Comprehensive assessment of the association between XPC rs2228000 and cancer susceptibility based on 26835 cancer cases and 37069 controls

The human XPC (xeroderma pigmentosum complementation group C) gene is located on chromosome 3p25 and contains 16 exons and 15 introns [1,2]. The human XPC protein with 940 amino acids, encoded by XPC, serves as an essential member within the NER (nucleotide excision repair) pathway [3–5]. The XPC protein is important for the early damage site recognition and DNA repair initiation of NER [3,6,7]. The abnormal expression of the XPC protein was also reportedly linked to the progression of the cancer [3,8].

Within the XPC gene, three common variants, including rs2228000 (C21151T) of exon 8, rs2228001 (A33512C) of exon 15, and poly-AT insertion/deletion polymorphism (PAT^−/+) of intron 9, were identified [4,9–11]. XPC rs2228000 results in a substitution of alanine for valine in position 499 (Ala499Val), while rs2228001 leads to a transversion from lysine to glutamine in position 939 (Lys939Gln) [4,9–11]. The present study investigated the potential genetic role of nonsynonymous XPC rs2228000 in the risk of different clinical types of cancer by pooling published studies with inconclusive conclusions.

After retrieving these studies, only three previous meta-analyses with no more than 15 studies in 2008 [12–14] and one meta-analysis with 33 studies in 2013 [15] were performed to assess the genetic association of XPC rs2228000 and the risk of overall cancer. Thus, we enrolled more sample sizes (71 case–control studies) and utilized different analysis strategies for an updated comprehensive evaluation in 2019 through meta-analysis and TSA (trial sequential analysis).

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) was utilized for our pooling analysis. In September 2019, we used a series of search terms (shown in Supplementary Table S1) to retrieve from four databases [PubMed, Embase, CChia National Knowledge Infrastructure (CNKI)) and WOS (Web of Science)] to obtain potentially relevant articles. We also designed a group of criteria for the inclusion/exclusion and eligibility assessment of the article. Inclusion criteria were the following: (1) case/control studies; (2) cancer; (3) XPC rs2228000; and (4) genotypic frequency data within both the case and control groups. Exclusion criteria were the following: (1) review; (2) meeting abstract; (3) case reports or family data; (4) meta-analysis; (5) cell, mice, horse, or other species; (6) other gene, disease or variant; (7) lack of specific data; (8) lack of normal group; (9) not in line with HWE (Hardy–Weinberg equilibrium); and (10) cohort.

We extracted some basic information, including author name, publication year, country, race, genotypic frequency, cancer type, control source, genotyping assay, and sample size, from the selected eligible case–control studies. The P-value of HWE based on the genotypic distribution in the control group was calculated.

We utilized two approaches, including the NOS (Newcastle–Ottawa quality assessment scale) system (Supplementary Table S2) [16,17] and the risk-of-bias score system (Supplementary Table S3) [18,19] for the assessment of article quality. The article with an NOS score > 5 and a risk-of-bias score > 9 was considered to be high quality.

We used STATA software (Stata Corporation, U.S.A.) to perform the association test in the overall and subgroup meta-analysis, heterogeneity assessment, Begg’s/Egger’s tests (for the publication bias evaluation) and sensitivity analysis (for data stability assessment) [16,17]. The OR (odds ratio), 95% CI (confidence interval) and P-value in a series of association tests under the five genetic models, including T vs. C (allele), TT vs. CC (homozygote), CT vs. CC (heterozygote), CT+TT vs. CC (dominant), and TT vs. CC+CT (recessive), were obtained. In addition, six factors, including race, country, control source, article quality, genotyping assay, and cancer type, were considered in our subgroup analysis.

The high heterogeneity was considered when the I² value in the I² test was larger than 50% and the P-value in the Q statistical test was less than 0.05, which led to the use of the DerSimonian–Laird method of the random-effect model. If not, a Mantel–Haenszel method of a fixed-effect model was used for the relatively low heterogeneity between studies.

Targeting the positive findings, we also calculated the false-positive report probability (FPRP) and statistical power, as suggested by Wacholder et al. [20]. During analysis, an FPRP cut-off value of 0.2, a power OR of 1.5, and different prior probability levels (0.25, 0.1, 0.01, 0.001, 0.0001) were established. After assessing the research status regarding the association between XPC rs2228000 and cancer risk and referencing the similar publications [21,22], the FPRP value of the positive results less than 0.2 under the prior probability level of 0.1 indicates a noteworthy outcome.

We also performed the TSA test to evaluate whether further research was needed, referring to some similar publications [23–26]. For the TSA parameter, a type I error probability of 5%, a statistical test power of 80%, and a low bias-based risk ratio reduction were established. Trial Sequential Analysis Viewer software (http://www.ctu.dk/tsa/) was utilized.

In total, we obtained 1186 potential eligible articles [PubMed (n=266), Embase [n=687], CNKI (n=28), and WOS (n=205)] and then ruled out another 412 duplicates and 646 improper articles according to our exclusion criteria (detailed information listed in Figure 1). Furthermore, we excluded 64 articles due to the question of ‘lack of specific data or normal group’, ‘not in line with HWE’ or ‘cohort’. Finally, we identified a total of 71 eligible case–control studies from the 64 retrieved articles [1,2,4,10,11,27–85] for pooling analysis. We summarized some basic information in Table 1 and presented the flow chart in Figure 1. All the genotypic distribution of the control group in all studies followed the principle of HWE. Although the NOS scores in all studies were larger than 5 (Supplementary Table S2), the risk-of-bias scores of nine articles (Supplementary Table S3) were less than 9.

As shown in Table 2, our overall meta-analysis included a total of 71 studies with 26835 cases and 37069 controls. We observed high between-study heterogeneity (Table 2, all I² > 50%, P_{heterogeneity}<0.001) and thus utilized the random-effect model for the pooling analysis. After pooling the different studies together, we only detected an increased risk of overall cancers under the TT vs. CC+CT model [Table 2, P_association=0.023, OR = 1.11, 95% CI = (1.01, 1.22)] but not other models (all P_association>0.05). These results indicated that XPC rs2228000 does not seem to be statistically associated with susceptibility to cancer.

Next, we performed a series of subgroup analyses by the factors of race, control source, country, article quality, and genotyping assay. As shown in Table 3, a total of 38 studies (12118 cases/18124 controls) were included for the subgroup analysis of ‘Asian’, while 22 studies with 9371 cases and 12338 controls were included for the ‘Caucasian’ subgroup. We did not observe a significant difference between cancer cases and negative controls under the most genetic models (Table 3, P_association>0.05), only apart from the Asian subgroup under the TT vs. CC+CT model [P_association=0.005, OR = 1.13, 95% CI = (1.04, 1.23)]. Within the subgroup analysis by the factor or control source (PB/HB (population/hospital-based)), an increased risk of cancer was only detected in the ‘HB’ subgroup under TT vs. CC [Table 3, P_association=0.010, OR = 1.17, 95% CI = (1.04,1.32)] and TT vs. CC+CT [P_association=0.006, OR = 1.18, 95% CI = (1.05, 1.32)] models but not others (P_association>0.05). Similarly, we observed negative results in the majority of the subgroup analyses by country, article quality and genotyping assay (Supplementary Table S4). As examples, we presented the forest plots of the subgroup analysis data by the factor of race (Figure 2), control source (Supplementary Figure S1), country (Supplementary Figure S2), article quality (Supplementary Figure S3), and genotyping assay (Supplementary Figure S4) under the T vs. C model.

Additionally, we performed a subgroup analysis using the specific cancer type. As shown in Table 4, in the subgroup of ‘bladder cancer’ with 3460 cases and 3613 controls, enhanced susceptibility was detected in bladder cancer cases under T vs. C [Table 4, P_association=0.004, OR = 1.25, 95% CI = (1.07, 1.45)], TT vs. CC [P_association=0.001, OR = 1.68, 95% CI = (1.25, 2.26)], CT+TT vs. CC [P_association=0.016, OR = 1.26, 95% CI = (1.04, 1.53)], TT vs. CC+ CT [P_association= 0.001, OR = 1.49, 95% CI = (1.18, 1.90)] compared with the negative controls. Additionally, there is an increased risk of breast cancer under T vs. C [Table 4, P_association=0.018, OR = 1.11, 95% CI = (1.02, 1.21)], TT vs. CC [P_association=0.003, OR = 1.33, 95% CI = (1.10, 1.60)], and TT vs. CC+ CT [P_association= 0.001, OR = 1.29, 95% CI = (1.12, 1.48)]. Nevertheless, we observed a decreased risk of gastric cancer in the Chinese population under T vs. C [Table 4, P_association=0.020, OR = 0.92, 95% CI = (0.85, 0.99)], CT vs. CC [P_association=0.001, OR = 0.83, 95% CI = (0.73, 0.93)], CT+TT vs. CC [P_association=0.003, OR = 0.84, 95% CI = (0.76, 0.94)]. The relevant forest plots under different genetic models are presented in Figure 3 (T vs. C), Supplementary Figure S5 (TT vs. CC), Supplementary Figure S6 (CT vs. CC), Supplementary Figure S7 (CT+TT vs. CC), and Supplementary Figure S8 (TT vs. CC+ CT).

Moreover, we performed subgroup analysis data for different system cancers. As shown in Supplementary Table S5 and Figure S9 (forest plot data under the allelic model), we observed the same result in the subgroup of ‘urinary system cancer’ as the subgroup of ‘bladder cancer’. There is a reduced cancer risk in the subgroup of ‘reproductive system cancer’ under the models of CT vs. CC [P_association=0.006, OR = 0.81, 95% CI = (0.70, 0.94)] and CT+TT vs. CC [P_association=0.041, OR = 0.82, 95% CI = (0.68, 0.99)] and an increased risk in the subgroup of ‘head and neck cancer’ under the TT vs. CC+CT [P_association=0.024, OR = 1.58, 95% CI = (1.06, 2.34)]. However, no positive association was observed in other subgroups (Supplementary Table S5, P_association>0.05).

The above results indicated that the TT genotype of XPC rs2228000 seems to be related to a high risk of bladder and breast cancer, whereas the CT genotype is more likely to be associated with reduced susceptibility to gastric cancer in the Chinese population.

As shown in Table 2, we did not observe a notable publication bias among these comparisons, in that all the P_Begg>0.05, P_Egger>0.05 apart from the P_Egger=0.031 (T vs. C), P_Egger=0.046 (CT vs. CC), P_Egger=0.023 (CT+TT vs. CC). Figure 4A presents the publication bias plot of Egger’s test under the T vs. C model. In addition, as shown in Figure 4B (allelic model data as example), we also observed relatively stable pooling data through the performance of sensitivity analyses.

An FPRP test was conducted to confirm the above positive findings for bladder, breast, and gastric cancers. The FPRP values of positive results at different prior probability levels are shown in Supplementary Table S6. We found that at a prior probability of 0.1 with an OR of 1.5, all the FPRP values were less than 0.2 (Supplementary Table S6, FPRP = 0.028, T vs. C; FPRP = 0.023, TT vs. CC; FPRP = 0.155, CT+TT vs. CC; FPRP = 0.022, TT vs. CC+ CT), indicating a noteworthy association between XPC rs2228000 and the risk of bladder cancer. Similar true positive associations were observed for breast and gastric cancer (Supplementary Table S6, all FPRP < 0.02) at a prior probability of 0.1.

In addition, we also performed the TSA test to assess the robustness of our significant findings. As shown in the TSA data of breast cancer under the TT vs. CC+CT model (Figure 5) and gastric cancer under the CT+TT vs. CC models (Supplementary Figure S10), we found that the cumulative number of participants (Z-curve) met the TSA monitoring boundary and required information size. With regard to the bladder cancer under the TT vs. CC+CT model (Supplementary Figure S11), the cumulative Z-curve crossed with the TSA monitoring boundary, even though it did not reach the required information size. These data therefore indicated the robustness of our conclusions.

There is a controversial conclusion regarding the genetic impacts of the XPC rs2228000 SNP in the risk of clinical cancer diseases in different publications. For example, XPC rs2228000 was reportedly related to susceptibility to bladder cancer cases in Iraq [27], Sweden [30], or India [67] but not the U.S.A. [85] or Spain [35]. Likewise, XPC rs2228000 was also significantly associated with the risk of breast cancer in a Chinese population [37,78] but not Caucasians or African-Americans in the U.S.A. [72]. Although several meta-analyses of XPC rs2228000 and certain specific cancer types exist [86–92], differences in study enrolment, data extraction, analysis strategy, and result descriptions were observed. We thus conducted a meta-analysis and TSA for a comprehensive assessment regarding the genetic influence of the XPC rs2228000 in the risk of various types of cancer, including bladder cancer, lung cancer, gastric cancer, melanoma, esophageal cancer, breast cancer, pancreatic cancer, and colorectal cancer.

Only three prior meta-analyses with fewer than 15 studies in 2008 [12–14] and one meta-analysis with 33 articles in 2013 [15] were reported to detect the genetic association between XPC rs2228000 and overall cancer risk. In our study, we retrieved four databases (updated till September 2019) to include the potential publication for the pooling analysis. After employing our strict screen strategy, we finally included 64 eligible articles, which contained 71 case–control studies, for the overall meta-analysis and the following subgroup analyses by the factors of race, country, control source, article quality, genotyping assay, and cancer type. Five genetic models, including allelic, homozygotic, heterozygotic, dominant, and recessive models, were utilized. We excluded the improper studies according to the strict requirement of full genotype frequency data and the HWE principle. For instance, there are a total of 33 articles with 14877 cases and 17888 controls [1,28,30–32,36,39,42,44,45,47,48,50,54–56,60,64–66,68–70,72,75–79,82,93–95] for the prior meta-analysis of He et al. in 2013 [15]. In this study, we excluded two articles regarding bladder cancer [95] and cutaneous melanoma [94] because the genotype distribution in the control group is not in line with the HWE, and we added 32 other published articles [4,10,11,27,29,33–35,37,38,40,41,43,46,49,51,52,57–59,61–63,67,71,73,74,80,81,83–85]. Our pooling data from eight case–control studies showed the genetic correlation between XPC rs2228000 and increased risk of bladder cancer under the allelic, homozygotic, heterozygotic, dominant, and recessive models, which is partly consistent with the positive data of He et al. (2013) [15] under the homozygotic and recessive models from four case–control studies. A similar result was obtained for breast cancer, even though four new case–control studies were added, compared with the pooling results of He et al. (2013) [15]. Moreover, we provided assessment evidence regarding the potential impact of XPC rs2228000 on the reduced susceptibility to gastric cancer in the Chinese population. Nevertheless, we did not detect a significant association between XPC rs2228000 and other types of cancer, such as lung cancer, melanoma, pancreatic cancer, or colorectal cancer.

In our study, we performed the FPRP test with a prior probability of 0.1 and an FPRP threshold of 0.2 to check whether the positive findings of breast, bladder, and gastric cancers are noteworthy, considering the potential presence of ‘false positives’. After the FPRP estimation, the genetic association between XPC rs2228000 and the risk of bladder, breast, and gastric cancers risk remain significant at the prior probability level of 0.1. Furthermore, we observed the robustness of our conclusions through the performance of TSA test and sensitivity analyses and the absence of large publication bias by Begg’s/Egger’s test.

Despite these findings, some limitations to this research may still influence the statistical power of analyses of certain types of cancer. Although more than 70 case–control studies were enrolled in the overall meta-analysis, small sample sizes were still included in some subgroup analyses. For example, only two case–control studies [56,74] were included for the subgroup of ‘blood system cancer’, while only two studies [81,83] were enrolled for ‘nervous system cancer’. Therefore, we still cannot rule out the possible genetic role played by XPC rs2228000 in the risk of cancers of the blood or nervous systems. A similar uncertainty also exists in the subgroup analysis of ‘lung cancer’, ‘melanoma’, ‘esophageal cancer’, ‘pancreatic cancer’ and ‘CRC’.

We observed clear between-study heterogeneity, even if articles with low quality are removed. Regarding the available sample size, more factors, such as gender, age, environmental exposure, drinking/smoking status, tumor situations, characteristics, antiepileptic agents, or drug resistance, should be adjusted in future pooling analyses. It would be valuable to carry out an integrated analysis to evaluate the combined role of more XPC polymorphic loci (e.g., rs2228001, PAT^−/+) in susceptibility to different types of cancer based on the available evidence.

In general, the TT genotype of XPC rs2228000 may be linked to an increased risk of bladder and breast cancers, whereas the CT genotype is more likely to be associated with a reduced susceptibility to gastric cancer in the Chinese population. Considering the limitations of our study, we need to analyze more publications to verify the genetic impact of XPC rs2228000 in other types of cancer.

The authors are grateful to American Journal Experts for providing English language editing.

Y.D., Z.S., and J.Z. conducted the database search and study screening. Y.D., Z.S., and W.G. summarized the evidence, performed the pooling analysis, FPRP, and TSA tests. Y.D. and J.Z. wrote the manuscript. All the authors reviewed and approved the final version.

The authors declare that there are no competing interests associated with the manuscript.

This work was in part supported by the National Natural Science Fund Grant [grant numbers 81471761, 81501568]; the Tianjin Science and Technology Support Plan Key Projects [grant number 15ZCZDSY00890]; and the Tianjin Medical University Cancer Hospital Project [grant number 1711].

CI

confidence interval

CNKI

China National Knowledge Infrastructure

FPRP

false-positive report probability

HB

hospital-based

HWE

Hardy–Weinberg equilibrium

NER

nucleotide excision repair

NOS

Newcastle–Ottawa quality assessment Scale

OR

odds ratio

PAT^−/+

poly-AT insertion/deletion polymorphism

TSA

trial sequential analysis

WOS

Web of Science

XPC

xeroderma pigmentosum complementation group C