Genetics of rheumatoid arthritis

As with any complex disease, RA represents a collection of symptoms, and the diagnosis is consensus based [9]. This may generate some uncertainty at initial stages of the disease and may cause misclassification due to overlap of symptoms between different rheumatic diseases. On the other hand, the study of only well-established chronic disease is likely to be biased towards individuals with more severe RA and to non-responders to treatment. The availability of specific biomarkers for early and accurate diagnosis significantly improves the quality of phenotyping for genetic research.

Rheumatoid arthritis is a prototype autoimmune disease that, together with other multiple impairments of the immune system, is characterized by development of different types of autoantibodies. Rheumatoid factor (RF) and anti-citrullinated peptide/protein antibodies (ACPA, usually measured by anti-CCP ELISA) are most commonly detected [2] and due to high predictive scores are a part of the classification criteria for RA [9]. Other types of autoantibodies to modified proteins are detected mostly in high correlation to RF and ACPAs [14‐16]. In systemic screening of human population biobanks for these autoantibodies in different populations, it was found that they may be detected in blood long before the first symptoms of RA [17, 18]. Later, with development of symptoms, the concentration of ACPAs increases significantly and in the majority of future RA patients, ACPAs are present at least a few months before the onset of symptoms [19]. In an attempt to dissect the importance of different ACPA reactivities, several studies have been performed on a peptide array with broad selection of citrullinated and non-citrullinated peptides [20]. There is a great degree of cross-reactivity between different types of autoantibodies, although some of them cross-correlate less strongly. It became an accepted practice in rheumatology research to consider autoantibody status as an important covariate in predictive models. However, it is also true that there is only a moderate difference between ACPA-positive and ACPA-negative RA in important clinical features such as disease activity scores at baseline, age of onset, and sex distribution. The greatest difference is probably in the risk of bone erosion [15]. Regarding ACPA-positive RA, it is also characteristic that clinical phenotypes are very similar in subgroups of patients with significantly different concentrations of autoantibodies [21]. There have also been several attempts to correlate the presence of different autoantibodies to clinical phenotypes. Discrimination between ACP-positive and ACPA-negative RA was especially efficient in genetic studies. It has become evident in recent years that there is only a moderate overlap between genetic risk factors between these two subgroups of RA.

Misclassification due to similar clinical phenotypes is not uncommon in rheumatology research and may cause patients with other rheumatic diseases to be included in a study group. This is more common for study designs that lack accurate information about autoantibody status and, in general, is more common for ACPA-negative RA when this autoantibody status information is available. In these cases, the diagnosis may sometimes convert, with time, to spondyloarthritis, which is known to have a different genetic association pattern [22]. On the other hand, multiple pleiotropic effects have been detected between rheumatic diseases [23, 24].

Discoveries of genetic associations in RA arose in the past through investigation of large cohorts with relatively poor sub-phenotyping of RA. Better clinical characterization of RA subgroups should now be pursued.

A genetic association between RA and variations at HLA was discovered early. With the growing understanding of the architecture of the HLA locus, in particular HLA-DRB1, it became evident that this association relates to a group of so named shared epitope (SE) alleles and is strongest for ACPA-positive RA [25]. The SE hypothesis was introduced in an attempt to find common features between different alleles or haplotypes within HLA-DRB1. Initially, a homologous amino acid sequence at positions 69–74 of the beta chain coded by the HLA-DRB1 gene was identified as the protein structure responsible for association with RA. By following this hypothesis, several DRB1 haplotypes were combined in a single SE allele group in multiple studies and compared with a reference group of other haplotypes. HLA haplotypes from most of the alleles of DRB1*01, *04, and *10 groups represent SE alleles, with *14:02 later suggested as an important contributor for American populations. In contrast, HLA-DRB1*13 alleles were found to bring strong protection against RA [26]. With further sub-phenotyping of RA using autoantibodies, first RF and later with ACPA, it was found that the strongest associations are mainly with autoantibody-positive RA [27, 28].

Several associations outside the HLA locus were detected before the GWAS era, including genetic variants at PTPN22, CTLA4, and PADI4 genes [43‐45]. With expanding data from GWAS from different populations, it became evident that the effect size of non-HLA associations is far lower in comparison to the major associations with HLA-DRB1 SE alleles. Development of affordable genotyping methods, together with the use of transethnic meta-analysis, expanded this list to 151 loci and coverage of all human chromosomes except chromosome Y (Table 1, Fig. 1) [5‐8, 34, 43, 46‐56]. Not surprisingly, most associated hits (64% from the current list) could be annotated to genes with a known function in the immune system. The annotation to immune system is based on previous knowledge about specific function of the gene or a pathway in cells related to innate or adaptive immunity, or data on expression of this gene in immune cells. Many of these genes belong to immune system-specific or more ubiquitous cell signaling pathways. The involvement of these genes in cell function for non-immune cells, therefore, potentially extends the importance of genetic associations with RA to the function of cell types, tissues, and organs outside the immune system. One should keep in mind, however, that these annotations are usually based on the physical position of SNP within the gene location and that how gene function is linked to associated SNPs is a matter for future intensive research. Possible approaches will be discussed further in other articles in this issue.

As is the case with the HLA locus, genetic association studies that focus on autoantibody status reveal remarkable differences between ACPA-positive and ACPA-negative arthritis. However, previous investigations were most often performed in RA patients without known autoantibody status and the differences remain non-confirmed by direct testing in the subgroups of RA. According to follow-up analysis in different cohorts all over the world, it is likely that ACPA-positive RA is more prevalent and therefore most of the association hits in studies without discrimination by serology represent associations with autoantibody-positive RA. There have been several attempts to perform GWAS separately for these two groups that often end up with replication of hits for ACPA-positive RA and only few association signals for ACPA-negative RA (Table 1).

The number of known genetic associations with RA is ever growing due to an increase in the number of observations and the resulting increase of statistical power. To expand this further, the complementary approach is in the investigation of population groups to capture differences due to ancestral, most often ethnic, background. In a similar way to the situation for HLA genetics until recently, studies of Europeans have been dominant, and it is important to extend investigation of the scope of human polymorphisms specific for different populations in association studies. These studies are in progress for major Asian and African populations and will help to map true genetic associations for RA.

From the available data in European and East Asian populations, there are several genetic associations that show population differences. It was found that the second-best association with RA in the PTPN22 gene with exonic SNP rs2476601 is not detectable in East Asian populations due to very low frequency of the polymorphism in these populations [50]. Similarly, several other SNPs associated with RA in European populations close to the IL20RB, CUL5, TYK2, and ILF3 loci are not polymorphic or have very low allelic frequency in East Asian populations [6, 7]. In contrast, association with RA for the SNPs rs12026490 close to SLAMF6 was not reproduced in Europeans due to a < 1% frequency of the risk allele [7]. One may expect more population-specific association hits to be discovered in the future. Currently, 18% of known associations in European populations are not confirmed for Asian populations, while only 16% independently reach convincing genome-wide significance in both populations. On the other hand, comparing the effect size for the majority of associated SNPs from trans-ethnic meta-analyses reveals no differences, not only in the presence of significant associations in two different population, but also when one of the groups did not reach genome-wide significance for association [6, 7].

As mentioned previously, the data regarding non-HLA genetic associations with seronegative RA is less consistent and robust. This subgroup is usually a smaller fraction of large GWAS of RA and until recently, only three confirmed associations have demonstrated genome-wide significance (Table 1), close to loci ANKRD55, IRF4, and LINC01898 [34, 46, 55]. At least some of these associations match to hits for association with ACPA-positive RA and, therefore, may also occur due to misclassification within the seronegative group, as discussed above. Most recent trans-ancestry meta-analysis did not confirm genome-wide significant association for seronegative RA [8]. Therefore, GWAS of ACPA-negative RA will be an emerging area of research, although it requires an increase in the size of studies and better phenotypic data for this RA subpopulation.

Common genetic risk factors for different autoimmune diseases were found previously and investigated specifically with Immunochip genotyping [62]. Subsequently, several studies confirmed pleiotropic effects from these genes and by cross-investigation of autoimmune diseases discovered possible new associations for RA [23, 24].

Further integration of genetic association data is required, together with available omics data. Our PPI analysis using inBio Discover (INTOMICS, Denmark) of the gene list from Table 1 generated at least two comprehensive networks with only direct PPI and demonstrated possible involvement in the generation of RA risk for several signaling pathways in immune cells (Figs. 2 and 3).

More analyses in serologically stratified RA cohorts and in populations with different ancestries are required to elucidate genetic risk of RA using data from genetic association studies.

An investigation of genetic risk factors for common complex diseases should consider the phenomena of gene-environment and gene–gene interaction. There are several statistical approaches for detecting interactions and when discussing interaction, it is important to clarify which definition of interaction is being used. Most often, interaction is part of a statistical model that tests for multiplicativity of contribution for selected parameters in the development of disease risk. The null hypothesis is that the coefficient for interaction mode in the regression model is null. In terms of odds ratio to develop disease, this transfers to the expectation that in a group of individuals with a combination of two parameters, the odds ratio will represent the product of odds for separate parameters. Deviation from the null hypothesis will be the indication for “multiplicative interaction.” In epidemiological studies following the Rothman hypothesis [63], the null hypothesis based on expectation for additivity of odds ratios for a group of individuals with a combination of two parameters was compared to groups with only one of these risk factors. Therefore, “additive interaction” arises in the case of significant deviation from the sum of contribution of each risk factor.

In previous studies, considerable interaction between cigarette smoking and SE alleles was detected in generation of risk, first for RF positive [10], and later for ACPA-positive RA [64]. Further analyses of other environmental factors for interaction with these alleles were performed and it was confirmed for independent Asian, African American, and European populations [65‐71]. Several other studies of environmental effects, including exposure to textile dust and alcohol consumption, also found interaction with SE alleles [72, 73]. Additionally, two genetic loci, genes GSTT1 and HMOX1, at chromosome 22 were identified to increase risk of RA due to interaction with smoking. These genes are functionally related to protection from oxidative stress [74].

Several studies of gene–gene interaction in the development of risk of ACPA-positive RA were performed previously, with PTPN22 risk allele and variation close to MAP2K4 gene found to interact with SE alleles [75, 76]. In another study, gene–gene interaction between variations in the BANK1 and BLK genes from chromosomes 4 and 8, respectively, was shown to increase risk of RA [77]. Based on our dominion hypothesis [72], we performed a study showing the global effect of RA risk alleles in interaction with SE alleles on the development of additional risk for ACPA-positive RA [11]. Although this study was initially criticized for statistical biases [78], careful statistical modeling has confirmed that it does indeed describe true effects and could be applied to study gene–gene interactions in other binary phenotypes with a known dominant risk factor [79]. Therefore, the conclusion of this study is that significant interaction between the HLA-DRB1 SE alleles and the group of SNPs associated with ACPA-positive RA is a global feature that increases risk of ACPA-positive RA. It was also shown in this study that gradual decrease of the number of interacting risk alleles in a group of individuals with HLA-DRB1 SE alleles will decrease the risk for RA. No gene–gene or gene-environment interactions in development of the risk of ACPA-negative arthritis have so far been found. Further analysis of statistical interactions may help to integrate functionally relevant associations involved in the development of RA.

Development of tools for fast analysis of a massive number of methylation sites, like Illumina 450 and Illumina EPIC arrays, opened the possibility for genome-wide epigenetic study (GWES). To date, the largest GWES of RA was performed on whole blood genomic DNA of 354 ACPA-positive RA and 357 healthy controls [80]. Epigenetic changes at disease baseline are massive and obviously non-specific, reflecting basic inflammatory reactions. In an attempt to identify differentially methylated sites (DMS) that are more specifically related to RA development, genetic association signals for RA were integrated with data for DMS in genomic DNA from whole blood in ACPA-positive RA. This approach helped to focus only on a few DMS that are regulated by SNPs with association to the disease. Interestingly, almost all these DMS, 9 out of 10, were located within the HLA locus. Additionally, following known association of seropositive RA with cigarette smoking, one of these DMS, close to the HLA locus, was attributed as a major contributor to the risk of ACPA-positive RA due to gene-environment interaction [81]. However, DNA methylation profile reflects contribution of the multiple environmental factors and of the ongoing inflammation in RA patients and it is difficult to identify RA-specific casual relations between these factors and DMS changes. It was later shown that methylome of peripheral mononuclear cells can be used to anticipate the evolution of undifferentiated arthritis to RA, and these modifications could be annotated to a number of inflammatory pathways and transcription factors [82]. The significance of ACPAs during RA development was analyzed in serologically distinct RA subgroups in a study of twins discordant for ACPA status [83]. However, only marginal differences in the methylation profile of whole blood DNA of twins were found when data was corrected for cell composition; this highlights the importance of blood cell profiling in epigenetic studies of RA development. With the goal of determining RA-specific methylation events, several research groups conducted epigenetic studies with profiling of specific cell subpopulations. These studies demonstrated the following: significant changes in methylome of CD4 + memory cells compared to CD4 + naïve cells [84]; specific methylation-induced regulation of CTLA4 promoter [85] and FOXP3 enhancer in Tregs [86]; evidence for epigenetic regulation of inflammatory cytokines in monocytes [87]; and multiple immune-related pathways in fibroblast-like synoviocytes (latterly, in comparison with osteoarthritis) [88] in RA. Interestingly, a common pattern of epigenetic changes was found in a regulation of interferon-related genes for different autoimmune diseases, including RA, with hypomethylation of these genes in CD4 + cells being highly predictive to disease development [89].

While epigenetic studies most commonly focus on the available pool of peripheral blood cells, significant diversity was shown for the tissue from anatomically different joints. Synovial fibroblasts of different anatomical origins represent different transcriptomes that translate into unique joint-specific phenotypes of these cells, which is likely to influence methylation profile in RA with different localization [90].

Coming studies of methylome in RA should consider the important differences observed in genetic studies between the RA subgroups. It is desirable to build the experimental design either on stratification of seropositive and seronegative RA, or to adjust the differential methylation analysis for serological status. The major challenge for the analysis is in a study design in which healthy controls are not necessarily the optimal reference group. Epigenetic studies should be integrated with genetics, transcriptomics, and proteomics to reveal a true pattern of pathway modulation in the development of RA.