Introduction

Systemic lupus erythematosus (SLE) is an autoimmune disease typically characterized by the generation of multiple autoantibodies to nuclear antigens, and by complement activation and immune complex deposition, which results in diverse clinical manifestation and the involvement of multisystem [85]. It has been confirmed that genetic and environmental factors play important role in the development of SLE [27, 34, 85]. The morbidity rate of SLE is diversiform in different populations [39], with 31–70/100,000 in Chinese [88], and 7–71/100,000 in European [12] and up to 200/100,000 in African populations [12]. Although, abundant studies have been carried out, the etiology and pathogenesis of SLE have not been clearly illustrated. Genome-wide linkage studies have evidenced susceptibility area in SLE, such as 1q23 [80], 4p16-15.2 [24], 11p13 [53], 11q14 [73], 12q24 [54], 16p13 [90] and 16q12 [80], and dozens of valuable susceptible genes have been found, such as IKBKE [72], IL8 [72], MAMDC1 [28], PTPN22 [78], STAT4 [62], FCGR2B [95], FCGR3B [95] and TLR5 [70]. Recently, genome-wide association studies (GWAS) have provided more clues for the studies of complex diseases, which assisted us in discovering the susceptibility genes and SNPs for SLE [26, 49, 56, 86].

PTPN22 (protein tyrosine phosphatase non-receptor 22) gene locates on chromosome 1p13.3-p13.1 and encodes lymphoid-specific phosphatase (Lyp) [67]. Lyp is known to participate in antigen receptor signaling and its impact on the immune response and role in the development of autoimmunity have been a focus of current research [67]. Lyp inhibits T cell activation by binding to C-terminal Src tyrosine kinase (Csk). PTPN22 participates in the inhibition of T cell activation, inactivates kinase of T cell receptor and limits to the reaction to the antigen [10]. The mutation of PTPN22 rs2476601 single-nucleotide polymorphism (SNP) could disrupt the physiological interaction between Lyp gene and Csk kinase, and cause signal transduction disorder and reduce the threshold for T cell receptor signaling, leading to the occurrence of autoimmune diseases [11, 82], including SLE, rheumatoid arthritis, type 1 diabetes and autoimmune thyroid disease [9]. To date, the GWAS have reported the association between PTPN22 rs2476601 and autoimmune diseases, such as type 1 diabetes, rheumatoid arthritis, myasthenia gravis and Crohn’s disease [3, 22, 25, 58]. The GWAS have also confirmed the significant association between PTPN22 rs2476601 and SLE in Europeans [6, 49]. Although results from GWAS were sometimes taken as the golden criterion for the association study, conflicting findings still existed in different studies, which might have resulted from greater variability of allele frequencies in different populations.

Toll-like receptor (TLR) family is a transmembrane pattern recognition receptor, which plays an important part in the activation and regulation of both innate and adaptive immune response. TLR9 is a kind of specific receptor, which can recognize cytosine phosphate guanosine (CpG) located in bacterial DNA [29]. The aberrant activation of TLR9 may lead to the production of type 1 interferon, which has been closely related to SLE pathophysiology [27]. Study has confirmed that TLR9 plays an important role in the generation and development of SLE by combining with the CpG motif [37].

Migita et al. [47] showed that TLR9 expression in lymphocytes was increased in patients with SLE. However, Hur et al. [33] observed that four polymorphisms within a 4334-bp segment of the TLR9 gene demonstrated no significant association with susceptibility to SLE in a Korean population. De Jager et al. [13] examined a much larger 68,742-bp segment that contains the entire TLR9 gene and 60 kb of flanking sequence, and no evidence of association was found.

Although a large number of studies have investigated the associations between PTPN22 SNPs (rs2476601), TLR9 SNPs (rs187084, rs352140, rs5743836, rs352139) and SLE risk, the results were inadequate or discordant. To check the association in different ethnics, we performed a comprehensive ethnicity-specific meta-analysis to illuminate the association of PTPN22 and TLR9 with SLE susceptibility by enhancing statistical power. We also tried to find the best-fit association model among the dominant, recessive and additive models for the polymorphisms.

Methods

Identification of eligible studies

A comprehensive literature research was conducted by the PubMed databases (National Center for Biotechnology, National Library of Medicine), CNKI (Chinese National Knowledge Infrastructure) and Wanfang databases (Wanfang Data Knowledge Service Platform) for the association between PTPN22 or TLR9 polymorphisms with SLE before December 2016, with the following key words: ‘The protein tyrosine phosphatase non-receptor 22’, ‘PTPN22’, ‘Toll-like receptor 9’, ‘TLR9’, ‘polymorphism’, ‘Systemic lupus erythematosus’ and ‘SLE’.

Inclusion and exclusion criteria

The included study should satisfy the following criteria: (1) case–control study; (2) providing raw data, from which the odds ratio (OR) and 95% confidence interval (CI) could be calculated; (3) all patients should fulfil the American College of Rheumatology (ACR) criteria [31]. Exclusion criteria were as follows: (1) data from conference or meeting abstracts; (2) family-based study; (3) study without genotype distributions.

Data extraction

The following information is extracted from each study: journal, first author, year of publication, ethnicity, the sample size, genotype distributions and the allele frequency of cases and controls. We calculated the number of genotype by allele frequency with Hardy–Weinberg equilibrium (HWE) law, when the data was insufficient in the articles.

Statistical analyses

We calculated the Hardy–Weinberg equilibrium (HWE) of control group separately in each of the study, and we considered that the study population did not conform to HWE when the P value for HWE <0.05. Allele frequencies of PTPN22 and TLR9 polymorphisms from each study were determined by allele counting. OR and 95% CIs were calculated to evaluate the association between PTPN22 SNP or TLR9 SNPs and SLE; meanwhile, Cochran’s Q-statistic was used to estimate the effect of variations and heterogeneities [30]. If a significant Q-statistic indicated heterogeneity across studies (P < 0.10), then the random effect model was applied to the meta-analysis; otherwise, the fixed effect model was used [15, 30]. The effect of heterogeneity was quantified with the following formula: I 2 = 100% × (Q − df)/Q [30]. The I 2 value represented the proportion of between-study variability attributable to heterogeneity, and I 2 values of 25, 50 and 75% were considered to be of low, moderate and high heterogeneity, respectively. If study groups showed no heterogeneity, fixed and random effects models produced similar results and, if not, the random effects model usually produced wider CIs than the fixed effects model. Thus, random effects model was applied to this analysis of significant between-study heterogeneity.

The potential publication bias was evaluated using Begg’s linear regression test [4], which measured the funnel plot asymmetry using a natural logarithm scale of ORs. A sensitivity analysis was performed to assess the influence of each study on the total effect. The statistical analysis was carried out using STATA 12.0 (Stata Corporation, College Station, TX, USA).

Results

Studies included in the meta-analysis

In this meta-analysis, totally 221 relevant articles (68 for PTPN22 and 153 for TLR9) were found. After reading titles and abstracts, we selected 66 articles (36 for PTPN22 and 30 for TLR9) for further reading. Then, we excluded 29 articles (19 for PTPN22 and 10 for TLR9), because of no data, insufficient data, repeated date, family-based studies, not referring to SLE or reviews. Thus, 37 articles (17 for PTPN22 and 20 for TLR9) met the study inclusion criteria [1, 2, 14, 16, 18,19,20,21, 23, 32, 33, 35, 38, 45, 46, 48, 52, 55, 57, 59,60,61, 63,64,65,66, 68, 69, 74, 76, 77, 79, 82, 84, 89, 93, 94]. The flowchart of selecting article process is presented in Fig. 1. Totally, there were 9349 cases and 12,390 controls to evaluate the relationship between PTPN22 rs2476601 and SLE. For TLR9, there were 2927 cases and 3509 controls for rs187084, 2834 cases and 3398 controls for rs352140, 2317 cases and 3291 controls for rs5743836 and 2090 cases and 2390 controls for rs352139, respectively. The basic information of these included studies is shown in Table 1a, b.

Fig. 1
figure 1

Flow diagram of the included and excluded articles in this meta-analysis

Full size image
Table 1 Characteristics of the (a) PTPN22 and (b) TLR9 polymorphism allelic and genotype distribution among SLE patients and control subjects
Full size table

Lastly, after excluding two groups of data in which the control populations deviated from HWE, 17 publications providing 9120 cases and 11,724 controls were pooled to evaluate the relationship between PTPN22 rs2476601 and SLE in the meta-analysis (Table 1a) [35]. Similarly, for TLR9 rs187084, rs352140, rs5743836 and rs352139, after exclusion of the studies of nonconformity with HWE, there were 2760 cases and 3298 controls, 2808 cases and 3386 controls, 1640 cases and 2282 controls and 1978 cases and 2290 controls (Table 1b), respectively.

Meta-analysis results

PTPN22 rs2476601 polymorphism and SLE risk

17 relevant studies including 9120 cases and 11,724 controls were analyzed in this meta-analysis after excluding two separate comparisons [35] of nonconformity with HWE. An outstanding association was found in the overall population, under the random effect model (T vs. C: OR = 1.511, 95% CI 1.338–1.706, P = 2.931 × 10−11, Table 2; Fig. 2a). Ethnicity-specific meta-analysis indicated significant association between rs2476601 and SLE in American and European populations (T vs. C: OR = 2.566, 95% CI 1.796–3.665, P = 2.219 × 10−7; OR = 1.399, 95% CI 1.261–1.552, P = 2.153 × 10−10, Table 2; Fig. 2d). We also identified an association between this polymorphism and SLE in Africans (T vs. C: OR = 4.14, 95% CI 1.753–9.775, P = 0.001, Table 2; Fig. 2d). However, no significant association was found between rs2476601 and SLE in Asians (T vs. C: OR = 1.323, 95% CI 0.550–3.181, P = 0.532, Table 2; Fig. 2d).

Table 2 Meta-analysis of PTPN22 rs2476601 polymorphisms and SLE association
Full size table
Fig. 2
figure 2

Forest plot for the meta-analysis of the association between PTPN22 rs2476601 and SLE a rs2476601 and SLE (T vs. C), b rs2476601 and SLE (TT+CT vs. CC), c rs2476601 and SLE (TT vs. CT+CC), d rs2476601 and SLE after stratification by ethnicity (T vs. C)

Full size image

Accordingly, we used the dominant and recessive models to estimate the relationship between PTPN22 rs2476601 and SLE risk. We observed the association between this polymorphism and SLE in the dominant and recessive model of T allele in the overall population (TT+CT vs. CC: OR = 1.531, 95% CI 1.346–1.742, P = 9.172 × 10−11; TT vs. CT + CC: OR = 2.03, 95% CI 1.519–2.712, P = 1.753 × 10−6, Table 2; Fig. 2b, c). We also performed the stratification analysis by ethnicity of dominant and recessive models; similarly, there was a significant association between rs2476601 and SLE in the dominant model in the American, European and African populations. However, we found no association in the dominant and recessive models in Asians (Table 2; Supplementary Fig S1a, b).

TLR9 polymorphisms and SLE risk

The association between TLR9 rs187084 polymorphism and SLE was conducted in ten independent studies after the exclusion of one study [32] on the discrepancy of HWE. The test of heterogeneity in the overall population was not significant (P = 0.732, I 2 = 0%), suggesting that the fixed effect model could be used. No association between the allele of TLR9 rs187084 and SLE was identified (C vs. T: OR = 0.929, 95% CI 0.860–1.003, P = 0.061, Table 3; Fig. 3a). We tested the dominant and recessive models of C allele in the overall population, but found no significant association in the two models (CC + CT vs. TT: OR = 0.913, 95% CI 0.817–1.019, P = 0.104; CC vs. CT+TT: OR = 0.899, 95% CI 0.776–1.042, P = 0.158, Table 3; Supplementary Fig S2a, b).

Table 3 Meta-analysis of TLR9 gene polymorphisms and SLE association
Full size table
Fig. 3
figure 3

Forest plot for the meta-analysis of the association between TLR9 polymorphisms and SLE a TLR9 rs187084 and SLE (C vs. T), b TLR9 rs352140 and SLE (T vs. C), c TLR9 rs5743836 and SLE (C vs.T) and d TLR9 rs352139 and SLE (G vs. A)

Full size image

Similarly, we analyzed the cases and controls of TLR9 rs352140, rs5743836 and rs352139 to assess their association with SLE, after performing the HWE test and heterogeneity test. However, no association was found between rs352140, rs5743836, rs352139 and SLE susceptibility (Table 3; Fig. 3b–d). Furthermore, we analyzed the dominant and recessive models, respectively, which indicated that no statistically significant association was discovered in all of the dominant and recessive models (Table 3; Supplementary Figs S3a, b, S4a, b, S5a, b).

Allele frequency of the five SNPs and comparison to the 1000 Genomes Phase 3 population

We displayed the allele frequencies in different ethnicities in our meta-analysis and 1000 genomes of the five SNPs in Table 4. In view of the sample size and population, the allele frequencies of PTPN22 rs2476601, TLR9 rs187084 and TLR9 rs352139 in this meta-analysis were consistent with the allele frequencies in the 1000 Genome Project AMR (Admixed American), EUR (European ancestry), AFR (African ancestry), EAS (East Asian ancestry), respectively, however, there was distinction between the allele frequencies of TLR9 rs352140, TLR9 rs5743836 and 1000 Genomes Project (Table 4).

Table 4 The allele frequency comparison between the meta-analysis and 1000 Genomes Project
Full size table

Publication bias and sensitivity analysis

Begg’s funnel plot was used to evaluate the publication bias. There was no obvious evidence of symmetry from the shapes of the funnel plots (Fig. 4a–e) and no evidence of publication bias of all study subjects (P > 0.05). Accordingly, we performed sensitivity analysis to assess the influence of individual study on the pooled OR. The analysis showed that no individual study significantly affected the pooled OR; therefore, no study was deleted (Fig. 5a–e).

Fig. 4
figure 4

Begg’s funnel plot of publication bias in the meta-analysis of the association of PTPN22 and TLR9 polymorphisms with SLE risk under allele genetic model. a PTPN22 rs2476601 and SLE (T vs. C), b TLR9 rs187084 and SLE (C vs.T), c TLR9 rs352140 and SLE (T vs. C), d TLR9 rs5743836 and SLE (C vs. T) and e TLR9 rs352139 and SLE (G vs. A)

Full size image
Fig. 5
figure 5

Sensitivity analysis to assess the stability of the meta-analysis. a PTPN22 rs2476601 in SLE, b TLR9 rs187084 in SLE, c TLR9 rs352140 in SLE, d TLR9 rs5743836 in SLE and e TLR9 rs352139 in SLE

Full size image

Discussion

In this study, we conducted a meta-analysis of the association between PTPN22 rs2476601 and TLR9 polymorphisms with SLE susceptibility. We validated that the T allele of PTPN22 rs2476601 apparently increased the risk of SLE, but its effect was diverse in different ethnics. Although we found no significant association between TLR9 polymorphisms and SLE, we summed up and verified previous studies.

SNP rs2476601 is a missense mutation in the PTPN22 gene on chromosome 1p13.3-13.1, which was identified to be associated with multiple autoimmune diseases [10]. In 2004, PTPN22 rs2476601 was first indicated to be associated with type 1 diabetes [7] and rheumatoid arthritis [5]; afterward, studies showed the relevance between PTPN22 and other diseases, such as SLE, myasthenia gravis and Crohn’s disease [17, 38, 81]. Recently, GWAS have discovered significant association between PTPN22 rs2476601 and SLE in Europeans [49]. To check this association in different ethnicities, we carried out a comprehensive ethnicity-specific meta-analysis involving 9120 cases and 11,724 controls with 19 separate comparisons, and found that the association between PTPN22 rs2476601 and SLE achieved genome-wide significance in the overall population and reached significant correlation in the American and European, and a certain relevance in the African, but not in the Asian population. Zheng et al. [92] previously carried out a meta-analysis with 14 case–control studies and testified a significant association between PTPN22 rs2476601 and SLE, acquiring a more significant result than those in our meta-analysis. However, after careful reading, we found that Zheng et al. included a study of rheumatoid arthritis [20] and a study with data deviated from HWE [35]. Papers of this kind were not supposed to pass the inclusion criteria of meta-analysis; we therefore excluded these two studies and included an additional seven newly published studies after Zheng’s meta-analysis. Meta-analysis by Lee et al. [43], Shi et al. [75] and Lea et al. [40] included only 6, 4 and 11 studies, mainly in European, respectively. Our meta-analysis was more comprehensive and more diverse in population.

There were only three studies [19, 35, 48] for the African population included in this meta-analysis (one paper was excluded because of HWE deviation [35]), including only 210 effective cases and 261 controls. However, the effect of the T allele in the African population is in the same direction as that in the overall population. It was easy for us to see that rs2476601 T allele frequency was low in all the populations, and it was monomorphic in the African and Asian populations in the 1000 Genomes Phase 3 (Table 4). We could notice that the T allele was not monomorphic in the two African studies, and the T allele frequency of SLE case from Moez et al. [48] was much higher than that from Elghzaly et al. [19] (0.263 vs 0.029) (Table 1a). Because only 40 SLE cases were included in the study from Moez et al., the frequency was not reliable. Additionally, Sanchez et al. [71] have reported that no association was found between PTPN22 rs2476601 and SLE in African-Americans. These phenomenon and discrepancy needed further investigation on the basis of large sample size.

Previous studies have showed that genotype TT was not detected in the Chinese population [8, 91], similarly, no rs2476601 T allele and structural variant R620W of the PTPN22 protein were found in the Asian population [36, 41]. As shown in the 1000 Genomes Phase 3, the T allele was also not existent in the East Asian population (Table 4). Therefore, theoretically speaking, no association should be detected between rs2476601 and SLE in Asians. Although our result is consistent with the theory, more studies need to evaluate their association.

TLR9 could modulate autoimmunity by inducing inflammatory cells and by the production of cytokines or antibodies [79]. The expression and mRNA level of TLR9 on B cell were up-regulated in SLE patients, and SLE activity index was significantly associated with TLR9 expression [51]. Some studies indicated that TLR9 expression was obviously higher in SLE cases than in healthy controls. Wang et al. [83] performed a meta-analysis of TLR9 polymorphisms and SLE risk, and 21 studies from 19 articles with 10,273 subjects were included in that meta-analysis. The results indicated that TLR9 rs187084 could increase the risk of SLE (T vs. C: OR = 1.15, 95% CI 1.02–1.30, P = 0.020) in Asians with marginal significance and no evidence was found for the association between TLR9 rs352140, rs5743836, rs352139 and SLE risk. Similarly, Lee et al. [42] carried out a meta-analysis and the results showed that only TLR9 rs187084 was associated with SLE risk (OR = 0.869, 95% CI 0.762–0.992, P = 0.038). Although they verified the relativity of SLE and TLR9 rs187084, under the circumstances of P = 0.02 and P = 0.038, we considered the relevance was marginal rather than significant. However, Li et al. and Yang et al. indicated that none of these three polymorphisms (rs187084, rs352139, rs352140) showed any significant association with SLE risk in Asian populations [44, 87].

To precisely explore the association between TLR9 polymorphisms (rs187084, rs352140, rs5743836, rs352139) and SLE, we analyzed the data that were consistent with HWE and used the random effect model to deal with the heterogeneity. Ultimately, 18 articles with 10,690 subjects were analyzed comprehensively. Although TLR9 has been evidenced to be responsible for the maturation of dendritic cells (DCs) and the production of pro-inflammatory cytokines in knockout mice [29], we have no sufficient information to identify the significant relationship between TLR9 and SLE in the study population. Then, we compared our study with the meta-analysis from Wang et al. [83] and found that they included two groups of repeated data [76, 77], which could lead to imprecise results. On comparing our data with that of Lee et al. [42], we found that a study with data deviating from HWE (P < 0.05) was included in Lee et al., and we included one additional newly published study [93] which led to the diversity of results. Simultaneously, we realized that marginal significance was reached under the allele model and dominant model for TLR9 rs187084 if we included the data of nonconformity with HWE in the population (C vs. T: OR = 0.877, 95% CI 0.783–0.982, P = 0.022;CC + CT vs. TT: OR = 0.875, 95% CI 0.787–0.974, P = 0.014, Supplementary Fig S6a and Fig S6b) and in Asians (C vs. T: OR = 0.881 95% CI 0.780–0.996, P = 0.043; CC + CT vs. TT: OR = 0.883, 95% CI 0.791–0.986, P = 0.027, Supplementary Fig S7a, b), which indicated that the association between rs187084 and SLE was suggestive.

Although there were new discoveries revealed by this study, several potential limitations existed. Firstly, we need a large number of data to validate the relationship between PTPN22 and SLE for further study in the African and other populations. For TLR9, most studies included in this meta-analysis were carried out in Asian populations, so we had insufficient data to conduct subgroup meta-analysis. In other words, available data could not confirm the significant association between TLR9 and SLE. Secondly, we had to point out that significant between-study heterogeneity that was detected. We used the random effect model to deal with the heterogeneity, but this might introduce an insensible statistical illusion because fixed effect model and random effect model address different research questions [50]. Despite these deficiencies, we had searched for information and analyzed data as comprehensively as possible.

In conclusion, our meta-analysis indicated that the missense SNP rs2476601 in PTPN22 was significantly associated with SLE susceptibility in the overall population, and the association had reached genome-wide significance. However, the relationship between Africans and Asians needs further investigation. TLR9 polymorphisms rs187084, rs352140, rs5743836 and rs352139 may not be associated with SLE.