Background
The monoclonal antibody inhibitor of epidermal growth factor receptor (EGFR)- cetuximab adds modest benefit to front-line chemotherapy in patients with previously untreated advanced non-small cell lung cancer (NSCLC)[
1,
2]. This benefit is cost-ineffective by reference standards [
3,
4]. The role of tumor EGFR expression in predicting benefit from adding cetuximab in NSCLC is under investigation [
5]–[
7]. For the small molecule EGFR inhibitor (EGFRi) erlotinib in NSCLC, skin rash and serum proteomic assays were initially identified as candidate predictive biomarkers [
8]–[
10]. These findings suggested that non-tumor biomarkers might predict subsets of patients likely to benefit from administration of EGFRis, including cetuximab. A retrospective landmark time-point analysis in the BMS099 trial [
2] and a pre-specified subset analysis in the FLEX [
7] trial demonstrated better outcomes for patients who developed rash by the end of cycle 1 with cetuximab added to front-line therapy. These findings suggested that rash is a candidate predictive clinical biomarker for cetuximab in front-line NSCLC therapy but to validate the marker would require another large, randomized prospective study [
11].
The scales for severity of rash from antineoplastic drugs are crude instruments for determining therapeutic decisions. The scales are structured to equate severity across the full spectrum of adverse events and are designed as tools for safe conduct of clinical trials. They are not validated or intended as biomarkers for treatment effects [
12]. To ascertain cetuximab-induced rash, Gatzemeier, et al. collapsed 20 different terms from the MEDRA 10.0 adverse event database [
7]. The prospective use of clinician detection of rash as a biomarker might require more structured assessment. The pathophysiology of the rash associated with EGFR inhibitors is most reminiscent of acne rosacea (rosacea)[
12], for which a validated severity scale has already been developed [
13].
A serum mass spectrometry-based classifier assay segregates candidates for EGFRi therapy into “good” and “bad” prognosis groups [
9]. In multiple studies of EGFRis, the patients in the two groups had divergent survival patterns, but not in NSCLC patients treated with cytotoxic therapy, suggesting that the serum proteomic profile is a predictive rather than prognostic marker. In studies of erlotinib in NSCLC: for first-line treatment [
14] and when combined with bevacizumab therapy [
15] the assay again predicted survival differences. The assay performed similarly in multiple clinical studies of patients with squamous cell carcinoma of the head and neck and in colorectal cancer [
16], including subsets of these patients treated with cetuximab. Although commercially available as VeriStrat® (Biodesix, Broomfield, USA), the assay has not been prospectively qualified as a predictive marker for EGFRi therapy.
We conducted a randomized, phase II trial in previously treated NSCLC patients that compared the concurrent and sequential administration of cetuximab and pemetrexed. A component of the trial was to conduct preliminary development of biomarkers for cetuximab. Specifically for this trial, we adapted the validated rosacea severity scale as an EGFRi-induced rash (EIR) scale for better differentiation than the ordinal CTCAE scale of rash assessments. Additionally, serial serum samples were collected and stored at -80°C. The randomization and 2-week run-in design of the trial enabled concurrent analysis of rash rating and the serum assay as candidate predictive markers for cetuximab therapy in NSCLC.
The trial closed prior to achieving its initially intended accrual goals. The original hypothesis to be tested by the trial was to determine in patients with NSCLC refractory to previous chemotherapy whether concomitant treatment with cetuximab and pemetrexed improved progression-free survival compared with cetuximab monotherapy. We estimated a sample of 40 patients per treatment group based on many assumptions, including a relatively modest difference in outcomes between the study arms. Our estimates were inaccurate, and patients who were randomized to the combination therapy arm had much better outcomes than patients in the monotherapy arm. Additionally, large clinical trials of pemetrexed revealed advantages to early initiation of pemetrexed after first-line therapy, and single-arm studies of cetuximab in unselected patients revealed no benefit to cetuximab monotherapy in this population. Therefore the study was closed after reaching half of its intended accrual. The preponderance of available evidence now supports that the administration of pemetrexed in the second-line treatment setting would be superior to cetuximab monotherapy. We report the results of this trial not to demonstrate a definitive assessment of the difference in the randomized treatment arms, but rather to provide the foundation for this biomarker development exercise.
The trial was intended to study pharmacodynamic biomarkers in unselected patients and collection of tumor tissue was not incorporated into the study design. But the study called for all patients to provide pre-treatment serum samples and to undergo standardized, prospective rash evaluations. There are many limitations to the successful development of valid clinically relevant biomarkers concurrently with novel anticancer agents. The need to collect material for markers prospectively, combined with the observation time required to accumulate outcome data with which to qualify biomarkers makes the process slow and expensive. In this manuscript we have examined the data from this trial with 3 goals: 1) to explore the use of quantitative changes in tumor burden at an early time-point in a trial as a measure of drug effect (rather than RECIST-based response or time to progression), 2) to determine whether the prospectively collected biomarker data in this trial are consistent with retrospective analyses of larger clinical trials, and 3) to perform a preliminary estimate of the value of intensive, quantitative rash assessments as a candidate biomarker for future studies of EGFR inhibitors.
Discussion
This exploratory randomized phase II trial with a 2-week run-in of cetuximab monotherapy could inform future development of predictive biomarkers for EGFRi therapy in NSCLC. The biomarker development concepts implemented included: 1) piloting a modified version of the validated rosacea scale for EIR-rating, 2) use of modified quantitative assessment of tumor size changes to assess marker relationships, and 3) concomitant assessment of candidate markers, rash and serum proteomic profiling in the same patients. The EIR scale distributed rash ratings across a larger ordinal range than the conventional CTCAE rash scale; however this scaling did not strengthen the association with treatment outcomes beyond that of the categorical “rash vs. no rash” approach employed by Gatzemeier, et al. [
7]. The serum proteomic predictor was associated with treatment outcomes in this study of cetuximab and pemetrexed in the second-line treatment of NSCLC, although this finding might be due to confounding with treatment arm.
As a small trial, not meeting its intended accrual goals, there are clear limitations to this study. The setting of advanced NSCLC meant many patients deteriorated during or immediately after investigational treatment. As the value of information was not clear at the time of conducting the trial, we did not assess candidate molecular markers on patients’ tumors and we did not stratify the randomization by tumor histology. We had competing trials at that time that required information on tissue type for enrollment and a PS of only 0 or 1. This study unintentionally enrolled a typically more ill population with few never smokers (10%), fewer than usual women (under 40%), and many patients with non-adenocarcinoma or unknown histology (70%). The actual sample size meant only large effects, such as the benefit of immediate initiation of pemetrexed, could be detected with conventional measures such as progression-free and overall survival. The randomization did not effectively free the assessment from bias. This is highlighted by the unusually poor median-survival time in Arm A, 3.5 months is even low for a best-supportive care only trial. Arm A had fewer women, had patients with a higher median age, and more patients with serum proteome-profile-predicted “poor” outcomes with cetuximab monotherapy. It is therefore unclear the extent to which the treatment assignment versus the small sample size have biased the outcomes of this trial by study arm. The results are consistent with prior evidence of the value of timely treatment with pemetrexed in second-line treatment of NSCLC [
29,
30]. Despite not reaching original accrual goals for the study, Arm B subjects, treated concurrently with pemetrexed and cetuximab, had better progression free and overall survival than Arm A subjects treated with sequential cetuximab followed by pemetrexed, reaching statistical significance for the latter. Notably, of the 20 evaluable patients who were randomized to Arm A, 10 did not maintain sufficiently good performance status to receive the pemetrexed after progression.
The EIR scale distributed ratings more broadly than CTCAE, but there is no evidence that this broader distribution will improve upon use of rash severity as a predictive marker for cetuximab efficacy in NSCLC. Serial skin biopsies were performed on a subset of patients in this study, but the heterogeneity of the specimens in terms of estimated volume, estimated total number of cells, enumerated EGFR-expressing cells and ratio of EGFR-expressing to total cells, as well as relative mRNA expression of candidate normalization molecules (data not shown) made semi-quantitative and correlative analyses impossible. Dose-to-rash studies with EGFRi have not revealed a significant benefit for increasing the severity of rash [
31,
32]. Our data are consistent with the observations of others in larger trials, that incidence and not relative severity of EIR appears to be the pharmacodynamically relevant biomarker [
7,
10].
Cetuximab monotherapy has no evident value in the treatment in unselected NSCLC patients in the second-line setting. Some could argue that cetuximab therefore has limited if any role in combination therapy. However, the data from Gatzemeier, et al. [
7], suggest that a biomarker-based selection of patients who should receive cetuximab added to standard chemotherapy could yield improved outcomes over those reported for the cetuximab arm in the FLEX trial. This circumstance is increasingly common in oncology therapeutics, an agent that has limited but evident benefit in combination cannot be used ethically as monotherapy. Therefore, the early development monotherapy trials become an important opportunity with which to characterize candidate biomarkers and conduct preliminary validation and comparative estimate studies. In this “pure” setting, investigators can determine the typical time course, intensity, and interindividual variance in candidate markers. In a single disease, investigators can conduct preliminary comparisons to make estimates regarding which markers represent the best opportunities for future validation and qualification studies in large trials, including combination therapy trials.
For future qualification of candidate markers of EGFRi in the treatment of NSCLC, we propose that either the serum proteomic assay or incidence of rash be further evaluated as a means to exclude patients from receipt of cetuximab therapy. In patients who have the “poor” proteomic profile and those who fail to develop rash by 21 days of cetuximab therapy the likelihood of benefitting from cetuximab therapy appears low. In NSCLC, these markers, similar to the use of
KRAS mutations in colorectal cancer, have reproducibly associated with absence of benefit from EGFRi therapy [
14]–[
16,
33]. Our findings suggest future strategies to qualify these biomarkers for clinical use would be to demonstrate prospectively in a randomized trial that either or both markers effectively reduces the unnecessary, toxic, ineffective, and expensive use of cetuximab [
11]. Ideally, this study should help to identify safe and more effective alternatives for the patients who will not benefit from cetuximab therapy.
Competing interests
CHC participated in an ad hoc advisory board meeting for and received compensation from Biodesix during the conduct of this investigation.
Authors’ contributions
MLM conceived of the initial protocol design with EEV and together drafted the protocol. In the first year of the study EEV served as principal investigator and MLM served the remaining years, coordinated efforts of the co-authors on sample and statistical analyses and interpretation and with MRL organized data, drafted all figures, and the first draft of the manuscript. MEL conceived and developed the EIR rating scale, contributed to the design and conduct of the trial and performed serial skin biopsies on initial patients enrolled in the study and KS assumed those responsibilities for the remainder of the study. KEW performed statistical analyses and drafted part of the manuscript. CHC supervised all analyses with the serum proteomic predictor, interpreted study results and modified the manuscript. IOG supervised analyses of skin biopsies and participated in interpretation of rash rating results. LS and GR provided patient care and ensured adherence to the study protocol. MFK, PCH, and RS enrolled patients, provided patient care, performed rash ratings and disease response assessments. DPC contributed to study design, provided funding and technical support on serum proteomic predictor analyses. TGK developed the initial study design and supervised KEW in all study-related analyses and interpretations. All authors read, commented upon and approved the final manuscript.