Induction Chemotherapy with or without Erlotinib in Patients with Head and Neck Squamous Cell Carcinoma Amenable for Surgical Resection

Head and neck squamous cell carcinoma (HNSCC) accounts for more than 550,000 cases and 380,000 deaths annually worldwide and is the 6th most common cancer type (1). Currently, oral squamous cell carcinoma (OSCC) is the most common type of head and neck cancer (2), and more than half of OSCCs are diagnosed with locally advanced stages III–IV, with a high rate of local-regional recurrence and death from disease (3–5). While surgery has been the standard first-line treatment option for patients presenting with resectable OSCC, the adverse functional impact of major resections on speech and swallowing and poor oncologic outcomes have led head and neck oncologists to examine neoadjuvant systemic therapy. This approach could reduce the extent of resection and enhance local-regional control. Studies by our group and others have shown that induction chemotherapy can be safely administered to patients with locally advanced OSCC, and major responses can be seen, with responders having excellent disease control and survival. However, in randomized trials of induction chemotherapy with docetaxel, cisplatin, and 5-fluorouracil followed by surgery versus up-front surgery in locally advanced, resectable OSCC, the addition of systemic chemotherapy showed no significant benefit in terms of local-regional control or survival. With the advent of targeted agents directed against important growth pathways in OSCC, we were interested to investigate whether the addition of erlotinib, a small molecule inhibitor of EGFR, to chemotherapy could increase response, disease control, and survival.

Our rationale to target the EGFR in OSCC was based on several decades of preclinical and clinical investigations that have shown EGFR signaling pathways to be important in HNSCC pathogenesis and progression (6). Early studies showed that EGFR overexpression was linked to worse clinical outcomes (7). EGFR inhibitors had been evaluated in clinical trials and found to have activity in HNSCC (8, 9). Erlotinib is an oral tyrosine kinase inhibitor that competes with ATP for the intracellular tyrosine kinase domain of EGFR, which impairs receptor phosphorylation and downstream signaling (10). Erlotinib is FDA approved for the treatment of patients with EGFR-mutated lung cancer and it has been extensively studied in this patient population. In a window-of-opportunity study, erlotinib was able to reduce tumor size compared with controls (11). We have previously demonstrated activity of erlotinib in HNSCCs as monotherapy prior to surgical resection (12), or in combination with chemotherapy for recurrent/metastatic disease (13). On the basis of these results, we developed the current randomized trial to determine whether the addition of erlotinib to induction chemotherapy with a platinum-taxane regimen could increase response, disease control, and survival for patients with locally advanced resectable OSCC.

In this study, we randomized patients with stage III–IV OSCC to receive induction chemotherapy prior to definitive surgery, with erlotinib or with placebo. We performed whole-exome sequencing (WES) and RNA sequencing (RNA-seq) on pretreatment biopsy specimens to identify potential biomarkers of response that could be used to guide treatment selection.

This is a randomized, placebo-controlled, single-center, phase II trial of induction chemotherapy (cisplatin 75 mg/m² or carboplatin AUC 6 with docetaxel 75 mg/m² every 3 weeks for 3 cycles) with erlotinib (150 mg oral daily) or placebo in patients with OSCCs stage III–IVB amenable for surgical resection (NCT01927744). This study was approved by the Institutional Review Board of The University of Texas MD Anderson Cancer Center (MDACC; protocol ID 2013–0179), which complied with the Declaration of Helsinki, and it was conducted according to these ethical guidelines. Eligible patients must have been 18 years and older, able to consent, and have adequate organ function. Surgeries were performed within 3 to 6 weeks after the completion of chemotherapy.

This randomized phase II clinical trial was designed to be divided in two stages: Stage I (biomarker discovery/testing of previous findings) and Stage II (biomarker-guided therapy), with 50 patients in each stage.

Stage I had two parts. The first 30 patients would be equally randomized between two arms, stratified by lymph node status (N0/1 vs. N2/3). Subsequently, patients 31 to 50 would be adaptively randomized between the two arms on the basis of major pathologic response (MPR) and nodal status. MPR was defined as complete pathologic response or a partial pathologic response to therapy with minimal residual disease/near total effect, that is ≤10% viable tumor cells, as scored by the trial pathologist. MPR was originally designed to exclude any patient with lymph node positivity, however, studies demonstrated that the response to chemotherapy in lymph nodes can be delayed relative to the tumor (14). Therefore, lymph node status was removed from our definition of MPR. The primary endpoint was MPR. Secondary endpoints included safety and long-term efficacy outcomes [progression-free survival (PFS) and overall survival (OS)] by treatment groups. For all patients, long-term PFS was defined as the time interval between date of randomization and dates of local recurrence, regional recurrence, distant recurrence, or death, whichever occurred first, and was censored at the last follow-up date for patients who neither recurred nor died. For patients who had definitive treatment, progressions that occurred after definitive treatment were considered as events. OS was defined as the time interval between date of randomization and date of death and was censored at the last follow-up date for patients who were alive.

In the initial trial plan, Stage II would be carried out upon the identification of a predictive biomarker of Chemotherapy/Erlotinib, named as the erlotinib-associated biomarker (EAB); the next 50 patients would be adaptively randomized on the basis of response, nodal status, and the predictive biomarker based on the Bayesian probit model to maximize the response probability. The primary endpoint would be the MPR.

An interim analysis would be carried out after all 50 patients enrolled in the first stage were evaluated. Three possibilities could emerge from the interim analysis: Interim result A (erlotinib-based therapy not beneficial), Interim result B (erlotinib-based therapy beneficial, but EAB could not be defined during stage I), and Interim result C (erlotinib-based therapy beneficial with corresponding EAB defined during stage I). With Interim result A, the subsequent 50 patients in Stage II would be treated with open-label cisplatin/carboplatin and docetaxel. With Interim result B, the subsequent 50 patients in Stage II would be adaptively randomized between the two arms on the basis of MPR and nodal status. With interim result C, the subsequent 50 patients in Stage II would be adaptively randomized between the two arms on the basis of MPR, nodal status, and EAB. Simulation studies were carried out in each setting. With the estimated MPR rates in the chemo arm versus chemo + erlotinib arm are 0.1 versus 0.3, 0.09 versus 0.31, and 0.05 versus 0.35 for Interim results B or C, respectively; the statistical powers for detecting the difference in MPR rate between the two arms are at least 74%, 81%, and 97% with the one-sided 10% type I error by the Fisher exact test. The minimum sample size of the smaller of the two arms is assumed to be 37. In addition, the overall powers for testing the marginal treatment effect and the treatment by EAB interaction are 93% and 71% for the Interim result B or C.

The interim analysis identified Interim result A (erlotinib-based therapy not beneficial). Although the trial was designed to continue recruiting following this result, due to the absence of efficacy, relatively slow enrollment, and the overall lack of enthusiasm for erlotinib-based induction therapy following the emergence of immunotherapy as a treatment modality for HNSCC, it was decided to end the trial at the interim analysis point. Stage II was not carried out.

Patients were randomized to either chemotherapy/placebo or chemotherapy/erlotinib using a website housed on a secure server at MDACC and maintained by the MDACC Department of Biostatistics. This was a double-blinded randomized trial; placebo was a pill with the same physical appearance as the erlotinib tablet, without active ingredient. When a patient was enrolled on the study, the patient's information regarding stratification factors was entered into the database by the study personnel. The patient was assigned electronically to a treatment arm.

Information on the treatment arm was not available to the research personnel that had direct contact with the patient (e.g., research nurses and investigators). Through the web interface, the MDACC pharmacy entered the patient medical record number and the information on the treatment arm was available on the screen for proper medication dispensing.

Descriptive statistics were used to summarize the outcomes. Continuous variables were summarized by median and range, while categorical variables were summarized by frequency and percentage. The χ² test or Fisher exact test were used to evaluate the association between two categorical variables. Kaplan–Meier method was used for time-to-event analysis including PFS and OS. Log-rank test was performed to assess the difference of time-to-event outcomes among different groups. For other binary secondary endpoints, logistic regression model was used to assess the effect of treatment and biomarker. Cox proportional hazards models were used to assess the effect of treatment and biomarker on time-to-event outcomes (PFS and OS).

All patients who received at least one dose of any of the study drugs were considered evaluable for all safety analyses. All randomized patients were considered evaluable for all efficacy analyses according to the intent-to-treat principle.

For identification of potential predictive biomarkers, we collected the tumor tissue during the diagnostic biopsy and blood drawn prior to patient randomization. The fresh tumor tissue and blood were processed for nucleic acid extraction, which was later used for transcriptome and WES.

Total RNA was isolated from fresh tumor using the RNeasy Kit (Qiagen) and libraries were prepared using the TruSeq Stranded Total RNA Sample Prep Kit (Illumina, Inc., San Diego, CA). The libraries were pair-end sequenced in an Illumina HiSeq3000. Raw sequencing BCL files were processed using the CASAVA tool and converted into FASTQ format. Raw reads were aligned to the hg19 human genome using the STAR aligner (15) and transcripts abundance were estimated using Cufflinks (16). Data quality control was accessed using Multi-QC. Gene expression data was analyzed with the R package edgeR and gene expression differences were accessed using the quasi-likelihood F test (17, 18).

Genomic DNA extracted from fresh tumor and peripheral blood was used for library preparation and exome capture with the SeqCap EZ Exome Probes v3.0 Kit (Roche). Exome libraries were sequenced on a HiSeq 4000 (Illumina Inc.) using Cycle Sequencing v3 Reagents (Illumina). Raw sequencing BCL files were processed using the CASAVA tool and converted into FASTQ format. FASTQ files were aligned to the reference genome (human Hg19) using BWA 17. The aligned BAM files were subjected to mark duplication, realignment, and recalibration using Picard and GATK (19) before any downstream analyses. Mutation calls were made using the MuTect, and insertions and deletion (indel) calls using Pindel (20). Combined results were manually filtered prior to downstream analysis. WEX filtered variants organized in an MAF file were analyzed using the R package Maftools (15, 17, 18, 21).

Pathway enrichment analyses were performed in web server KOBAS 3.0 which parses 4 pathway databases, including Reactome, PANTHER, KEGG Pathway, and BioCyc, and the functional category database Gene Ontology (22). Enrichment significance is assessed by Fisher exact test and FDR correction using the Benjamini and Hochberg method.

Biomarker analyses were performed using JMP 15.0 Software. The same statistical methodologies employed for the trial analyses were applied for the biomarker investigation. Gene expression patterns were investigated by hierarchical clustering analysis performed using the Ward minimum variance method for defining distances between clusters. Cluster distances were computed on standardized expression values.

Raw data for this study were generated at Sheikh Zayed Bin Sultan Al Nahyan Institute for Personalized Cancer Therapy Lab (IPCT Lab) Department of Genomic Medicine MDACC. Derived data supporting the findings of this study are available within the article and its Supplementary Data files.

A total of 55 patients were screened and 52 met the study criteria. The first phase of the trial enrolled 30 patients with 1:1 randomization. The remaining 22 patients were adaptively randomized, with 9 to erlotinib and 13 to placebo (Supplementary Fig. S1 Consort). Among the 52 randomized patients, 47 underwent definitive surgery and their tumor samples were available for biomarker analysis.

The median age of patients was 58 years, with 60% male patients. Most patients had a performance status of 0 to 1. In this cohort, 52% of the patients were never smokers, which is different than the expected proportion (Table 1). All tumors were squamous cell carcinoma, from various sites in the oral cavity with about half from the oral tongue. Patients at stage IVA represented 77% of the cohort, and 54% of the patients were N2/3 (Table 2). As the trial enrolled only oral cavity tumors, which are not commonly associated with human papillomavirus (HPV) infection, p16 or HPV status was only tested in 11 tumors, with 2 being positive for p16, but none for HPV.

A total of 45 (87%) patients completed all 3 cycles of chemotherapy with erlotinib or placebo. Ninety two percent (22/24) of the patients in the erlotinib arm completed treatment, whereas 68% (19/28) in the placebo group completed treatment (P = 0.046) (Supplementary Table S1).

Overall, the treatment toxicities were as expected. Treatment-emerged adverse events (Supplementary Table S2) were observed in all patients. Ninety-six percent of patients in both groups experienced treatment-related adverse events (TRAE) of any grade. Grade 3 to 4 TRAEs were reported in 33% of patients in the erlotinib group and in 36% of patients in the placebo group. The most common TRAEs in the erlotinib group were metabolic and nutritional (25%), nausea/vomiting (21%), and gastrointestinal disorders (13%). In the placebo arm, the most common TRAEs were fatigue (18%), nausea/vomiting (7%), gastrointestinal disorders (7%), cytopenia (7%), and pain (7%). Rash (with combined terms of rash, acnieform, rash maculo-papular, rash pustular, palmar-plantar erythrodysesthesia syndrome, and paronychia) were seen in 79% of patients in the erlotinib group and in 50% of the placebo group, but all events were grade 1 to 2 (Table 3).

MPR was achieved in 30% (7/23) of the cases in the erlotinib group and in 41.7% (10/24) of patients in the placebo group (P = 0.55, Fisher exact test). In the first 30 patients, 26.7% (4/15) MPR was observed in the erlotinib group and 45.4% (5/11 and 2 not evaluable) MPR was observed in the placebo group. In the subsequently enrolled patients, 9 were adaptively allocated to the erlotinib group and 13 to the placebo group. In the entire study, among the 47 patients who underwent surgical resection, 14 had recurrence at data cut-off, with 9 in the placebo group (9/24, 37.5%) and 5 in the erlotinib group (5/23, 21.74%; P = 0.341; Fisher exact test). There was no distinct pattern in loco-regional versus distant metastasis between the two treatment groups (Supplementary Table S3). With the median follow up of 26.5 months, the 2-year PFS probabilities in the two groups were 75.0% (erlotinib) and 58.6% (placebo), respectively. There was no significant difference in PFS distribution between the two groups (HR = 0.535; 95% confidence interval (CI), 0.197–1.451; Fig. 1A). The 2-year OS rates in the two groups were 76.0% (erlotinib) and 77.2% (placebo). There was no difference on OS distribution between the two groups (HR = 1.11; 95% CI, 0.36–3.43; Fig. 1B).

As expected, patients who achieved MPR had a trend towards improved PFS outcome, although the statistical power was probably limited by sample size (Fig. 2A). Patients who received erlotinib and achieved MPR had the best PFS outcomes (Fig. 2B), albeit not significant.

Next, we explored whether genetic alterations were associated with MPR in each study arm. Pretreatment WES data was generated for 34 patients (those with adequate tissue and sequencing quality): 19 treated with placebo (in which 7 achieved MPR and 4 did not receive surgery) and 15 treated with erlotinib (in which 5 achieved MPR and 1 did not receive surgery). Their mutation profile is depicted in the oncoplot in Fig. 3A and in Supplementary Table S4. The mutation profile of these samples was similar to the pattern observed in HNSCC from other cohorts such as those available at the The Cancer Genome Atlas (TCGA) database, which is characterized by a high frequency of TP53, TTN, FAT1, CDKN2A, PIK3CA, and NOTCH1 mutations. However, MUC16 and LRP1B mutations (observed in 20% and 16.6% of HNSCC in the TCGA database, respectively) were not detected among our samples, and CSMD3 and SYNE1 (19.2% and 16.8% of HNSCC-TCGA cohort, respectively) mutations were quite rare (2.9% of our samples for both genes). On the other hand, we detected higher frequencies of IGFN1 (11.8%), C17orf97 (11.8%), and DNAJC11 (11.8%) mutations than in the HNSCC-TCGA cohort (0.6%, 0.4%, 0.4%, respectively) as well as high frequency of MUC12 (26.5%), ENDOV (8%), HMCN2 (8%), and TUBB8P7 (8%) mutations that were not reported among HNSCC TCGA tumors. There was no gene mutation frequency differences associated with pathologic response status. Presence of MUC12 mutations were significantly associated with poor OS and PFS among patients in the erlotinib arm (Log-rank test P = 0.009 and P = 0.015, respectively; Supplementary Fig. S2). However, this result should be interpreted with caution because MUC12 mutations were identified in a region of the protein with low complexity (first 5,000 amino acids) and unclear function.

We compared the tumor transcriptome of patients that developed MPR with those that did not develop minimal pathologic response (NR) in each study arm to identify potential biological characteristics associated with treatment responsiveness (normalized transcriptome data for patients in both study arms are available at Supplementary Table S5). In the erlotinib arm, 472 genes were differentially expressed between patients with MPR and NR (140 downregulated and 332 upregulated; Supplementary Table S6). In the placebo arm, 642 genes were differentially expressed among patients with MPR and NR (301 upregulated in MPR and 341 downregulated in MPR; Supplementary Table S7). Only 9 genes (MREG, DCUN1D3, GTSF1, CT83, IL1B, NLRP10, MIR3171, MIR767, CALB1) were significantly upregulated and only 2 (FDCSP, IL17D) were significantly downregulated in patients with MPR in both study arms (Fig. 3B), suggesting that induction chemotherapy pathologic response is driven by a distinct biology in erlotinib and placebo-treated patients.

Considering that patients in the erlotinib arm who developed MPR had a better PFS and OS, we further explored their tumor molecular phenotype. Pathway analysis (Fig. 3C; Supplementary Table S8) showed that genes upregulated in tumors from patients with MPR significantly enriched pathways associated with signal transduction, specifically GPCR signaling pathways. However, genes associated with these upregulated pathways demonstrated poor discriminatory power to differentiate patients according to their pathologic response status by principal component analysis (PCA; Fig. 3D). Genes downregulated in tumors from patients with MPR enriched pathways associated to posttranslational modification, especially glycosylation, as well as Cadherin and Wnt signaling pathways (Fig. 3E; Supplementary Table S9). Genes associated with these top downregulated pathways showed good discriminatory power to differentiate patients according to their pathologic response status, suggesting they might represent better predictive biomarker candidates (Fig. 3F).

Most of the downregulated pathways linked to protein metabolism were enriched by the genes ST3GAL4, B4GALT1, and B3GALT5, which encode enzymes involved in protein glycosylation. Considering that the literature suggests a direct link between protein glycosylation, Wnt signaling, and EGFR-signaling activity, we searched for glycosylation enzymes and Wnt signaling genes (based on GO terms and KEGG pathways) among those downregulated in the MPR group. We found that 5 genes with putative glycosylation activity (ALG1L, B3GAT3, B3GALT5, B4GALT1, ST3GAL4) and 4 genes related to Wnt signaling (MMP7, NKD1, PORCN, SOSTDC1) were significantly downregulated among tumors from patients that developed MPR in the erlotinib arm. Hierarchical clustering analysis with these 9 genes revealed the existence of two distinct molecular groups among erlotinib-treated patients. One group was characterized by higher expression of these glycosylation and Wnt signaling genes (named as Group 1) compared with the second group (named as Group 2; Fig. 3G). The Group 1 (n = 8) was formed only by patients with NR, while the Group 2 (n = 11) was formed by all MPR and 4 patients with NR. Patients from the molecular Group 1 showed worse OS and PFS survival compared with those from Group 2 (Log-rank P value = 0.064 and 0.048, respectively; Figs. 3H–I). Although these genes could identify two populations of tumors within the placebo arm (Fig. 3J), they did not stratify MPR, OS, or PFS with statistical significance (Figs. 3K–L).

To evaluate whether this molecular classification was an independent prognostic factor in the erlotinib arm, we assessed whether clinical variables were associated with patient outcome. However, univariate analysis evaluating gender, age, smoking status, and clinical T and N stages showed that none of these clinical variables were associated with OS and PFS (Supplementary Table S10), likely because the sample size was not powered for clinical subgroup analysis.

Our study is a randomized clinical trial evaluating the benefit and safety of the addition of erlotinib to induction chemotherapy in stage III–IV OSCC prior to surgery. This trial was originally designed to include two stages, where the second stage would prospectively test a predictive biomarker identified in the first stage. During the interim analysis at the end of the first stage, the lack of efficacy for erlotinib and the emergence of immunotherapy as a treatment approach in HNSCC prompted an early end to the trial.

Our results showed that the addition of erlotinib to chemotherapy was well tolerated, but failed to confer additional benefit to induction chemotherapy. Overall, the clinical outcomes for this group of 52 patients who received induction chemotherapy were good with a 2-year recurrence-free survival at ∼60% and OS above 75%. The results were comparable to our single-center's historical data with induction chemotherapy (23). Our cohort had a high portion of never-smoker patients, which may have contributed to the good outcome.

Our randomized trial utilized a two-stage biostatistical strategy to allow maximal benefit to the patients. In the first stage of 30 patients, all patients were randomized at 1:1 ratio to each of the treatment arms. With the primary endpoint of MPR being a relatively rapid read-out (within 3 months of enrollment of a subject), the second stage of the trial was able to use an adaptive randomization design to assign patients to the arm with favorable outcome, without unblinding the treatments or pausing the trial enrollment. In our trial, the first 30 patients showed a slight favorable outcome towards the placebo group; therefore, our second stage assigned more patients to the placebo group (13 patients) than the erlotinib group (9 patients). This trial design maximally protects patients’ benefit in a placebo-controlled, randomized, phase II trial, without sacrificing the blinding and sample size. The overall conclusion of our study was consistent in both first and second stages.

Previous studies were not able to identify strong molecular signatures linked to erlotinib response in HNSCC (24). We also did not find a strong gene expression profile associated with erlotinib response, which may be due to the limited statistical power caused by the small sample size of the erlotinib-treated arm. However, it is also possible that the molecular profile associated to erlotinib responsiveness is not fully defined at the transcriptional level. This hypothesis can be supported by the significant enrichment of glycosylation-related pathways that are differentially expressed between responders and nonresponders.

Glycosylation alterations, usually characterized by glycosyltransferase transcriptional alterations, are key molecular events in cancer and may have an important therapeutic role (25).

The function of transmembrane glycoproteins are highly regulated by their glycosylation state. EGFR glycosylation defines its localization/sorting through the membrane, ligand binding activity/affinity, and structural conformation. It also controls ligand-independent oligomerization, EGFR endocytosis dynamics and its ectodomain orientation relative to the plasma membrane, interfering with its function (26–30). EGFR glycosylation state has been associated with erlotinib sensitivity. Inhibition of N-glycosylation of EGFR led to increased erlotinib effect in lung cancer cell lines (31). We identified a significant reduction in the expression of glycosyltransferase genes (B4GALT1, B3GALT5, ST3GAL4, and ALG1L) among patients that developed MPR in the erlotinib arm.

The B1,4-Galactosyltransferase 1, the enzyme coded by the B4GALT1 gene, directly interacts with EGFR, affecting its functionality and reducing its expression (32, 33). We did not observe an association between EGFR mRNA levels with pathologic response status in patients treated with erlotinib. However, glycosyltransferases such as B4GALT1, can reduce EGFR protein expression without affecting its transcript levels (33). Another glycosyltransferase, coded by the B3GALT5 gene, can also interfere with EGFR signaling by participating in the production of the CA19–9 glycan, which is responsible for increased EGFR downstream signaling activity independent of EGF stimulus (34). These molecular differences suggestive of altered glycosylation activity between patients with MPR and NR treated with erlotinib may indicate EGFR function variability in OSCC. Considering these findings, it is plausible to hypothesize that posttranslational modifications related to the function of the EGFR receptor are potential regulators of erlotinib response in HNSCC.

EGFR and Wnt/B-catenin pathway activities are directly related in epithelial tissues, including in oral cancer (35, 36). It has been suggested that elevated Wnt/B-catenin pathway genes are surrogate markers of resistance to EGFR-targeting drugs (37, 38). Our findings indicated a significant reduction in Wnt signaling genes among MPR patients treated with erlotinib. The gene PORCN is essential for Wnt signaling by promoting Wnt palmitoylation, which is important for its transport and ligation to Frizzled receptors. Thus, PORCN inhibition impairs Wnt-signaling pathway (39). MMP7, also downregulated among patients with MPR, is positively regulated by Wnt/B-catenin pathways (40). Lower expression of these genes is suggestive of downregulation of Wnt signaling among responders. Interestingly, increased Wnt/B-catenin signaling has been showed as an important mechanism of resistance to EGFR inhibition (41). Beyond that, we showed that concomitant lower expression of glycosylation and Wnt signaling genes was associated with improved survival outcome, irrespective of the pathologic response status.

There are a few limitations to our study. Since the conception of this trial, the therapeutic landscape for HNSCC has evolved significantly, particularly with the introduction of anti–PD-1/L1 immune checkpoint inhibitors that significantly prolong patients’ survival, especially in the metastatic setting (42, 43). EGF signaling with targeted therapy failed to establish efficacy, but immunotherapy has demonstrated survival benefit both alone or in combination with chemotherapy. Recent clinical trials evaluating induction therapies in HNSCC have incorporated immunotherapy components (44). Our study confirms that chemotherapy-based induction therapy is well tolerated and provides reference efficacies for future investigations. The neoadjuvant strategies in solid tumors have also evolved since the designing of this trial; with accumulated knowledge in defining response, the lymph node responses were recognized to often delay the tumor response, therefore, we modified our MPR definition according to exclude the lymph node status to best reflect current knowledge. In our biomarker analysis, because WES and RNA-seq were prioritized, most of the pretreatment tissue was depleted and IHC validation could not be performed. Furthermore, due to the limited sample size, our results are suggestive but require future validation. All together, our data support the hypothesis that erlotinib responders have an EGFR-dependent phenotype, in light of novel EGF signaling targeting strategies emerging clinically, future analysis and trials are warranted to explore this phenomenon to improve patients’ clinical outcomes.

X. Le reports grants and personal fees from EMD Serono, Eli Lilly, Regeneron, and Boehringer Ingelheim, as well as personal fees from AstraZeneca, Spectrum Pharmaceuticals, Novartis, Hengrui, Janssen, and AbbVie outside the submitted work. G. Blumenschein Jr reports grants and personal fees from Genentech during the conduct of the study. G. Blumenschein Jr also reports grants and personal fees from Amgen, CytomX Therapeutics, Daiichi Sankyo, AstraZeneca, Bristol Myers Squibb, Celgene, MedImmune, Merck, Novartis, Sanofi, Roche, Xcovery, Regeneron, and BeiGene; grants from Bayer, Adaptimmune, Elelixis, GlaxoSmithKline, Immatics, Immunocore, Incyte, Kite Pharma, MacroGenics, Tmunity Therapeutics, Repertoire Immune Medicines, and Verastem Oncology; and personal fees from Instil Bio, Genzyme, Gilead, Lilly, Janssen, TymeOncology, Virogin, Maverick Therapeutics, AbbVie, Adicet, Ariad, and BeyondSpring Pharma outside the submitted work. In addition, G. Blumenschein Jr reports an immediate family member is employed by Johnson & Johnson/Janssen. K. Marcelo-Lewis reports other support from Caris Life Sciences outside the submitted work. M. Gillison reports personal fees from Sensei, Mirati, Coherus, Debiopharm, Kura, Shattuck, Nektar, Ipsen, EMD Serono, Gilead, Eisai, Istari, LLX Solutions, Seagen, BioNTech, Merck, Bicara, Bayer, and Roche outside the submitted work, as well as clinical trial funding support from Genocea, Kura, Genentech, Bristol Myers Squibb, Agenus, RTOG, and Seagen. F. Meric-Bernstam reports personal fees from AbbVie, Aduro Biotech Inc., Alkermes, AstraZeneca, Debiopharm, eFFECTOR Therapeutics, F. Hoffman-La Roche Ltd., Genentech Inc., IBM Watson, Infinity Pharmaceuticals, Jackson Laboratory, Kolon Life Science, Lengo Therapeutics, OrigiMed, PACT Pharma, Parexel International, Pfizer Inc., Samsung Bioepis, Seattle Genetics Inc., Tallac Therapeutics, Tyra Biosciences, Xencor, Zymeworks, Black Diamond, Biovica, Eisai, Immunomedics, Inflection Biosciences, Karyopharm Therapeutics, Loxo Oncology, Mersana Therapeutics, OnCusp Therapeutics, Puma Biotechnology Inc., Seattle Genetics, Silverback Therapeutics, Spectrum Pharmaceuticals, Chugai Biopharmaceuticals, and Zentalis, as well as grants from Aileron Therapeutics, Inc., AstraZeneca, Bayer Healthcare Pharmaceutical, Calithera Biosciences Inc., Curis Inc., CytomX Therapeutics Inc., Daiichi Sankyo Co. Ltd., Debiopharm International, eFFECTOR Therapeutics, Genentech Inc., Guardant Health Inc., Klus Pharma, Takeda Pharmaceutical, Novartis, Puma Biotechnology Inc., and Taiho Pharmaceutical Co. outside the submitted work. G.B. Mills reports SAB/consultant fees from AstraZeneca, BlueDot, Chrysallis Biotechnology, Ellipses Pharma, ImmunoMET, Infinity, Ionis, Lilly, Medacorp, Nanostring, PDX Pharmaceuticals, Signalchem Lifesciences, Tarveda, Turbine, and Zentalis Pharmaceuticals; stock/options/financial from Catena Pharmaceuticals, ImmunoMet, SignalChem, Tarveda, and Turbine; and licensed technology from HRD assay to Myriad Genetics and DSP patents with Nanostring during the conduct of the study. W.N. William Jr reports grants from Kadoorie Foundation, as well as other support from Astellas Pharmaceuticals during the conduct of the study. W.N. William Jr also reports personal fees from Roche/Genentech, Boehringer Ingelheim, Bristol Myers Squibb, Eli Lilly, Pfizer, Takeda, Janssen, Sanofi-Aventis, Merck KGA and Novartis, as well as grants and personal fees from AstraZeneca and Merck outside the submitted work. J.N. Myers reports grants from Michael Kadorrie Foundation during the conduct of the study. C.R. Pickering reports grants from NIH and Kadoorie Charitable Foundation during the conduct of the study. No disclosures were reported by the other authors.

X. Le: Supervision, validation, investigation, writing–original draft, project administration, writing–review and editing. F.O. Gleber-Netto: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.L. Rubin: Formal analysis, visualization. Y. Qing: Formal analysis, visualization. R. Du: Investigation. M. Kies: Methodology. G. Blumenschein Jr: Investigation. C. Lu: Investigation. F.M. Johnson: Investigation. D. Bell: Investigation. J. Lewis: Data curation. J. Zhang: Data curation, software, formal analysis, visualization. L. Feng: Formal analysis. K. Wilson: Investigation. K. Marcelo-Lewis: Investigation. J. Wang: Data curation, software, formal analysis. L. Ginsberg: Investigation. M. Gillison: Investigation. J.J. Lee: Conceptualization, resources, data curation, software, formal analysis, supervision, methodology, project administration, writing–review and editing. F. Meric-Bernstam: Investigation. G.B. Mills: Investigation. W.N. William Jr: Conceptualization, resources, supervision, methodology, project administration, writing–review and editing. J.N. Myers: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. C.R. Pickering: Conceptualization, resources, supervision, investigation, methodology, project administration, writing–review and editing.

The Kadoorie Charitable Foundation, the Sheikh Khalifa Bin Zayed Al Nahayn Institute for personalized cancer therapy, and the UT MDACC supported the operation of the trial and translational analysis.

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.