Biomarkers predictive of response to pembrolizumab in head and neck cancer

Abstract Background We performed an integrated biomarker evaluation in pembrolizumab‐treated patients with R/M HNSCC enrolled in KEYNOTE‐012 or KEYNOTE‐055. The relationship between biomarkers and HPV status was explored. Methods We evaluated PD‐L1 (combined positive score [CPS]), TMB, T‐cell‐inflamed gene expression profile (TcellinfGEP), and HPV status. Associations between biomarkers were evaluated by logistic regression (ORR) and Cox regression (PFS, OS). Results Two hundred and fifty‐seven patients (KEYNOTE‐012, n = 106; KEYNOTE‐055, n = 151) had TMB data available; of these, 254 had PD‐L1 and 236 had TcellinfGEP. TMB, PD‐L1, and TcellinfGEP were each significantly associated with ORR (p < 0.01). Kaplan–Meier curves at prespecified cutoffs generally showed PFS and OS separation in the anticipated direction for these biomarkers, except for OS and TMB. TMB did not correlate with PD‐L1 or TcellinfGEP (Spearman ρ = −0.03 and ρ = −0.13, respectively); PD‐L1 and TcellinfGEP were moderately correlated (Spearman ρ = 0.47). In multivariate models, TMB, PD‐L1, and TcellinfGEP were each independently predictive for ORR (p < 0.001). ORR was higher in patients with high versus low levels of biomarkers when dichotomized using prespecified cutoffs; patients with higher versus lower levels of TMB and PD‐L1 or TMB and TcellinfGEP had the highest ORRs. Within HPV subgroups, higher versus lower distributions of biomarkers (PD‐L1, TMB, and TcellinfGEP) were associated with response. HPV detection by p16‐immunohistochemistry and WES showed good concordance (81%); results were generally similar by HPV status, regardless of the detection method. Conclusions TMB and the inflammatory biomarkers PD‐L1 and TcellinfGEP, assessed alone or together, may be useful for characterizing clinical response to pembrolizumab in R/M HNSCC.


| INTRODUCTION
Immunotherapies that target the programmed death-1 (PD-1) axis can provide durable antitumor responses in multiple cancer types; however, the durable benefit is limited to specific patient subpopulations. 1 Biomarkers indicative of tumor antigenicity, including tumor mutational burden (TMB) and microsatellite instability (MSI), as well as inflammatory biomarkers related to a T-cellinflamed tumor microenvironment, such as programmed death-ligand 1 (PD-L1) expression and T-cell-activated gene expression signatures, may help characterize patient subpopulations who can benefit from these therapies. 2 Biomarkers approved by regulatory authorities to predict the likelihood of response to the anti-PD-1 monoclonal antibody pembrolizumab include PD-L1 expression for several tumor types and TMB or MSI regardless of specific cancer. 3 An 18-gene T-cell-inflamed gene expression profile (Tcell inf GEP) can independently predict response to pembrolizumab in multiple tumor types, including head and neck squamous cell carcinoma (HNSCC). 2,4 Oncogenic viruses such as hepatitis B virus, Merkel cell polyomavirus, Epstein-Barr virus, and human papillomavirus (HPV) generate viral antigens distinct from tumor-specific neoantigens that arise from somatic mutation and can lead to T-cell responses. 5,6 In some virus-induced cancers, PD-L1 expression is increased. Higher response rates with immunotherapy have also been noted in virusinduced versus non-virus-induced cancers 6 ; however, genetic determinants related to response to immuno-oncology agents in these tumor types are not well understood. HPV is etiologic for a subset of HNSCC, 7 and this subset is associated with greater survival than HPV-negative tumors. 8 Although HPV status is usually assessed through the detection of p16-antigen by immunohistochemistry (IHC), HPV genomes can also be detected through genomic methods (i.e., whole-exome sequencing [WES]). 8,9 Pembrolizumab demonstrated durable antitumor activity with a manageable safety profile in recurrent and/or metastatic (R/M) HNSCC in the phase Ib KEYNOTE-012 and phase II KEYNOTE-055 studies. [10][11][12] In a recent analysis of KEYNOTE-012, the expression of TMB, Tcell inf GEP, and PD-L1 each was found to be an independent predictor of clinical response to pembrolizumab. 13 Patients with high rather than low levels of TMB and inflammatory biomarkers (PD-L1 expression or Tcell inf GEP score) had between biomarkers were evaluated by logistic regression (ORR) and Cox regression (PFS, OS).
Results: Two hundred and fifty-seven patients (KEYNOTE-012, n = 106; KEYNOTE-055, n = 151) had TMB data available; of these, 254 had PD-L1 and 236 had Tcell inf GEP. TMB, PD-L1, and Tcell inf GEP were each significantly associated with ORR (p < 0.01). Kaplan-Meier curves at prespecified cutoffs generally showed PFS and OS separation in the anticipated direction for these biomarkers, except for OS and TMB. TMB did not correlate with PD-L1 or Tcell inf GEP (Spearman ρ = −0.03 and ρ = −0.13, respectively); PD-L1 and Tcell inf GEP were moderately correlated (Spearman ρ = 0.47). In multivariate models, TMB, PD-L1, and Tcell inf GEP were each independently predictive for ORR (p < 0.001). ORR was higher in patients with high versus low levels of biomarkers when dichotomized using prespecified cutoffs; patients with higher versus lower levels of TMB and PD-L1 or TMB and Tcell inf GEP had the highest ORRs. Within HPV subgroups, higher versus lower distributions of biomarkers (PD-L1, TMB, and Tcell inf GEP) were associated with response. HPV detection by p16-immunohistochemistry and WES showed good concordance (81%); results were generally similar by HPV status, regardless of the detection method.
Conclusions: TMB and the inflammatory biomarkers PD-L1 and Tcell inf GEP, assessed alone or together, may be useful for characterizing clinical response to pembrolizumab in R/M HNSCC.

K E Y W O R D S
biomarker, head and neck squamous cell carcinoma, immunotherapy, pembrolizumab, tumor microenvironment, tumor mutational burden higher responses to pembrolizumab. The present analysis extends those observations and evaluates the relationships of these biomarkers with response to pembrolizumab and survival using data pooled from KEYNOTE-012 and KEYNOTE-055 in patients with R/M HNSCC, including a descriptive evaluation of the relationship between these biomarkers and HPV status.

| Study design and patients
The study design and eligibility criteria of the multicohort phase Ib KEYNOTE-012 (Clini calTr ials.gov, NCT01848834) study and the phase II KEYNOTE-055 (Clini calTr ials.gov, NCT02255097) study have been reported. [10][11][12] In brief, eligible patients in KEYNOTE-012 cohorts B and B2 had confirmed R/M HNSCC by Response Evaluation Criteria in Solid Tumors, version 1.1, by investigator review and Eastern Cooperative Oncology Group performance status (ECOG PS) ≤1; cohort B enrolled 60 patients with PD-L1-positive tumors (≥1%, QualTek IHC), and 132 patients with PD-L1-unselected tumors were enrolled in cohort B2. Eligible patients in KEYNOTE-055 had confirmed R/M HNSCC of the oral cavity, oropharynx, hypopharynx, or larynx resistant to platinum and to cetuximab; had an ECOG PS of 0 or 1; and provided newly obtained core or excisional biopsy for PD-L1 expression analysis (n = 171). Concurrent treatment with platinum and cetuximab was not required; however, patients must have experienced progressive disease (PD) or recurrence <6 months after the last dose of each therapy. For both studies, key exclusion criteria included previous treatment with an anticancer monoclonal antibody <4 weeks of the initiation of study drug; previous chemotherapy, small molecule-targeted therapy, or radiation therapy <2 weeks of the initiation of study drug; known active central nervous system metastases; and a diagnosis of immunodeficiency, autoimmune disease, interstitial lung disease, or active infection that required systemic therapy.
In KEYNOTE-012, patients received pembrolizumab 10 mg/kg every 2 weeks in cohort B and 200 mg every 3 weeks in cohort B2. In KEYNOTE-055, patients received pembrolizumab 200 mg every 3 weeks. In both studies, treatment continued for ≤2 years or until confirmed PD or unacceptable toxicity, investigator decision, or withdrawal of patient consent.
The study protocols were approved by regulatory boards or ethics review committees at each study center. The studies were conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All patients provided written informed consent before study entry.

| Assessments
TMB was assessed using WES as previously described. 2 PD-L1 expression level was assessed via IHC using PD-L1 IHC 22C3 pharmDx (Agilent) and measured by combined positive score (CPS), defined as the number of PD-L1positive cells (tumor cells, lymphocytes, macrophages) divided by the total number of viable tumor cells, multiplied by 100. 14 4 HPV status was determined by p16-IHC for patients with cancer of the oropharynx. p16-IHC testing was not conducted for primary tumor locations outside the oropharynx, and such tumors were categorized as HPV negative. HPV status was also assessed by WES regardless of the tumor site.

| Statistical analysis
Statistical testing of biomarker and response relationships, prespecified in the statistical analysis plans for KEYNOTE-012 and KEYNOTE-055, was performed using data pooled from both studies. TMB, PD-L1 expression, and Tcell inf GEP score relationships with objective response rate (ORR; defined as the proportion of patients achieving complete response or partial response), progression-free survival (PFS), and overall survival (OS) with pembrolizumab were assessed in all patients and by HPV status. Associations between TMB, PD-L1, Tcell inf GEP, neoantigen load, and clonality-weighted TMB (TMB × clonality as calculated using a published algorithm 15 ) and ORR were assessed by logistic regression. Associations of TMB, PD-L1, and Tcell inf GEP with PFS and OS were assessed by Cox regression. Associations of neoantigen load and tumor clonality (an estimate of TMB restricted to clonal mutations) with ORR were also evaluated. An appropriate transformation for TMB, PD-L1, neoantigen load, and clonality-weighted TMB, either log or square root, was applied when needed. Models were adjusted for ECOG PS and study or cohorts. One-sided nominal p values were reported for ORR, PFS, and OS because a positive association was hypothesized. Significance was determined at the 0.05 level, unadjusted for multiplicity. The area under the receiveroperating characteristic (AUROC) curve was used as a measure of the discriminatory ability of TMB, PD-L1, and Tcell inf GEP biomarkers to distinguish responders from nonresponders. The following cutoffs were used to assess the clinical utilities of the biomarkers with the response (1) 175 mutations/exome (mut/exome) for TMB, derived from TMB and GEP data across multiple cohorts 16 (previously shown to be concordant with 10 mut/Mb [FoundationOne®CDx]) 17 ; (2) CPS 1 for PD-L1 expression 14 ; and (3) -0.318 for GEP score, defined using multiple tumor types 18 and preceding the availability of gene expression data from KEYNOTE-055.
Correlations between TMB and inflammatory biomarkers (PD-L1 and Tcell inf GEP) were assessed using Spearman correlation. A contingency table was used to evaluate the concordance of HPV status (WES versus p16 IHC). Boxplots were used to descriptively illustrate the distribution of each biomarker (TMB/Tcell inf GEP/PD-L1) by HPV status; the mean difference of each biomarker in HPV-positive versus HPV-negative subgroups was tested using a two-sample t test, and the adjusted p values are reported for multiple testing across the three biomarkers (TMB/Tcell inf GEP/PD-L1). The Hochberg step-up procedure was used for multiplicity to control the family-wise error rate. Testing for differential biomarker relationships according to HPV status was performed with an interaction term between the biomarker (TMB/Tcell inf GEP/PD-L1) and HPV status in a logistic regression model (other terms were ECOG PS, study/cohort, the biomarker itself [TMB/Tcell inf GEP/PD-L1], and HPV status) and was similarly adjusted for multiplicity.

HPV status
Of 363 patients, 256 (71%) had evaluable WES and p16-IHC data. Despite p16-IHC analysis being restricted to oropharyngeal tumors and WES analysis performed on all tumor sites, detection of HPV status by p16-IHC and WES showed good concordance (208/256 [81%]) (Table S2) in the analysis population of patients with evaluable WES data. Similar proportions of patient tumors were classified as HPV negative (78% and 69%) and HPV positive (22% and 31%) by p16-IHC and WES, respectively.
The distribution of biomarkers (TMB/PD-L1/ Tcell inf GEP) was comparable in HPV-positive and HPVnegative subgroups, defined by either WES or p16-IHC ( Figure S1A-C). Two-sample t testing showed no significant difference between HPV-positive and HPV-negative subgroups, defined by either WES or p16-IHC for any of the biomarkers (adjusted p > 0.3).

| Association of biomarkers with ORR by HPV status
Evaluating trends for TMB, PD-L1, and Tcell inf GEP within HPV status suggested consistency in the positive association between each biomarker and response in HPVpositive and HPV-negative subgroups ( Figures S2A-C). Some associations with outcome appeared to be stronger in one HPV subgroup than another. For example, TMB distribution was more separated for responders than nonresponders in the HPV-negative group and the HPVpositive group. Conversely, PD-L1 CPS and Tcell inf GEP distributions showed stronger trends with response status for the HPV-positive group than the HPV-negative group. Statistical interaction testing was conducted for HPV-specific associations for these three biomarkers with ORR, albeit in a manner outside the statistical analysis plan of each study, to gauge the evidence supporting differences in the relationship between TMB and inflammatory biomarkers with clinical outcome according to HPV status based on a more powered analysis offered by the combined two-study data package. The results of this interaction testing, with multiplicity adjustment for the three biomarkers within HPV assay type, provided some indication that the association between the inflammatory markers PD-L1 and Tcell inf GEP might have been stronger in the HPV-positive subgroup, although the two versions of the HPV assays were not consistent in their testing conclusions with regard to adjusted P values achieving significance at the 0.05 level (Table S3).

| DISCUSSION
In this exploratory analysis of patients with R/M HNSCC enrolled in KEYNOTE-012 and KEYNOTE-055 who had evaluable WES data, TMB and the inflammatory biomarkers PD-L1 CPS and Tcell inf GEP were independently predictive of ORR to pembrolizumab. Patients with high levels (based on a prespecified cutoff) of TMB, PD-L1, and Tcell inf GEP versus low levels of these biomarkers had higher response rates to pembrolizumab, and patients

-Specificity
whose tumors expressed high levels of dual biomarkers (both TMB and PD-L1 CPS and TMB and Tcell inf GEP) had the best responses to pembrolizumab. Taken together, the data suggest that these biomarkers reflect complementary measures of tumor antigenicity and an inflamed tumor microenvironment and that, used alone or jointly, have the potential to characterize responses to anti-PD-1 therapy in HNSCC. However, outlier responses, such as patients whose tumors express low levels of TMB and PD-L1, suggest that additional biology or dynamic changes may also contribute to these effects, necessitating further evaluation in additional studies. The results of this analysis are consistent with those of previous studies showing that TMB, PD-L1, and Tcell inf GEP are related to response to anti-PD-1 therapy in multiple cancers 2 and with those of a recent analysis of the KEYNOTE-012 trial in patients with HNSCC. 13 PD-L1 is a US Food and Drug Administration-approved diagnostic biomarker that is related to response to pembrolizumab monotherapy in several cancers, including HNSCC when used as first-line therapy, 3 and is considered to be driven by interferon-gamma signaling as partly indicative of a T-cell-inflamed tumor microenvironment. In a similar fashion, Tcell inf GEP, a signature composed of genes related to multiple cell types involved in the cytolytic process, including PD-L1 and PD-L2 as member genes whose levels are coexpressed with those of other genes in the signature, is related to response to pembrolizumab. 4 Hence, as expected and as demonstrated previously, 19 PD-L1 and Tcell inf GEP were moderately  positively correlated in this analysis. Similar to other published data, 2,13 this combined HNSCC cohort shows TMB did not correlate with PD-L1 or Tcell inf GEP but rather indicates mutational load acted as an additional explanatory axis for objective response to pembrolizumab in this pooled analysis. Pembrolizumab was recently approved by the US Food and Drug Administration to treat patients with unresectable or metastatic TMB-high (≥10 mut/Mb) solid tumors who experienced PD after previous treatment. 3,20 Using prespecified biomarker cutoffs, trends of longer median PFS were observed with higher versus lower levels of TMB but not PD-L1 or Tcell inf GEP. Evaluation of the Kaplan-Meier curves shows a fairly similar level of progression early, regardless of biomarker status, but later separation in PFS curves in the anticipated direction  according to the biomarker cutoff ( Figure 4). Longer median OS was associated with higher rather than lower levels of PD-L1 and Tcell inf GEP but not of TMB. However, OS data from single-arm studies should be interpreted with caution; a more informative evaluation of the relationship between OS and TMB will require larger randomized studies. HPV infection is a risk factor for some HNSCC subtypes, 21 and its presence can be determined by p16-IHC, DNA/RNA-in situ hybridization, or genomic methods (e.g., WES). In our study, both p16-IHC and WES showed good concordance (81%) in HPV detection, and results in the HPV subgroups identified by either method were generally similar. When evaluating the response to pembrolizumab by HPV status in patients with HNSCC in KEYNOTE-012 and KEYNOTE-055, both studies previously reported that response rates were similar regardless of HPV status. [10][11][12] In the current analysis, the distributions of TMB, PD-L1, and Tcell inf GEP were similar among the HPV-positive and HPV-negative subgroups. The inflammatory biomarkers PD-L1 and Tcell inf GEP, TMB, neoantigen load, and clonality-weighted TMB were each associated with response to pembrolizumab regardless of HPV status detected by p16 or WES. Higher response rates were observed in patients with high levels of TMB or either of the inflammatory biomarkers in both HPV subgroups. It is possible that biomarker trends with the clinical outcome may vary by HPV status. For the trends observed here, statistical significance and clinical relevance were not clear but may be worth additional follow-up as further data accumulate.
In conclusion, TMB and the inflammatory biomarkers PD-L1 CPS and Tcell inf GEP were each significantly and independently predictive of response to pembrolizumab in patients with HNSCC. Greater responses to pembrolizumab were associated with higher levels of the inflammatory biomarkers and TMB than were lesser responses, an observation that was consistent regardless of HPV status, suggesting that biomarkers representing complementary measures of tumor antigenicity and a T-cell-inflamed tumor microenvironment may be useful in characterizing clinical response to pembrolizumab in HNSCC. Larger randomized studies are required to better identify and understand biomarkers of response and resistance to pembrolizumab monotherapy and combination therapy in patients with HNSCC.

DATA AVAILABILITY STATEMENT
Merck Sharp & Dohme LLC, a subsidiary of Merck & Co., Inc., Rahway, NJ, USA (MSD) is committed to providing qualified scientific researchers access to anonymized data and clinical study reports from the company's clinical trials for the purpose of conducting legitimate scientific research. MSD is also obligated to protect the rights and privacy of trial participants and, as such, has a procedure in place for evaluating and fulfilling requests for sharing company clinical trial data with qualified external scientific researchers. The MSD data-sharing website (available at: http://engag ezone.msd.com/ds_docum entat ion.php) outlines the process and requirements for submitting a data request. Applications will be promptly assessed for completeness and policy compliance. Feasible requests will be reviewed by a committee of MSD subject matter experts to assess the scientific validity of the request and the qualifications of the requestors. In line with data privacy legislation, submitters of approved requests must enter into a standard data-sharing agreement with MSD before data access is granted. Data will be made available for request after product approval in the US and EU or after product development is discontinued. There are circumstances that may prevent MSD from sharing requested data, including country-or region-specific regulations. If the request is declined, it will be communicated to the investigator. Access to genetic or exploratory biomarker data requires a detailed, hypothesis-driven statistical analysis plan that is collaboratively developed by the requestor and MSD subject matter experts; after approval of the statistical analysis plan and execution of a data-sharing agreement, MSD will either perform the proposed analyses and share the results with the requestor or will construct biomarker covariates and add them to a file with clinical data that are uploaded to an analysis portal so that the requestor can perform the proposed analyses.

ETHICAL APPROVAL STATEMENT
The study protocols were approved by regulatory boards or ethics review committees at each study center. The studies were conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All patients provided written informed consent before study entry.