SUPREME-HN: a retrospective biomarker study assessing the prognostic value of PD-L1 expression in patients with recurrent and/or metastatic squamous cell carcinoma of the head and neck

Background Programmed cell death ligand-1 (PD-L1) expression on tumor cells (TCs) is associated with improved survival in patients with head and neck squamous cell carcinoma (HNSCC) treated with immunotherapy, although its role as a prognostic factor is controversial. This study investigates whether tumoral expression of PD-L1 is a prognostic marker in patients with recurrent and/or metastatic (R/M) HNSCC treated with standard chemotherapy. Methods This retrospective, multicenter, noninterventional study assessed PD-L1 expression on archival R/M HNSCC tissue samples using the VENTANA PD-L1 (SP263) Assay. PD-L1 high was defined as PD-L1 staining of ≥ 25% TC, with exploratory scoring at TC ≥ 10% and TC ≥ 50%. The primary objective of this study was to estimate the prognostic value of PD-L1 status in terms of overall survival (OS) in patients with R/M HNSCC. Results 412 patients (median age, 62.0 years; 79.9% male; 88.2% Caucasian) were included from 19 sites in seven countries. 132 patients (32.0%) had TC ≥ 25% PD-L1 expression; 199 patients (48.3%) and 85 patients (20.6%) had TC ≥ 10% and ≥ 50%, respectively. OS did not differ significantly across PD-L1 expression (at TC ≥ 25% cutoff median OS: 8.2 months vs TC < 25%, 10.1 months, P = 0.55) or the ≥ 10% and ≥ 50% cutoffs (at TC ≥ 10%, median OS: 9.6 months vs TC < 10%, 9.4 months, P = 0.32, and at TC ≥ 50%, median OS 7.9 vs TC < 50%, 10.0 months, P = 0.39, respectively). Conclusions PD-L1 expression, assessed using the VENTANA PD-L1 (SP263) Assay, was not prognostic of OS in patients with R/M HNSCC treated with standard of care chemotherapies. Trial registration ClinicalTrials.gov, NCT02543476. Registered September 4, 2015.


Background
Approximately 60% of patients with head and neck squamous cell carcinoma (HNSCC) are diagnosed with locally advanced disease, which has a 5-year overall survival (OS) rate of approximately 30% [1]. Most patients with HNSCC will eventually experience either local or distant recurrence [2], while approximately 10% of patients with HNSCC will initially present with metastatic disease [3]. Patients with recurrent and/ or metastatic (R/M) HNSCC have historically had a poor prognosis [4]. Traditional standard of care for first-line therapy in patients with R/M HNSCC is platinum-based chemotherapy plus cetuximab and 5-fluorouracil [5,6], yielding a median OS of approximately 10 months [7]. However, this is usually only appropriate for patients who have an acceptable Eastern Cooperative Oncology Group performance status (ECOG PS) and are able to tolerate platinum-based therapy. Patients with R/M HNSCC treated in the second-line setting have a poorer prognosis, with median OS of approximately 4-8 months [3,8,9]. Standard therapy in this setting includes single-agent therapies (e.g. methotrexate, docetaxel, or cetuximab) which yield objective response rates (ORRs) of 4-13% in the platinum-refractory setting [3,8,9]. More recently, phase III studies have demonstrated that immuno-oncology (IO) agents targeting programmed cell death-1 (PD-1)/programmed cell death ligand-1 (PD-L1) improve OS in both the first-line and second-line settings, with median OS of approximately 13-15 months and 7-8 months, respectively [10][11][12][13][14].
PD-L1 is expressed on antigen-presenting cells and other immune cells (ICs) and is upregulated on HNSCC tumor cells (TCs) [15,16]. The presence of PD-L1 can be readily detected by immunohistochemistry (IHC) staining [16]. Evidence is building that PD-L1 expression on TCs is associated with improved survival in patients with HNSCC treated with IO agents and yet the role of PD-L1 in outcomes irrespective of treatment (i.e. prognosis) is still unclear, with conflicting reports of PD-L1 as both a negative and positive prognostic factor [17][18][19][20][21][22][23]. Therefore, the SUPREME-HN study was conducted to investigate the possible prognostic role PD-L1 expression on TCs has in patients with R/M HNSCC. Here, we describe patient characteristics, OS, and other clinical outcomes related to PD-L1 expression independent of treatment choice [20,24].

Study design
SUPREME-HN was a retrospective, international, multicenter, noninterventional cohort study based on data derived from established medical records and analysis of archival tumor samples (ClinicalTrials.gov identifier: NCT02543476); for the purposes of this study and for patient selection, the index date was defined as the date of diagnosis of R/M disease not amenable to local therapy.

Patient population
Patients aged ≥ 18 years with histologically confirmed HNSCC of the oral cavity (tongue, gum, floor of mouth, or other/unspecified part of the mouth), oropharynx, larynx, or hypopharynx were eligible if they had R/M disease not amenable to local therapy with curative intent (surgery, radiation therapy, chemo-radiation). Patients with locally advanced disease amenable to curative local therapy were excluded as were patients who had received prior IO treatment with anti-cytotoxic T-lymphocyteassociated antigen 4, or anti-PD-1, anti-PD-L1, or anti-PD-L2 antibodies for HNSCC.

Procedures
Archival tumor samples (< 5 years old) were obtained anytime during the disease history from patients who were diagnosed between March 1, 2011 and June 30, 2015. Biopsies or resections from the primary site, lymph node, or distant metastatic sites were provided for analysis as formalin-fixed, paraffin-embedded (FFPE) blocks or sections < 60 days old.
For patients with more than one tissue sample, the most recent sample from the index date was used to determine PD-L1 expression. PD-L1 IHC staining of FFPE tissue samples was performed using the VENTANA PD-L1 (SP263) Assay on the automated Ventana Bench-Mark ULTRA ® platform (Ventana Medical Systems Inc., Tucson, AZ, USA) [25]. PD-L1 expression was scored by pathologists trained by the manufacturer, at an approved central testing laboratory. PD-L1 expression was evaluated for a cutoff of ≥ 25% of TCs with membrane staining for PD-L1 at any intensity (TC ≥ 25%). Exploratory scoring was assessed at TC ≥ 10% and TC ≥ 50%. Patient characteristics were collected including ECOG PS at the index date, smoking habits, alcohol consumption, human papilloma virus (HPV) status, HIV status, and medical Keywords: Biomarker, Head and neck squamous cell carcinoma, Immuno-oncology, PD-L1, Programmed cell death ligand-1, Prognosis history. Tumor characteristics, treatment patterns, and outcome measures were recorded.

Study endpoints
The study primary endpoint was OS as defined from the date of diagnosis of R/M HNSCC (index date) to time of death due to any causes. OS was reported separately in predefined subgroups based on baseline characteristics (e.g. HPV status, anatomical site of tumor). Secondary endpoints included descriptive analyses of demographics and clinical characteristics distribution with PD-L1 as well as investigator-assessed ORR, duration of response, and progression-free survival (PFS). ORR (complete response + partial response) was based on Response Evaluation Criteria In Solid Tumors (RECIST) v1.1. PFS was assessed from the start of first-line therapy for R/M disease to progression on or after therapy, or death due to any cause (whichever came first), and from the start of second-line therapy to first documented disease progression or death due to any cause (whichever came first).

Statistical analyses
The sample size to support the primary endpoint was not known a priori and was driven by the number of patients at selected sites with available tissue samples. Based on assumptions of a PD-L1 high prevalence of 25% (TC ≥ 25%), a median OS of 10 months, uniform accrual over 52 months with 10 months' follow-up from the last patient entering, and exponentially distributed survival times, it was determined post hoc that the study statistics could be powered to the 80% level (two-sided alpha 0.05) to detect a hazard ratio (HR) of 0.7 for PD-L1 high versus low/negative patients for a total of 396 patients and 278 deaths. Time-to-event endpoints were described using the Kaplan-Meier method. Two-sided 95% confidence intervals (CIs) were provided for the main statistical estimators. OS and PFS were compared between patients with PD-L1 high and low/negative expression for the different cutoffs using a log-rank test at a 5% level of significance. Prognostic value of PD-L1 expression in terms of OS was investigated using a multivariable Cox proportional hazards model where covariates were selected by biological and clinical significance and included age, race, smoking status, alcohol use, metastatic disease, platinum-based therapy, and anatomical site as baseline covariates. Due to the retrospective design of the study, some data were unavailable for collection.

Baseline characteristics
Nineteen sites in seven countries screened 513 patients with R/M HNSCC tumors not amenable to local therapy (e.g. surgery or radiation) or at stage IVC between March 1, 2011 and June 30, 2015. The majority of patients (n = 213; 51.7%) were from the United States, with the remainder from Greece (n = 57; 13.8%), Spain (n = 49; 11.9%), Germany (n = 35; 8.5%), Italy (n = 33; 8.0%), Japan (n = 15; 3.6%), and South Korea (n =10; 2.4%). Of the 513 patients, 412 met all eligibility criteria and comprised the full analysis set; PD-L1 expression was unknown in 16 (3.9%). The 16 patients with unknown PD-L1 expression were not included in prevalence assessments or outcome assessments unless otherwise stated. Most patients (n =400; 97.1%) provided one tissue sample, with 12 patients providing two samples for a total of 424 tissue samples. For patients who provided two samples, PD-L1 expression was determined independently on each sample, and the sample obtained closest to the index date was used to assess PD-L1 expression. Tumor samples were obtained from the primary site in 162/424 cases (38.2%), from recurrent disease in 179/424 cases (42.2%), and from distant sites in 83/424 cases (19.6%).

Discussion
In this study, we investigated if PD-L1 expression was associated with survival in patients treated with standard chemotherapy.
In the entire population of this study, PD-L1 was not prognostic for survival in patients with HNSCC who received standard chemotherapy regimens. This finding was consistent with observations in randomized controlled trials of similar patients with R/M HNSCC [10,26,29]. In CheckMate 141, for patients treated with investigator's choice the median OS in PD-L1 TC ≥ 1% was slightly lower than in PD-L1 TC < 1% [4.6 months (95% CI 3.8-5.8) vs 5.8 months (95% CI 4.0-9.8)] [30]. In KEY-NOTE-040 the survival of patients treated with investigator's choice of standard of care (methotrexate, docetaxel, or cetuximab) did not increase with increasing PD-L1 expression [12]. Similar results have also been observed a b in an evaluation of commercially obtained patient samples with stage I-IV HNSCC, in which PD-L1 expression was not prognostic for OS based on a TC ≥ 25% cutoff [31]. Currently accepted prognostic markers in HNSCC include HPV status in patients with oropharyngeal carcinoma and smoking status [32]. Other researchers have identified prognostic factors including age, race, ECOG PS, prior treatments [33], C-reactive protein, leukocyte levels, and time from diagnosis to relapse [34]. In a multivariable analysis of the SUPREME-HN study we found age, platinum therapy, primary tumor location, and metastatic disease to be associated with survival. It is not surprising that metastatic disease is associated with poorer survival, this variable has been incorporated in prognostic models of survival in advanced cancers [35]. Similarly, patients healthy enough to tolerate a platinum-based therapy might be expected to survive longer. The observation here of improved survival in older patients (≥ 60 years) compared with younger patients is somewhat counterintuitive; it is generally considered that older adults have comparable survival outcomes but with increased toxicity [36]. However, a non-significantly higher survival in patients > 65 years versus < 65 years has also been shown in patients treated with investigator's choice in a retrospective analysis of CheckMate 141 [37]. In both the SUPREME-HN and the CheckMate 141 studies, investigator's choice of standard of care was used. It is possible that elderly patients were treated with taxanes, rather than cisplatin and cetuximab, due to the higher toxicities associated with the latter therapies. Later publications have indicated that docetaxel improves OS over cisplatin [38]. One could speculate that investigators selected therapies for older patients based on the toxicity Table 3 Response and survival by PD-L1   profile, which were later demonstrated to be more efficacious. Urba identified race (Caucasian vs other) as prognostic for OS and PFS. In the SUPREME-HN study an association was observed that was only significant for PFS from first-line therapy; possibly because there was a smaller non-Caucasian population in this study. In a univariate analysis, Urba identified primary tumor location as negatively prognostic for survival (oral cavity vs "other", HR 1.37, 95% CI 1.15-1.63, P = 0.01) and associated with reduced PFS [33]. In the multivariable analysis of the SUPREME-HN study, patients with primary tumor locations of oropharynx and hypopharynx had improved OS compared with patients with oral cavity carcinoma and survival was significantly longer in patients with tumors in laryngeal versus oral cavity sites (HR 0.63, 95% CI 0.46-0.86, P = 0.003). Currently smoking and HPV status are considered to be major independent prognostic factors in patients with oropharyngeal cancer [32] and recent HNSCC randomized clinical trial studies have been stratified using PD-L1 and HPV, smoking status, and performance status [39]. The SUPREME-HN study shows meaningful survival differences by primary tumor location, raising the question whether site of tumor origin should also be considered in study design and patient treatment.
The PD-L1 prevalence at TC ≥ 25% was consistent across biopsy locations: 32.1% (primary tumor), 31.8% (recurrent site), and 32.5% (metastatic site). These data suggest that any tumor lesion can be used for PD-L1 testing for HNSCC, although in this study the primary and metastatic lesions were not from the same patient. Additionally, PD-L1 expression seems to be stable across the primary versus metastatic setting, only the punch biopsy gave lower PD-L1 expression.
The prevalence of PD-L1 varied according to a number of other factors; gender (higher in females), race, region, ECOG PS 0, oral cavity cancers, and never smokers. High PD-L1 prevalence has previously been significantly associated with females, never smokers, and oral cavity in other studies of second-line patients with HNSCC [23]. The PD-L1 TC ≥ 25% prevalence varied substantially depending on the primary tumor location; from 43.5% in oral cavity to 9.5% in hypopharyngeal (see Table 1). The median OS for patients with oral cavity carcinoma was lower in PD-L1 TC ≥ 25% than PD-L1 TC < 25% patients; poor prognosis in PD-L1 TC ≥ 25% oral cavity patients has been observed by others [18]. Likewise, for oropharyngeal primary site patients, median OS in patients with PD-L1 TC ≥ 25% was less than that seen for patients with PD-L1 TC < 25% (log-rank test; P = 0.03; Fig. 1d). Conversely, longer survival was seen in PD-L1 TC ≥ 25% than PD-L1 TC < 25% patients with hypopharyngeal primary tumors (21 months vs 12.2 months). These data indicate that for patients with tumors of oral cavity and oropharyngeal origin, PD-L1 expression is linked to shorter survival, whereas those with PD-L1 high hypopharyngeal primary tumors live longer.
Therefore, although PD-L1 was not prognostic in the entire SUPREME-HN cohort, our data indicate PD-L1 can be both positively and negatively prognostic depending on the primary tumor location. This finding may help to explain historical conflicting views of the prognostic value of PD-L1; for example, the finding that PD-L1 expression was positively prognostic in laryngeal squamous cell carcinoma [22] but conversely associated with poor prognosis in oral squamous cell carcinoma [17].
Study limitations pertain mainly to the retrospective study design, and hence, the reliance of available information in medical charts. Quantitative analyses of risk factors were limited due to missing information on performance status, HPV status, and small sample size. This study used an assay validated for PD-L1 expression on TCs and did not investigate the prognostic value of IC PD-L1 expression. PD-L1 expression in other cellular compartments of the tumor microenvironment may be indicative of survival. The variety of scoring methods used for determining PD-L1 positivity (TCs and/or ICs) may also contribute to the apparent contradictory publications regarding its prognostic value.
Tumor stage and grade at initial diagnosis were not available for all patients since they may have received initial care in a hospital other than the investigating site. Furthermore, the definition of R/M status may have led to the exclusion of patients who received local therapies for palliative purposes, as the treatment intent was not always mentioned in the patient's medical records. Additionally, evaluations of tumor response and progression were not evaluated via blinded, independent committee review as would be the case in clinical trials, which can lead to some variability in results. PD-L1 expression was assessed using available tissue that was not necessarily obtained at the time of initial diagnosis or at the same stage of disease for all patients. Findings from additional exploratory analyses suggest that PD-L1 expression was lower in tissue samples obtained after a patient's prior exposure to chemotherapy than prior to initiation of chemotherapy, irrespective of tissue origin (primary tumor, recurrent site, or metastatic site). A similar finding was observed for the subset of samples from the primary tumor obtained after exposure to radiotherapy.
Since starting the SUPREME-HN study a number of immunotherapies have been approved for use in R/M HNSCC. The approvals of PD-L1 assays as companion diagnostics demonstrates the predictive nature and the value of this biomarker. As the use of immunotherapy increases the opportunity diminishes to perform a prospective study in patients treated with non-immune based treatments and thus SUPREME-HN represents a unique historical record of the prognostic value of PD-L1.

Conclusion
There have been conflicting results reported regarding the prognostic value of PD-L1 expression on TCs. Early reports did not demonstrate any association between PD-L1 expression and OS, whereas other studies have suggested that PD-L1 expression may be associated with improved survival [17][18][19][20][21][22][23]. In the SUPREME-HN study, PD-L1 expression using TCs at cutoff values of 10%, 25%, and 50% was not prognostic for survival in patients with HNSCC treated with standard therapies; however, PD-L1 expression may be positively or negatively prognostic when anatomic subsites within the head and neck are considered.
In evaluating the correlation of PD-L1 and survival, previous studies did not always account for confounding factors. Based on our analysis these factors, specifically HPV status, primary tumor location, and demographic factors, may be highly relevant to OS in patients with R/M HNSCC.