Abstract
Purpose: The clinical impact of next-generation sequencing (NGS) in patients with head and neck squamous cell carcinoma (HNSCC) has not been described. We aimed to evaluate the clinical impact of NGS in the routine care of patients with HNSCC and to correlate genomic alterations with clinical outcomes.
Experimental Design: Single-center study examining targeted NGS platform used to sequence tumor DNA obtained from 213 HNSCC patients evaluated in outpatient head and neck oncology clinic between August 2011 and December 2014. We correlated tumor genomic profiling results with clinical outcomes.
Results: PI3K/RTK pathway activation occurred frequently [activating PIK3CA mutation or amplification (13%), PTEN inactivation (3%), RAS activation (6%), EGFR or ERBB2 activation (9%)]. Alterations in pathways affecting cell-cycle regulation [CCND1 amplification (9%), CDKN2A inactivation (17%), BRCA2 inactivation (2%)] and squamous differentiation [NOTCH1 inactivation (8%) andEP300 inactivation (6%)] were identified. PIK3CA amplification (n = 43), not PIK3CA mutation, was associated with significantly poorer progression-free survival (P = 0.0006). Oncogenic RAS mutations (n = 13) were associated with significantly poorer progression-free survival (P = 0.0001) and lower overall survival (P = 0.003). Eight patients with advanced, treatment-refractory HNSCC enrolled on clinical trials matched to tumor profiling results, and 50% achieved a partial response.
Conclusions: Incorporation of NGS clinical assays into the routine care of patients with HNSCC is feasible and may readily facilitate enrollment into clinical trials of targeted therapy with a higher likelihood of success. Data can be utilized for discovery of genomic biomarkers of outcome. PIK3CA amplification and RAS mutations were frequently identified and associated with poorer prognosis in this cohort. Clin Cancer Res; 22(12); 2939–49. ©2016 AACR.
Precision medicine approaches for patients with head and neck squamous cell carcinoma (HNSCC) are lacking. Unlike many other cancer types, no validated biomarkers exist to predict response to targeted therapies in HNSCC, and no targeted therapies have been approved by the Food and Drug Administration for HNSCC since 2006. Several large genomic surveys of HNSCC have been published; however, the lack of clinical annotation of the studied samples has limited the ability to determine the clinical significance of genomic variants. Here, we report for the first time, the incorporation of targeted NGS in the routine clinical care of 213 individuals with HNSCC, and correlate genomic alterations with clinical outcomes. For the first time, we identified that PIK3CA amplification and RAS mutations were associated with inferior outcome, suggesting candidate biomarkers for prognostic validation and biomarker-driven clinical trials. We demonstrate examples of personalized approaches in advanced HNSCC patients in which genomic information was used to select clinical trials of targeted therapy resulting in a higher likelihood of response. Our findings suggest the incorporation of molecular analyses into the clinical care of patients with HNSCC may enable biomarker discovery efforts and enhance enrollment of biomarker-driven clinical trials.
Introduction
Head and neck squamous cell carcinoma (HNSCC) is the sixth most common cancer worldwide (1). Effective personalized medicine approaches, including the evaluation of biomarker-driven targeted therapies, are lacking in this disease. Risk stratification and treatment selection are largely determined by traditional American Joint Committee on Cancer (AJCC) staging parameters. The only molecular marker currently used routinely for risk stratification is the presence of high-risk human papilloma virus (primarily HPV-16) in oropharyngeal cancer, although its impact is tempered by smoking history (2). The sole FDA-approved targeted therapy in HNSCC is cetuximab, a monoclonal antibody against the epidermal growth factor receptor, which is associated with a response rate of 10% to 15% in the recurrent or metastatic setting (3), and for which there is no validated biomarker to guide patient selection. Recent characterization of the genomic landscape of HNSCC has offered insight into underlying heterogeneity and discovery of potential new therapeutic targets (4–9). However, these studies largely consisted of retrospective surgical cohorts without long-term follow-up or data on response to conventional or investigational therapy. These studies have identified frequent mutations of several genes including TP53, PIK3CA, CDKN2A, NOTCH1, and MLL2 among others as well as copy-number alterations in EGFR, CCND1, PIK3CA, and FGFR1 (4–9). These genes, however, have yet to translate into clinically useful prognostic or predictive biomarkers in HNSCC, although some such as TP53 mutations may be associated with poor postoperative outcomes (10).
The application of next-generation sequencing (NGS) to the clinical setting has not been fully described in the care of patients with HNSCC, and the clinical significance of the most frequently recurrent genomic alterations is largely unknown. The objective of this study was to examine the clinical utility of NGS in the routine clinical care of patients with HNSCC and correlate significantly mutated, deleted, and amplified genomic alterations with clinical outcomes. Here, we report our findings from NGS data generated in a Clinical Laboratory Improvement Amendments (CLIA)-certified clinical laboratory obtained from tumor specimens of 213 HNSCC patients evaluated at our center. For the first time, we demonstrate how incorporation of tumor genomic profiling into routine clinical care could readily facilitate HNSCC patient enrollment into clinical trials of targeted therapy leading to objective responses and offer insight into candidate prognostic biomarkers.
Materials and Methods
Patient selection and data collection
Patients with HNSCC who received outpatient care in the Head and Neck Oncology Program at the Dana-Farber Cancer Institute between August 2011 and December 2014 and consented to an Institutional Review Board (IRB)–approved, institute-wide research protocol that aimed to assess feasibility and utility of genomic profiling and potential targetable alterations in cancer patients were identified. Data analysis of tumor specimens and abstraction of pertinent clinical data from patient chart review was conducted in accordance with an IRB-approved protocol. All tumor specimens underwent secondary histopathologic review by a dedicated pathologist for confirmation of the diagnosis and identification of optimal areas of the section for isolation of tumor DNA. Human papillomavirus (HPV) status was determined by p16 immunohistochemistry and classified as negative (weak or no staining) or positive (strong and diffuse). Progression-free survival (PFS) was measured as the time from initial diagnosis of HNSCC until the time of first disease relapse, progression or death, or until last contact if none of these events occurred. Overall survival (OS) was defined as the time from diagnosis until the time of death or until last contact.
Genetic analysis
All sequencing assays were performed in a CLIA-certified laboratory within the Center for Advanced Molecular Diagnostics Laboratory of the Brigham and Women's Hospital. DNA was isolated from 5 to 10.5 μM FFPE slides containing at least 20% tumor nuclei using routine extraction methods as previously described (11, 12) and analyzed by massively parallel sequencing using a solution-phase Agilent SureSelect hybrid capture kit and an Illumina HiSeq 2500 sequencer. MuTect and GATK were used to detect single-nucleotide variants (SNV) and indels, and VisCap was used to detect copy-number variations (CNV).
Samples were initially sequenced utilizing the first-generation platform, Oncomap, a multiplex mass spectrometry–based platform capable of detecting 471 loci from 41 cancer genes which was instituted in August 2011 (12). In August 2013, Oncomap was replaced by a second-generation platform, Oncopanel, a multiplexed targeted assay which surveys exonic DNA sequences of 275 cancer genes to identify somatic mutations and CNVs, in addition to 91 introns from 30 genes to detect rearrangements (11). Alterations are tiered based on clinical significance, and results are communicated in a report to the treating physician with a turnaround time on the order of weeks.
Variant calling
Alterations were classified into five tiers using the following guidelines by an expert panel at our institution. Tier 1 designation was used for alterations with published evidence confirming clinical utility in HNSCC in predicting response to FDA-approved therapy, prognosis, diagnosis, or increased inherited risk of cancer. Tier 2 classification applied to alterations with clinical utility in selection of an investigational therapy in clinical trials for HNSCC, or proven association of response to treatment with an FDA-approved therapy in a different type of cancer. Tier 3 encompassed alterations of uncertain clinical significance but with association with treatment response in preclinical HNSCC studies, or selection of an investigational therapy for a different cancer type. Tier 4 was used for novel alterations or those that have not been studied in cancer. Tier 5 described alterations that have no clinical utility.
Copy-number changes were detected as per previously described methods (11): amplifications were defined as mean sequence coverage >3-fold greater than reference normal, and deletions were defined as mean sequence coverage >3-fold lower than the reference normal.
Statistical methods
Statistical analyses to test for correlation between genomic and clinical features were performed using standard R packages. We used the Fisher exact test for discrete variables, the log-rank test for continuous variables, and the Bonferroni method of multiple testing correction. Median follow-up time was calculated using a reverse Kaplan–Meier estimate. Mutual exclusivity significance was calculated using the muex R package.
Results
Clinical characteristics of the cohort
Our cohort consisted of 158 males and 55 females, with a median age of 59 years at diagnosis of HNSCC (Table 1). Primary tumor sites were predominantly of the oropharynx (n = 97, 46%), and oral cavity (n = 60, 28%). Most patients (n = 123, 58%) had stage IVA disease at diagnosis. Forty-eight percent of patients had p16-positive tumors, and the vast majority of oropharynx cancer cases (89%) were p16-positive. The majority of patients (52%) had a smoking history of 10 pack years or more, whereas only one third had never smoked. Heavy smoking history was associated with larynx cancer (P = 0.002) and age over 55 years old (P = 0.002). Tumor DNA was sequenced from 213 patients with available HNSCC tumor specimens, which were obtained from the primary tumor (n = 155), neck lymph node (n = 15), locoregional recurrence (n = 35), or distant metastasis (n = 12). Of the 213 specimens tested, 97 were tested by Oncomap, and 116 were tested by Oncopanel. A total of 140 specimens had at least one genomic alteration and 73 tested negative [negative rate was higher for Oncomap (66/97=68%) compared with Oncopanel (7/116=6%)].
Characteristics of 213 patients with HNSCC
Characteristic . | N . |
---|---|
Age, median (range) in years | 59 (22–88) |
Sex | |
Male | 158 |
Female | 55 |
Primary site | |
Oral cavity | 60 |
Oropharynx | 97 |
Larynx | 24 |
Hypopharynx | 6 |
Nasopharynx | 7 |
Paranasal sinus | 5 |
Unknown primary | 14 |
Tumor stage | |
T1-2 | 145 |
T3-4 | 54 |
Unknown (Tx) | 14 |
Nodal stage | |
N0 | 64 |
N1 | 27 |
N2 | 111 |
N3 | 8 |
Unknown (Nx) | 3 |
AJCC stage | |
I | 20 |
II | 29 |
III | 29 |
IVA | 123 |
IVB | 6 |
IVC | 5 |
Unknown | 1 |
Tobacco exposure | |
Never smoked | 70 |
≤10 pack-years | 31 |
>10 pack-years | 110 |
Unknown | 2 |
P16 expression across all tumors | |
Positive | 103 |
Negative | 34 |
Unknown | 76 |
P16 expression in oropharyngeal primary tumor | |
Positive | 86 |
Negative | 8 |
Unknown | 3 |
Treatment | |
Upfront surgery | 91 |
Definitive radiation | 9 |
Definitive chemoradiation | 71 |
Sequential therapy | 31 |
Palliative systemic therapy | 2 |
Unknown | 9 |
Type of first treatment failure | |
Locoregional disease | 58 |
Distant metastases | 18 |
Both | 8 |
Assay | |
Oncomap positive | 31 |
Oncomap no variants identified | 66 |
Oncopanel positive | 109 |
Oncopanel no variants identified | 7 |
Characteristic . | N . |
---|---|
Age, median (range) in years | 59 (22–88) |
Sex | |
Male | 158 |
Female | 55 |
Primary site | |
Oral cavity | 60 |
Oropharynx | 97 |
Larynx | 24 |
Hypopharynx | 6 |
Nasopharynx | 7 |
Paranasal sinus | 5 |
Unknown primary | 14 |
Tumor stage | |
T1-2 | 145 |
T3-4 | 54 |
Unknown (Tx) | 14 |
Nodal stage | |
N0 | 64 |
N1 | 27 |
N2 | 111 |
N3 | 8 |
Unknown (Nx) | 3 |
AJCC stage | |
I | 20 |
II | 29 |
III | 29 |
IVA | 123 |
IVB | 6 |
IVC | 5 |
Unknown | 1 |
Tobacco exposure | |
Never smoked | 70 |
≤10 pack-years | 31 |
>10 pack-years | 110 |
Unknown | 2 |
P16 expression across all tumors | |
Positive | 103 |
Negative | 34 |
Unknown | 76 |
P16 expression in oropharyngeal primary tumor | |
Positive | 86 |
Negative | 8 |
Unknown | 3 |
Treatment | |
Upfront surgery | 91 |
Definitive radiation | 9 |
Definitive chemoradiation | 71 |
Sequential therapy | 31 |
Palliative systemic therapy | 2 |
Unknown | 9 |
Type of first treatment failure | |
Locoregional disease | 58 |
Distant metastases | 18 |
Both | 8 |
Assay | |
Oncomap positive | 31 |
Oncomap no variants identified | 66 |
Oncopanel positive | 109 |
Oncopanel no variants identified | 7 |
For the whole cohort, median PFS was 15 months (0.5–86 months), and median OS was 20 months (0.5–96 months). Three-year PFS and OS were 68% and 69%, respectively. Median follow-up was 20.4 months. Clinical factors, such as age, gender, alcohol, smoking, and primary site, were not significantly correlated with OS and PFS. As expected, patients with p16-positive oropharynx cancers (n = 86) had better outcomes compared with their p16-negative counterparts (n = 8), and this was evident for PFS and OS despite the low number of p16-negative cases [3-year PFS 79% vs. 53% (P = 0.13), 3-year OS 61% vs. 42% (P = 0.06)].
Overview of SNVs and CNVs: Molecular profiling identifies frequent mutations (SNVs) in TP53, NOTCH1, MLL2, PIK3CA, and CDKN2A, and frequent CNVs in CCND1, CDKN2A/B, and EGFR
We first identified mutations that were sequenced in both Oncomap and Oncopanel. A genomic overview of the most frequent somatic mutations of interest (Supplementary Table S1A) in all Oncomap and Oncopanel cases with at least one mutation (n = 140) is shown in Fig. 1A according to anatomic site, p16 status, and smoking status. Twenty patients had canonical PIK3CA mutations (E542K, E545K, H1047R), 15 patients had CDKN2A mutations (13 nonsense, 2 missense), 13 patients had TP53 mutations (1 nonsense, 12 missense), 5 patients had HRAS mutations (5 missense), 4 patients had JAK3 mutations, 2 patients had AKT1 mutations, 2 patients had FGFR3 mutations, 2 patients had MET mutations, 2 patients had NRAS mutations, 2 patients had PTEN mutations, and 2 patients had RB1 mutations. The observed variants were largely consistent with previously profiled nonclinical cohorts (4–6).
Mutational spectrum of HNSCC tumors detected by clinical NGS sequencing and copy-number alterations. A, specific mutations of interest detected across the HNSCC cohort (n = 213), identifying PIK3CA, CDKN2A, TP53, and HRAS as the most recurrently mutated genes. Samples with at least one detected alteration are shown (n = 140), stratified by sequencing platform, tumor anatomic site, p16 expression, and smoking level. B, Oncopanel mutations detected across the HNSCC cohort (n = 116), showing the most recurrently mutated genes in samples with at least one detected alteration (n = 109). C, Oncopanel copy-number alterations detected across the HNSCC cohort (n = 109) for samples with at least one detected copy-number event (n = 94). Genes with at least two high-level events (high gain or homozygous loss) are shown, arranged by loci.
Mutational spectrum of HNSCC tumors detected by clinical NGS sequencing and copy-number alterations. A, specific mutations of interest detected across the HNSCC cohort (n = 213), identifying PIK3CA, CDKN2A, TP53, and HRAS as the most recurrently mutated genes. Samples with at least one detected alteration are shown (n = 140), stratified by sequencing platform, tumor anatomic site, p16 expression, and smoking level. B, Oncopanel mutations detected across the HNSCC cohort (n = 116), showing the most recurrently mutated genes in samples with at least one detected alteration (n = 109). C, Oncopanel copy-number alterations detected across the HNSCC cohort (n = 109) for samples with at least one detected copy-number event (n = 94). Genes with at least two high-level events (high gain or homozygous loss) are shown, arranged by loci.
A more comprehensive mutational analysis utilizing the second-generation platform, Oncopanel, revealed TP53 was the most common mutational event and was identified in 42 patients, followed by NOTCH1 mutations (n = 24), MLL2 mutations (n = 20), PIK3CA mutations (n = 16), and CDKN2A mutations (n = 16; Fig. 1B). As expected, TP53 mutations were more frequently observed in HPV-negative patients than in HPV-positive patients (4) and were enriched in heavy smokers.
A genomic overview of high-level CNVs identified by Oncopanel is depicted in Fig. 1C and ordered according to loci. Copy-number analysis of the Oncopanel cohort identified focal regions of amplification in PIK3CA, SOX2, BCL6, EGFR, CDK6, MYC, JAK2, CCND1, ERBB, GATA6, and SRC. Focal regions of deletion were found in CDKN2A, CDKN2B, NOTCH1, STK11, TCF3, and SRC. Genes most frequently affected by high-level gain or homozygous loss were CCND1 gain (n = 10), CDKN2A/B loss (n = 8), and EGFR gain (n = 8).
Clinical correlation: Oncogenic driver events and PI3K pathway alterations and pan-RAS alterations are associated with patient outcomes
We performed an exploratory analysis in the OncoPanel cohort and correlated gene aberrations (SNVs and CNVs) with clinical factors (age, primary site, tumor–node–metastasis stage, AJCC stage, tobacco exposure, p16 status, treatment, and type of first treatment failure; Table 1) and outcomes including OS and PFS. The total number of SNVs (upper quartile was ≥8 and lower quartile was ≤3 across the cohort) or CNVs (upper quartile was ≥45 and lower quartile was ≤1) did not stratify PFS or OS (Supplementary Fig. S1). There was a trend for higher number of SNVs (greater than the median number of 5 SNVs) to be associated with heavy smoking history (P = 0.001), and CDKN2A mutations to be associated with nonoropharynx sites (P = 0.002). Amplifications of IKZF1 and MCL1 were associated with lower PFS (P = 0.000002 and P = 0.0002, respectively; Supplementary Figs. S2 and S3). Genomic events associated with poorer OS were PIK3CA amplification (P = 0.004; Supplementary Fig. S4), NRAS mutation or CNV gain (P = 0.0016; Supplementary Fig. S5), and GATA4 amplification (P = 0.003; Supplementary Fig. S6). However, we note that aside from PIK3CA, the other genes had relatively few events in the cohort, which may lead to inaccurate estimates of survival curves.
Patients with tumors harboring previously reported markers of PI3K pathway activation that were tested in the OncoPanel cohort (Tier 1/2 mutations in PIK3CA, PIK3C2B, PIK3R1, PIK3CA amplification or PTEN loss) appeared to have poorer outcomes and lower PFS (P = 0.19; Fig. 2A). Specifically, patients with PIK3CA-amplified tumors (40 patients had CNV gain, 3 patients had high gain) had significantly worse PFS (P = 0.0006; Fig. 2B) and worse OS (P = 0.004; Supplementary Fig. S4). In contrast, PIK3CA mutations alone were not associated with PFS (Supplementary Fig. S7; P = 0.79). The reason for this correlation with outcome with PIKC3A amplifications but not mutations is unclear, and no other obvious confounding clinical variable, including p16 status, was associated with PIK3CA mutation to explain this finding (data not shown). PI3K pathway mutation or copy-number alteration was not significantly associated with OS (P = 0.17).
PIK3CA and RAS alterations are associated with survival in HNSCC patients. Kaplan–Meier estimates of PFS and OS. A, patients with PIK3-family (PIK3CA, PIK3C2B, PIK3CR1) or PTEN Tier 1/2 mutation had poorer PFS across the Oncopanel cohort (n = 109, 15 with PI3K/PTEN mutations). B, patients with any PIK3CA copy-number amplification events significantly associated with poorer PFS in the Oncopanel cohort (n = 109, 43 with PIK3CA amplification). Patients with Tier 1/2 activating RAS mutations associated with both poorer PFS in C and poorer OS in D across the entire cohort (n = 213), 10 HRAS-, 2 NRAS-, and 1 KRAS-mutated patients.
PIK3CA and RAS alterations are associated with survival in HNSCC patients. Kaplan–Meier estimates of PFS and OS. A, patients with PIK3-family (PIK3CA, PIK3C2B, PIK3CR1) or PTEN Tier 1/2 mutation had poorer PFS across the Oncopanel cohort (n = 109, 15 with PI3K/PTEN mutations). B, patients with any PIK3CA copy-number amplification events significantly associated with poorer PFS in the Oncopanel cohort (n = 109, 43 with PIK3CA amplification). Patients with Tier 1/2 activating RAS mutations associated with both poorer PFS in C and poorer OS in D across the entire cohort (n = 213), 10 HRAS-, 2 NRAS-, and 1 KRAS-mutated patients.
RAS alterations were also associated with poorer survival outcomes. Notably, patients with tumors harboring previously reported activating RAS mutations [Tier 1/2 mutations for HRAS (n = 10), NRAS (n = 2), and KRAS (n = 1)] appeared to have a significantly poorer PFS (P = 0.0001; Fig. 2C) and lower OS (P = 0.003; Fig. 2D). RAS and PIK3CA events tended to be mutually exclusive.
Pathway overview and potential genotype-driven targeted therapies for HNSCC
The most frequent and clinically significant somatic alterations that we identified belong to key oncogenic pathways affecting cell signaling, cell-cycle regulation, or squamous differentiation, and for which potentially relevant targeted agents are in clinical development. We utilized previously described functional annotations of somatic alterations of squamous cell carcinomas to classify mutations and copy-number alterations in this cohort (6, 9). Notably, activation of the PI3K/RTK pathway occurred frequently as evidenced by recurrent activating PIK3CA mutation or amplification (13%), PTEN inactivating mutation or loss (3%), RAS activation (6%), and EGFR or ERBB2 activating mutation or high amplification (9%; Fig. 3A). Cell-cycle deregulation via tumor suppressor loss-of-function occurred frequently via TP53 inactivating mutations (30%) or via retinoblastoma tumor suppressor pathway inactivation [CDKN2A deletion or mutation (17%), CCND1 amplification (9%)] or BRCA2 inactivation (2%; Fig. 3B). Squamous differentiation pathway deregulation was often observed via significant homozygous loss or inactivating mutations in NOTCH1 (8%), NOTCH2 (3%), EP300 (6%), and FBXW7 (3%; Fig. 3C). Together, these findings confirm the sufficient frequency of targetable alterations in HNSCC to support the development of biomarker-driven clinical trials for HNSCC patients.
Recurrently altered pathways in HNSCC based on integrated analysis of genomic alterations. A–C, key affected signaling pathways in the HNSCC cohort. Pathway genes with recurrent alterations detected in Oncopanel are classified as activating (high-level amplification or Tier 1/2 activating mutation colored red), inactivating (homozygous loss or truncating mutation colored blue), or potentially cancer associated (Tier 3/4 mutation colored white). The frequency of Tier 1/2 mutations across the entire cohort (Oncomap and Oncopanel) is shown in gray if different from Oncopanel alone. For each pathway, integrated heatmaps show the detailed alteration pattern of each gene, with shades of red and green representing copy-number events and black text representing mutations. Note that each heatmap is sorted independently across the samples, to best illustrate the pattern of mutations, such as mutual exclusivity or concurrence. A, PI3K/RTK. B, cell cycle. C, squamous differentiation.
Recurrently altered pathways in HNSCC based on integrated analysis of genomic alterations. A–C, key affected signaling pathways in the HNSCC cohort. Pathway genes with recurrent alterations detected in Oncopanel are classified as activating (high-level amplification or Tier 1/2 activating mutation colored red), inactivating (homozygous loss or truncating mutation colored blue), or potentially cancer associated (Tier 3/4 mutation colored white). The frequency of Tier 1/2 mutations across the entire cohort (Oncomap and Oncopanel) is shown in gray if different from Oncopanel alone. For each pathway, integrated heatmaps show the detailed alteration pattern of each gene, with shades of red and green representing copy-number events and black text representing mutations. Note that each heatmap is sorted independently across the samples, to best illustrate the pattern of mutations, such as mutual exclusivity or concurrence. A, PI3K/RTK. B, cell cycle. C, squamous differentiation.
Examples of NGS profiling of tumors leading to clinical trial enrollment
In this cohort, 8 patients with recurrent or metastatic HNSCC enrolled on clinical trials utilizing agents prospectively matched to their tumor profiling results (Supplementary Table S1B). Of these, 4 patients achieved partial response, 1 patient had stable disease, and 3 patients had progressive disease as their best response. Specifically, a 56-year-old female presenting with widely metastatic HPV-positive HNSCC harboring a PIK3CA mutation (p.E545K) enrolled on a clinical trial of an oral PI3K/mTOR inhibitor and achieved dramatic symptomatic clinical benefit within days of starting therapy and a significant partial response on first assessment scans (71% tumor shrinkage; Fig. 4A). Another patient with HPV-negative oral cavity squamous cell carcinoma (SCC) who progressed after surgery and chemoradiation and developed painful locoregional recurrence and lung metastases refractory to multiple lines of palliative systemic therapy, including cetuximab, was found to have high copy-number gain in EGFR in her tumor. She enrolled on a clinical trial of an EGFR inhibitor and PI3K/mTOR inhibitor resulting in complete resolution of pain and also significant partial response (47% tumor shrinkage) on first assessment scans which was maintained for 8 months (Fig. 4B). A 55-year-old male patient with recurrent HPV-positive oropharynx SCC was found to have a tumor harboring an AKT1 (p.E17K) mutation and enrolled on a clinical trial of a PI3K/mTOR inhibitor and also achieved a sustained partial response. Another 52-year-old female with EGFR-amplified recurrent and metastatic oral cavity SCC was treated with a novel EGFR inhibitor and achieved a radiographic partial response (39% shrinkage) at initial treatment assessment with symptom improvement.
Examples of NGS profiling leading to enrollment on clinical trials of matched targeted therapy resulting in objective responses in HNSCC patients. A, 56-year-old female with metastatic HPV-positive HNSCC harboring a PIK3CA mutation (p.E545K) matched to PI3K/mTOR inhibitor on trial achieved a significant partial response on first assessment PET/CT scans (71% tumor shrinkage) accompanied by reduction in SUVmax from 9.6 to 2.7 in bilateral hilar and left axillary nodes. B, another patient with recurrent and metastatic HPV-negative HNSCC harboring high copy-number gain in EGFR treated on a clinical trial of an EGFR inhibitor and PI3K/mTOR inhibitor resulting in partial response on first assessment CT neck scans (47% tumor shrinkage).
Examples of NGS profiling leading to enrollment on clinical trials of matched targeted therapy resulting in objective responses in HNSCC patients. A, 56-year-old female with metastatic HPV-positive HNSCC harboring a PIK3CA mutation (p.E545K) matched to PI3K/mTOR inhibitor on trial achieved a significant partial response on first assessment PET/CT scans (71% tumor shrinkage) accompanied by reduction in SUVmax from 9.6 to 2.7 in bilateral hilar and left axillary nodes. B, another patient with recurrent and metastatic HPV-negative HNSCC harboring high copy-number gain in EGFR treated on a clinical trial of an EGFR inhibitor and PI3K/mTOR inhibitor resulting in partial response on first assessment CT neck scans (47% tumor shrinkage).
In contrast, 12 patients with recurrent or metastatic HNSCC were enrolled on clinical trials of novel therapies unmatched to their tumor profiling results. The best response was stable disease for 4 patients and progressive disease for 8 patients (Supplementary Table S1B). In addition, there were 6 patients for whom tumor profiling results were used to match therapy on a clinical trial; however, the patients either declined enrollment or were deemed ineligible. Of these 6 patients, 1 patient achieved stable disease as best response on further chemotherapy and then died within 5 months, whereas the remaining 5 patients had progressive disease (4 died within 4 months and 1 died within 6 months). Taken together, 50% of patients on matched therapy experienced a partial response, whereas no partial responses were observed in patients who received unmatched therapy on clinical protocols at our institution from late 2011 through 2014.
Discussion
This study represents the first comprehensive characterization of the use of NGS in the routine clinical care of HNSCC patients, including the identification of candidate biomarkers for further prognostic validation, and examples of the application of NGS results to personalize therapy. While the molecular landscape of HNSCC has been described in several cohorts (4–7), the clinical significance of genomic alterations in HNSCC is underreported and is a major impediment to the realization of personalized medicine in HNSCC. Initial research efforts to genomically characterize HNSCC include three large studies employing whole-exome and whole-genome sequencing of fresh-frozen primary HNSCC tumors (4–6). The Cancer Genome Atlas study reflects the largest cohort to date (comprised of 279 cases), and is biased toward a surgically resected oral cavity or laryngeal squamous cell carcinoma population (only 15% were HPV-positive; 6). To characterize HPV-positive tumors, a University of Chicago cohort of 120 locally advanced HNSCCs (43% were HPV-positive) was examined utilizing a targeted approach evaluating 617 selected cancer-associated genes in the frozen primary tumors and did not include gene-rearrangement data (9). Recently, 252 clinical HNSCC FFPE samples were evaluated in aggregate utilizing the Foundation Medicine NGS platform (13). The genomic profiles in this Foundation Medicine cohort (33% were HPV-positive) were comparable with prior reports obtained from The Cancer Genome Atlas (TCGA) and University of Chicago; however, treatment and survival data were limited. In our study, the most common genomic alterations were identified at frequencies in concordance with prior reports (4–6), confirming that the genomic landscape obtained from clinical HNSCC FFPE specimens appears consistent with research-derived frozen specimens.
We utilized more complete outcome data to identify candidate genomic alterations for further prognostic validation based on our exploratory analysis with clinical correlations. PIK3CA was the most frequently altered oncogene in our cohort (13% amplified or mutated) and was enriched in HPV-positive oropharynx cancers, consistent with prior reports (6, 8, 9, 13–16). For the first time, we found that PIK3CA amplification was associated with significantly decreased PFS, whereas PIK3CA mutation was not associated with survival outcomes. Prior smaller studies were unable to identify a significant correlation between PI3K pathway activation and survival outcomes (9, 15, 17) and did not distinguish the relative contribution of various mechanisms of activation of the PI3K pathway (e.g., PIK3CA mutation vs. amplification vs. PTEN loss; 15). One study identified that PIK3CA amplification in 32% of 115 surgical HNSCC cases was associated with earlier relapse in a subset of patients without lymph node metastases (n = 59, P = 0.026; 17). Another study reported PIK3CA mutations were potentially correlated with poorer prognosis in an HPV-negative subset of locally advanced HNSCC (n = 69; 9). PIK3CA overexpression (not mutation) has been associated with shorter OS in other cancers (18). Preclinical studies have demonstrated that patient-derived PIK3CA mutant HNSCC tumorgrafts are sensitive to PI3K/mTOR inhibitors (8), and PI3K pathway inhibition may sensitize cancer cells to radiation (19). Clinical trials of a PI3K inhibitor in combination with chemoradiation (NCT02113878) or chemotherapy alone (NCT01816984) are underway. In our cohort, 2 patients with PIK3CA mutant tumors received a PI3K/mTOR inhibitor on study and one achieved partial response while another had progressive disease as best response. For the purpose of clinical trials enrollment, clinically restricting testing to PIK3CA hotspot mutations is a feasibly attractive option (14, 20); however, our findings suggest that PIK3CA amplification may also bear clinical significance. Taken together, our data support further evaluation of PIK3CA amplification as a potential negative prognostic indicator in HNSCC, independent of HPV status, and preclinical and clinical investigation of biomarker-driven studies in patients with PI3K pathway–activated tumors is warranted.
RAS mutations [Tier 1/2 mutations for HRAS (n = 10), NRAS (n = 2), and KRAS (n = 1)] were also associated with significantly poorer survival in our cohort. This has not been previously described in HNSCC. In contrast to our findings, in the TCGA dataset, patients with HRAS mutant oral cavity tumors and few copy-number alterations had better PFS. HRAS mutations occur in 5% to 10% of HNSCC (4, 5, 21) and are enriched in tobacco-related SCC up to frequencies as high as 35% (22). RAS mutations may also predict lack of response to EGFR inhibitors in HNSCC (23, 24). Preclinical studies suggest that PIK3CA and RAS mutations may predict intrinsic resistance to cetuximab, which may be prevented by concomitant administration of cetuximab and PI3K and/or mTOR inhibitors in HNSCC patients (23). Taken together, RAS mutations warrant further prognostic evaluation in larger cohorts, and biomarker-driven clinical trials in RAS mutant HNSCC patients are warranted. RAS-directed therapies in HNSCC are lacking (25); however, an HRAS biomarker–driven trial is currently open to HNSCC patients (NCT02383927). We also confirmed the presence of other targetable alterations, including ErB2 and EGFR amplification, FGFR3 mutations, CCND1 amplification, and CDKN2A mutations. Early clinical development of novel HER2 and EGFR inhibitors, FGFR inhibitors, CDK inhibitors, and p53 modulators are all under way, and may be open to HNSCC patients who undergo appropriate genotyping studies. In the closely related disease, lung squamous cell carcinoma, genomic stratification for clinical trials is increasingly being adopted as evidenced by the recently initiated Lung-MAP study (26).
Notably, our study illustrates how the application of NGS results may be used to guide selection of targeted therapy for advanced HNSCC patients on clinical trials. In our cohort, 8 patients with advanced disease enrolled on clinical trials of targeted agents matched to their NGS results. Given that 84 patients had recurrent disease (26 of these had metastatic disease), it is possible that NGS may be used to direct clinical trial selection in approximately 10% to 31% of advanced cases. Patients matched to therapy appeared to have a higher objective response rate than patients unmatched to therapy in our cohort; however, these results need to be interpreted with caution as many patients could not be matched to therapy because no targetable alterations were identified, and many genomic alterations in HNSCC are not readily “druggable.” In our cohort, approximately half of the patients were tested with the Oncomap panel (n = 97), and half were tested with the Oncopanel platform (n = 116). However, a comparison of the Oncomap and OncoPanel platforms clearly demonstrates a higher negative result rate for the older version, Oncomap (68% vs. 6%), and also differences in mutation rates between the two tests, suggesting that the detection of actionable alterations may be improved with newer, more comprehensive NGS panels which are becoming increasingly available. To date, the development of targeted therapies in HNSCC is fraught with negative clinical trials which have largely tested targeted agents in unselected patient populations (27–31). Our findings would suggest that a greater likelihood of success may be achieved with the incorporation of genomic stratification into clinical trials evaluating targeted agents in HNSCC, and such approaches are consistent with national initiatives, such as NCI-MATCH (32).
In summary, we demonstrate that it is feasible to incorporate NGS into the routine clinical care of patients with HNSCC. For the first time, we identify an association of PIK3CA amplification and RAS mutations with poorer survival outcomes. This warrants further evaluation and validation in larger cohorts, but lends support for the pursuit of further preclinical evaluation and biomarker-driven clinical trials in HNSCC patients. We demonstrate examples of personalized medicine approaches in advanced HNSCC patients by using genomic profiling results to guide selection of targeted therapy on clinical trials leading to objective responses. Our experience suggests the potential value of incorporating NGS into the clinical care of patients with HNSCC to better define genomic markers associated with outcome and to enhance accrual to biomarker-driven trials.
Disclosure of Potential Conflicts of Interest
R.I. Haddad is a consultant/advisory board member for Bristol-Myers Squibb, Celgene, Eisai, Merck, and Pfizer. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: N.G. Chau, V.Y. Jo, L.A. Goguen, N.I. Lindeman, R.I. Haddad, P.S. Hammerman
Development of methodology: N.G. Chau, Y.Y. Li, L.A. Goguen, N.I. Lindeman, P.S. Hammerman
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.G. Chau, Y.Y. Li, V.Y. Jo, J.H. Lorch, R.B. Tishler, D.N. Margalit, J.D. Schoenfeld, D.J. Annino, L.A. Goguen, T. Thomas, H. Becker, T. Haddad, J.F. Krane, N.I. Lindeman
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.G. Chau, Y.Y. Li, V.Y. Jo, J.H. Lorch, D.N. Margalit, T. Thomas, N.I. Lindeman, G.I. Shapiro, P.S. Hammerman
Writing, review, and/or revision of the manuscript: N.G. Chau, Y.Y. Li, V.Y. Jo, G. Rabinowits, J.H. Lorch, R.B. Tishler, D.N. Margalit, J.D. Schoenfeld, D.J. Annino, L.A. Goguen, T. Thomas, J.F. Krane, N.I. Lindeman, G.I. Shapiro, R.I. Haddad, P.S. Hammerman
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.G. Chau, L.A. Goguen, H. Becker, T. Haddad, N.I. Lindeman, R.I. Haddad, P.S. Hammerman
Study supervision: N.G. Chau, L.A. Goguen, R.I. Haddad, P.S. Hammerman
Acknowledgments
The authors would like to acknowledge all of the patients and their families.