Genetic variants in ERCC1 and XPC predict survival outcome of non-small cell lung cancer patients treated with platinum-based therapy

Nucleotide excision repair (NER) plays a vital role in platinum-induced DNA damage during chemotherapy. We hypothesize that regulatory single nucleotide polymorphisms (rSNPs) of the core NER genes modulate clinical outcome of patients with advanced non-small cell lung cancer (NSCLC) treated with platinum-based chemotherapy (PBS). We investigated associations of 25 rSNPs in eight NER genes with progression free survival (PFS) and overall survival (OS) in 710 NSCLC patients. We found that ERCC1 rs3212924 AG/GG and XPC rs2229090 GC/CC genotypes were associated with patients’ PFS (HRadj = 1.21, 95% CI = 1.03–1.43, P adj = 0.021 for ERCC1 and HRadj = 0.80, 95% CI = 0.68–0.94, P adj = 0.007 for XPC), compared with the AA and GG genotypes, respectively. The association of XPC rs2229090 was more apparent in adenocarcinoma than in squamous cell carcinoma patients. Additionally, ERCC4 rs1799798 GA/AA genotypes were associated with poorer OS (HRadj = 1.32, 95% CI = 1.04–1.69, P adj = 0.026), compared with the GG genotype. The expression quantitative trait loci analysis revealed that ERCC1 rs3212924 and XPC rs2229090 might regulate transcription of their genes, which is consistent with their associations with survival. Larger studies are needed to validate our findings with further functional studies to elucidate the mechanisms underlying these observed associations.

among the patients, with an overall response rate of 26~60% 9 . It is speculated that this may be related to individual variability in repairing DNA damage induced by PBC 10,11 . Increasing body of evidence highlights the importance of genetic factors, such as single nucleotide polymorphisms (SNPs), and gene expression in individual response to the treatment, which have an impact on subsequent survival 12 , particularly for genetic variations in nucleotide excision repair (NER) genes [13][14][15][16] .
The DNA repair pathways are the safeguard of genomic stability by restoring damaged DNA induced by mutagens (i.e. UV, tobacco or chemicals), of which NER is the major mechanism removing bulky DNA lesions caused by chemicals. NER has been frequently associated with survival in NSCLC patients treated with PBC 13 . NER functions by repairing platinum-DNA (Plt-DNA) adducts, involving the coordination of 20-30 proteins that replace the bulky adduct DNA segment with a newly synthesized DNA segment using the intact complementary strand as the template 17 . The hypotheses of NER genes affecting lung cancer prognosis are two-folds, a double edged sword: on one side, suboptimal DNA repair may promote carcinogenesis by weakening mutation-fixation of DNA damage induced by both exogenous and endogenous carcinogens and subsequent development of tumours 18 and the other side, efficient DNA repair in the tumour may lead to fast removal of plt-DNA adducts, reducing the efficacy of PBC 13,19 .
NER comprises of three main events: recognition of base damage, the bimodal incision of DNA, and excision of DNA fragments 17,20,21 . The specific recognition of substrate sites consists of several key proteins: the initial step involves the XPC-HHRAD23 complex, which recognizes the base damage caused by exogenous carcinogens 22 . The XPE/DDB1 protein has been studied for its auxiliary role for the recognition of cyclobutane pyrimidine lesions, due to its affinity for UV-damaged DNA 23 . The XPC/HHRAD23 complex further binds to several other proteins (i.e. XPA, RPA, TFIIH and XPG/ERCC5), in which transcription factor IIH (TFIIH) is a subcomplex of the RNA polymerase II transcription initiation machinery, and XPB/ERCC3 and XPD/ERCC2 are two central DNA helicases that unwind the DNA duplex in the close vicinity of the base damage; XPG and ERCC1-XPF heterodimeric protein are two endonucleases that cut the damaged DNA strand 3′ and 5′ to the site of the base damage, respectively 20,22 . These core proteins work in concert to maintain NER function, and hence their respective roles in the NER pathway have been more extensively studied. In the present study, we undertook a hypothesis-based approach to evaluate the impact of regulatory SNPs (rSNPs) in the core NER genes (ERCC1, ERCC2, ERCC3, ERCC4, ERCC5, RAD23B, XPA, XPC and XPE) on survival of NSCLC patients treated with PBC by analysing a pool of 25 rSNPs in 710 patients with advanced disease stages. All these 25 rSNPs were predicted by bioinformatics tools to be potentially functional in regulating their gene expression (Table 1 and Supplemental S1).

Results
Characteristics of the study population. The present study consisted of 710 patients diagnosed with NSCLC 24 , who had DNA samples, complete data on demographic, clinical characteristics, progression free survival (PFS) and overall survival (OS). Of all the patients, 508 were males and 202 were females, with a median age at diagnosis of 58 (a range of 23-83) years, and 334 (47%) were never, 41 (5.8%) former, and 335 (47.2%) current smokers. All subjects had an advanced TNM stage (III or IV) cancer, with 478 (67.3%) being adenocarcinoma. For different chemotherapy combinations, 237 (33.4%) received platinum-docetaxel/paclitaxel, whereas 300 (42.3%) received platinum-pemetrexed treatment. Furthermore, 219 (30.8%) and 257 (36.2%) of the patients received palliative radiotherapy and tyrosine-kinase inhibitor (TKI) treatment, respectively. The associations of these demographic characteristics and the known risk factors with NSCLC survival were also described in a previous publication 24 . The characteristics of demographic and clinical variables are described in Supplemental  Table S2.
NER rSNPs and NSCLC survival. The details of the eight (after excluding DDB1/XPE that does not have any rSNPs) core NER genes. We selected 25 rSNPs that are located in a regulatory region in either of the eight genes, and those rSNPs under investigation are shown in Table 1. We then performed the genotyping with DNA samples extracted from the whole blood cells. Call rates of the majority of the SNPs were >95%, except for three rSNPs (rs2607735, rs1007616 and rs7507745), which were then excluded from further analysis. In the univariate analysis without and multivariate analysis with adjustment for clinical variables, three rSNPs (ERCC1 rs3212924, XPC rs2229090 and ERCC4 rs1799798) consistently showed a significant association with either PFS or OS in NSCLC patients (Tables 2). Further subgroups analysis was performed for adenocarcinoma and squamous cell carcinoma patients, as well as by the dominant chemotherapy treatment, for these two histological types.
Specifically, the ERCC1 rs3212924 G allele was found to be significantly associated with a poor PFS [AG/GG vs. AA: median survival time (MST) 6.5 vs. 7.6 months, P log-rank = 0.030; adjusted hazards ratio (HR adj ) = 1.21,  Fig. 1D). This variant has a borderline association with OS in adenocarcinoma patients, but not in squamous cell lung cancer patients, which is likely due to sample size reduction in the subgroup analysis (Supplemental Table S3). When we combined all risk genotypes into the number of risk genotypes (NRGs, i.e., the number of ERCC1 rs3212924 GG/AG and XPC rs2229090 GG genotypes) for assessing their joint effect on PFS, the frequencies of patients with a score of 0, 1 or 2 for NRGs were 205, 347 and 156, respectively ( Table 2). A dose-dependent trend was observed for patients carrying at least one of these genotypes and patients carrying two of these genotype had the highest risk for disease progression, compared with those carrying zero risk genotypes (HR adj = 1.50, 95% CI = 1.19-1.90, P log-rank = 0.017, P adj = 0.0006) ( Table 2; Fig. 1A). After dichotomizing patients into a low-risk (0 risk genotype) (LRi) or a high-risk (1-2 risk genotypes) (HRi) group, patients in the HRi group exhibited a significant shorter survival time before progression (HR adj = 1.32, 95% CI = 1.10-1.58, P adj = 0.003), compared to those in the LRi group (Table 2).
Stratified analysis between the risk genotypes and NSCLC survival. Stratified analysis was also performed to assess differential effects of demographic or clinical variables (such as tumour histological type and treatment) on survival risk associated with genotype groups (LRi or HRi) or risk genotypes. Overall, the risk genotype group carriers (ERCC1 rs3212924 AG/GG and XPC rs2229090 GG) tended to have a significantly increased risk of disease progression in subgroups of younger (≤58 years old), males, current smokers, TNM stage III, no radiotherapy, ECOG status 2, poorly differentiated, platinum-docetaxel/paclitaxel recipients. Most homogeneity tests did not provide any evidence to support for differences in HRs between the strata, except for the performance status (P = 0.006), which may be caused by unbalanced distribution of risk genotype groups between different subgroups. For ERCC4 rs1799798 GA/AA carriers, an increased risk of death was observed in older patients (>58 years), non-smokers or former smokers, well-moderately differentiated tumours, and recipients of carboplatin-based or TKI chemotherapies, compared with the GG carriers (Supplemental Table S5).
Correlations between ERCC1 and XPC risk genotypes and mRNA expression levels. To examine genotypic effect of the survival-associated rSNPs on gene expression, the eQTL analysis of the three NER rSNPs was further performed by using two publically available datasets. One included the GTEx samples of normal lung tissues, in which the ERCC1 rs3212924 G allele was associated with a significantly higher ERCC1 mRNA expression level (P = 0.038, effect size = 0.13) (Fig. 2A). The XPC rs2229090 protective C allele was associated with a lower expression level of XPC (z-score = −6.83, P = 8.39E-12) and a nearby gene TMEM43 in peripheral blood cells (z-score = −6.29, P = 3.17E-10, Fig. 2D). Therefore, it is biologically plausible that the associations between those variants and NSCLC survival may be explained by the difference in gene expression levels regulated by those variants. That is, an increased expression of ERCC1 was associated with a poor survival, whereas a decreased expression of XPC was associated with a better survival, and these support the notion that DNA repair is a double-edged sword.

Discussion
According to the American Society of Clinical Oncology and National Comprehensive Cancer Network (NCCN) guidelines, lung cancer patients with a performance status of 0 or 1 should be treated with a combination of a platinum drug (cisplatin or carboplatin) and a non-platinum drug (e.g. paclitaxel) in the first-line therapy 25 . Cytotoxicity of platinum compounds results from formation of Plt-DNA adducts, leading to bulky distortion of DNA, destabilization of the double helix, inhibition of DNA replication, transcription and ultimately death of tumour cells 26 . Better clinical outcome was observed in patients with higher levels of Plt-DNA adducts in the tumours 13 . DNA repair capacity, particularly of the NER pathway, has been associated with the PBC efficacy. This is because NER primarily repairs bulky DNA adducts caused by mutagens and guanine-cisplatinium adducts formed during PBC 17 . Likewise, in vitro studies have also shown that NER is the major DNA repair pathway responsible for the repair of cisplatin-DNA damage 10 . Previous association studies on SNPs of NER genes and the survival of NSCLC have mainly focused on missense variants or coding regions of individual genes, with very few studies focusing on all the core genes in the pathway, linkage disequilibrium (LD) blocks or non-coding variants 27 . We adopted a hypothesis-based approach with a main focus on regulatory variants predicted to be biologically functional in NER. In the present study, we found that two rSNPs (ERCC1 rs3212924 and XPC rs2229090) were associated with PFS and one rSNP (ERCC4 rs1799798) associated with OS of NSCLC patients, and these associations were not previously reported for lung cancer. The rs3212924 variant resides at the upstream or an intron of different ERCC1 transcripts, with a predicted function of altering transcription factor binding, which may further affect gene expression. Additional evidence from the eQTL analysis also indicated a significantly higher mRNA expression level in lung tissues containing the risk ERCC1 G allele. Difference in gene expression by the rs3212924 G allele has been observed not only in lung tissue, but also in artery, skin, and ovary tissues, suggesting a genetically determined regulatory role of this variant in its gene expression. In the rs3212924 LD block, none of the other SNPs in high LD (r 2 > 0.8) have been previously reported to be associated with cancer survival (Fig. 3A). Taken together, the associations between this variant with high tumour tissue levels of ERCC1 mRNA may have led to cisplatin resistance 28,29 , which may have independently affected disease progression in NSCLC patients.
The rs2229090 variant is located at the 3′UTR of XPC, and the G to C allele substitution is predicted to affect miRNA binding. In fact, the eQTL analysis indicated a genotypic effect of rs2229090 on expression of a pseudogene (Vomeronasal 1 Receptor 20 Pseudogene, VN1R20P) downstream of XPC in tibial artery tissues (Fig. 2B) and XPC expression in peripheral blood cells (Fig. 2D). In the corresponding LD block (Fig. 3B), only variant rs2228000 (Ala499Val) (r 2 = 0.875) was reported to be associated with survival of patients with various cancers including lung cancer [30][31][32][33] . Prior evidence indicated that subjects carrying rs2228000 CT/TT genotypes exhibited a better DNA repair capacity, and a poorer survival or risk of recurrence in oropharynx squamous cell carcinoma 34 and acute myeloid leukaemia 35 . However, no detailed molecular mechanism of how rs2228000 T allele functions in these associations has ever been reported. It is likely that the phenotypic change of XPC function associated with rs2228000 may have been responsible for the observed association with rs2229090 that is within the same LD block (Fig. 3B). Because XPC plays a key role in recognizing DNA damage and initiation of the NER process, these collective findings suggest that XPC variants at the rs2229090 block may have an impact on PFS in NSCLC patients treated with PBC by affecting PBC outcome through changing XPC expression and thus the DNA repair capacity. These findings call for further functional studies to reveal the biological mechanisms underlying those associations.
In the subgroup analysis by histological type and chemotherapy treatment, XPC rs2229090 GC/CC exhibited a significant association with a longer PFS, while ERCC4 rs1799798 GA genotype was significantly associated with a shorter OS in 477 adenocarcinoma patients, but not in 138 squamous cell carcinoma patients, suggesting a potential histological difference in genetic regulation of lung cancer survival outcome in response to the treatments. Although XPC rs2229090 GC/CC genotypes were significantly associated with PFS in squamous cell carcinoma patients who received docetaxel-cisplatin, the sample size of this treatment group was relatively small (n = 56); hence, this result needs to be interpreted with caution (Supplemental Table S4). It is also possible that the sample size in most of the subgroups was not large enough to reveal the real associations, suggesting that future larger validation studies are required to substantiate our findings.
The ROC curve prediction model for PFS incorporating ERCC1 and XPC risk genotypes exhibited a statistically significant improvement in discriminatory power, compared with that of the clinical factors only (I/C AUC 0.59 vs. 0.58, P = 0.019) (Supplemental Fig. S1A and B; Table S7). There was a trend towards a higher AUC of ROC and C index in the genotype-inclusive prediction models for the five-year overall survival (Supplemental Fig. S1C and D; Table S7).
There are inherent limitations in the present study. First, the recruitment of patients treated in the same hospital may lead to selection bias in generalization of the results to the general population; therefore, additional results of patients from other hospitals of other populations are necessary to confirm our findings. Second, with the aim of studying potentially functional SNPs in the regulatory regions of eight NER core genes, we did not incorporate the other known effect of non-synonymous SNPs on survival outcome of NSCLC patients, although they are not in the LD block with the ones under investigation in the present study (except for rs2228080). Third, multiple testing correction was not conducted in the present study, because this was an exploratory study with a limited study power. Prospective studies in larger populations are warranted to substantiate the findings in the present study.

Conclusions
The present study provided evidence that rSNPs in the core NER genes may modulate PBC-related survival outcome in Chinese NSCLC patients with an advanced stage disease. Potential gene regulation by rSNPs of two NER genes associated with outcomes of patients with NSCLC call for further functional studies to unravel the molecular mechanisms underlying the observed associations, which will also allow for further development of predictive biomarkers to facilitate personalized chemotherapy regime.

Material and Methods
Study populations. The present study was conducted on patients diagnosed with histologically advanced NSCLC from Fudan University Shanghai Cancer Centre (FUSCC) between February 1, 2009 and November 30, 2013. The recruitment criteria included the following: (1) unrelated Han Chinese with inoperable TNM stages III to IV tumours of NSCLC without prior history of cancer other than in situ carcinoma; (2) received PBC as the first-line treatment; (3) having Eastern Cooperative Oncology Group performance (ECOG) status 0 to 2 with laboratory testing for blood tests and uronoscopy in normal range; (4) no active infection and serious medical or psychological conditions that might prevent patients from adhering to treatment; and (5) patients with recent myocardial infarction, cardiac arrhythmia, active congestive heart failure or cerebral apoplexy, crankiness or depression were excluded from this study. The clinical data including age at treatment, sex, smoking history, ECOG performance, TNM stage, histological type and grade, chemotherapy regimens, radiotherapy, tyrosine-kinase (TKI) treatment were collected from patients' medical records.
Survival data. Survival data were collected from patients' next of kin through a telephone follow-up and inpatient and outpatient clinical medical records. OS time was calculated from the starting date of the treatment until the date of the last follow-up or death. PFS time was measured from the starting date of the treatment until the last follow-up, progression of disease or death. Patients without progression were censored at the date of last contact. The median follow-up time was 32.1 months. The Institutional Review Board of FUSCC approved this study, with all methods performed in accordance with the guidelines and regulations of FUSCC. All participants provided an informed consent for using their blood samples in future research.

SNP selection.
To specifically explore the association between rSNPs in core NER genes and survival of NSCLC in response to PBC, all rSNPs were queried from the NER gene regions under the study by using SNP/ GeneView in dbSNP database (http://www.ncbi.nlm.nih.gov/snp/) using the GRCh38 reference build of the human genome. A total of 25 rSNPs in eight (out of nine) core NER genes were chosen, with detailed characteristics of all investigated genes and rSNPs shown in (Table 1 and Supplemental Table S1). The selection criteria were based on the following: minor allele frequency (MAF) ≥ 5% in Han Chinese, in the regulatory region (5′ near gene, 5′UTR, intron, 3′ near gene, or 3′UTR), in low LD with each other (r 2 < 0.8), have predicted functions (transcription factor binding site, splicing, miRNA binding site or significant eQTL) by SNPinfo (http://snpinfo. niehs.nih.gov/snpfunc.htm) and GTEx portal (http://www.gtexportal.org/home/). A full list of the NER genes analysed in this study, their region coordinates, their start sites and stop sites, and the characteristics of genotyped variants are summarized in Table 1.
SNPseq genotypin. Genomic DNA was extracted from the whole blood of all study subjects by using DNA Blood Mini Kit (Qiagen, Valencia, CA). The purity [optical density (OD) 260/280 at 1.7~2.0] and concentration ( >20 ng/μl) that met the sequencing requirements. Genotyping of all rSNPs was conducted by FastTarget, a next generation sequencing-based method using Illumina Miseq. 2000 Platform (2 × 250 bp, Illumina, CA, USA). Prior to sequencing, 5% of the samples were randomly selected and subjected to 1% agarose gel electrophoresis quality control. Genomic regions containing the investigating rSNPs were amplified using the FastTarget TM technology (Genesky Biotechnologies Inc, Shanghai, China). A total of 25 amplicons were amplified, with the primers information attached in Supplemental Table S6. After multiple PCR reactions, DNA fragments were ligated with the adaptor by using Q5 DNA polymerase Kit (New England Biolabs, MA, USA), and further purified by Agencourt AMPure XP (Beckman Coulter, CA, USA). Next-generation sequencing of the amplification products was carried out by MiSeq 2000 Sequencer (Illumina, Inc., San Diego, CA, USA), following the manufacturer's standard protocols. Sequencing depth of more than 30x was achieved for over 90% of the samples. Output sequence data were trimmed and then compared with fragment reference sequences (hg19) using the Blat program 28 . Burrows-Wheeler Aligner (BWA, V 0.7.5a) was used to map the reads 36 , followed by Sequence Alignment/Map (SAM)-to-BAM conversion, sorting, and removal of duplicates using SAM tools (v0. 1.19) 37 . Combined rSNP calling was performed on the resulting BAM files using Genome Analysis Toolkit (GATK, https://software.broadinstitute.org/gatk/best-practices/) and VarScan programs 38 . Finally SNP annotation was done by the Annovar program 39 .
Statistical analysis. The association between each genetic variant and PFS/OS was estimated by Cox proportional hazards regression model, calculated as HRs with their corresponding 95% CIs. The covariates used for adjusted HR for PFS included age-at-treatment, sex, smoking status, TNM stage, histological type, histologic grade, ECOG performance status, chemotherapy regimens, grade 3/4 chemotherapy toxicity and palliative radiotherapy, whereas TKI treatment was included for adjusted HR for OS in addition to the covariates mentioned above. Kaplan-Meier test was used to assess each genetic variant on the cumulative probability of PFS and OS 40 . Log-rank test was used to examine the difference in survival between groups. The observed associations were stratified by selected demographic and clinical variables. The heterogeneity between subgroups was assessed by the χ 2 -based Q test. For survival prediction model construction, independent predictors including selected clinical variables and genetic variants were included. ROC analysis was used to compare sensitivity and specificity of the OS and PFS prediction by the included parameters. Predictive values of selected variables were evaluated by I/D AUC of the ROC curves for censored data and C index for comparison of survival models. The I/D ROC and I/D AUC were calculated and plotted by RisksetROC package of R software (version 3.2.3; The R Foundation for Statistical Computing) 41 . All statistical analyses were performed by SAS software (version 9.4; SAS Institute, Cary, NC). Unless stated otherwise, all P values were two-sided with a significance level of P < 0.05.
The eQTL analysis. Two large-scale eQTL datasets were used to assess the correlation between survival-related genetic variants and NER gene expression levels: one is the GTEx project using 278 lung tissue samples, and the other is the blood eQTL browser (http://www.genenetwork.nl/bloodeqtlbrowser/) encompassing 5,311 individuals and 2,775 replicates 42 .