Excision Repair Cross-Complementation Group 1 (ERCC1) Status and Lung Cancer Outcomes: A Meta-Analysis of Published Studies and Recommendations

Purpose Despite discrepant results on clinical utility, several trials are already prospectively randomizing non-small cell lung cancer (NSCLC) patients by ERCC1 status. We aimed to characterize the prognostic and predictive effect of ERCC1 by systematic review and meta-analysis. Methods Eligible studies assessed survival and/or chemotherapy response in NSCLC or SCLC by ERCC1 status. Effect measures of interest were hazard ratio (HR) for survival or relative risk (RR) for chemotherapy response. Random-effects meta-analyses were used to account for between-study heterogeneity, with unadjusted/adjusted effect estimates considered separately. Results 23 eligible studies provided survival results in 2,726 patients. Substantial heterogeneity was observed in all meta-analyses (I2 always >30%), partly due to variability in thresholds defining ‘low’ and ‘high’ ERCC1. Meta-analysis of unadjusted estimates showed high ERCC1 was associated with significantly worse overall survival in platinum-treated NSCLC (average unadjusted HR = 1.61, 95%CI:1.23–2.1, p = 0.014), but not in NSCLC untreated with chemotherapy (average unadjusted HR = 0.82, 95%CI:0.51–1.31). Meta-analysis of adjusted estimates was limited by variable choice of adjustment factors and potential publication bias (Egger's p<0.0001). There was evidence that high ERCC1 was associated with reduced response to platinum (average RR = 0.80; 95%CI:0.64–0.99). SCLC data were inadequate to draw firm conclusions. Conclusions Current evidence suggests high ERCC1 may adversely influence survival and response in platinum-treated NSCLC patients, but not in non-platinum treated, although definitive evidence of a predictive influence is lacking. International consensus is urgently required to provide consistent, validated ERCC1 assessment methodology. ERCC1 assessment for treatment selection should currently be restricted to, and evaluated within, clinical trials.


Introduction
Lung cancer is the commonest cause of cancer death globally, accounting for around 1.3 million deaths per year [1]. Despite advances in therapeutics, survival from both major subtypes (nonsmall cell lung cancer (NSCLC) and SCLC) remain poor, with only around 5% of all patients reaching 5 years. Standard of care for both advanced NSCLC and SCLC is platinum-based doublet chemotherapy, with non-platinum doublets inferior [2,3]. In NSCLC platinum doublets are associated with response rates of 25-30% [4]. A number of tumour biomarkers have been investigated for prognostic and predictive utility when considering systemic therapy, and prominent amongst these is excision repair cross-complementation group 1 (ERCC1) protein.
ERCC1 is the rate-limiting member of the nucleotide excision repair pathway (NER), one of at least 5 overlapping biochemical pathways by which altered DNA sequences can be restored to base-line. Abrogation of these pathways has been both associated with carcinogenesis [5], and targeted as a therapeutic mechanism [6]. The NER pathway functions to remove bulky DNA lesions [5], including tobacco-associated adducts formed by carcinogen exposure [7]. Mechanisms of platinum cytotoxicity include forming bulky DNA adducts leading to both inter-and intrastrand cross-link generation, which results in apoptosis unless repaired.
The critical role of ERCC1 in carcinogen and platinum adduct removal by NER has led to a number of studies reporting the relationship between ERCC1 status and survival in lung cancer patients, predominantly in NSCLC. ERCC1 has been investigated both as a prognostic biomarker, and for a predictive influence in determining benefit from platinum-directed therapy, with estimates between studies differing considerably.
We performed a systematic review and, where possible, metaanalysis of study outcomes to produce evidence-based results on the prognostic and predictive utility of ERCC1 status in lung cancer, and identify further research needs.

Eligibility Criteria
English language published studies were eligible if they assessed association of ERCC1 expression with survival or tumour response in NSCLC or SCLC patients. The primary outcomes of interest were overall survival (OS), event-free survival (EFS) and tumour response to chemotherapy, as defined by the contributing studies. For publications with overlapping datasets the smaller series was excluded.

Identification of Studies
The search for studies was performed in duplicate (SP and RAH) using the electronic database PubMed (http://www. pubmed.com) until 10th August 2009. The search strategy used the keywords ''lung cancer,'' ''NSCLC'', or ''SCLC'', and ''ERCC1''. Bibliographies of eligible studies, review articles and other relevant publications were also hand-searched to identify additional studies. Data from review articles, abstracts, and letters were not included.

Data extraction
Study characteristics were extracted from the full published article and summarized in a consistent manner to aid comparison. Methodology of ERCC1 analysis was categorized, including the threshold used to dichotomize ERCC1 as ''high'' and ''low''.
For platinum or non-platinum based chemotherapy treatment groups within each study, the log hazard ratio (log(HR)) estimate and its variance were extracted (log ratio of survival risk in ERCC1 high group versus ERCC1 low group). When not directly reported the log(HR) and its variance were estimated [11,12], from other data, such as log rank test statistics and p values, and number of patients and events in each group, using the methods of Parmar et al [13,14]. Where relevant effect estimates were not obtainable using the methods above, or through direct contact with corresponding authors, the study was excluded from the metaanalysis. Both unadjusted and adjusted HR estimates were sought for each study, and the choice of adjustment factors recorded.
In addition, we sought to extract the log relative risk (log(RR)) estimate and its variance indicating the log of the ratio of the risk of tumour response (response versus non-response) to platinumbased systemic therapy in the ERCC1 ''high'' group versus ERCC1 ''low'' group.

Statistical Analysis
Direct evidence of ERCC1 as a predictive biomarker for platinum-based chemotherapy, defined by the interaction between platinum-based versus non-platinum-based chemotherapy and ERCC1 status within randomised controlled trials, was summarised where available. Indirect evidence of predictive influence was obtained by collating evidence on the prognostic effect of ERCC1 in platinum-based treated groups and non-platinum-based treated groups available separately from studies, and on the relationship between ERCC1 and radiological response to platinum-based chemotherapy.
Heterogeneity between studies was expected, hence extracted study log(HR) or log(RR) estimates and their variances were pooled using a random effects meta-analysis which accounts for such heterogeneity; estimates the average ('summary') HR across studies and its confidence interval (CI); and provides a prediction interval for the true hazard ratio in an individual study setting [15,16].
The impact of between-study heterogeneity in our metaanalyses was assessed by the I 2 statistic [17]. I 2 describes the proportion of total variation in meta-analysis estimates due to between-study heterogeneity, and is measured from 0-100% with increasing I 2 values indicating a larger impact of between-study heterogeneity in the meta-analysis.
For meta-analyses including 10 or more studies we assessed the possibility of small study effects (which indicates potential publication bias) by performing Egger's test, with a 10% significance level due to the low power of this test [18].
In four studies [28,32,36,38] outcome data were presented separately for patients who underwent different therapeutic strategies, and these groups were treated as separate datasets (Table 1). Specifically, Rosell et al. [38] stratified patients into groups who received three different chemotherapy regimes, one of which overlapped with a larger cohort [39] and was thus excluded; Olaussen et al. [36] reported separately on patients randomised to either chemotherapy or observation following resection; Okuda et al. [28] presented separate outcome data from non-randomised cohorts who underwent either peri-operative chemotherapy or surgery alone; and Fujii et al. [32] separately analysed outcomes for patients undergoing neoadjuvant chemotherapy or chemoradiotherapy, but datapoints were not extractable for the chemoradiotherapy group and this dataset was therefore excluded. Two studies [23,27] reporting patients with SCLC presented separate outcome data by stage (limited or extensive), and these datasets were treated separately (Table 1).

Study Characteristics
Characteristics of the 23 eligible studies are summarised in Table 1. All studies assessed ERCC1 expression retrospectively. Six studies [31,32,35,36,38,39] assessed ERCC1 in tumours from unselected patients enrolled into clinical trials, whilst in the remainder patients were not accrued to a trial. The median percentage of male patients was 76%, whilst the median of the study age means was 61 years. Three studies [28,31,42] did not report mean age.
The observed median proportion of NSCLCs with high ERCC1 expression was 46% (range 17-65%), and 50% (range 41-56%) respectively in studies using IHC and RTqPCR. In Table 1. Summary of studies reporting ERCC1 expression and outcomes in non-small cell and small-cell lung cancer patients. SCLC, the two studies using IHC reported 17% [23] and 28% [40] high expression, and 51% in the study using RTqPCR [27].

Indirect evidence for predictive influence of ERCC1
i) Relationship between ERCC1 status and survival in NSCLC without systemic therapy. Seven datasets [24,28,29,30,34,36,37] assessing 896 patients, with high ERCC1 expression observed in 48% of tumours, were available for pooling estimates of survival in patients who underwent surgery alone without systemic therapy. Neither the meta-analysis of unadjusted nor meta-analysis of adjusted estimates provided evidence that ERCC1 status has prognostic value for either OS or EFS in these patients (Table 3 and Figure 2).

ii) Relationship between ERCC1 status and survival in
NSCLC with platinum-based systemic therapy. OS HR estimates obtained from 1,391 NSCLC patients who received platinum were available for meta-analysis, 640 (46%) of whom had tumours with high ERCC1 expression. Meta-analysis of the unadjusted estimates indicated a significantly poorer OS in patients with high ERCC1 expression (average HR = 1.61,  95%CI: 1.23-2.10). This effect was maintained when pooling adjusted estimates (average HR = 1.57, 95%CI: 1.24-1.99), and was observed in patients treated in both the adjuvant and advanced disease settings ( Figure 3). However, large heterogeneity was observed in both unadjusted and adjusted analyses (I 2 = 52% and I 2 = 83%, respectively), resulting in wide prediction intervals for the prognostic effect in an individual clinical setting, and whilst there was no evidence of small study effects using unadjusted estimates (Egger's test: p = 0.72), there was clear evidence of such using adjusted estimates (Egger's test: p,0.0001). Meta-analysis for EFS indicated a relationship with ERCC1 status using adjusted estimates (average HR = 1.46, 95%CI:1.09-1.95, I 2 = 32%), but not with unadjusted (Table 3), and investigation of small study effects was not possible due to the small number of studies.
iii) Relationship between ERCC1 status and survival in SCLC. Five studies were available in regard SCLC patients receiving platinum-based chemotherapy ( Table 2), contributing data on 292 patients, of whom 92 (32%) had high ERCC1 expressing tumours. Meta-analysis of neither unadjusted nor adjusted estimates indicated an association between OS and high ERCC1 expression (Table 3). In an exploratory analysis stratifying by stage of disease, there was evidence of association between high ERCC1 expression and poor OS in limited stage patients, although this was based on only 2 studies.

Discussion
Somatic molecular phenotype has become well established as a key determinant in NSCLC in terms of both outcome and efficacy of systemic therapy [46,47,48]. Previous studies have suggested that somatic ERCC1 expression level is both prognostic in lung cancer, and predictive of outcome to platinum-based chemotherapy. Indeed, trials have already reported randomising patients contingent on somatic ERCC1 status [49,50], with others actively accruing. However, our analysis has shown that many of the published series assessing the ERCC1 utility have been small in size, reported conflicting outcomes, and used widely differing expression methodology and, in particular, threshold levels.
Only one study [36] directly assessed the predictive utility of ERCC1 status for platinum-based chemotherapy in NSCLC, and this was retrospective in design and in a subgroup of patients recruited to a clinical trial. A relationship between ERCC1 and survival in platinum-treated patients but not in non-platinum treated patients provides indirect evidence that ERCC1 is predictive, although the robustness of this conclusion is compromised due to the non-randomised nature of the treatment groups. Based on data from NSCLC patients untreated with systemic therapy, our analyses do not support the hypothesis that ERCC1 expression is prognostic in NSCLC. However, tentative evidence that, on average, high ERCC1 expression identifies poorer survival in NSCLC patients treated with platinum-based chemotherapy, provided indirect evidence of a predictive influence, although this may be merely a prognostic effect in this patient group. The lower likelihood of response to platinum-based chemotherapy in high ERCC1 expressing tumours observed further supports the notion that ERCC1 might be predictive of lack of platinum benefit.
Our study has identified several methodological weaknesses that must be addressed. Firstly, we observed notable variation in cutoffs in ERCC1 expression, which account in part for heterogeneity observed. Second, there has been considerable debate as to whether the 8F1 antibody clone specifically binds ERCC1 alone or other non-ERCC1 epitopes [51,52]. Third, it is unclear whether ERCC1 expression analysis by IHC and RTqPCR stratify patients into similar groups [34], and in the absence of a proven correlation between these two methods we recommend the use of IHC since it is more readily available and more easily standardised across laboratories. For ERCC1 to be a useful predictive biomarker in clinical practice, a single clearly defined protocol needs to be developed and validated to allow comparison of outcomes across studies. Inadequate sample size was also a frequent problem in the studies included in our analyses, with only eight of the 23 studies reporting outcomes from over 100 patients. Whilst pooling data may in part address deficiencies in individual study sample sizes, smaller studies are more likely to generate heterogeneity, as we observed. Several independent studies, each an order of magnitude greater than most published series, are required to allow accurate estimation of the true associations between ERCC1 expression and outcomes in lung cancer patients, particularly in SCLC.
Strengths of our study include the analysis of both survival and chemotherapy response endpoints, the separate meta-analysis of studies in which patients did or did not receive chemotherapy allowing an assessment of both predictive and indirect prognostic influences, calculation of prediction intervals to estimate the range of potential effects in any one individual study, and extensive comparison of study methodologies. In contrast, a similar recently published meta-analysis [53] made no assessment of the influence of ERCC1 status in patients who did not receive chemotherapy or SCLC patients, did not account for adjusting and made no assessment of methodological differences between contributing studies, failed to use the published HR and CIs for one study which compromised assessment of publication bias, used fixed effect meta-analyses despite significant between-study heterogeneity, and included potentially overlapping datasets. In our study we did not assess the quality of the primary studies, however quality assessment tools for examining prognostic and predictive biomarker studies do not currently exist, and are only beginning to be discussed for prognosis studies in general [54].
Large heterogeneity, methodological concerns, and potential for publication bias revealed by our analyses indicate that although ERCC1 shows considerable promise as a predictive biomarker in platinum-treated NSCLC patients, it is not ready for 'prime-time'. International consensus is urgently required to mandate homogeneous ERCC1 assessment methodology, as are prospective trials sufficiently powered to detect an interaction between platinum chemotherapy and ERCC1 expression, and in the interim, initiation of large, prospectively planned individual patient data meta-analyses. Until then, ERCC1 expression should not be routinely used in clinical decision-making.

Author Contributions
Conceived and designed the experiments: RAH SP. Performed the experiments: RAH SP. Analyzed the data: RAH RDR LJB SP. Contributed reagents/materials/analysis tools: RAH RDR LJB SP. Wrote the paper: RAH RDR LJB SP.