Objectives Psoriatic arthritis (PsA) susceptibility is associated with several different class I alleles, suggesting separate patterns of MHC effect. This exploratory study was based on the hypothesis that heterogeneity of the clinical phenotype of PsA might be explained by differing associations of clinical features with these susceptibility genes.
Methods The clinical phenotype of 282 PsA patients in a cohort previously studied for associations with human leukocyte antigen (HLA)-B and HLA-C genotypes was first preliminarily assessed by univariate associations of susceptibility genes with specific clinical characteristics. To explore the potential genotypic effects of pairwise combinations of different HLA-B and C alleles/haplotypes, we created a series of allele/haplotype risk scores combining single alleles/haplotypes separately associated with being in the highest PsA severity propensity tertile based on the features studied by univariate analysis.
Results In exploratory univariate analyses, B*27:05:02 was positively associated with enthesitis, dactylitis and symmetric sacroiliitis, whereas B*08:01:01-C*07:01:01and its component alleles were positively associated with joint fusion and deformities, asymmetrical sacroiliitis, and dactylitis. HLA-C*06:02:01 was negatively associated with asymmetrical sacroiliitis. The highest propensity score for severe PsA was with B*27:05:02-C*02:02:02, B*08:01:01-C*07:01:01 and B*37:01:01-C*06:02:01, but not the B*27:05:02-C*01:01:01 or B*57:01:01-C*06:02:01 haplotypes. In contrast, B*44 haplotypes were associated with presence of milder disease, and in univariate analysis with a decreased frequency of enthesitis, joint fusion, deformities and dactylitis.
Conclusions Different HLA susceptibility genes were associated with particular features that defined the PsA phenotype of a given patient. Additive interactions between different susceptibility HLA alleles defined the propensity for a more severe or milder musculoskeletal phenotype.
- Psoriatic Arthritis
- Gene Polymorphism
Statistics from Altmetric.com
Using improved psoriatic arthritis (PsA) case ascertainment based on ClASsification criteria for Psoriatic ARthritis (CASPAR) classification criteria and more precise methods of allele determination, susceptibility to PsA has recently been strongly associated with inheritance of several class I major histocompatibility complex (MHC) alleles of both human leukocyte antigen (HLA)-B and HLA-C loci and their haplotypes.1 ,2 The number of these alleles, and that they differed in frequency between PsA cases and those with isolated cutaneous-only psoriasis (PsO) provided strong evidence of genetic heterogeneity for the general psoriasis phenotype, and for PsA in particular, as well as resolving several important points of divergence in the literature.3 The genes associated with increased PsA susceptibility were primarily present in five classic ancestral haplotypes, for example, B*27:05: 01-C*02:02:02 and B*27:05:02-C*01:02:01, B*08: 01:01-C*07:01:01, B*37:0:01-C*06:02:01 and B*57:01:01-C*06:02:01. B*44:02:01-C*05:01:01 and B*44:03-C*16:01:01 were associated with significantly decreased susceptibility.
We also presented evidence that the interval between PsO onset and development of PsA is a quantitative phenotypic trait determined by HLA type.1 Precedent for an effect of susceptibility alleles on phenotype has been present since the initial study by Brewerton et al,4 where HLA-B*27 was associated with the presence of axial disease in PsA. However, in subsequent studies this association has been inconsistent.5 ,6–9 Analogously, HLA-C*06:02 was reported to be associated with fewer involved or damaged joints.10
PsA is heterogeneous in presentation, with considerable between-patient variability in the number of phenotypic features present. In the present study, we sought to explore whether certain phenotypic features were associated with particular genes associated with PsA susceptibility.
We clinically phenotyped 282 patients of the previously reported study,1 with the assessors blinded to the previously reported HLA typing results. All patients belonged to the relatively homogeneous Irish population and had Irish parents. Their ethnic homogeneity and lack of cryptic relatedness were confirmed by single-nucleotide polymorphism (SNP) analysis in 260 of the cases (J. Bowes, University of Manchester, personal communication). All were previously genotyped for HLA-B and HLA-C alleles to at least four digits using sequence-based HLA typing as described.1 All fulfilled CASPAR criteria.11 The cohort consisted of subjects referred to a rheumatology department to confirm musculoskeletal disease.1 The cohort follow-up was a median of 19 years from the initial symptoms of PsA.
The study was approved by the institutional ethics committee, and following written informed consent, patients underwent a detailed assessment, including recording the number of tender and swollen joints using the 68 tender/66 swollen joint counts, ascertaining dactylitis, enthesitis and the number of clinically deformed joints. Deformed joints were defined as the presence of fixed deformities, flail joints, fused joints and surgically replaced joints.12 ,13 Moreover, usage of non-biological disease-modifying antirheumatic medications (DMARDs) and tumour necrosis factor alpha (TNFα) inhibitors (TNFi) was documented for either skin or PsA indications. We used radiographic evidence of sacroiliitis (SI) to define axial disease, because this is less subjective and more reproducible. The criterion for identifying SI was presence of ≥grade 2 radiographic changes (unilateral or bilateral). Asymmetrical SI was diagnosed when the grades were different between the two sacroiliac joints. Hand and foot radiographs were assessed for the presence/absence of peripheral joint erosions or osteolysis. All radiographs were scored by a consultant musculoskeletal radiologist blinded to the HLA results.
We first explored univariate associations of HLA-B and C alleles and haplotypes, where there was >10 occurrences in the cohort previously associated with susceptibility, with clinical characteristics using the asymmetrical two-tailed Pearson χ2 method. These exploratory analyses were used to inform the subsequent development of allele/haplotype risk scores. While susceptible to spurious associations due to multiple comparisons, we eschewed adjusting the p value threshold to avoid rejecting potentially informative allele/haplotype contributors to the development of the allele scores, according to the rationale of Rothman14 We then ranked PsA patients according to the number of PsA features reflecting more severe disease (dactylitis, enthesitis, type of sacroiliitis, joint deformity, joint fusion, erosion, osteolysis) using a propensity score model.15 Variance inflation factors were calculated for the components of the propensity model and did not reveal significant collinearity. The PsA patient with the smallest aggregate of these features received the lowest score, whereas the patient with highest aggregate received the highest score (referred to as the severity propensity score). We then explored the differences in the distribution of all the clinical characteristics according to the severity propensity score tertile using t tests for normally distributed continuous variables, the Kruskal–Wallis test for non-normally distributed continuous variables, and the χ2 goodness-of-fit or Fisher's exact test for categorical variables.
To explore the potential genotypic effects of pairwise combinations of different HLA-B and C alleles or their haplotypes on the propensity for PsA features, we created a series of allele/haplotype risk scores combining single alleles/haplotypes separately associated with being in the highest PsA severity propensity tertile. For the first allelic risk score (alleles only, unweighted), we included all of the HLA-B and C alleles associated with the highest severity tertile at the p<0.20 level from the univariate modelling. In an individual each allele that was positively associated with the outcome contributed a score of +1 and each allele negatively associated contributed −1, creating a range of scores between −4 and +4 for each individual. The association of the risk score with PsA severity was explored using ordinary logistical regression, adjusting for potential confounders: age, smoking, duration of PsO and PsA, time between the onset of PsO and PsA diagnosis and TNFi use. Receiver operator characteristics (ROC) were calculated for the crude and adjusted models, including the effect on the area under the receiver operator curve (AUC) of excluding the risk score from the multivariable model.
Additional risk scoring scenarios were explored and their performance characteristics tested. Constructing a second allele risk score (alleles only, weighted and reduced), we excluded non-contributory alleles from the logistical model using the likelihood ratio test for nested models and weighted the scores based on the magnitude of association for each allele (ie, the β coefficient from the logistical model for each allele was used as its weighting factor in the risk score). Crude and adjusted logistical regression models were constructed to explore the association of the risk score with PsA severity propensity, and ROCs calculated as described for risk score 1. Difference in the performance of risk score 2 versus risk score 1 was assessed by comparing the AUCs for each in the context of their fully adjusted logistical models. In a similar fashion, we constructed and tested two additional risk scores. For the third (alleles+haplotypes, unweighted), each allele or haplotype contributing to the multivariable logistical regression model (ie, not excluded based on the Likelihood ratio test) was assigned a weight of −1 or +1 according to its association with the outcome. In a fourth risk score (alleles+haplotypes, weighted), each allele or haplotype received a weight based on its magnitude of association with the outcome similar to risk score 2. Intercooled STATA12 and SPSS version 12 were used for analyses.
Exploratory univariate associations of HLA-B and C alleles and their haplotypes with PsA features
Sacroiliitis was present in 25% of the patients, with 18% having asymmetrical SI and 6.4% symmetrical, whereas all had history of backache. In patients with symmetrical SI, table 1 shows the frequency of B*27:05:02 is significantly increased to 61.1% versus 12.9% in those without the trait (OR=10.6). Both the B*27:05:02-C*02:02:02 haplotype (26.3% vs 5.3%, OR=6.3) and the B*27:05:02-C*01:02:01 haplotype (31.6% vs 7.2%, OR=6.0) were comparably and significantly increased. A total of 24.4% of those with HLA B*27:05:02 had symmetrical SI. Male sex strongly influenced the development of symmetrical SI in B*27 patients (OR=11.91, 2.33 to 60.83) versus asymmetrical SI, data not shown. The B*08:01:01-C*07:01:01 haplotype (15.8% vs 37.5%, OR=0.31) and its component alleles were not increased in frequency in symmetrical SI (see online supplementary table S1).
In contrast, in asymmetrical SI, B*27:05:02 or either of its haplotypes were not significantly increased in frequency (see online supplementary table S1), whereas the frequency of the haplotype B*08:01:01-C*07:01:01 in asymmetrical SI was increased to 58.4% versus 31% (OR=3.1), along with its constituent alleles (table 1). Among those with the B*08:01:01-C*07:01:01 haplotype, 29.4% had asymmetrical SI, whereas only 2.4% had symmetrical SI. In asymmetrical SI, the frequency of C*06:02:01 was significantly decreased (17.6%, OR=0.45), as was the B*57:01:01-C*06:02:01 haplotype.
Enthesitis was positively associated with the presence of B*27:05:02 (28.1%, vs 9.7%, OR= 3.7; table 1). However, the association was only significant in patients with the B*27:05: 02-C*01:02:01 haplotype (17.5% vs 4.3%, OR=4.7), but not in those with the B*27:05:02-C*02:02:02 haplotype (see online supplementary table S1). A significant decrease in the likelihood of enthesitis was conferred by the B*44:03:01-C*16:01:01 haplotype (table 1). This and subsequent instances of a potentially protective effect were confirmed by conditional association on major positively associated alleles, and no significant interaction with them was found, data not shown. Neither B*08:01:01, C*06:02:01 nor other alleles or haplotypes exhibited a significant association with enthesitis (see online supplementary table S1).
Dactylitis was primarily associated with B*27:05:02, being found in 71.1% of those with the allele, and only significantly with the B*27:05:02-C*01:02:01 haplotype that was also associated with enthesitis (table 1). The B*27:05:02-C*02:02:02 haplotype was not significantly associated with dactylitis (see online supplementary table S1). However, dactylitis differed from each of the other phenotypical traits in that it was also positively associated with the haplotype B*08:01:01-C*07:01:01 and with its constituent alleles, which were found in 61.8% of those with dactylitis. Dactylitis was negatively associated with B*44:02:01:01-C*05:01:01:01 and B*44:03:01-C*16:01:01 haplotypes and most of their component alleles.
Osteolysis was positively associated with C*02:02:02, a component of the B*27:05:02-C*02:02:02 haplotype (table 1), but a significant association with the B*27:05:02-C*02:02:02 haplotype B*27:05:02 itself or the B*27:05:02-C*01:02:01 haplotype was not seen nor was there a negative association of B*44-containing haplotypes (see online supplementary table S1). Reciprocally, erosions were associated with B*27:05: 02-C*01:02:01, but not C*02:02:02.
In marked contrast, joint deformity was not associated with B*27:05:02 or its haplotypes (see online supplementary table S1), but rather was present in 75.5% of those with B*08: 01:01-C*07:01:01 (42.1% vs 25%, OR=2.3) and its constituent alleles, similar to the association of asymmetrical SI. A significant decrease in the likelihood of joint deformity was conferred by the B*44:03:01-C*16:01:01 haplotype (3.3% vs 12%, OR=0.24) and its component alleles (table 1). Joint fusion was also positively associated with B*08:01:01-C*07:01:01 and its component alleles and negatively associated with B*44:02:01-C*05:01:01 and B*44:03:01-C*16:01:01 haplotypes and most of their component alleles.
Development of a propensity score for severity of the PsA phenotype
Table 2 summarises the cohort characteristics. A propensity score model for 8 PsA features (table 1) was used to make a continuous variable score to order the cohort and delineate tertiles of mild, moderate or more severe disease. As anticipated, characteristics that formed the basis of the score, with the interesting exception of enthesitis, were significantly more prevalent in the group with a severity propensity score in the third tertile compared with those in the first tertile (table 2). Additionally, among variables not comprising the score, those in the highest PsA tertile had longer duration PsA, less time between onset of PsO and development of PsA, more peripheral and less oligoarthritis, lower Psoriasis Area Severity Index (PASI) scores, less non-plaque psoriasis, were more likely to be receiving a TNFi and had been exposed to more prior DMARDs.
Association of HLA-B and C alleles/haplotypes with PsA severity
The associations of HLA-B and C alleles and haplotypes with being in the top tertile of the PsA severity propensity score are summarised in table 3. B*27:05:02-C*02:02:02, B*37:01-C:06:02 and B*08:01:01-C*07:01:01 haplotypes were most strongly associated with an elevated severity propensity score, and B*44:02:0101-C*05:01:0101 was most strongly associated with a decreased score. Table 3 shows four HLA-B alleles associated with increased propensity and four with decreased propensity (p<0.20), whereas 2 HLA-C alleles were similarly associated with increased propensity and 3 with decreased propensity.
Positive and inverse indicators at the p<0.20 level were included in a calculated allele risk score (score 1: alleles only, unweighted; table 4). Each one-unit increase in this score, on average, was associated with a 30% increase in the odds of being in the highest tertile of PsA severity (p<0.001). Adjustment for age, smoking, duration of psoriasis and PsA, time between psoriasis and PsA and TNF inhibitor use did not substantially alter the association (adjusted OR=1.35, p<0.001). The AUC for the prediction model including only risk score 1 as a predictor was 0.656 (p<0.05) and increased to 0.750 when other PsA severity indicators were added to the model. Excluding risk score 1 from the model significantly reduced the AUC to 0.703 (p=0.038 for the comparison of the AUC for the full vs reduced model) indicating that risk score 1 adds to the prediction of PsA severity over and above the other covariates. Each of the additional risk scores were also highly associated with the severity propensity score and significantly added to the prediction of severity in the context of other indicators. However, no manipulation of the risk score (ie, excluding non-contributory alleles, including haplotypes, weighting the contributions of alleles/haplotypes based on their magnitudes of association) resulted in a score that was superior to risk score 1 in the prediction of PsA severity (ie, the AUC for risk scores 2–4 was not significantly different from that of risk score 1). As an illustration, the crude and adjusted associations of HLA risk scores 1 and 4 with severity propensity are graphically depicted in figure 1. Because table 2 showed that additional features not used to construct the propensity score were significantly positively associated with the score, such as the presence of polyarthritis, deformed joint count, lower current PASI, increased frequency of plaque psoriasis, the percentage of patients requiring TNFi and the number of DMARDs used, the presence of these features are similarly associated with HLA risk score (table 3, figure 1). For example, in the case of model 1, the likelihood that a PsA patient receives a TNFi is associated with the presence of higher risk alleles/haplotypes shown in table 3 and each unit increase in the HLA risk score for model 1 was also associated with a 30% higher odds of being in the highest tertile of TNFi usage.
This exploratory study provided evidence that the HLA genes associated with susceptibility were also associated with particular features that collectively define the PsA phenotype of a given patient. This was evident in the preliminary univariate associations between phenotypical traits and HLA alleles and haplotypes, where each gene associated with susceptibility appeared to define a particular subphenotype. This action of a susceptibility gene did not appear stereotyped, but was often modified by the presence of additional HLA susceptibility alleles, such as those found on an ancestral haplotype. Similarly in the propensity analysis, additive interactions were evident between different susceptibility HLA alleles comprising the person's genotype, which were associated with clinical features defining the propensity for a more or less severe musculoskeletal phenotype.
HLA-B*27:05:02 was strongly associated with the development of symmetrical SI as were each of the haplotypes containing B*27:05:02, suggesting that HLA-B*27:05:02 itself was the driving allele. The combination of B*27:05, male sex and symmetrical SI suggests that the action of B*27:05:02 in this subset of PsA resembles that of B*27:05:02 in ankylosing spondylitis. A recent study supports this concept.16 In enthesitis, the pattern of association differed slightly, appearing to be predominantly driven by the B*27:05:02-C*01:02:01 haplotype, and less so by B*27:05:02. The presence of erosions exhibited a similar pattern, whereas osteolysis was predominantly conferred by C*02:02:02. Because osteolysis is one of the cardinal features of the mutilans phenotype, it was of interest to note that 31.8% of osteolysis patients have C*02:02:02 (table 1, see online supplementary table S1) compared with 9.7% that have C*01:02:01.
In contrast, the B*08:01:01-C*07:01:01 susceptibility haplotype and its alleles conferred a different phenotypical pattern. B*08:01:01 was associated with the presence of predominant form of asymmetrical SI, joint deformity and joint fusion, with no contribution from B*27:05:02 and its haplotypes.
The HLA-C*06:02:01 allele and B*57:01:01-C*06:02:01 that are primary susceptibility elements of PsO susceptibility were not positively associated with any feature and were negatively associated with asymmetrical SI. However, the B*37:01-C*06:02 haplotype was positively associated with joint fusion, and an increased severity index in the propensity analysis, again suggesting the epistatic effect of other genes on this C*06:02:01 haplotype in PsA.
We also found support for concluding that the character and severity of PsA phenotype are determined at the genotype level with genes of both chromosomes contributing additively to the composite phenotype measured by the propensity score. The finding of both positive and negative associations of phenotypical traits with different alleles at the same locus, for example, B*27 and B*44 alleles, was consistent with this (figure 1, table 3).
Enthesitis is often a striking feature of PsA. McGonagle et al,17 based on the finding of prominent bone marrow oedema at ligamentous attachments, proposed PsA should be considered an entheseal-based disease with secondary involvement of synovial structures. Subsequent studies confirmed the marrow oedema but showed it is non-PsA-specific and found in only a proportion of PsA patients.18 The present study showing that a genotype containing B*27:05:02-C*01:02:01 more strongly predicts the development of enthesitis in PsA, whereas the B*08:01:01-C*07:01:01 haplotype and its alleles are conversely associated with more synovial-based pathology such as joint deformity and joint fusion, is more consistent with the latter report. Furthermore, as emphasised by results in table 2, enthesitis was not more prevalent in those with a severity score in the third tertile, indicating an unlinking of enthesitis with other elements in the score. Dactylitis was associated with the B*27:05:02–01:02:01 haplotype, consistent with the contribution of enthesitis to this feature. However interestingly, dactylitis was also associated with B*08:01:01-C*07:01:01, implying possible heterogeneity in the mechanism of this trait.
Clearly, despite the relatively large size of this cohort, with 282 PsA cases clinically phenotyped that were also high resolution sequence-based HLA-typed to at least 4 digits, the study must be considered exploratory. A much larger study size and different study design would be required to address any possible spurious associations and provide results of clinical utility. Additionally, the ability to explore which gene in a haplotype was responsible for the association was often limited by small numbers of HLA-B and C alleles occurring outside of haplotypes common in the Irish population. Further confirmation of these observations especially in more admixed populations having different principal ancestral haplotypes will be required together with studies aimed to uncover pathogenic mechanisms underlying these clinical phenotypes.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
Handling editor Tore K Kvien
Correction notice This article has been corrected since it was published Online First. Minor changes have been made to table 1 and formatting in the right hand column of the first page has been corrected.
Contributors MH carried out the work, collection and interpretation of data and manuscript drafting. RW, JTG, EH, OF conceived the study, its design, coordination, data interpretation and manuscript drafting and editing.
Competing interests MH: Unrestricted educational grant from Abbvie; OF: has received honoraria and grant support and has been a member of advisory boards for Pfizer, Abbvie, MSD, Roche, UCB, Janssen and Cellgene.
Patient consent Obtained.
Ethics approval St. Vincent's Healthcare Group, Ethics and Medical Research Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.