Introduction

Rheumatoid arthritis (RA) is a chronic autoimmune disease of progressive joint destruction and impacts substantially 1 % of the world’s population [1]. Although the pathogenesis of RA is still unknown, the genetic factors seem to play an important role [2]. Human leukocyte antigen (HLA) region is the best-characterized gene susceptible to RA [3]. Nonetheless, the HLA locus only accounts for about one third of the total genetic factors of RA susceptibility [4]. It is indicated that non-HLA genes may take a prevalent part in RA.

Interleukin-23 (IL-23) has been classified as a necessary factor for the development of T cell-dependent inflammation and regulates the differentiation of naive CD4+ T cells into T helper cells [5, 6]. IL-23 plays a crucial role in the development of pathogenic Th17 cells which produce the cytokine IL-17. IL-17 can stimulate the production of several inflammatory cytokines, such as IL-6, and tumor necrosis factor (TNF)-α [7, 8]. IL-23 also leads to activation of janus kinases (Jak2) and tyrosine kinase (Tyk2) when combining with a complex, including IL-23 receptor (IL-23R) and IL-12 receptor beta-1 (IL-12Rβ1). Jak kinases phosphorylate IL-23R at discrete positions and emerge docking sites for the signal transducers and activators of transcription (STATs). Then the STATs are phosphorylated, dimerized and translocated to the nucleus where they effect the transcription of crucial pro-inflammatory genes such as IL-17 [9, 10]. IL-23/IL-17 axis is believed to be a new pathway which is involved in inflammatory responses [11]. Moreover, the serum and synovial fluid levels of IL-23 correlate with the IL-17 concentration in RA [12].

IL-23R, encoding on chromosome 1 (1p31.3), is a type I transmembrane protein and consists of 629 amino acids [9]. Recent studies suggest that the IL-23R variants might be affected with serum cytokine concentrations and subsequently impact on dependent cytokines [13]. There are four single nucleotide polymorphisms (SNPs) which have been studied most in RA (rs10489629, rs7517847, rs11209026, rs1343151). Although meta-analysis of the four SNPs had been reported before [14–16], new researches on the role of rs10489629 and rs7517847 in RA have been published and provided new proofs which were not involved in the previous meta-analysis. Due to small sample size of the former studies, the results of the new articles were contradictory with the previous ones. To better understand the relation of IL-23R gene polymorphisms (rs10489629, rs7517847) and RA risk, we performed the meta-analysis of these relevant studies to explore more accurate results.

Materials and methods

Identification of eligible studies

Available articles about the relation between IL-23R polymorphisms and RA were cautiously collected. A literature search was made using the key words “Interleukin 23 receptor’’ (IL-23R), “rheumatoid arthritis” (RA) and “polymorphism”, and only fully published studies until February 2012 were selected. All data were gathered from the following electronic databases: Pubmed, Medline, China National Knowledge Infrastructure database (CNKI), Chinese Biomedical database (CBM). When the studies contained overlapping data, we just chose the most recent subjects with largest sample. Eligible studies should approach to the following inclusion criteria: (a) all patients with RA were diagnosed by the 1987 American College of Rheumatology criteria, (b) it was a case–control design, (c) the study demonstrated the association between the IL-23R gene polymorphisms and RA, (d) the frequencies of alleles were available.

Data extraction

All information was extracted cautiously by two authors. Discrepancy was addressed by discussion. From each study, we extracted the following information: first author’s name, year of publication, country, ethnicity, the numbers of cases and controls, genotype method and allele frequencies of the IL-23R SNPs.

Quality assessment

Two authors evaluated the quality of the selected studies with the Newcastle–Ottawa Scale (NOS) [17]. In the case–control study, this scale describes three aspects of quality: selection of cases and controls, comparability of cases and controls, and ascertainment of exposure. Maximums of four stars, two stars and three stars are distributed to selection, comparability and exposure respectively. So, nine stars are the highest. The score higher than six is defined as high quality. Disagreement was settled by discussion.

Statistical analyses

We assessed the heterogeneity between studies with Q statistic as well as I2 statistic which measures the degree of heterogeneity (I2 = 100 % × (Q − df)/Q). I2 value varies from 0 to 100 %. Below 25 % is representative of no heterogeneity. From 25 % to 50 % stands for moderate heterogeneity. There is a large heterogeneity over 50 % [18]. p ≤ 0.10 was suggested to a statistically significance. We obtained the pooled odds ratio (OR) and 95 % confidence interval (CI) of all studies from fixed or random effect model. When there was no heterogeneity, we used the fixed effect model for meta-analysis [19]. Otherwise, a random effect model was used [20].

Evaluation of publication bias

The funnel plot was conducted on publication bias, and Egger’s linear regression test was to estimate funnel plot asymmetry [21]. The statistical analyses were completed by Stata 10.0 (Stata Corporation, College Station, TX, USA). All the p values were two-sided. p values less than 0.05 were regarded as significant.

Result

Selection and characteristics of studies

The process of study selection was showed in Fig. 1. After an initial search, we got 17 reports. According to the inclusion criteria, six reports were identified (excluded seven other autoimmune diseases, two reviews, and two other design reports). Three were excluded because two were not related to the SNPs (rs10489629, rs7517847) and one was not in Hardy–Weinberg equilibrium (HWE) [22]. In the remaining three reports, one comprised four different populations of each SNP, another contained two different populations of each SNP, and the other included one population of rs7517847. At last, six studies of rs10489629 and seven studies of rs7517847 were available in our meta-analysis [14–16]. All cases and controls in the studies were from European. Characteristics of all studies were illustrated in Table 1. The quality assessments of the studies were shown in Table 2. All studies were high quality for NOS score above six.

Fig. 1
figure 1

Process of study selection

Full size image
Table 1 Characteristics of each study in the meta-analysis
Full size table
Table 2 Quality assessment according to the NOS scale
Full size table

Associations between polymorphisms and RA

Eligible studies were all pooled in this meta-analysis. A summary of the result was presented in Table 3:

Table 3 Meta-analysis of the polymorphisms and RA association
Full size table
  1. 1.

    rs10489629. Six studies were involved with 5,786 patients and 7,326 controls from European. Compared with T allele, C allele was risk for RA (OR = 1.092, 95 %CI = 1.038–1.149, p = 0.001). Distribution of the ORs was presented in the forest plot (Fig. 2). We also discovered significant associations in the dominant and recessive modeling (C/C+C/T vs. T/T, OR = 1.146, 95 % CI = 1.059–1.240, p = 0.001, Fig. 3; C/C vs. C/T+T/T, OR = 1.099, 95 % CI = 1.008–1.199, p = 0.032, Fig. 4). In addition, a fixed effect model was applied to the studies for no significant heterogeneity between the comparisons (C vs. T, p = 0.122, I2 = 42.5 %, Fig. 2; C/C+C/T vs. T/T, p = 0.183, I2 = 33.8 %, Fig. 3; C/C vs. C/T+T/T, p = 0.303, I2 = 17.1 %, Fig. 4) (Table 3).

    Fig. 2
    figure 2

    Forest plot for meta-analysis of C versus T (fixed effect model) with rs10489629

    Full size image
    Fig. 3
    figure 3

    Forest plot for meta-analysis of C/C+C/T versus T/T (fixed effect model) with rs10489629

    Full size image
    Fig. 4
    figure 4

    Forest plot for meta-analysis of C/C versus C/T+T/T (fixed effect model) with rs10489629

    Full size image
  2. 2.

    rs7517847. Pooling the seven studies, a significant heterogeneity was found (p = 0.078, I2 = 47.2 %), so a random effect model was used (OR = 1.087, 95 % CI = 1.009–1.170, p = 0.027). Owing to a significant heterogeneity in overall studies, we calculated the combined ORs repeatedly after removing one study at a time. Excluding the Norway research, the result showed no significant heterogeneity (p = 0.483, I2 = 0.0 %, Fig. 5). The remaining six studies comprised of 5,089 cases and 6,711 controls from Europe. The G allele was risk for RA (G vs. T, OR = 1.121, 95 % CI = 1.063–1.183, p < 0.001, Fig. 5). The dominant and recessive modeling were also susceptible to RA (G/G+G/T vs. T/T, OR = 1.184, 95 % CI = 1.092–1.283, p < 0.001, Fig. 6; G/G vs. G/T+T/T, OR = 1.133, 95 % CI = 1.030–1.246, p = 0.010, Fig. 7). For no significant heterogeneity between comparisons after omitting the Norway population, we chose a fixed effect model (G/G+G/T vs. T/T, p = 0.506, I2 = 0.0 %, Fig. 6; G/G vs. G/T+T/T, p = 0.509, I2 = 0.0 %, Fig. 7) (Table 3).

    Fig. 5
    figure 5

    Forest plot for meta-analysis of G versus T (fixed effect model) with rs7517847

    Full size image
    Fig. 6
    figure 6

    Forest plot for meta-analysis of G/G+G/T versus T/T (fixed effect model) with rs7517847

    Full size image
    Fig. 7
    figure 7

    Forest plot for meta-analysis of G/G versus G/T+T/T (fixed effect model) with rs7517847

    Full size image

Publication bias

The method of Egger’s linear regression test was used for accessing the funnel plot asymmetry. When publication bias does not exist, the regression line will pass through the origin. The results of Egger’s linear regression test shown in Table 3 indicated that there was no publication bias.

Discussion

The standard form of IL-23R is encoded by about 12 exons. It is suggested that at least six spliced isoforms could be generated (IL-23R1 to 6) via alternative splicing [23]. The SNPs rs10489629 located in intron seven and rs7517847 was in intron six [24]. The IL-23R gene expressed at least six alternatively spliced mRNAs, which lead to various isoforms of the receptor protein. The intronic polymorphisms (rs10489629, rs7517847) might exert their influence through regulating the specific splicing [25]. A genome-wide association study suggested a strong association with Crohn’s disease (CD) and IL-23R gene polymorphisms in 2006. Theoretical considerations implied potential relation of these SNPs with other autoimmune diseases such as RA [26]. A series of researches have been undertaken to replicate this association in RA. It has been confirmed that rs1343151 is risk for RA but rs11209026 is not associated with RA susceptibility [14, 16]. However, the associations of rs10489629 and rs7517847 with RA are still conflict. In this present study, we updated previous studies of the association between the IL23R polymorphisms (rs10489629, rs7517847) and RA in European population.

In this meta-analysis, we performed three genetic models for each SNP (rs10489629: C vs. T, C/C+C/T vs. T/T, C/C vs. C/T+T/T; rs7517847: G vs. T, G/G+G/T vs. T/T, G/G vs. G/T+T/T). Our research indicated C allele of rs10489629 was risk for RA which was consistent with the previous meta-analysis, and G allele of rs7517847 was susceptible to RA, while there was no significant association between rs7517847 and RA in the previous one [16]. Compared with the previous researches, we eliminated the study which exhibited a small deviation from HWE in the controls. Perhaps it contributed to the contradiction of rs7517847 between previous and present meta-analysis. In addition, the quality assessments of the studies were evaluated by the NOS. Although the number of the combined studies is rather small for a meta-analysis, these studies are all high quality.

However, the two SNPs are protective factors in ankylosing spondylitis (AS) [27–30]. All noninfectious inflammatory diseases depend on a spectrum from autoinflammatory (inducing by the innate immune system) to autoimmune (mediating by the adaptive immune system). Relative actions of each can define the type of diseases [31]. AS is considered to be a typical autoinflammatory disease, while RA is proposed to be autoimmune in nature [32, 33]. It is similar to PTPN22 C1858T which is associated with systemic autoimmune diseases such as RA, but not associated with organ-specific autoimmune diseases like AS [34, 35]. The same genetic variant may not share a common mechanism in different autoimmune diseases. Further researches are needed to clarify the exact mechanisms of IL-23R polymorphisms in RA.

By the way, some limitations of our study should be paid attention to. Primarily, the number of studies is small, and the sample was limited to European ancestry. Perhaps the allelic frequencies are different from European to others. Consequently, the relation in other ethnic groups can not be conducted. Secondly, our literature searching was based on English and Chinese, and language bias might be occurred. Next, significant heterogeneity was detected among several comparisons. The differences of experimental methods may cause the heterogeneity. Ultimately, owing to only extracting published studies, publication bias might occur, even if statistical test implied no significance.

In a word, this meta-analysis suggests the IL-23R polymorphisms rs10489629 and rs7517847 are susceptible to RA in European populations. Further gene–gene and gene–environment interactions researches on large case–control design in all races are still required.