immune phenotypes, such as psoriasis, infections, cancers, and arrhythmias were nominally associated with the wGRS for RA (P<0.05).

Replication of the significant results in the PheWAS showed that the wGRS for RA remained significantly associated with RA-related phenotypes and with hypothyroidism-related phenotypes in eMERGE (Figure 1) at a P-value threshold of <0.0036 (0.05/14 phenotypes), but the association with MS was nominal (P-value=0.02).

IVWR meta-analyses

A genetic predisposition toward RA was consistently associated with T1D (Figure 2) in the IVWR, weighted median, and Egger analysis (Supplementary Table S5), with evidence for horizontal pleiotropy (y-intercept was significantly different from zero in the Egger regression, P=0.001) and likely driven by rs2476601 (Table 2, Supplementary Figure S1).

Figure 2: IVWR meta-analysis for RA risk alleles.

#SNPs indicates the number of SNPs that were selected for the meta-analysis, OR [95%CI]: odds ratio [95% confidence interval]. AF: atrial fibrillation, IS: ischemic stroke, CAD: coronary artery disease, T2D: type 2 diabetes, SBP: systolic blood pressure, CKD: chronic kidney disease, LDL: low density lipoprotein-cholesterol, WC: waist circumference, T1D: type 1 diabetes, MS: multiple sclerosis, CRP: C-reactive protein, IL6: Interleukin 6:

Table 2:

Genetic association analysis between RA and IVWR phenotypes

Phenotypes	# SNPs	IVWR		Egger regression
		OR [95%CI]	P-value	OR [95%CI]	P-value	I.E.	P-valuea
T1D
All RA SNPsb	*	*	*	*	*	*	*
Excluding MHC SNPs	99	1.67 [1.47, 1.90]	1.15×10−14	2.61 [1.96, 3.49]	6.92×10−11	−0.056	0.001
Excluding rs2476601	98	1.47 [1.29, 1.68]	9.53×10−9	1.50 [0.90, 2.48]	0.117	−0.002	0.939
T2D
All RA SNPsb	230	1.03 [1.02, 1.04]	1.11×10−9	1.04 [1.03, 1.06]	6.94×10−8	−0.003	0.055
Excluding MHC SNPs	136	1.03 [1.00, 1.05]	0.020	1.04 [0.98, 1.09]	0.170	−0.001	0.612
CRP
All RA SNPsb	226	1.02 [1.01, 1.03]	1.72×10−6	1.02 [1.01, 1.04]	0.001	−0.001	0.566
Excluding MHC SNPs	136	1.02 [1.00, 1.04]	0.033	1.01 [0.97, 1.06]	0.620	0.001	0.753
CAD
All RA SNPsba	230	1.02 [1.01, 1.03]	3.83×10−4	1.02 [1.00, 1.04]	0.031	0.000	0.924
Excluding MHC SNPs	136	1.01 [0.99, 1.04]	0.258	1.01 [0.95, 1.08]	0.706	0.000	0.979

RA: rheumatoid arthritis; T1D: Type 1 diabetes; T2D: Type 2 diabetes; CRP: C-reactive protein; CAD: coronary artery disease; SNP: single nucleotide polymorphism; IVWR: Inverse weighted variance regression; OR: odds ratio; 95%CI: 95% confidence interval. I.E.: Intercept estimate for the Egger regression.

^aP-value for the intercept.

^bIncludes independent SNPs (R²<0.05) that were previously associated with rheumatoid arthritis (RA) at P <5×10⁻⁶.

^*Summary statistics for T1D already excluded the MHC region.

T2D, CRP, and CAD were significantly associated with RA in the IVWR (Figure 1, Table 2); but discrepancies were observed in the weighted median approach for CAD (Supplementary Table S5), suggesting the association observed with CAD in the IVWR is less robust compared to T2D and CRP. A nominal association was found with SBP and CKD, with evidence of horizontal pleiotropy for CKD (P<0.001 for the y-intercept in the Egger regression, Supplementary Table S5).

To examine the possibility that SNPs in the MHC region were responsible for the associations observed with T2D, CRP and CAD, we excluded this region from the analyses and none of the associations remained significant (Table 2, Supplementary Table S7). The MHC region was not included in the summary statistics for T1D thus an analysis excluding this region was not necessary. For T1D, ninety-nine independent SNPs were included in the IVWR analysis and 22 of them were associated with RA and T1D at P<5×10^-5. Twenty of these 22 SNPs had the same direction of effect in RA and T1D (Supplementary Table S6) and the most significant SNP associated with both phenotypes was rs2476601, a SNP in PTPN22. The other concordant SNPs mapped to SH2B3, IL2RA, CTLA4, PTPN2, TYK2, CD226, CD28, and AFF3 among other loci (Supplementary table S6). To assess if rs2476601 was responsible for the horizontal pleiotropy observed for T1D, we excluded this SNP from the IVWR analysis, and the association remained significant with no evidence of horizontal pleiotropy in the Egger regression (Table 2).

To validate these findings, we used summary statistics from UK Biobank for self-reported T1D, self-reported T2D, and CRP. As a proxy for CAD, we used summary statistics from self-reported myocardial infarction /heart attack. In T1D, a similar pattern of associations was observed with evidence of horizontal pleiotropy when the MHC region and rs2476601 were included in the analysis (Supplementary Table S8). CAD, T2D and CRP showed a smaller but still significant associations with RA (Supplementary Table S8), with evidence of horizontal pleiotropy for CRP (P=0.02 for the y-intercept in Egger regression, Supplementary Table S8). All the associations were attenuated when the MHC region was excluded from the analysis (Supplementary Table S8).

DISCUSSION

The increased prevalence of several autoimmune and cardiometabolic conditions in patients with RA has raised the possibility of shared genetic predisposition. However, little work has addressed this possibility. We found evidence for pleiotropy between SNPs contributing to a genetic predisposition for RA and other autoimmune disorders including T1D.

The wGRS for RA was associated with increased risk for T1D and related phenotypes, and in the IVWR analysis genetic predisposition to RA was also associated with increased risk of T1D, a finding independent of the PTPN22 variant. Genetic association between RA and T1D has been suggested previously. For example, a previous study comparing underlying genetic susceptibility of T1D and 15 other immune disorders used a variant set enrichment method to show that RA was strongly correlated with T1D (29). In the largest meta-analysis in RA, more than two thirds of non-MHC RA risk loci showed pleiotropic effects across several disorders, including T1D (11), and seven loci were significantly associated with both diseases (AFF3, CTLA4-CD28, BACH2, RASGRP1, PTPN2, TYK, and PTNP22) in a fine-mapping study of RA and T1D (30). We observed that SNPs in several of these loci were not only significantly associated with T1D and RA, but the direction of effect was consistent in both phenotypes.

The genetic association between RA and T1D was independent of rs2476601, a missense variant in PTPN22. Although PTPN22 is the most influential non-MHC gene for autoimmunity (with OR of ~1.7 and 2.0 for RA and T1D, respectively) (31), our findings suggest that the genetic association observed between RA and the increased risk of T1D results from the aggregate effect of the other RA risk alleles.

The wGRS for RA was also associated with a decreased risk of MS and this association disappeared when the single MHC region SNP in the wGRS was excluded. A similar trend was observed in the IVWR analysis where genetic predisposition to RA was nominally associated with a decreased risk for MS, and the exclusion of MHC region attenuated the association. However, the protective signal for MS was only nominally significant in the eMERGE replication set. Previous clinical and population studies have suggested an inverse relationship between RA and MS (10), and genetic studies have found that some risk alleles are significantly associated with both diseases, several with effects in the opposite direction so that the allele that increased risk for RA decreased risk for MS and vice versa (32, 33). The same risk alleles can have opposite directional effects in different autoimmune diseases; for example, a risk allele in CD40 increases the risk of MS and inflammatory bowel disease but decreases risk of RA, lupus, autoimmune thyroiditis and Kawasaki disease (34). Thus, we cannot exclude that genetic predisposition for RA protects against for MS, a finding that needs further exploration.

To our knowledge, this is the first study that explores the potential genetic pleiotropy of RA using a PheWAS with a wGRS for RA. In the PheWAS, we found that the wGRS for RA was associated with several autoimmune disorders besides T1D and MS including RA, T1D-related diagnoses, and autoimmune thyroiditis-related phenotypes. However, only the association with RA-related phenotypes and hypothyroidism were significant in the replication analysis. Interestingly, when the HLA SNP was excluded from the wGRS there were additional significant associations with other autoimmune disorders such as lupus-related phenotypes (nephritis and proteinuria), biliary cirrhosis, systemic sclerosis, and polyarteritis nodosa, among others.

The finding of an association between the non-HLA RA wGRS and other autoimmune disorders is not surprising. Genome-wide genotyping studies have found several risk loci that are shared among autoimmune disorders (35).For example, several non-MHC RA risk alleles have shown pleiotropic effects in several disorders, including T1D, MS, and lupus (11), and a cross phenotype meta-analysis of 7 common autoimmune diseases (RA, T1D, MS, lupus, psoriasis, Crohn’s and celiac disease) found that approximately 44% of disease-associated SNPs were shared by some of these disorders (36). Similarly, a study that explored interconnections across multiple diseases in a large EHR system found that the strongest disease network (indicated by the highest number of shared SNPs) was composed of RA, T1D, MS and psoriasis (37). The SNPs linking these conditions mapped to 18 loci in the MHC region and one in PTPN22.

The association between the wGRS for RA and T2D was not significant; however, genetic liability for RA was associated with increased risk of T2D in the IVWR analysis, probably through variants in the MHC region since its exclusion from the analysis attenuated the association. The differing results in the wGRS and the IVWR analyses is likely explained by the number of SNPs from the MHC region included in each approach; the wGRS includes one variant whereas the IVWR includes 94 independent MHC SNPs. Even though RA and T2D share several risk alleles (38), previous genetic studies using GWAS summary statistics have not found a genetic correlation between RA and T2D (11, 39), likely because the MHC region was omitted in these studies. Thus, it is possible that variants in the MHC region may explain the shared association observed in the IVWR analysis. Indeed, GWAS studies identified several MHC SNPs associated with T2D (40), but the biological mechanisms through which they affect the risk of T2D have not been defined. Also, a recent PheWAS using high resolution typing of 33 human leukocyte antigen (HLA) genes in Japanese individuals found that MHC variability was associated with several cardiometabolic phenotypes including T2D (28), while a study in people of European ancestry found a nominal association (P>1×10⁻⁵) between variation in HLA class II genes (HLA-DRB1, HLA-DQB1 and HLA-DPA1) and T2D complications (41).

Similar to the findings with T2D, we found no association between the wGRS for RA and CAD; however, genetic liability for RA was associated with increased risk of CAD only when the MHC region was included in the IVWR analysis. While RA and CAD share several risk alleles (38) that may affect common disease pathways (e.g. NF-kB signaling (42)), genetic pleiotropy between both phenotypes has not been reported in previous studies that excluded MHC genes (11, 39). Since genetic variability in MHC region, particularly in the HLA-DRB1 gene is associated with susceptibility for RA (43) and with CAD in studies that performed direct HLA genotyping (44, 45), it is possible that HLA haplotypes known to affect immune reactions may influence the risk of CAD by affecting presentation of oxidized LDL (46). Nevertheless, discrepancies observed on the weighted median analysis compared to the IVWR and Egger analysis suggest that the genetic association between RA and CAD is not robust and bias needs to be rule out.

We found a positive association between RA and CRP in the IVWR which was driven by variability in the MHC region. Because HLA-DQA1 and HLA-DRB1 play a major role in T cell activation for cytokine production, variation in both genes have been associated with susceptibility for RA (47) and with CRP levels (44, 48). A previous study found no association between genetically-determined variance in CRP and RA (48) using 52 SNPs of which only one was in the MHC region. Although our findings suggest that common susceptibility genes in the MHC region are responsible for the genetic association observed between RA and CRP, this needs to be studied further. In contrast to CRP, we found no genetic association between RA and IL6. All of these associations were replicated in the UK Biobank.

In the IVWR, we found nominal associations between genetic liability for RA and two cardiometabolic phenotype (SBP and CKD) when the MHC region was included. Clinical and animal studies have identified a role for the immune system in blood pressure regulation (49) and there is evidence that SBP shares risk alleles with RA (38). A previous genetic study that included the MHC region did not find genetic overlap between blood pressure and RA (50), and a PheWAS of HLA variants in ~37,000 Caucasians did not find any significant association with any blood pressure-related phenotypes (41). However, a recent PheWAS of 33 HLA genes in ~166,000 Japanese individuals reported a significant association between HLA-B gene and SBP (28). Thus, it is possible that SNPs in the MHC region are responsible for the nominal genetic association observed between RA and SBP through pleiotropic alleles or through the immune cell activation and inflammation that leads to endothelial dysfunction and hypertension (51). As for the nominal inverse association between CKD and genetic liability to RA, genes in the MHC region have been associated with both increased and decreased risk of kidney disorders (52)

Although low LDL-C levels and increased prevalence of central obesity, AF, and ischemic stroke have been associated with RA in epidemiological studies, the genetic association between these disorders and RA has not been well defined. Okada extracted 311 phenotypes from the NHGRI catalog (including LDL-C, obesity, AF, and stroke) to assess region-based pleiotropy for non-HLA RA risk SNPs and none these traits were associated with RA (11). Others have also found no genetic correlation between RA and cardiometabolic phenotypes such as BMI and LDL (53), but a study of 567 patients with RA and 979 controls found that a GRS for RA that excluded the HLA region was associated with decreased LDL-C concentrations (54). Although inflammation is an important component of these cardiometabolic traits and many inflammatory genes have been associated with susceptibility to some of these traits (55, 56), we did not find significant genetic pleiotropy between RA and these phenotypes. An alternative explanation is that it is not a genetic predisposition toward RA, but development of the inflammatory milieu that results from disease which causes the associations with these phenotypes (11,36,37,57).

The study has some limitations. First, because we studied a large cohort of patients of European ancestry from a tertiary care hospital, our findings may not be generalizable to other populations. Second, because PheWAS uses billing codes to assemble phenotypes, the quality of the case-control definition varies across phenotypes and can lead to misclassification (58). However, many of our findings are consistent with what is known about the risk of RA and other related disorders, suggesting the approach was robust to such misclassification. Nevertheless, further replication will be needed to generalize our findings. Third, for the IVWR analyses we extracted summary statistics already available for RA and the selected outcomes, which limited the number of SNPs included in some analyses. For example, while the summary statistics for RA and some of the outcomes were generated using imputed data from GWAS arrays, summary statistics for T1D and MS were generated from non-imputed data from the ImmunoChip, and LDL from non-imputed data from the Metabochip. Thus, we cannot exclude the possibility that the reduced number of SNPs available for some outcomes may have affected our ability to identify shared genetic predisposition. Fourth, although the IVWR analysis suggested the MHC RA risk alleles are associated with increased the risk of T2D and CAD, we cannot exclude the possibility that these risk alleles regulate an unobserved trait that causally influences the risk of these cardiometabolic phenotypes.

Summary conclusions:

In summary, we found that the wGRS for RA was associated with an increased risk of T1D as well as with susceptibility to other autoimmune disorders such as autoimmune thyroiditis and systemic sclerosis. The IVWR meta-analysis showed that genetic predisposition to RA is associated with an increased the risk of T1D, independent of PTPN22. These findings suggest pleiotropy between genetic predisposition for RA and other autoimmune disorders including T1D.