|Home | About | Journals | Submit | Contact Us | Français|
Background: Lipoma preferred partner (LPP) and T-cell activation Rho GTPase activating protein (TAGAP) polymorphisms might influence the susceptibility to celiac disease. Therefore, we performed a meta-analysis by identifying relevant studies to estimate the risks of these polymorphisms on celiac disease. Methods: The PubMed, Web of Science and Embase databases were searched (up to October 2016) for LPP rs1464510 and TAGAP rs1738074 polymorphisms. Results: This meta-analysis included the same 7 studies for LPP rs1464510 and TAGAP rs1738074. The minor risk A allele at both rs1464510 and rs1738074 carried risks (odds ratios) of 1.26 (95% CI: 1.22–1.30) and 1.17 (95% CI: 1.14–1.21), respectively, which contributed to increased risks in all celiac disease patients by 10.72% and 6.59%, respectively. The estimated lambdas were 0.512 and 0.496, respectively, suggesting that a co-dominant model would be suitable for both gene effects. Conclusions: This meta-analysis provides robust estimates that polymorphisms in LPP and TAGAP genes are potential risk factors for celiac disease in European and American. Prospective studies and more genome-wide association studies (GWAS) are needed to confirm these findings, and some corresponding molecular biology experiments should be carried out to clarify the pathogenic mechanisms of celiac disease.
Celiac disease (CD) is a chronic and immune-mediated enteropathy that is induced by dietary protein gluten (from wheat, barley and rye) in genetically predisposed individuals . It is a small-intestine disorder, affecting approximately 1% of the European population with some regional variations  and causing malnutrition and severe complications. Celiac patients have a greater burden of disease than the general population, and a long-term gluten-free diet (GFD) is the only therapy for this disease [1,3]. HLA-DQ2 and HLA-DQ8 molecules are responsible for only approximately 40% of genetic predisposing factors in the pathogenesis of CD , which is necessary but not sufficient to cause disease [5,6]. Thus, many more risk loci outside the HLA region should be identified as disease markers.
In recent years, genome-wide association studies (GWAS) have expanded our understanding of genetic makeup and revealed several possible inherited risk factors in celiac disorders [7,8,9,10]. Many of the non-HLA loci overlap with Crohn’s disease, type 1 diabetes, rheumatoid arthritis and juvenile idiopathic arthritis [11,12,13,14,15], such as lipoma preferred partner (LPP) and T-cell activation Rho GTPase activating protein (TAGAP). Alterations of the actin cytoskeleton and cell shape can be observed in the CD patients’ intestinal mucosa [16,17], while the cell shape is maintained through the actin cytoskeleton and focal adhesion . LPP is localized with paxillin in focal adhesions, and the number of paxillin focal adhesions with LPP is increased in CD fibroblasts. A constitutive alteration in cell shape and adhesion involving LPP occurs in CD fibroblasts, suggesting a correlation between LPP and CD pathogenesis . In addition, LPP is considered a substrate of the protein-tyrosine-phosphatase 1B (PTP1B) . Of note, loss of PTP1B can attenuate the activation of extracellular signal regulated kinase (ERK) , which is activated in the CD patients’ mucosa on a GFD or a gluten-containing diet (GCD). Only when ERK is phosphorylated can it transduce to the nuclei, and it has been found that more nuclei of the enterocytes from CD patients were positive for ERK compared with controls. Inhibition of ERK phosphorylation normalizes crypt enterocyte proliferation of CD atrophic mucosa . When PTP1B is sufficient or excessive, there may be more ERK activity in the celiac enterocytes, resulting in the progression of CD.
TAGAP is involved in the Rho GTPase cycle [23,24], which is between the inactive GDP-bound and the active GTP-bound states. The exchange of GDP-bound for GTP-bound is catalyzed by GEFs, while GAPs increase the intrinsic GTPase activity of Rho GTPases to accelerate the return of the proteins to the inactive state [25,26,27]. In the active state, GEF-catalyzed activation of Rho interacts with ROCK, which can activate the myosin light chain (MLC) and LIM domain kinase (LIMK), and both of them play an important role in focal adhesion and regulate the rearrangement and stabilization of the actin cytoskeleton . However, TAGAP propagates the inactive form of the RHO molecule; and it increases the activity of Rho GTPases via phosphorylation, enhancing their intrinsic activity up to fivefold . TAGAP negatively regulates downstream effects; thus, the actin cytoskeleton rearrangement is dysfunctional and lack of unstable .
Mutation of LPP and TAGAP may interfere with their original function and even promote the progress of CD. In recent years, a number of studies, including GWAS, have reported the association of LPP and TAGAP polymorphisms with CD susceptibility, and many have focused on LPP rs1464510 (A/C) and TAGAP rs1738074 (A/G). However, those studies have drawn inconsistent conclusions due to the limited regions and small numbers of articles. For example, Dubois et al.  reported that rs1464510 was positively associated with CD in the Netherlands, whereas there was no relationship in a Dutch population according to Coenen et al.  and Hunt et al. . Similarly, results for rs1738074 differed from country to country in the studies by Plaza-Izurieta et al.  and Sperandeo et al. . Therefore, we decided to carry out this meta-analysis on all the available case-control studies to accurately assess the relationship between the LPP rs1464510/TAGAP rs1738074 and CD risk.
Relevant studies were searched in PubMed, the Web of Science and Embase up to October 2016. The search strategies were as followed: (((LPP or 3q28 or rs1464510 or “lipoma preferred partner”) or “lim domain containing preferred translocation protein”) and celiac disease) or ((TAGAP or 6q25 or rs1738074 or “T-cell activation GTPase activating protein”) and celiac disease). The search was limited to English-language and human studies. Only published studies were considered. We scanned the title and abstract of all relevant articles, manually examined reference lists for additional relevant publications and obtained the full text of all possibly relevant studies. If multiple articles were published on the same topic, the most complete and recent study was used.
A reviewer independently examined the titles and abstracts of the identified articles. Any human population-based association study was included regardless of subjects’ ethnicity if it met the following criteria: (1) it showed an association between LPP (rs1464510) or TAGAP (rs1738074) polymorphism, (2) the outcome was celiac disease and there was a control group, (3) there were sufficient data for extraction (i.e., minor allele frequency and genotype frequency) and (4) there was a clear diagnosis of celiac disease. Studies were excluded if: (1) the case and control subjects were biologically related; (2) the insufficient data that were failed to ask for supplementary information from the authors; (3) the studies comprised unrelated data, family studies, animal studies, reviews, or meeting abstracts; or (4) the studies were written not in English.
Summary data were extracted independently by reviewers using a standardized data extraction form. We extracted general information as follows: name of first author, year of publication, region of study population, source of controls, genotype method, diagnostic criteria, the number of cases and controls, and the minor allele frequency in cases and controls. Any disagreement was resolved by consensus.
Study quality was assessed independently by the same reviewers using a risk-of-bias score for genetic association studies that was developed by Thakkinstian et al.  (Supplementary Materials Table S1). The score considered 5 domains: information bias (ascertainment of outcome and gene), confounding bias, selective reporting of outcomes, population stratification, and Hardy-Weinberg equilibrium (HWE) assessment in the control group. Each item was scored “yes”, “no” or “unclear”, representing low risk, high risk and insufficient information, respectively. Disagreement between the two reviewers was solved by a senior reviewer (C.X.J). Additionally, the MOOSE checklist was used to measure the quality of our study (Supplementary Materials Table S2).
We used Stata software (version 12.0, StataCorp LLC, College Statopm, TX, USA) and the Comprehensive Meta-Analysis software (version 2.0, Biostat, Englewood, NJ, USA) for all statistical analyses. All tests with a p value less than 0.05 were considered statistically significant, except for the heterogeneity tests, in which a p value less than 0.10 was used. It was tested whether the distribution of genotypes in the controls was compliant with Hardy-Weinberg equilibrium (HWE) by a Fisher’s exact test to estimate the quality of studies. If the study was found not to be in HWE with a p value less than 0.05, it was considered to be in disequilibrium. We used both per-allele and per-genotype analysis to estimate the strength of the association between the polymorphism of LPP rs1464510 or TAGAP rs1738074 and CD risks.
Per-allele analysis: Suppose that A and a are risk and non-risk alleles, respectively, and AA, Aa and aa are minor homozygous, heterozygous, and common homozygous genotypes, respectively, for each polymorphism. The risk allele frequency in each group was estimated according to the reported genotype data, and overall prevalence along with 95% confidence intervals were estimated for each single nucleotide polymorphism (SNP). The Mantel-Haenszel method was used to determine the statistical significance of the pooled OR, and its p value was used to determine whether the overall gene effect was significant (p = 0.05). The heterogeneity of allele effects across studies was checked using a Q test, and the degree of heterogeneity was quantified by I2 (I2 < 25%, no heterogeneity; 25% < I2 < 50%, moderate heterogeneity; 50% < I2 < 75%, large heterogeneity; I2 > 75%, extreme heterogeneity). If heterogeneity was present (i.e., if the Q test was significant or I2 was greater than 25%), the cause of heterogeneity was explored using sensitivity analysis. We chose a random-effects model if I2 was greater than 50%; otherwise, a fixed-effects model was used. The population attributable risk (PAR) for the risk allele was calculated based on results from a discrete-time model. If the main effect of the genotype was statistically significant and had the appropriate effect model selection, further comparisons of OR1 and OR2 were explored. Per-genotype analysis: We used the model-free approach to estimate the genotype effect, and two odds ratios—AA vs. aa (OR1) and Aa vs. aa (OR2)—were estimated for each study. The model of the genetic effect, measured by the parameter lambda (λ), which is defined as the ratio of logOR2 to logOR1, was then estimated using the model-free Bayesian approach. Lambda (λ) represents the heterozygote effect as a proportion of the homozygote variant effect. The value of lambda ranges from 0 to 1. We obtained information about the genetic mode of action as follows: If λ = 0, a recessive (Aa + aa vs. AA) model is suggested; if λ = 1, a dominant model (AA + Aa vs. aa) is suggested; and if λ = 0.5, a co-dominant model (AA vs. aa, Aa vs. aa) is suggested. If λ > 1 or λ < 0, then a homozygous or heterosis model is likely, although this is rare. Once the best genetic model is identified, this model is used to collapse the three genotypes into two groups and to pool the results again. For lambda, WinBugs 1.4.2 was used with vague prior to distributions for the estimation of parameters (i.e., lambda and odds ratio). The models were run with a burn-in of 1000 iterations, followed by 10,000 iterations for parameter estimates. The Begg and Mazuma rank correlation and Egger’s test were adopted to assess and quantify the publication bias. A sensitivity analysis was performed, and we removed studies one by one to reflect the influence of each study on the pooled OR of the others. In addition, we calculated the classic fail-safe N value using Comprehensive Meta-Analysis software (version 2.0) to quantitatively evaluate the reliability of the results.
Twenty-five, twenty-one and twenty-five studies were identified from PubMed, Web of Science, and Embase, respectively; an additional three studies were identified from references in the included studies (Figure 1). After duplicates were removed, there were forty-eight studies, thirty-nine of which were ineligible. The ineligible records consisted of seventeen other studies, one animal study, three review articles, three family studies, six meeting articles, one meta-analysis of inflammatory bowel disease, two studies without the target SNPs, and six studies aimed at other immune diseases. After retrieving and reviewing the nine remaining studies, we excluded two studies without sufficient data, leaving seven studies to be used for further data extractions (Table 1).
The results of bias assessment are presented in Table 2. Each study was compliant with HWE. All studies had a low risk of bias from population stratification, selective outcome reports, ascertainment of celiac disease and ascertainment of control. The risk of bias was highest in quality control for genotyping and confounding bias (both unclear in 1 study, 14.29%).
The seven studies reported an association between LPP rs1464510 polymorphism and CD, with 14,936 cases and 24,788 controls (Table 3). The pooled OR (A vs. C) showed moderate heterogeneity (p = 0.106, and I2 = 29.52%) across the studies, with a pooled OR of 1.26 (95% CI: 1.22, 1.30) (part A of Figure 2), suggesting that individuals carrying the risk A allele had a 26% higher risk of developing CD than those carrying the C allele. The PAR for risk allele A was 10.72%. The sensitivity analysis suggested that, if we excluded the study by Coenen et al. , I2 was reduced from 29.52% to 11.64% and the pooled odds ratio was 1.27 (95% CI: 1.23, 1.31) (Supplementary Materials Table S3). The Egger test (p = 0.100) and Begg and Mazumdar rank correlation (p = 0.284) suggested that no publication bias existed. Publication bias was also tested using a funnel plot (Supplementary Materials Figure S1). The classic fail-safe N value was 1032 (Z = 14.21; p = 0.00), which suggested that 1032 unpublished negative studies would have to be included to convert the combined p value to a non-significant value.
The genotype frequency and estimated ORs of LPP rs1464510 are presented in parts B and C of Figure 2. The OR1 (AA vs. CC) (p = 0.097; I2 = 30.45%) was moderately heterogeneous, and the OR2 (AC vs. CC) (p = 0.979; I2 = 0.0%) was homogenous. The pooled OR1 (1.58; 95% CI: 1.49, 1.68; p < 0.001) and OR2 (1.26; 95% CI: 1.19, 1.32; p < 0.001) were statistically significant, which indicated that persons with AA and AC genotypes in LPP rs1464510 had an approximately 58% and 26% higher risk, respectively, of developing CD than persons with the CC genotype. The Egger test did not suggest any asymmetry for both ORs (p = 0.133 for OR1, p = 0.054 for OR2). The λ was 0.512 (95% CI: 0.388, 0.660), suggesting that a co-dominant effect was most likely.
The seven studies reported an association between TAGAP rs1738074 polymorphism and CD, with 14,936 cases and 24,788 controls (Table 4). The pooled OR (A vs. G) was 1.17 (95% CI: 1.14, 1.21), estimated by the fixed-effects model (p = 0.974, and I2 = 0.00%) (part A of Figure 3), which suggested that individuals carrying the risk A allele had a 17% higher risk of developing CD than those carrying the G allele. The PAR for risk allele A was 6.59%. The Egger test (p = 0.440) and Begg and Mazumdar rank correlation (p = 0.315) suggested that no publication bias existed. Publication bias was also tested using a funnel plot (Supplementary Materials Figure S2). The classic fail-safe N value was 513 (Z = 10.11; p = 0.00), which suggested that 513 unpublished negative studies would have to be included to convert the combined p value to a non-significant value.
The OR1 (AA vs. GG, 1.37; 95% CI: 1.29, 1.46; p < 0.001) and the OR2 (AG vs. GG, 1.17; 95% CI: 1.11, 1.22; p < 0.001) were homogenous, and estimated by a fixed-effects model in parts B and C of Figure 3. The results can be interpreted as indicating that persons with AA and AG genotypes in TAGAP rs1738074 had approximately 37% and 17% higher risks, respectively, of developing CD than persons with the GG genotype. Egger’s test did not suggest any asymmetry for both ORs (p = 0.425 for OR1, p = 0.611 for OR2). The λ was 0.496 (95% CI: 0.310, 0.711), which suggested that a co-dominant effect was most likely.
Our meta-analysis suggests that both LPP rs1464510 and TAGAP rs1738074 polymorphisms contribute to the susceptibility to CD in European and American.
The pooled OR (A vs. C) of LPP suffered from moderate heterogeneity, but I² decreased significantly (from 29.52% to 11.63%) when we eliminated The Netherlands data from Coenen et al. , indicating that heterogeneity originated mainly from this study. The results between different studies are often heterogeneous, and there are three feasible reasons for such heterogeneity in genetic association studies: association in one population rather than in another, different studies without comparable measures of phenotype, or deviation from HWE . Therefore, we speculate that the main underlying cause of heterogeneity might be populations of various ethnicities.
LPP, which is strongly expressed in the small intestine, participates in the regulation of cell adhesion, cytoskeletal remodeling and maintenance of cell shape and motility [35,36], and it seems to be activated more strongly in biopsy specimens from CD patients than in those from non-CD controls . We infer that mutations in the LPP lead to the PTP1B becoming sufficient or even excessive, so more ERK may be activated, and that it may play a functional role in CD enterocyte proliferation. Our results suggested a powerful relationship between CD and the LPP of rs1464510 (p < 0.001, OR = 1.26, 95% CI: 1.22–1.30).TAGAP is a Rho GTPase-activating protein crucial for modulating cytoskeletal changes [9,11,12], and it is thought to be a negative regulator of cell signaling and relevant to the regulation of the Rho GTPase cycle . Therefore, we hypothesize that mutations in the TAGAP rs1738074 might increase GTPase activity, which propagates the inactive form of the Rho molecule in the Rho GTPase cycle and leads to negative regulation of downstream effects, thus promoting the development of CD. Our meta-analysis confirmed the involvement of rs1738074 in CD susceptibility (p < 0.001, OR = 1.17, 95% CI: 1.14–1.21), so pathway analysis should be implemented to generate hypotheses for clarifying the biological link between TAGAP and CD .
There are some limitations of our study. First, we only included European (38197/39725) and American (1528/39725) populations; nonetheless, our results provide a comprehensive overview of the association between LPP rs1464510/TAGAP rs1738074 and CD in European populations. Second, all included studies were case-control studies, which might have overestimated the genetic association; a population-based nested case-control study is needed to avoid this bias. Finally, because only English-language literature was retrieved, we may have missed relevant articles written in other languages.
In summary, our meta-analysis reveals that both LPP rs1464510 and TAGAP rs1738074 are associated with CD susceptibility. Furthermore, the gene–gene and gene–environment interactions should be evaluated, and studies with larger and more diverse samples should be performed to confirm the results of this meta-analysis.
This work was supported in part by Training Program of the Major Research Plan of the National Natural Science Foundation of China (Grant numbers: 91543132), National Natural Science Foundation of China (Grant numbers: 30901249, 81101267 and 81541070), Guangdong Natural Science Foundation (Grant numbers: 10151063201000036,S2011010002526and 2016A030313089), Guangdong Province Medical Research Foundation (Grant number: A2014374 and A2015310) and Project from Jinan university (Grant number: 21612426, 21615426, JNUPHPM2016001, JNUPHPM2016002).
The following are available online at www.mdpi.com/1660-4601/14/2/171/s1, Figure S1: Funnel plot for LPP rs1464510 (A vs. C) with CD. Figure S2: Funnel plot for TAGAP rs1738074 (A vs. G) with CD. Table S1: Risk of bias assessment for genetic association studies of CD of studies included in the meta-analysis. Table S2: MOOSE checklist: The association of LPP and TAGAP genes with CD risks: a meta-analysis. Table S3: The sensitivity analysis of LPP rs1464510 and CD risk (A vs. C).
Shi-Qi Huang and Na Zhang contributed equally to writing of this paper. Chun-Xia Jing, Guang Yang and Eddy Y. Zeng contributed to study conception and design. Zi-Xing Zhou, Chui-Can Huang, Cheng-Li Zeng, Di Xiao, Cong-Cong Guo, Ya-Jing Han, Xiao-Hong Ye, Xing-Guang Ye, Mei-Ling Ou, Bao-Huan Zhang, Yang Liu performed the part of analysis and the interpretation. All authors approved the final version to be published.
The authors declare no conflict of interest.