ERCC2 polymorphisms and radiation-induced adverse effects on normal tissue: systematic review with meta-analysis and trial sequential analysis

Background The relationship between ERCC2 polymorphisms and the risk of radiotoxicity remains inconclusive. The aim of our study is to systematically evaluate the association between ERCC2 polymorphisms and the risk of radiotoxicity. Methods Publications were identified through a search of the PubMed and Web of Science databases up to August 15, 2015. The pooled odds ratios (ORs) with corresponding 95 % confidence intervals (CIs) were calculated to evaluate the association between ERCC2 polymorphisms and radiotoxicity. Trial sequential analysis (TSA) and power calculation were performed to evaluate the type 1 and type 2 errors. Results Eleven studies involving 2584 patients were ultimately included in this meta-analysis. Conventional meta-analysis identified a significant association between ERCC2 rs13181 polymorphism and radiotoxicity (OR = 0.71, 95 % CI: 0.55-0.93, P = 0.01), but this association failed to get the confirmation of TSA. Conclusions The minor allele of rs13181 polymorphism may confer a protect effect against radiotoxicity. To confirm this correlation at the level of OR = 0.71, an overall information size of approximate 2800 patients were needed. Electronic supplementary material The online version of this article (doi:10.1186/s13014-015-0558-6) contains supplementary material, which is available to authorized users.


Introduction
Radiotherapy is commonly used in cancer treatment. At least half of cancer patients will require radiotherapy with either curative or palliative intent [1]. However, the adverse effect induced by radiotherapy restricts this modality from playing a larger role in the multidisciplinary therapy of cancer. It has been widely noticed that patients were not homogeneous in the reaction of normal tissue following radiotherapy [2,3]. The standard radiotherapy schedule was recommended treating the cancer patients as a whole, which was actually miscellaneous with patients of different radio-sensitivity. So the radio-resistive patients who can bear more doses of radiotherapy were confined in the protocol of standard radiotherapy with the radio-sensitive patients who may even fail the cost-benefit evaluation of receiving radiotherapy. Consequently, the likelihood of a cure was to some extent reduced for some patients. On the other hand, for some others, the standard radiotherapy was still too harmful to the balance between the therapeutic effect and the normal tissue injury. It is believed that a genetic basis plays an important role in this heterogeneous response to radiation [4,5].
'Radiogenomics' is the study of genetic variation associated with response to radiotherapy, with a main purpose of establishing single nucleotide polymorphism (SNP) based risk models that can stratify patients according to radio-sensitivity [6,7]. In the last decade, candidate gene association studies have identified several potential predictors for radio-sensitivity. Due to the insufficient sample size of these studies and relatively small effects conferred by relevant SNPs, it makes much sense to systematically synthesize the previous evidence. Recently, we have reviewed several SNPs in radiogenomics, including ATM [8], XRCC3 [9] and XRCC1. The aim of the present meta-analysis is to evaluate the effect of excision repair cross-complementing 2 (ERCC2, also known as XPD, xeroderma pigmentosum group D ) polymorphisms on radiotoxicity.

Search strategy and inclusion criteria
Two investigators independently searched the PubMed, and Web of Science databases through a comprehensive search strategy including the terms "SNP", "ERCC2/ XPD", "radiotherapy" and "toxicity" (see Additional file 1 for the specific strategy). The searching result was last updated on August 15, 2015. Studies satisfying the following criteria were eligible for inclusion: (1) case-control study or cohort study; (2) evaluated the effect of ERCC2 polymorphisms on radiotoxicity; (3) adequate information provided to calculate the odds ratio (OR) and the corresponding 95 % confidence interval (CI). There were no limitations on the language of publication. To avoid exaggerating the representation of certain study, the effect of each study should be cumulated only one time in the synthetical result. If a single study evaluated multiple toxicities, we included the most common toxicity to reduce the potential heterogeneity between studies.

Data extraction
Two investigators independently extracted data from each included study using a standard data collection form. The following information was extracted: first author's surname, year of publication, country of origin, ethnicity, cancer type, radiotoxicity, assessment criteria, and ORs with the corresponding 95 % CI for the association between ERCC2 polymorphisms and radiotoxicity. Study authors were contacted when the data provided was insufficient. Disagreements were resolved by discussion among all investigators.

Statistical methods
The pooled OR and 95 % CI were calculated to assess the strength of the association between ERCC2 polymorphisms and the risk of radiotoxicity. Meta-analysis was first performed on each included polymorphism under 3 genetic comparison models: dominant model (mutant homozygote/heterozygote vs. wild type), recessive model (mutant homozygote vs. wild type/heterozygote) and allelic model (minor allele vs. common allele). Subsequently, subgroup analyses were conducted by adverse effect. The heterogeneity between studies was assessed with the chi-squared based Q-test and I 2 statistics [20,21]. When the chisquared P was <0.10 or the I 2 statistic was ≥50 %, the heterogeneity was considered statistically significant, and a random-effects model (DerSimonian-Laird method) was applied [22]; otherwise, a fixed-effects model (Mantel-Haenszel method) was accepted [23]. Sensitivity analysis was performed to evaluate the stability and reliability of the pooled results by excluding each study individually and reanalyzing the remaining studies. Publication bias was evaluated via Begg's funnel plot and Egger's test [24,25]. If publication bias existed, the "trim and fill" method was applied to estimate the number of missing studies and to adjust the pooled result [26]. Statistical power calculation was performed to evaluate the potential type 2 errors for the primary and subgroup analyses. A two-sided P <0.05 was considered significant for all the analyses except the heterogeneity tests. Meta-analyses were performed using Stata (Version 14.0, StataCorp LP, College Station, TX, USA).
Trial sequential analysis (TSA) was performed for the analyses involving more than six studies. A two sided α = 0.05, β = 0.2 and a relative risk reduction (RRR) of 10 % were used. A required information size was estimated with the adjustment by diversity (D 2 ) between studies [27,28]. An α-spending boundary and a futility boundary were constructed accordingly. The cumulative z-curve was produced by plotting a series of z-values of cumulative metaanalyses [29]. If the accrued number of included patients surpassed the required information size, it means that a sufficient level of evidence has been reached. If the curve crossed the α-spending boundary, it was considered that the conclusion of significant association was confirmed. When the z-curve crossed the futility boundary, the conclusion of indiscrimination between two groups was accepted under the given conditions [30]. TSA was performed using trial sequential analysis software version 0.9 beta [29].

Eligible studies
A flow diagram summarizing the literature review process and reasons for exclusion is presented in Fig. 1. A total of 11 studies involving 502 cases and 2082 controls were eventually included in this meta-analysis. The baseline characteristics of the included studies are presented in Table 1. The studies were published from 2005 to 2013, and the sample sizes ranged from 60 to 698. Three SNPs of ERCC2 were involved in the present meta-analysis: ERCC2 Lys751Gln (rs13181) (nine studies), Asp312Asn (rs1799793) (five studies) and Asp711 (rs1052555) (two studies). Cancer categories included breast cancer (six studies), lung cancer (one study), esophageal cancer (one study), prostatic carcinoma (one study), bladder cancer (one study) and rectal cancer (one study).

Meta-analysis results
Significant associations between rs13181 and risk of radiotoxicity were identified in dominant model (OR = 0.71, 95 % CI: 0.55-0.93, P = 0.01) and allelic model (OR = 0.78, 95 % CI: 0.64-0.97, P = 0.02) by conventional meta-analysis (Fig. 2, Table 2). TSA suggested that more than half (1486/2804) of the required information size has been accrued (Fig. 3). The cumulative z-curve (dominant model) has transcended the conventional significance boundary, which was in accordance with the result of conventional meta-analysis. No trial sequential monitoring boundary was further crossed, leaving the meta-analysis inconclusive of a 10 % relative risk reduction. For rs1799793 and rs1052555 polymorphisms, no significant association was identified.

Heterogeneity and sensitivity analyses
The heterogeneities between studies of all analyses were not significant. The pooled results present stable in the sensitivity analysis (Fig. 4). The study Terrazzino 2012 possesses the greatest influence to the direction favors wild type, while Zhang 2010 owns the greatest influence to the direction of favors variant type. After omitting the

Publication bias
The distribution of all the included studies of rs13181 in Begg's funnel plot was visually symmetrical (Additional file 1: Figure S1), and the p-value of egger's test was 0.891, which indicated that no obvious publication bias exists in our meta-analysis of rs13181.

Discussion
Radiogenomics has entered the era of big data [31]. However, for the last decade candidate gene approach was predominant, and inconsistent results have been reported due to most studies were underpowered with a relatively small information size. Besides, some single SNP may confer only slightly elevated risk of radiotoxicity, so it is difficult to identify this true effect without an enough sample size. Hence, systematically summarizing the previous data benefits of identifying a relatively small but significant effect of relevant SNPs. In fact, meta-analysis has played an important role in radiogenomics. The present meta-analysis systematically summarized the previous data of ERCC2 in radiogenomics. A significant association between rs13181 and radiotoxicity was identified by conventional meta-analysis. Our data revealed that the major allele of rs13181 presents as a risk allele, which means the minor allele confers a protect effect against the appearance of radiotoxicity. However, we should notice that this association was still borderline (95 % CI: 0.55-0.93), and one study made the original significance vanish in leave-one-out sensitivity analysis. In addition, this conventional significance failed to get the confirmation of TSA. We applied TSA with the intention of drawing out more specific conclusions. TSA revealed that the z-curve failed to further cross the α-spending boundary after crossing the conventional boundary of z = −1.96, which means the correlation between rs13181 polymorphism and radiotoxicity risk still need the confirmation of subsequent studies. To safely conclude an effect of OR = 0.71, additional 1400 patients were needed.
We performed subgroup analyses by adverse effect. Radiation-induced adverse effects can be classified as early or late effects according to the time before the manifestation of relevant clinical symptoms. Most radiogenomics studies reported early and late effects separately, and some SNPs exert inconformity effect on early and late effects. For instance, the pooled data of XRCC1 revealed that Arg399Gln (rs25487) polymorphism significantly correlated with an elevated risk of early radiotoxicity, while this SNP was ruled out any clinical relevance with late radiotoxicity (Song YZ: The XRCC1 Arg399Gln Polymorphism and Radiation-Induced Adverse Effects on Normal Tissue: Systematic Review with Meta-analysis and Trial Sequential Analysis. Submitted). Meta-analysis of XRCC3 Thr241Met (rs861539) polymorphism also reported a similar result, that is a significant association with early radiotoxicity rather than late radiotoxicity [9]. In the present metaanalysis, most studies evaluated early radiotoxicity, and only one study involved late radiotoxicity, so it is noteworthy that the pooled result of the present meta-analysis mainly reflects the effect on early radiotoxicity. Within early radiotoxicity, a significant association between rs13181 polymorphism and acute esophageal toxicity was also identified. However, this association should not be over interpreted, due to only two studies were included in this subgroup. The most evaluated radiotoxicity was acute skin toxicity in the present meta-analysis, and four studies evaluated this reaction following radiotherapy. The association between rs13181 polymorphism and acute skin toxicity was not significant.
For interpreting the pooled result of a meta-analysis, adequate evaluation on the heterogeneity between studies is crucial (as was previously described in detail [9]). One of the most important source of heterogeneity derived from the heterogeneous treatment protocols among the included studies. For instance, the radiotherapy parameters, such as total dose, dose per fraction and irradiation volume, were not identical among the included studies. Some treatment protocols of included studies involved chemotherapy as a component of multidisciplinary therapy, while some were basing on radiotherapy alone. Toxicity evaluation was another important source of heterogeneity. Both the criteria applied and the division grade were not consistent among included studies. Nevertheless, statistical calculations did not identify obvious heterogeneity, we believed that the influence of these potential heterogeneity factors were at an acceptable level.
Based on the radiogenomics studies throughout the last decade, it is safe to conclude that no SNP alone possesses the power to accurately predict the radio-sensitivity prior radiotherapy [6,31]. A study aiming to validate the associations previously reported between candidate SNPs and radiotoxicity did not confirm any significant association [32]. To date, six genome-wide association studies (GWASs) have been published on radiogenomics [33][34][35][36][37][38]. The SNPs which were identified with a genome-wide significance were not located in the region supposed by candidate gene association studies. However, statically significant conclusions were constantly reported by meta-analyses of candidate gene association studies. Through combining information of all the relevant studies, more statistical power was acquired [39]. Evidence of meta-analysis has revealed that XRCC1 rs25487, XRCC3 rs861539, ATM rs1801516 and ERCC2 rs13181 polymorphisms significantly associated with early radiotoxicity, though the effect size was relatively small [8,9]. While an individual candidate SNP was not expected to confer a large effect on radiotoxicity. Instead, composing a synthesized risk model is the major modality how the relevant SNPs play a role on the prediction of radio-sensitivity. Despite a relatively small effect (with odds radios of 1.2 to 1.5) exerted on radiotoxicity by individual SNP, an enough prediction power can be accrued by involving multiple such SNPs.

Conclusions
Although the minor allele of rs13181 polymorphism was identified with a protect effect against radiotoxicity, it is noteworthy that the correlation was borderline, and one included study made the overall meta-analysis loss the statistical significance in leave-one-out sensitivity analysis. In addition, this significant association identified by traditional meta-analysis failed to get the conformation of TSA. More studies with additional 1400 patients were needed to draw the firm conclusion at the level of OR = 0.71.

Competing interests
The authors declare that they have no competing interests.
Authors' contributions DLH conceived of the study, participated in its design and critically reviewed the manuscript. SYZ, DMN and ZYY performed publication screening & data extraction and drafted the manuscript. SWY and XCC performed the statistical analysis and participated in the design of the study. All authors read and approved the final manuscript.