Somatic EGFR Mutation and Gene Copy Gain as Predictive Biomarkers for Response to Tyrosine Kinase Inhibitors in Non–Small Cell Lung Cancer

The epidermal growth factor receptor (EGFR) and members of its family play a considerable role in carcinogenesis through their involvement in proliferation, apoptosis, enhanced cell motility, and neoangiogenesis (1). Furthermore, the overexpression of EGFR has been correlated with advanced disease, a more aggressive phenotype and overall poor prognosis (2, 3). Therefore, the predictive and prognostic significance of EGFR overexpression in non–small cell lung cancer (NSCLC) has over the years become important, resulting in the development of numerous targeted therapies (1, 4–8). Further molecular analysis searching for aberrations in the gene using techniques such as fluorescence in situ hybridization (FISH) have also indicated that an increased EGFR copy number with balanced polysomy (present in a high proportion of cancer cells) can be detected in approximately 10% to 40% of patients with NSCLC, squamous cell carcinoma of the head and neck, and colorectal cancer (9–11).

When EGFR-targeted agents first entered clinical development, EGFR protein expression was viewed as a predictive biomarker of response (7, 12–17). This was the case for both the anti-EGFR monoclonal antibodies erbitux and panitumumab in colorectal cancer, and also for the anti-EGFR tyrosine kinase inhibitors (TKI) gefitinib and erlotinib in NSCLC. Several early studies also attempted to correlate certain phosphorylated forms of EGFR, or downstream molecules such as pAkt, as potential response biomarkers in an attempt to offer improved patient stratification toward more effective agents while circumventing unnecessary toxicities in patients that would otherwise have low response rates (18). Much of this changed in early 2004, when further research into the mechanisms of NSCLC sensitivity to TKIs identified the presence of somatic mutations clustered in and around the tyrosine kinase domain of EGFR as the most likely predictive biomarker (19–21).

Almost simultaneously, EGFR gene copy gain, in the form of gene amplification, was also proposed as a potential biomarker of TKI responsiveness (22–24). Since then a wealth of data has been published, mostly retrospectively analyzing the ability of different candidate biomarkers to predict response of NSCLC to TKI monotherapy (25). Some authors have also shown that patient preselection on the basis of EGFR mutational status leads to high response rates. Additional studies have also gone on to show that in the case of both colorectal cancer and NSCLC, somatic mutations in KRAS are a mechanism associated with resistance to anti-EGFR–targeted agents (26).

Considering that in the case of NSCLC, it is well recognized that somatic KRAS and EGFR mutations are mutually exclusive (25), the further understanding of the role of EGFR mutations is critically important in this debilitating disease. Increased response rates and potential improvements in survival have also been speculated for patients with EGFR gene gain receiving TKI monotherapy (27, 28). However, unlike the consistent reporting of somatic EGFR mutational status correlating with improved response rates, findings regarding EGFR gene copy number as a predictor of response to TKIs have been inconsistent (29, 30).

With this in mind, we summarized the existing literature to provide a quantitative overview of the diagnostic value of somatic EGFR mutations and EGFR gene copy number for predicting response of advanced NSCLC to monotherapy with TKIs supporting a systematic review and meta-analysis of published studies. We also attempted to explore potential sources of heterogeneity such as ethnicity, the category of diagnostic assay used for patient stratification, and patient selection criteria on the final diagnostic test performances.

We performed systematic computerized searches of the PubMed (MEDLINE) and EMBASE databases [from inception to December 31, 2008 for EGFR gene copy number, and from January 1, 2004 to December, 31, 2008 for somatic EGFR mutational status (last search conducted on February 10, 2009) and the Cochrane library (Issue 1, 2009)] to identify all published articles reporting on EGFR TKIs for advanced NSCLC using a previously described search strategy (19). Eligible studies had to report complete (CR) and partial response (PR) rates (CR+PR) stratified by EGFR mutation and/or EGFR copy number status. We have also hand searched journals known to publish data relevant to our search, wherein the reference lists of all retrieved articles and those of relevant review articles were also cross-referenced. Experts in the field of lung cancer treatment were contacted to broaden the search. Whenever multiple reports pertained to overlapping groups of patients, we retained only the report with the largest patient population (where appropriate) to avoid duplication of information. In all cases, corresponding authors were contacted to minimize this eventuality (see below). All study designs (prospective and retrospective; randomized or observational) were considered potentially eligible as long as they provided adequate response information, whether they were single or multiple arm trials. Studies examining EGFR TKIs in combination with any other agents such as cytotoxic agents or any investigational drug were excluded from the meta-analysis. Case reports, defined as studies reporting on 10 or fewer patients, were excluded. No abstracts or meeting proceedings were included, and no language restriction was imposed. Reports that did not contain sufficient information to extract all pertinent data fields were earmarked for author confirmation. Author confirmation by means of contact with the corresponding author of each article (or alternate lead author where necessary) was conducted over a period of 2 months. Details of the Author Contact Protocol are available on request.

The following information was recorded from each recovered article: first author, journal and year of publication, number of patients screened, number of patients with EGFR mutation or amplification, the number of patients treated with a TKI, treatment schedules and line of treatment, stage of disease, clinicopathologic and demographic data (smoking history, histology, gender, ethnicity), and data linking mutations and/or gene copy status to treatment outcome (CR, PR, CR+PR, stable disease, progressive disease, nonevaluable) with EGFR TKIs. The methods of mutation and/or gene copy number analysis were also recorded as follows: for analyzing EGFR mutations, we recorded whether bidirectional sequencing, allelic discrimination, or alternative approaches were used; similarly for EGFR gene copy number analysis, methods such as FISH, chromogenic in situ hybridization (CISH), or quantitative PCR–based strategies were used. Data extraction was done independently by two of the authors (I.J.D. and S.M.) and discrepancies were resolved by consensus including a third author (H.L.).

For the purposes of these analyses, a True Positive test was defined as a patient harboring an EGFR mutation or EGFR gene copy gain showing a response to EGFR TKI monotherapy (i.e., CR+PR). A True Negative test was defined as a patient wild-type for EGFR mutation or EGFR gene copy gain showing no response to EGFR TKI monotherapy (i.e., stable disease + progressive disease). Therefore, a False Positive result was a patient harboring an EGFR mutation or with EGFR gene copy gain showing no response.

The meta-analysis of sensitivity and specificity was conducted using a bivariate approach, as previously described (26). We plotted (per study) sensitivity and specificity within the receiver operating characteristic (ROC) plane and drew summary ROC curves (sROC) based on the hierarchical summary ROC curve model (31). The amount of heterogeneity was quantified using the I² statistic, which ranges from 0% to 100%. Likelihood ratios were calculated using the pooled estimates for sensitivity and specificity.

Subgroup analyses were done to evaluate the effect of ethnicity (East Asian versus Whites), patient recruitment criteria (clinical enrichment for higher prevalence of EGFR mutations or gene copy number versus unselected patients), response assessment criteria [Response Evaluation Criteria for Solid Tumors (RECIST) versus WHO criteria], method of EGFR aberration detection (sequencing versus high-sensitivity methods for EGFR mutations; quantitative PCR versus FISH/CISH for EGFR gene copy analysis), and the specific EGFR TKI used (gefitinib versus erlotinib) on the predictive usefulness of EGFR aberrations.

Considering that there are no generally accepted guidelines for meta-analysis of diagnostic markers, we attempted to conform to the Meta-Analysis of Observational Studies in Epidemiology guidelines and the Quality of Reporting of Meta-Analyses guidelines when appropriate, also taking into account recommendations on biomarker ontology as previously reported (26).

Statistical analyses were conducted with SAS (version 9.1 TS level 1M3, SAS Institute, Inc.) and STATA (version SE/10, StataCorp). P values for all comparisons were two tailed and statistical significance was defined as P < 0.05.

This was a literature-based study and as such no ethics approval was required.

There was no funding source associated with the study design, collection, and analysis and interpretation of the data or writing of the report. All authors had access to the raw data. The corresponding author had full access to all of the data and the final responsibility to submit for publication. The authors and their respective institutions assume no responsibility for any injury or damage to persons or property arising out of or related to use of these analyses, or for any omissions or errors.

Our initial search yielded 3,693 studies concerning TKI treatment in NSCLC. For the overall analyses, a total of 222 studies were considered potentially eligible; however, 158 were finally excluded. Fifty-two studies were excluded due to overlapping data availability in larger patient series; 10 additional studies were removed due to nonavailability of data (i.e., not supplied in the article, insufficient independent data for segregation, data not made available on author confirmation); 10 studies wherein patient populations were screened for the presence of EGFR mutations or EGFR gene copy number and only those patients received the TKI were also deemed ineligible; 7 studies were excluded as they had patient populations treated with drug combinations (not monotherapy); and 4 studies were excluded as they presented TKI rechallenge in TKI refractory populations. In addition 74 studies were excluded due to insufficient participant numbers (≤10 participants). A list of excluded studies is available on request.

Finally, as indicated in the search flow chart (Fig. 1), 59 studies were included in the meta-analysis for EGFR mutations (S1-8,S10-13,S15,S16,S18-41,S43-48,S50-63,S65), reporting on 3,101 patients of which 1,020 (32.9%) harbored EGFR mutations. The characteristics of eligible studies for EGFR mutation are summarized in Table 1. In addition, 21 studies reporting on 1,539 patients of whom 542 (35.2%) were considered EGFR gene copy gain positive (per study criteria) were included in the gene copy meta-analysis; their characteristics are summarized in Table 1 (S2,S9,S14,S15,S17,S18,S29,S30,S34,S35,S38,S42,S43,S48, S49,S51,S53,S57,S59,S60,S64). Fifteen studies had sufficient data for inclusion in both the EGFR mutation and the EGFR gene copy number analyses (S2,S15,S18,S29,S30,S34,S35,S38,S43,S48,S51,S53,S57,S59,S60,S64). References are provided in the online supplement.

Figure 2A and B illustrate how the (a) 59 EGFR mutation and (b) 21 EGFR gene copy number positive studies lay within the ROC plane. For EGFR mutations, the sensitivities ranged from 0.33 to 1, and specificities ranged from 0.50 to 1.0. The results of the meta-analysis of the pairs of sensitivities and specificities for studies investigating EGFR mutations are presented in Table 2A. There was significant difference between study heterogeneity [I² = 43.4; 95% confidence interval (95% CI), 25.9-61.0] for sensitivity and (I² = 75.0; 95% CI, 68.7-81.3) for specificity. The pooled sensitivity was 0.78 (95% CI, 0.74-0.82), and pooled specificity was 0.86 (95% CI, 0.82-0.89) for predicting a response to EGFR TKIs. Based on these estimates, EGFR mutations were associated with a positive likelihood ratio (+LR) of 5.6 and a negative likelihood ratio (−LR) of 0.26.

In studies evaluating EGFR gene copy number (amplification), the sensitivity ranged from 0.14 to 1 and specificity ranged from 0.48 to 0.93. There was significant heterogeneity both for sensitivity (I² = 66.5; 95% CI, 51.1-81.8) and specificity (I² = 66.4; 95% CI, 51.0-81.8). Compared with EGFR mutations, the pooled overall sensitivity was lower at 0.61 (95% CI, 0.49-0.71), as was the specificity, with a pooled estimate of 0.71 (95% CI, 0.66-0.76) for predicting a response to EGFR TKIs, Table 2B. EGFR gene gain was associated with a +LR of 2.1 and a −LR of 0.55.

Figure 3A to D indicate the sensitivities (A-C) and specificities (B-D) of each individual study in the form of forest plots for EGFR mutation (A-B) and EGFR gene gain (C-D).

The results of subgroup analyses and the respective study numbers per subgroup are also presented in Table 2A and B; EGFR mutation and EGFR gene gain, respectively, are also shown. There seems to be little if any differences in these outcomes on the examined subgroups, with the possible exception of Ethnicity where all diagnostic accuracy measures for EGFR mutation were superior in the Whites subgroup. A similar trend was also evident for EGFR gene gain.

The use of EGFR TKIs for the treatment of advanced NSCLC has been considered a revolution for an otherwise debilitating disease. Their introduction was greatly expected to provide a pathway for overcoming the long-term lack of progress in advanced NSCLC therapeutics that has resulted in a plateau in the survival improvement conferred by multiagent chemotherapy (32–36). Despite promising phase II results of single-agent TKIs in patients with heavily pretreated disease, showing significant responses and improvements in quality of life (12, 15), phase III trials of first-line combinations of chemotherapy incorporating TKIs (both gefitinib and erlotinib) provided disappointing results regarding response, disease-free, and overall survival (13, 14, 37, 38). The disappointments brought on by these results transformed into further confusion when second- or higher line, phase III trials investigating the efficacy of TKI monotherapy showed that erlotinib led to increased survival (39) whereas gefitinib seemed not to (40–43).

In 2004, intensive translational efforts culminated in the discovery of EGFR TKI binding domain mutations (19–21), which seemed to be functionally significant (25, 44). These mutations where detected in patients responding to gefitinib or erlotinib, and were shown to be more common among nonsmoking female patients of Asian ethnicity. Similarly, EGFR gene copy number gain has also been proposed as a potential mechanism modulating responsiveness to anti-EGFR TKIs (22, 35, 40–44). A number of phase II/III clinical studies have examined the influence of both mutations and also gene gain with respect to response and survival; however, due to the relatively low incidence of EGFR mutations (∼15% in Whites compared with 30% in Asians), much of the data from what are otherwise large trials remains underpowered to offer appropriate conclusions (13, 14, 37–40). On the other hand, there are a number of ongoing clinical studies addressing both the predictive and prognostic significance of EGFR mutations and gene copy number in TKI-treated NSCLC (including amongst others SATURN, TAILOR, TORCH, and RADIANT), the results of which are eagerly awaited.

To expedite our understanding on the predictive nature of EGFR mutations and gene copy number, we have systematically reviewed the literature in an effort to evaluate their sensitivity and specificity for predicting response to anti-EGFR TKIs in NSCLC. Overall, EGFR mutations seem to have high sensitivity and specificity for predicting response to TKIs. The corresponding +LR and −LR of 5.6 and 0.26, respectively, showed that EGFR mutations are an adequate response-prediction test, particularly for “ruling-in” response. In comparison, EGFR gene gain was a less sensitive and less specific marker, a fact reflected in the lower +LR and −LR of 2.1 and 0.55, respectively.

We have also conducted a thorough subgroup analysis to identify study level characteristics influencing the sensitivity and specificity of EGFR aberrations. Overall, the predictive value of mutations was high within each subgroup, and consistently higher when compared with similar subgroups according to EGFR copy number status (Table 2A and B, all subgroups). Of interest, EGFR mutations had a higher +LR in studies conducted in White subjects, compared with those of East Asian subjects. Also, high sensitivity detection methods (such as allele-specific PCR) seem to be somewhat inferior to sequencing-based detection methods (typically exons 18-21 inclusive) for ruling-in response; there was however, little difference regarding their ability to exclude the possibility of response. One possible interpretation is that the majority of nonstandard mutations (i.e., those other than the “classic” del exon19 and L858R) may still be associated with response to EGFR TKIs. We have recently supported this view in an analysis of a somatic EGFR mutation database (25, 44). As expected, the discriminatory value of mutational testing was increased when patients were not preselected for factors associated with TKI response (Table 2A, Recruitment). This may be an important factor if individual patient preselection is to be implemented into clinical practice. A number of small prospective studies have already proved the principle that high response rates can be obtained by molecular screening programs that address the whole population (26), or through preselection using clinicopathologic and molecular factors that are correlated with the presence of EGFR mutations (23). For EGFR copy number gain, what is most striking is the low sensitivity and specificity for predicting response, although it should be emphasized that this finding cannot be extrapolated to imply an inferior predictive ability for alternative outcomes, such as overall survival. However, the data indicate that there is a clear difference in the predictive accuracy of somatic EGFR mutations compared with gene gain. The increasing response rate to single-agent TKIs as one examines patients with KRAS mutations (lowest repose rate), wild-type, EGFR gene gain, and EGFR mutation (highest response rate) may have bearings on the anticipated differences in survival outcomes. Indeed, we have recently completed analysis for the effect of EGFR mutations on survival and, as has been expected by the community, the higher response rate among patients with mutations compared with wild-type patients seems to be translated into improved disease-free and overall survival.⁸

It is important to note that irrespective of whether these biomarkers are prognostic or predictive of response to EGFR TKIs, this cannot be established at this point, even after thousands of patients have been treated with TKIs and analyzed for EGFR mutation and/or gene copy number. The predictive or prognostic nature of a biomarker can only be established by the conduction of randomized trials. Interestingly, although several retrospective translational investigations based on phase III randomized studies were included in this review, they only contributed a small number of patients. It must, however, be noted that removal of these studies from this particular data set had no significant effect on the summary sensitivity and specificity, supporting the notion that it is unlikely that the estimated predictive accuracy of somatic EGFR mutations will be altered by the publication of additional studies. Several authors have extensively commented on the lack of tissue availability from large randomized studies (25, 44), and this is indeed one critical factor associated with a lack of guidance from those studies. The fact that even 5 years after the discovery of EGFR mutations their relevance to treatment selection has not yet been established supports the notion that the development of novel targeted agents should incorporate prospective tissue procurement and translational studies (45).

Although this study provides large-scale assessment of the predictive value of EGFR mutations and EGFR gene copy gain, several limitations should be taken into account when interpreting the results. First, response rates, the “gold standard” toward which mutations and gene gain were compared in this review, may be a poor surrogate for more clinically relevant outcomes such as overall or progression-free survival. Be that as it may, a large amount of evidence indicates that response rates can serve as surrogates for survival. Several meta-analyses in advanced lung, colorectal, and breast cancer (46–49) add support to this. A second limitation of this review is the unavailability of individual patient data that would have allowed for detailed subgroup analyses. This would have allowed an investigation of the effect of important covariates such as smoking, gender, and histology on the predictive value of EGFR mutations and EGFR gene copy gain. Finally, studies providing 3-fold stratification of responders and nonresponders, by KRAS, EGFR mutation, and gene copy status, are scarce, thus limiting the potential to evaluate the combined effects of different biomarkers. A further complication to this is that although KRAS and EGFR mutations are mutually exclusive, gene gain and mutational status are not (50). Therefore, the absolute potential of either EGFR mutations or EGFR gene copy number cannot be precisely examined without an individual patient data analysis. We have also recently presented data indicating that KRAS mutations are a mechanism associated with resistance to EGFR-targeted agents (26). Considering that these nonresponders fall into the wild-type population of this analysis, points to the necessity of a global collaborative individual patient data analysis.

In conclusion, we have reviewed the literature correlating EGFR mutations and gene copy number with response to single-agent erlotinib and gefitinib in the treatment of NSCLC. Overall, somatic EGFR mutations seem to be a good marker for predicting response to EGFR inhibitors independent of other clinical parameters. EGFR gene gain however seems less predictive of response to EGFR TKIs and, at least for this end point, cannot be considered clinically suitable for patient selection without further study. Additional studies are warranted for refining these predictive markers and their clinical correlates, for investigating their role in predicting survival improvement, and for clarifying their interaction with other molecular markers, such as KRAS mutations and HER2 gene status.

S. Murray, consultant or advisor for Merck, AstraZeneca; P. Kosmidis, consultant for AstraZeneca. The other authors declare no potential conflicts of interest.

Grant Support: I.J. Dahabreh is supported by a research fellowship provided by the “Maria P. Lemos” Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.