Genetics of rheumatoid arthritis susceptibility, severity, and treatment response

Although it is well recognised that ankylosing spondylitis, gout, and osteoarthritis have existed for several thousand years, the situation is less clear for rheumatoid arthritis (RA). Paleopathological evidence on skeletal remains dating back to the time of pre-Columbian Indians would suggest that RA is also an ancient disease, which might have been affecting the lives of people for at least 4000 years [1]. However, despite the expansion of archaeology and the accumulation of material to examine, the age and origin of RA are still a matter of debate and controversy, as the distinction of RA from other rheumatic diseases can rarely be made solely on the examination of bones [2]. Nonetheless, several indications from medical and non-medical literature and from the visual arts have suggested that RA has existed for many centuries, but it was not until 3 August 1800 that the unequivocal first diagnosis of RA was made by the French physician Landré-Beauvais of Paris [3]. He described in detail the clinical picture of the disease and originally called it “goutte asthénique primitive” (“primary asthenic gout”). His first nine patients were all women; an early suggestion of a genetic component to disease aetiology.

The nineteenth century saw the basic concepts of heredity and evolution being established with the publication of the theories of evolution by the British scientists, Charles Darwin and Alfred Wallace, in 1858 [4] followed by the discovery in 1865 of the laws of heredity by the Austro-German Augustinian Friar Gregor Mendel through his breeding experiments with peas [5]. These publications arose without any knowledge of the existence of DNA, which was only discovered later, in 1869, by a Swiss physician and biologist, Friedrich Miescher [6]. In this historical and scientific context, the observation that patients with RA aggregated occasionally in families supported the idea that the disease was at least partially heritable, even if it did not follow a Mendelian pattern of inheritance.

Research on the mechanisms of blood transfusion incompatibility and studies on skin transplant incompatibility followed by the discovery of the major histocompatibility complex (MHC) in the twentieth century further established the scientific and experimental context for ground-breaking discoveries in the genetics of RA. For example, the practice of mixing blood from two individuals before a transfusion to test their compatibility or the experimental practice of mixing only their lymphocytes to study their tissue compatibility or reactivity led to the development of so-called mixed lymphocyte cultures (MLC); in 1969, Gonzalo Astorga and Ralph Williams observed a reduced reactivity of lymphocytes from patients with RA when incubated together in MLC [7], indicating that the tissues from patients with RA were “more compatible” with each other than any pair of randomly selected healthy individuals—a further indication of a common genetic background between patients with RA. Further development of tissue typing using serological techniques in addition to MLC was used by Peter Stastny, who, between 1976 and 1978, described a strong association between MLC type Dw4 or serological type HLA-DRw4 and RA in white populations from the United States [8‐10]. This association was further found to be present with a limited number of other specific types, namely, DR4 Dw14.1, DR4 Dw14.2, and DR4 Dw15.

In the 1990s, DNA-based techniques superseded serotyping and provided much greater precision. The MLC and serologic markers listed above were found to correspond to different alleles of the HLA-DRB1 gene, each coding for a different amino acid sequence. These alleles were subsequently renamed HLA-DRB1*04:01 (for DR4 Dw4) and HLA-DRB1*04:04, *04:08, and *04:05, for DR4 Dw14.1, DR4 Dw14.2, and DR4 Dw15, respectively.

In animals and humans, the MHC region comprises a set of genes encoding proteins mainly involved in immune function. The human MHC is synonymous with the human leukocyte antigen (HLA) system, which lies on the short arm of chromosome 6 at position 6p21.3. The classical HLA region is highly variable (polymorphic) with, for example, over 1900 alleles for the HLA-DRB1 gene alone. A detailed description of the HLA region is available online through the European Bioinformatics Institute and the International Immunogenetics Project [11]. As the techniques used to type HLA genes have evolved over the years, the nomenclature for the different alleles at individual HLA genes has changed. Originally, immunological typing methods were used, including serotyping and cellular typing with MLC, later to be superseded by various DNA-based typing methods.

The clustering of RA cases within families has been a consistent observation across studies and can be measured either as λs, the sibling relative risk, or λr, the relative risk to first-degree relatives (parents, children, and siblings) [14]. Values for λs or λr have been reported to lie between 2 and 10, consistent with an increased prevalence of RA within the families of affected individuals as compared to the general population [15‐20].

A complementary and more popular approach to measure the genetic contribution to RA susceptibility is to determine the proportion of the variance of the disease which is explained by genetic variations (i.e., the genetic contribution to the disease), also called disease heritability [14]. Several different methodologies have been proposed to estimate heritability of RA, but all have important shortcomings, which have resulted in large variations in estimates.

One method to calculate heritability is based on the comparison between disease discordance in monozygotic and dizygotic twins. The most cited publications using this method estimate a disease heritability of around 60% in the UK and Finnish populations [21, 22], whilst a Danish study published in 2002 found a heritability of 0% [23], revised to 12% in a larger sample in a follow-up study in 2012 [24]. Interestingly, one study calculated heritability separately for the two main subsets of RA (anti-citrullinated protein antibodies (ACPA) positive and negative disease) and found no difference in heritability between the two subsets (68% and 66%, respectively). A main limitation of twin studies is sample size, as the number of twin pairs available in these studies was modest.

A second method consists of estimating heritability from familial aggregation using large population registers; this method was applied in the Swedish total population and concluded that heritability of ACPA-positive RA is around 50%, but that heritability is only 20% for ACPA-negative RA [25]. This study is likely to represent the most reliable heritability estimates to date given the sample size tested.

A third and increasingly popular method to calculate heritability requires genome-wide genetic profiles from large numbers of unrelated individuals [26]; one technique, called genome-wide complex trait analysis (GCTA), has been successfully applied to several complex traits [27], but results from RA studies have produced highly varying estimates; for example, heritability was estimated to be 52% in one study [28] but 0% for ACPA-negative RA and 19% for ACPA-positive RA in another [29].

Heritability can also be calculated for RA severity or response to treatment in order to determine the quantitative role of genetic factors on disease course or response to specific interventions. By contrast to the calculation of heritability for disease susceptibility, where healthy individuals have to be incorporated in the study design as controls, heritability calculations are performed in RA cases only (for example, comparing globally the genetics of patients with erosive disease with that of patients without erosions). Using such an approach, one study estimated the heritability of erosive disease in the Icelandic population to be 50% [30]. Three studies have attempted to calculate the heritability of response to anti-TNF treatment in RA in different populations: [31] found a heritability of 45% for the post-treatment reduction in the 28 joint Disease Activity Score (DAS28) and 60% for the change in Swollen Joint Count (SJC); Umićević et al. [32] reported a heritability of 71% for the change in DAS28 and 87% for change in SJC; however, Sieberts SK et al. [33] found only an 18% heritability for non-response, but no genetic contribution to the prediction of response. These results should be interpreted with caution, not only due to the limitations inherent to the calculation of heritability but also because there is no clear definition of “severe” disease or “response” to treatment.

In summary, large variations in the calculation of heritability are problematic as researchers cannot know how many more genetic factors for RA susceptibility, severity, or treatment response are yet to be identified, and therefore how much effort and money to further invest in susceptibility, severity, or treatment response gene identification.

The genetics of RA susceptibility has already been covered in detail in many reviews [34‐40] and the most comprehensive meta-analysis of all available GWAS datasets has now identified over 100 loci associated with RA susceptibility [41]; rather than presenting an exhaustive listing of genetic loci associated with RA susceptibility, we will summarise the key findings necessary to understand the translational potential of genetics and the future directions of research.

The identification of genetic markers of RA outcome is a much more complicated task than the identification of susceptibility markers, as several methodological challenges have to be overcome. First, the definition of disease severity is not standardised; second, disease outcome varies over time; third, the sample sizes in prospective cohorts of patients with good quality longitudinal data on disease outcome are modest and, finally, statistical modelling is complex, as several outcome variables are continuous, non-normally distributed and affected by time-varying confounders, including treatment.

Despite these challenges, multiple studies have identified HLA-DRB1 alleles as markers of radiological damage in RA [80]; for example, a well-powered study has recently shown that the risk hierarchy defined by the 16 HLA-DRB1 susceptibility haplotypes (i.e., defined by positions 11/13, 71, and 74, See Table 1) was correlated between disease susceptibility, erosive damage, and mortality: thus, the major genetic markers of disease susceptibility in the HLA-DRB1 gene are also markers of severity [45]. Valine at position 11 is the strongest genetic predictor for the development of erosions, radiographic damage, mortality, and poor outcome in general, including non-radiographic measures of disease activity/outcome [45, 81]. A serine at the same position is protective against radiographic damage and poor outcome [45, 81]. One interesting observation was that effect sizes for disease outcome were systematically smaller than those observed for susceptibility (Table 1). Classification of patients with RA into different prognostic categories could be performed using HLA-DRB1 susceptibility markers, but the proportion of the variance of radiographic damage explained by HLA markers remains too low to be clinically useful.

Although most of the effect of HLA-DRB1 on disease severity is mediated by ACPA, there is some evidence emerging that HLA-DRB1 amino acids may regulate the level of laboratory inflammation (as measured by CRP) and clinical inflammation (disease activity score at 28 joints (DAS28) or Swollen Joint Count) through different biological pathways, some of which are likely to be independent of ACPA [45, 81] (Fig. 2).

Few GWAS have been performed for RA severity [82, 83], and most studies reporting genetic associations with radiographic outcome are candidate gene association studies (Table 2). Very few studies have reported associations below genome-wide significance for loci located outside the HLA region [83], and the replication rate of these associations in independent datasets has been very low [83]. Apart from HLA loci and SNPs located near TRAF1, which have been consistently associated with radiographic damage and replicated in several independent cohorts by independent research groups across different populations, no other genetic locus can be regarded as a confirmed association with radiographic outcome in RA, as replications by independent groups of researchers in large cohorts are lacking. Interestingly, a few RA severity SNPs have been followed up in functional studies to increase confidence that they are true positive associations and to understand the pathogenesis of severity: a SNP (rs12212067: T>G) in the FOXO3A gene region, which is not associated with RA susceptibility, has been reported to be associated with disease course in several TNF-mediated conditions, including RA [94]. The mechanism of action involves a reduction in the production of pro-inflammatory cytokines, including TNFα, by monocytes. The differential response of monocytes in RA patients dependent on the carriage of the minor allele at FOXO3A is likely to be seen only in an inflammatory context, therefore explaining the lack of association with disease susceptibility.

The identification of the association of rs26232, a SNP located in the first intron of the chromosome 5 open reading frame 30 (C5orf30), with both RA susceptibility and severity, has prompted the functional characterisation of C5orf30, a gene with previously unknown functions [98]. C5orf30 was found to be expressed at high levels in the synovium of patients with RA compared with control synovial tissue. C5orf30 decreases the migration of synovial fibroblasts, so that loss of function increases joint inflammation and tissue damage [98].

The two examples of FOXO3A and C5orf30 illustrate the potential of genetic studies in the identification of important pathogenetic mechanisms of RA susceptibility and severity. Although our current understanding of the genetics of RA severity is much more limited than our understanding of the genetics of RA susceptibility, the emerging picture seems to be that there is only a partial overlap between genetic markers of susceptibility and severity. Moreover, it is also likely that genetic markers of severity are different between ACPA⁺ and ACPA⁻ RA [83, 116].

Identifying biomarkers to target the right treatments to the right patients would bring immediate patient benefit in RA because a number of treatment options exist and some patients will do well with each option whilst, in others, disease activity will remain uncontrolled leading to impaired quality of life for the patient and increasing the risk of long-term disability or will result in adverse effects in some patients [121, 122]. In addition, some of the treatment options are expensive thereby impacting on healthcare resources; for example, biologic drugs targeting inflammatory pathways cost between £5 and 10K per year per patient. If a stable biomarker could be used to select the best treatment option for individuals or groups of patients with RA, it would have the potential to improve health and costs of healthcare, simultaneously. Hence, RA is an ideal condition in which to apply such precision medicine approaches. Genetic biomarkers are stable and are easily assayed. In other disease areas, genetic biomarkers are being used to inform treatment selection decisions; for example, EGFR gene mutation screening in non-small cell lung cancer is undertaken to determine suitability for oral tyrosine kinase inhibitors (reviewed in Shea et al. [123]). However, no genetic variants have yet been robustly and consistently associated with response to therapies used in RA. Methotrexate is the most common first-line disease modifying drug choice, but results from candidate gene association studies have been conflicting and few GWAS have been undertaken to date [124, 125]. Similarly, biologic drugs targeting the TNF pathway (THF inhibitor (TNFi) drugs) are the most common first choice biologic treatment, but results of genetic association studies have often been conflicting. For example, an association of PDE3A-SLCO1C1 at genome-wide significance levels with TNFi response has been reported but not replicated [126, 127], whilst association of the PTPRC gene polymorphism has been associated with TNFi response in some [128‐130] but not all studies [131, 132].

The lack of success in identifying treatment response biomarkers is disappointing, but not unexpected, given that the studies face many of the same challenges as for disease severity studies: First, the outcome measure is a composite of both objective and subjective measures making reliable, consistent, and standardised measurement difficult. Furthermore, many of the subcomponents of the outcome scores are based on clinical assessments, which may not be truly reflective of response in terms of synovial inflammation. For example, the DAS28 score comprises a clinical assessment of 28 joints for swelling and tenderness, a serological marker of inflammation (ESR or CRP), and a score of global well-being provided by the patient; changes in the DAS28 score before and after treatment are used to assess treatment response. Third, disease outcome varies over time; fourth, the power of such studies is limited by sample size. To illustrate the latter, the most comprehensive study of RA susceptibility loci involved analysis of samples from over 100,000 individuals whereas the largest analysis of TNFi response was based on ~2700 patients [41, 133]. Finally, other factors, such as whether the patients actually take the drug prescribed (adherence) or whether antibodies to the drug develop have very important influences on response, but are not yet accounted for in studies [134‐136].

Given that genetic variants will act on specific biologic pathways, it is likely that genetic studies may be better correlated with changes in synovial inflammation, as that is the target of such treatments; however, synovial inflammation is poorly correlated with DAS28. Therefore, re-weighting of current measures or, better, new biological outcome measures are required that better reflect the synovial inflammatory response in order to better classify responders and identify factors that predict response pre-treatment. For example, giving higher weightings to the components of the DAS28 score that correlate better with synovitis (the swollen joint count and serological inflammatory markers) has been proposed [137], whilst others have used DCE-MRI scans to accurately quantify synovitis to determine treatment response [138].

Clinical prediction models incorporating genetic susceptibility loci to identify healthy individuals at high risk of disease have shown a very modest prediction performance and are insufficiently accurate for general population screening [139]. Also, genetic markers are not recommended for diagnosis. The clinical utility of a genetic stratification system for precision medicine based on HLA haplotypes correlated with disease course or outcome [45] remains to be evaluated, but its performance is likely to be equivalent or inferior to ACPA status, thus insufficient to guide clinical decisions, as the association of HLA with severe disease acts mainly through the presence of ACPA. Currently, the testing of patients for HLA-DRB1 is therefore only performed as a research tool. However, further methodological developments and the identification of an increasing number of susceptibility/severity/treatment response SNPs are ongoing. Together with the identification of other types of biomarkers (epigenetic, immunological, cellular, serological …), genetic markers might allow in the future the definition of combined genetic, demographic, laboratory, and clinical risk scores to accurately classify patients at diagnosis into different prognostic or treatment response categories for precision medicine.

GWAS have been extremely successful in the identification of a large number of genetic susceptibility polymorphisms associated with RA. Although the effect sizes of SNPs outside the HLA are modest, genetics has shed new light on pathogenetic mechanisms of disease susceptibility and has been hypothesis generating. Functional genomics approaches are now taking over from genetic association studies to identify the mechanisms of actions of susceptibility polymorphisms. The identification of genetic markers of disease outcome and response to treatment is still at its infancy, but bears the potential to contribute to the development of a precision medicine approach in the management of RA in the next 10 years.