Pooling-analysis on hMLH1 polymorphisms and cancer risk: evidence based on 31,484 cancer cases and 45,494 cancer-free controls

To elucidate the veritable relationship between three hMLH1 polymorphisms (rs1800734, rs1799977, rs63750447) and cancer risk, we performed this meta-analysis based on overall published data up to May 2017, from PubMed, Web of knowledge, VIP, WanFang and CNKI database, and the references of the original studies or review articles. 57 publications including 31,484 cancer cases and 45,494 cancer-free controls were obtained. The quality assessment of six articles obtained a summarized score less than 6 in terms of the Newcastle-Ottawa Scale (NOS). All statistical analyses were calculated with the software STATA (Version 14.0; Stata Corp, College Station, TX). We found all the three polymorphisms can enhance overall cancer risk, especially in Asians, under different genetic comparisons. In the subgroup analysis by cancer type, we found a moderate association between rs1800734 and the risk of gastric cancer (allele model: OR = 1.14, P = 0.017; homozygote model: OR = 1.33, P = 0.019; dominant model: OR = 1.27, P = 0.024) and lung cancer in recessive model (OR = 1.27, P = 0.024). The G allele of rs1799977 polymorphism was proved to connect with susceptibility of colorectal cancer (allele model: OR = 1.21, P = 0.023; dominate model: OR = 1.32, P <0.0001) and prostate cancer (dominate model: OR = 1.36, P <0.0001). Rs63750447 showed an increased risk of colorectal cancer, endometrial cancer and gastric cancer under all genetic models. These findings provide evidence that hMLH1 polymorphisms may associate with cancer risk, especially in Asians.


INTRODUCTION
As one of the pivotal pathways in maintaining genetic stability, MMR system is mainly in charge of repairing the replication-associated errors, including removing mistaken bases, correcting substitutions and rectifying insertion-deletion mismatches. Its defects may result in microsatellite instability (MSI), a type of genetic instability related to colorectal cancer, gastric cancer, and endometrial cancer, etc. [1][2][3] Interest in MLH1 has increased in the last few years because MLH1 was discovered as a key component in MMR for MSI, and its dysfunction is supposed to be implicated in cancer predisposition.
MLH1 not only takes part in the activities of MMR system, but also has other interesting cellular functions, such as participating in cell cycle arrest, triggering DNA damage-induced apoptosis to response to some chemical www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 54), pp: 93063-93078 Research Paper or physical agents [4], and interacting with tumorrelated signaling molecules like BRCA1 [5] and p53 [6]. Moreover, various polymorphisms were found in MLH1 gene, part of them were proved to influence the expression of functional MLH1. We selected three most common loci rs1800734, rs1799977, and rs63750447 in hMLH1 which may alter the function of the hMLH1 gene according to literature. Among these, the A allele of rs1800734 polymorphism could alter the methylation level of nine CpG sites mapped on the MLH1 promoter [7], while rs1799977 and rs63750447 were situated at the exons of hMLH1 [1,8]. Emerging inspiring evidences indicate these functional polymorphisms of hMLH1 may be potential candidates in mediating hereditary susceptibility to cancer, however, applying them in clinical application is still treated critically. Past decades witnessed numerous molecular epidemiological studies carried out worldwide to investigate the actual association between them, yet no coincident conclusion was reached so far.
For example, Nizam et al. [9] concluded that rs1800734 polymorphism had an influence on colorectal cancer (CRC) risk among Malaysians in 2013, while Zhang et al [10] found no obvious connection between rs1800734 and CRC risk in 2016. For rs1799977 polymorphism, Milanizadeh et al. [11] detected it could increase CRC risk particularly in female patients, but Peng et al. [1] hold a contrary opinion that no association existed between the two. The inconsistent conclusions also existed in the studies exploring the relationship between hMLH1 polymorphisms and other cancer types. Although rs63750447 polymorphism was accepted as a risk factor for east-Asian CRC patients [1,[12][13][14], no reliable conclusion reported on the possible relationship between rs63750447 and overall cancer or other kinds of tumors. To solved these controversies, a comprehensive and persuasive meta-analysis was excepted to conduct depending on complete published data and proper methodological tools, thus we carried out this metaanalysis to illuminate the objective connection between hMLH1 polymorphisms (rs1800734, rs1799977 and rs63750447) and cancer risk.

Characteristics of eligible studies
Finally, we obtained a total of 57 publications including 31,484 cancer cases and 45,494 cancer-free controls (all were from the databases and no study was identified by manual search of the references of the original studies or review articles). The detail selection process was shown in the flow diagram ( Figure 1). What needed illustration is that we abandon three studies contained in previous meta-analyses after comprehensive reading full text. The first one was the study performed by Chen et al [15], contained in the meta-analyses conducted in 2011 [16] and 2015 [17], which was excluded on account of both its cases group and controls group are women with cancers (cases with MLH1 methylation while controls not). Another study finished by van Roon et al. [18], also included in previous meta-analyses [17,19], has two controls groups collected from literature [20,21]. We excluded it after discussing with a senior author within us. And the third study we abandoned was due to deficiency of cancer-free control group [16].
Among the 57 eligible literatures, 26 were based on Caucasian background from, Poland, Spain, the United States, Denmark, the United Kingdom, Sweden, Portugal, Czech Republic and Canada. 27 were carried out in Asians from China, Kazakhstan, India, Iran, Malaysia, Japan and Korea, and four were based on mixed ethnic groups. All the publications involving rs63750447 polymorphism were carried out among the Chinese population. Three casecohort designed studies [22][23][24] and 54 case-controlled studies were involved in this meta-analysis. All cancer cases were confirmed by pathology or histology, involved cancer types covering colorectal, gastric, ovarian, head and neck, endometrium, lung, bladder, prostate, thyroid, breast, prostate, Non-Hodgkin lymphoma, acute myeloid leukaemia, and acute lymphoblastic leukaemia. The quality assessment of six studies obtained a summarized score less than 6 in terms of the Newcastle-Ottawa Scale (NOS), four of them are studying on rs1800734 [25][26][27][28] while one of them is for rs63750447 [29], and the other one focused on rs1800734 and rs1799977 polymorphisms [30]. Specially, two publications by Zhang et al. [8] and Wang et al. [29] contained four and three independent studies respectively. One study focused on rs1799977 polymorphism by Joshi et al. [31] did not provide complete genotype frequencies. Hence only the dominant model was evaluated. Detail characteristics of eligible publications are displayed in Table 1.

Quantitative synthesis
The distributions of genotypes frequencies of hMLH1 polymorphisms (rs1800734; rs1799977; rs63750447) for every single study are exhibited in Table  2. The minor allele frequencies (MAF) among cancer cases varied widely according to the included studies, ranging from 0.205 to 0.656 for rs1800734 polymorphism, 0.016 to 0.744 for rs1799977 polymorphism, and 0.032 to 0.069 for rs63750447 polymorphism. The average MAF of case-group for the three polymorphisms is 0.396, 0.233, 0.053, respectively. The meta-analysis results of these three polymorphisms were shown in Supplementary  Table 1.
When the subgroup carried out by ethnicity ( Figure  3B), a significant association was observed between rs1799977 and cancer risk among Asians in four genetic models (dominant comparison: OR = 1. 52 Figure 2C. The subgroup analysis by cancer type ( Figure  4B When we conducted the subgroup analysis by quality score, there was a significantly increased cancer risk for rs63750447 polymorphism in both high-quality studies and low-quality studies (shown in Supplementary Table 1).

Test of heterogeneity and sensitivity analysis
As shown in Supplementary Table 1, significant heterogeneities existed after pooled the data of rs1800734 and rs1799977 polymorphisms under different comparison models (P ≤ 0.10 or I 2 ≥ 50%), thus further subgroup analyses base on ethnicity, cancer type, source of control, and quality scores were performed. No obvious heterogeneity was found for rs63750447 polymorphism (P > 0.10 or I 2 < 50%). Subsequent sensitivity analysis proved the stability of our study, since no significant alteration was detected after removing each individual study and rechecking the pooled ORs and 95%CIs for the rs1800734 and rs1799977 polymorphisms ( Figure 5A, 5B). The third study performed by Zhang et al seemingly altered the pooled ORs significantly (Figure 5), and the detailed data from Stata 14.0 also showed us it was nearly approached to the upper limit. We guess it was due to the sample size of rs63750447 polymorphism was insufficient, only 11 studies from 6 articles were included. It indicated us the overall results of rs63750447 should be treated more carefully.

Publication bias
The possible publication bias in the eligible literature was evaluated by Egger's test and funnel plots. As shown in Figure 6, the Begg's funnel plots appear to be symmetrical. This symmetry was then confirmed by the statistical results of Egger's test (P > 0.05, shown in Table  3). These provided evidence for the absence of publication bias.

DISCUSSION
To elucidate the veritable relationship between three hMLH1 polymorphisms (rs1800734; rs1799977; rs63750447) and cancer risk, we performed this metaanalysis based on overall published data up to May 2017. We found all of these polymorphisms can enhance overall cancer risks, especially Asians, under different genetic comparisons (Supplementary Table 1). Further subgroup analyses were carried out according to cancer type, source of control, quality score, and study design, and results worth discussing were obtained.
Interestingly, we found a moderate association existing between rs1800734 and the risk of gastric cancer in three genetic models (OR = 1.14, P = 0.017; OR = 1.33, P = 0.019; OR = 1.27, P = 0.024) and lung cancer in recessive model (OR = 1.27, P = 0.024), while no connection was display with colorectal cancer. As far as we know now, microsatellite instability (MSI) often occurs when mismatch errors failed to be corrected or hMLH1 gene was epigenetic silencing. Campbell et al. [41] found rs1800734 polymorphism enhanced MSI-positive colorectal cancer, the association was proved by Mrkonjic et al. [42] due to the effects of rs1800734 on the MLH1 promoter methylation, immunohistochemistry (IHC) deficiency, or both. This indicated us when performing further studies focused on the relationship between rs1800734 and cancer risk, the MSI-statue of cancer patients should be evaluated fundamentally.
Rs1799977 was a nonsynonymous coding polymorphism in hMLH1, which leaded to an amino acid change from isoleucine to valine. The mutational G allele of rs1799977 polymorphism was proved to connect with susceptibility of colorectal cancer and prostate cancer. For rs63750447, the cancer-specific analysis showed an increased risk of colorectal cancer, endometrial cancer and gastric cancer. Recently, rs63750447 was observed overexpressed in patients with EGFR-TKI (epidermal growth www.impactjournals.com/oncotarget  rs1799977; (C) rs63750447). HWE: Hardy-Weinberg equilibrium. www.impactjournals.com/oncotarget factor receptor-tyrosine kinase inhibitor) resistance, which has a possible shorter progression-free survival [43]. Thus, it was speculated that MLH1 might be involved in EGFR signaling or other pathways (such as proliferation and survival) [1].
Compare with previous meta-analyses study on the association between hMLH1 and cancer risk, our study included a larger sample size and performed more detailed stratification analysis. Besides, our study has stricter inclusion criteria and exclude criteria, thus avoided omissive and false drop (refer to the section of Characteristics of eligible studies, paragraph one). Thus, we think our results are more reliable and convinced.
Moreover, we found rs1800734 was related to gastric cancer, while rs1799977 may have an influence on colorectal and prostate cancer. It may give us some hints for the further study.
There are still some limitations existing in this metaanalysis. Firstly, insufficiency of original data limited us to proceed more accurate analyses on the potential interaction between these polymorphisms and other risk factors such as age, sex, hereditary background, lifestyle, and MSI status, etc. Secondly, the studies involved in the rs63750447 analysis was insufficient, whose statistical significance was needed to verify by further well-designed study with larger sample sizes. Thirdly, we couldn't

MATERIALS AND METHODS
PRISMA statement was used to guide the process of this meta-analysis [44].

Search strategy
A comprehensive literature search was conducted using the following search terms: ("cancer", "carcinoma", "tumor", "tumour", or "neoplasm") and ("polymorphism", "variation", "variant", or "mutation") and ("hMLH1"). The PubMed, Web of knowledge, VIP, WanFang and Chinese National Knowledge Infrastructure (CNKI) databases were searched up to May, 2017. Additional studies were identified by manual search of the references of the original studies or review articles. This study was approved by the ethics committee of Xi'an Jiaotong University.
To be eligible for this meta-analysis, the included study was required to (1) be case-control or case-cohort studies; (2) focused on the relationship between hMLH1 polymorphisms and risk of any cancer; (3) have at least three articles for each studied hMLH1 polymorphism, and available information concerning the genotype frequency of each included SNP of hMLH1 (i.e., rs1800734; rs1799977; rs63750447); (4) be published in English or Chinese. The exclusion criteria were as follows: (1) studies were not focused on cancer risk or targeted hMLH1 SNPs (rs1800734; 2: rs1799977; 3: rs63750447); (2) studies failed to supply any data on genotype distribution, (3) studies were updated by a following study where a larger number of subjects were included, (4) studies were designed as a case-case or case-only study. If 2 or more studies contained overlapping data, we selected the paper included more samples. Studies containing two or more case-control groups were considered as two or more independent studies.

Data extraction and quality assessment
For each included study, two investigators independently extracted the raw data and demographic information, including publication year, first author, ethnicity and country or origin, the number of cases and controls, source of controls, genotyping methods, genetic distribution, and P value of Hardy-Weinberg equilibrium (HWE) among the controls. Studies not follow HWE were excluded in subgroup analysis. We applied the Newcastle-Ottawa Scale (NOS) to evaluate the methodological quality of the eligible studies according to Zeng et al [45]. Accumulated score ranges from 0 to 9 points, and a score of 0-5 and 6-9 is considered to suggest a low and high quality respectively, with higher quality representing lower risks of bias. A discussion or consultation with a senior author was conducted to settle controversy until a consensus was reached.

Statistical analysis
To evaluate the strength of association between hMLH1 polymorphisms (rs1800734; rs1799977; rs63750447) and cancer risk, we calculated the odds ratios (ORs) and 95% confidence intervals (CIs) based on the genotype and allele frequencies in cases and controls of each eligible study. We used the Z test to access the significance of all pooled ORs and it was considered statistically significant if the P value < 0.05. The Chisquare-based Q statistic test and I 2 statistic were applied to examine the statistical heterogeneity among studies. When no obvious heterogeneity existed across the studies (P>0.10 or I 2 <50%), we pooled the ORs using fixed-effect model (Mantel-Haenszel); otherwise, the random effects model (DerSimonian and Laird) was chosen. The potential publication bias was evaluated by funnel plot and Egger's test. To access the stability of the results in this meta-analysis, we performed sensitivity analysis by sequentially excluding each study and rechecked whether the pooled ORs were altered significantly.
The following genetic models were evaluated: allele comparison (B vs. A), homozygote comparison (BB vs. AA), heterozygote comparison (AB vs. AA), recessive model (BB vs. AA+ AB), and dominant model (BB+ AB vs. AA). "A" represents the wild allele, while "B" represents the mutation allele. After excluded studies not according to HWE, we conducted the subgroup analysis based on ethnicity (divided into Asian and Caucasian), cancer type, and source of control. All statistical analyses were calculated with the software STATA (Version 14.0; Stata Corp, College Station, TX).