Prognostic Significance of PD-L1⁺ and CD8⁺ Immune Cells in HPV⁺ Oropharyngeal Squamous Cell Carcinoma

Human papilloma virus–positive oropharyngeal squamous cell carcinoma (HPV⁺ OPSCC) is a clinically, genomically, and immunologically distinct subset of head and neck cancer that now represents the most prevalent form of oropharyngeal cancer in many parts of the world (1, 2). These tumors are located predominantly in the base of tongue and tonsil of the oropharynx, where they arise from the epithelium associated with the lingual and palatine tonsils. Unlike other head and neck squamous cell carcinomas (HNSCC) that are associated with smoking and alcohol consumption, HPV⁺ OPSCCs typically occur in young individuals that have never smoked and are associated with small primary tumors and larger nodal disease. Several randomized studies have shown improved prognosis in HPV⁺ OPSCC compared with HPV-negative tumors (3–5).

Given differences in outcomes between HPV⁺ tumors and HPV-negative tumors, accurate prognostic stratification is important to enable deintensification of therapy in patients with more favorable outcomes and, conversely, use of standard or intensified therapy in patients with worse outcomes. The current 7th edition Union for International Cancer Control/American Joint Committee on Cancer (UICC/AJCC) staging based on tumor size and extent, lymph node involvement, and metastasis [tumor–node–metastasis (TNM) staging] does not distinguish between HPV⁺ tumors and HPV-negative tumors and fails to accurately stratify HPV⁺ patients (6–8). Two recent prognostic models have been proposed specifically for HPV⁺ OPSCC, both of which appear to provide improved prognostication compared with the 7th edition of the UICC/AJCC classification. The first, which incorporates smoking status, was based on recursive partition analysis of RTOG 0129 (4), whereas the second was developed from a retrospective review at Princess Margaret Hospital (7). The latter system, with minor modifications, was validated by the International Collaboration on Oropharyngeal Cancer Network for Staging (ICON-S; ref. 9). The ICON-S classification, which has since been independently validated (10–12), has been adopted for the 8th edition of the UICC/AJCC staging system due to come into effect in 2018.

The presence of tumor-infiltrating lymphocytes (TIL) within tumors has been identified as a powerful prognostic factor in colorectal cancer, ovarian cancer, and other malignancies (13–16) with potential to improve prognostic stratification compared with TNM staging (13, 17). In HNSCC, CD8⁺ infiltrates are more common in HPV⁺ SCC and have been associated with prognosis in several studies (18). We hypothesized that quantifying the abundance and distribution of CD8⁺ TILs, together with expression of programmed death ligand 1 (PD-L1) on tumor cells (TC) or immune cells (IC), would be of prognostic importance in HPV⁺ OPSCC. We, therefore, determined if profiling the immune microenvironment could improve the prognostic stratification provided by existing clinical TNM-based staging systems.

Patients with OPSCC planned for curative treatment were retrospectively identified from an institutional database of patients with HNSCC treated at the Peter MacCallum Cancer Centre between 2002 and 2012. Formalin-fixed paraffin-embedded (FFPE) tumor blocks or unstained sections, together with clinicopathologic and outcome data, were collected. All FFPE blocks and unstained slides were stored at room temperature in the dark. HPV⁺ OPSCCs were identified by p16 immunohistochemistry (IHC). All tumor samples were from primary resections or from biopsies taken prior to the commencement of treatment. This study had Institutional Ethics Committee approval (PMCC HREC 12/144).

All IHC was performed on 4 μm whole tumor sections, which were baked at 60°C for 1 hour followed by dewaxing through three histolene baths, two 100% alcohol baths, a 70% alcohol bath, and distilled H₂O (dH₂O). Antigen retrieval was performed in a pressure cooker at 125°C for 3 minutes. After cooling to 90°C, slides were placed onto a Dako autostainer (Agilent). Slides were incubated with 3% H₂O₂ for 10 minutes, followed by incubation with the following primary antibodies: PD-L1 (clone SP142; Spring Bioscience) at a 1:500 dilution for 10 minutes; CD8 (clone 4B11; Leica Biosystems) at a 1:1,000 dilution for 60 minutes; and p16 prediluted antibody (CINtec Histology kit; Ventana Medical Systems) for 60 minutes. Secondary detection was performed using the following Dako EnVision+ systems: EnVision+ rabbit for 30 minutes (for PD-L1–incubated slides) and EnVision+ mouse for 60 minutes (for CD8- and p16-incubated slides). A 10-minute incubation with 3,3′-diaminobenzidine (DAB) was used for color detection. Slides were rinsed with buffer [50 mmol/L Tris-HCl (ph7.6) with 1% tween20] between incubations. IHC sections were then counterstained with hematoxylin, mounted, and coverslipped.

p16 expression was scored for both intensity and proportion of staining in the TC nucleus and cytoplasm. Intensity was scored as 0 (none), 1 (weak), 2 (moderate), or 3 (strong), with proportion scored as 0% to 100% to the nearest 10% for each intensity. p16 positivity was defined as ≥70% of TCs with moderate or strong staining (19). The abundance and compartmental location of CD8-expressing and PD-L1–expressing immune cells (IC) within tumors were determined by semiquantitatively scoring the proportion of positively stained cells located within each of two areas: (i) within the epithelial intratumoral (iT) nests and (ii) in the stroma (S) immediately adjacent to the tumor nests (see Supplementary Fig. S1).

The proportion of positively stained CD8⁺ TILs was scored as a proportion of the total number of cells within the intratumoral or stromal compartments as 0% to 100% to the nearest 10%. PD-L1 expression was scored separately for TCs and ICs by two observers. TC expression was scored for the proportion of membranous staining of any intensity as 0%, 1%, 5%, and 10% to 100% to the nearest 10%. PD-L1 IC expression was scored in the intratumoral and stromal compartments as a percentage of tumor area and categorized as 0 (<1%), 1 (1%–4%), 2 (5%–9%), or 3 (≥10%), based on criteria used in previous publications with this PD-L1 antibody (20–22). After the independent assessment of PD-L1 staining (TCs, intratumoral ICs, and stromal ICs), the two observers reviewed the discrepant cases and reached an agreement on discrepant cases. The agreed PD-L1 result was used in all analyses.

Dual stain immunohistochemistry for p16 and PD-L1 was performed to confirm the localization of PD-L1 expression on ICs within the intratumoral compartment. Slides were placed into a Ventana Benchmark Ultra autostainer (Ventana Medical Systems) using supplied reagents for the following steps: slides were warmed to 72°C for 8 minutes to bake sections, followed by deparaffinization, antigen retrieval in Cell Conditioner 1 for 56 minutes at 100°C, and incubation in preprimary peroxidase inhibitor for 4 minutes. Slides were then warmed to 36°C for the following incubations: PD-L1 antibody (clone SP142, Spring Bioscience; 1:500 dilution) with antibody detection using the Optiview Detection Kit (brown) and p16 (prediluted antibody, CINtec) with antibody detection using a red chromagen detection kit. Slides were rinsed with reaction buffer between each incubation. At completion, slides were counterstained with hematoxylin, mounted, and coverslipped.

Vectra/OPAL multispectral immunohistochemistry (MSI) was used to confirm the findings of the individual IHC. A four-color panel was optimized to enable simultaneous detection, imaging, and analysis of PD-L1, CD8, p16 (for TCs), and DAPI (for cell visualization) on the same single section. Briefly, slides were baked at 60°C for 30 minutes and then dewaxed to water followed by antigen retrieval in 1 mmol/L EDTA buffer (pH8) for 2.5 minutes at 125°C in a pressure cooker. After slides were allowed to cool, they were washed in dH₂0, followed by TBST (0.1M TRIS-HCl pH8, 0.15M NaCl, 0.05% Tween-20), and then the following incubations: the first primary antibody (CD8, Clone 4B11; Leica Biosystems) for 30 minutes; 0.1% H₂O₂ block for 5 minutes; secondary antibody (rabbit or mouse horseradish peroxidase–conjugated secondary antibodies at a 1:500 dilution) at room temperature for 10 minutes; TSA reagent (PerkinElmer; diluted 1:50 in TSA amplification diluent) for 10 minutes. All incubations were performed in a humidified chamber at room temperature. Three 2-minute washes in TBST were performed between each step above. At the completion of the above steps, microwave removal of 1^o and 2^o antibodies was performed, allowing for the above protocol to be repeated for p16 (CINtec) and PD-L1 (clone SP142; Spring Bioscience; multiplexing of antibodies). After the final antibody protocol was completed, slides were stained with DAPI and coverslipped with Vectashield Hard Set mounting media. Slides were stored at 4°C until imaging. With each run of test OPSCC slides, appropriate single-antibody control slides were included as imaging controls.

MSI slides were imaged with a Vectra microscope utilizing Vectra 3.0.3 software. For each test OPSCC slide, a low-power (magnification, ×4) overview image was captured, from which several high-power (magnification, ×20) multispectral images were captured, representing regions to be analyzed. Analysis software (Inform V2.2.1) was then used to analyze the images. Programming of the software to distinguish tissue categories (tumor, stroma, and other) and phenotyping of individual cells based on fluorescence intensities resulted in data that allowed for the quantification of different cell phenotypes within each tissue category.

The statistical analysis of prognostic factors occurred in three steps. In the first step, Cox proportional hazard models were used to assess the impact of three different staging systems (treated as ordinal) on OS [AJCC 7th edition, ICON-S (7, 9), and Ang (4) risk stratification] to decide which staging would be used in the multivariable models with PD-L1 and CD8. In the second step, Cox proportional hazard models were used to assess the impact of each PD-L1 and CD8 assessment, as described previously in text, on OS and to assess which PD-L1 and which CD8 assessment provides best prognostic value, adjusting for stage (ordinal), as per step 1, and age (continuous). In the third step, a multivariable analysis was performed with the PD-L1 and CD8 assessments selected from step 2 to assess the independent prognostic value of these biomarkers. A sensitivity analysis was performed, adjusting for smoking (>10 packs/year, Y/N), gender (M/F), and chemotherapy (Y/N), in addition to stage (ordinal) and age (continuous). HRs, 95% confidence interval (95% CI), likelihood ratio P values, and c-statistics were provided for each model. The c-statistic is a measure of overall model performance in discriminating patients with different survival times. It is a global assessment of goodness of fit of a particular model. The assumption of (i) proportional hazards and (ii) linearity (for continuous variables) was verified for each of the models. The presence of influential observations was assessed, and the Kaplan–Meier method was used to display the survival curves. All statistical analyses were performed in R v3.4.2 (39).

Clinical outcome data and FFPE tumor blocks were obtained from 355 patients with OPSCC treated at Peter MacCallum Cancer Centre between 2002 and 2012. The p16 status was determined in 354 of the 355 OPSCC patients, of which, 57% (202/354) were p16⁺. Clinical outcome in p16⁺ OPSCC (HPV⁺ OPSCC) patients was better than that of p16-negative OPSCC patients, using follow-up data collected in 2014 (Supplementary Fig. S2). Updated follow-up data were obtained from the patient records (up until a close-out date of August 31, 2017, patient death, or the last follow-up visit, whichever came last) for the p16⁺ patients to assess the prognostic significance of PD-L1 and CD8 in this group. One hundred and ninety of the 202 patients with HPV⁺ OPSCC were treated with curative intent, and this population was included in the prognostic factor analysis. The median age was 57, and 157 of 190 (82.6%) were male, consistent with the known demographics of HPV⁺ OPSCC. Thirty-four deaths occurred, and the median follow-up was 5.2 years. Patient characteristics and a CONSORT diagram describing this cohort are shown in Table 1 and Supplementary Fig. S3, respectively.

Given the limitations of current staging systems for HNSCC in providing accurate prognostic stratification for HPV⁺ OPSCC, the effect of different staging systems was assessed to decide which one would be used when assessing the independent prognostic value of immune profiles (CD8⁺ and PD-L1⁺ cells) on overall survival (OS). Three staging systems were considered: UICC/AJCC 7th edition, Ang risk classification (4), and the ICON-S classification (9) with Kaplan–Meier curves, as shown in Supplementary Fig. S4. The c-statistic was 0.65 for ICON-S, 0.52 for UICC/AJCCv7, and 0.60 for Ang risk classification. Based on this analysis and its adoption as the new staging system in the 8th edition of the UICC/AJCC, staging system ICON-S was selected as the most suitable to be used as a covariate for multivariable analysis of IHC PD-L1 and CD8 on OS.

The abundance and location of CD8⁺ TILs were evaluable by IHC in 181 of 190 patients. Examples of CD8⁺ TILs in the peritumoral stroma (hereafter described as stromal TILs) and within the epithelial TC nests (hereafter described as intratumoral TILs) are shown in Fig. 1A and B. High CD8⁺ stromal TIL abundance (defined as >30% positivity) was seen in 103 of 181 (56.9%) of tumors, whereas high CD8⁺ intratumoral TILs were seen in 51 of 181 (28.2%) of tumors. The correlation between the stromal TILs and intratumoral TILs in patients with HPV⁺ OPSCC was 0.56 (95% CI, 0.45–0.66).

The prognostic significance of CD8⁺ TILs in both stromal and intratumoral compartments in HPV⁺ OPSCC was evaluated using univariable analysis and also adjusted for stage and age with the purpose of assessing the prognostic value of different CD8⁺ TILs assessments (stromal or intratumoral; continuous or using 30% cutoff). The presence of abundant stromal CD8⁺ TILs (>30%) was associated with improved OS with an HR of 0.4 (95% CI, 0.2–0.8, P = 0.008) using univariate analysis and an HR of 0.50 (95% CI, 0.24–1.06, P = 0.066) when adjusted for stage and age (Kaplan–Meier plot; Fig. 1C). When the presence of CD8⁺ stromal TILs was assessed as a continuous variable, the HR for 10% increments was 0.8 (95% CI, 0.7–1.0, P = 0.030) using univariable analysis and 0.89 (95% CI, 0.75–1.04, P = 0.127) when adjusted for stage and age. The presence of abundant intratumoral CD8⁺ TILs (>30%) demonstrated a similar HR for OS and stromal CD8⁺ TILs and was also statistically significant by univariable (HR, 0.3; 95% CI, 0.1–1.0, P = 0.035) and multivariable analyses (HR, 0.36; 95% CI, 0.13–1.03, P = 0.032). The Kaplan–Meier plot for OS by intratumoral CD8⁺ TILs using the 30% cutoff is shown in Fig. 1D. When analyzed together, the combination of high CD8⁺ TILs in the stroma or the intratumoral compartment was associated with improved OS (HR, 0.4; 95% CI, 0.2–0.7, P = 0.003) by univariate analysis and an HR of 0.42 (95% CI, 0.20–0.87, P = 0.017) when adjusted for stage and age (Fig. 1E).

IHC for PD-L1 on TCs, expressed as a percentage of TCs showing positive membranous staining for PD-L1, was evaluable in 182 of the 190 HPV⁺ OPSCC patients. Eighty-six of 182 patients (47%) had 1% or more PD-L1 TC staining, and 14 of 182 had >50% PD-L1 TC staining. Although PD-L1 TC staining >50% suggested an inferior OS curve, the number of patients with PD-L1 TC staining >50% was small (n = 14), and results were not statistically significant by either univariable analysis (HR, 1.9; 95% CI, 0.7–5.4, P = 0.238) or after adjustment for stage (HR, 1.9; 95% CI, 0.7–5.6, P = 0.257). An example of PD-L1 TC staining and the Kaplan–Meier plot for OS by PD-L1 TC staining are shown in Fig. 2.

We then evaluated the abundance and prognostic significance of PD-L1 expression on both ICs located in the peritumoral stromal compartment (stromal ICs) and within the epithelial TC nests (intratumoral ICs; Fig. 3A and B). For stromal ICs, 41 (22.7%) tumors had an IC score of <1%, 69 (38.1%) had a score of 1% to 4%, 38 (21.0%) had a score of 5% to 9%, and 33 (18.2%) had a score of ≥ 10%. Abundant PD-L1 expression by stromal ICs (≥5%) was not associated with improved OS by univariable analysis (Fig. 3C) or when adjusted by stage (HR, 0.6; 95% CI, 0.3–1.4, P = 0.214). For intratumoral ICs, 46 (25.4%) of cases had an IC score of <1%, 65 (35.9%) had a score of 1% to 4%, 31 (17.1%) had a score of 5% to 9%, and 39 (21.5 %) had a score of ≥10%. Tumors with an intratumoral IC score of ≥5% had significantly better OS (HR, 0.4; 95% CI, 0.2–1.0, P = 0.039) using univariable analysis (Fig. 3D). Only 6 events were recorded in 70 patients with intratumoral IC PD-L1 expression ≥5%. In contrast, 25 events were recorded in 111 patients with PD-L1 expression <5%. This prognostic effect of intratumoral PD-L1 expression remained after adjustment for ICON-S stage (HR, 0.37; 95% CI, 0.15–0.93, P = 0.023). PD-L1 expressions on TCs and ICs were not strongly correlated (r_s = 0.38).

MSI and dual-color IHC were performed on a subset of tumors to confirm PD-L1 expression on TCs and intratumoral ICs, and to ensure they are distinct, given their colocalization within TC nests. Using Opal four-color fluorescent MSI with antibodies for p16, CD8, and PD-L1 and the VECTRA multispectral imaging platform, we demonstrated that tumors with PD-L1⁺ ICs by DAB IHC had PD-L1 expression on cells that were not p16⁺ TCs (Fig. 4). Similar findings were seen with dual-color IHC (Supplementary Fig. S5). Phenotyping also showed that although a proportion of the PD-L1⁺ ICs were CD8⁺ TILs, many were CD8⁻, indicating that these PD-L1⁺ ICs are not limited to CD8⁺ cells and may include other IC populations.

Following the selection of PD-L1 intratumoral ICs (<5% or ≥5%) and CD8⁺ ICs (<30% stromal and intratumoral or ≥30% stromal and intratumoral) as previously described, we proceeded to a multivariable model, including PD-L1 and CD8 adjustments by stage and age. Intratumoral IC PD-L1 staining provided powerful additional prognostic information to the ICON-S stage. The c-statistic for stage (ICON-S) and age alone was 0.69. The addition of intratumoral IC PD-L1 staining (<5% or ≥5%) improved this value to a c-statistic of 0.74, whereas the addition of CD8 ICs (<30% or ≥30%) resulted in a c-statistic of 0.72. The c-statistic for PD-L1, CD8, age, and stage was 0.76 with a PD-L1 HR of 0.39 (95% CI, 0.15–0.99, P = 0.048) and a CD8 HR of 0.43 (95% CI, 0.20–0.91, P = 0.027).

A sensitivity analysis was carried out reassessing the impact of PD-L1 and CD8 on OS but adjusting for smoking (>10 packs/year, Y/N), gender (M/F), and chemotherapy (Y/N), in addition to staging (ordinal) and age (continuous). Although the impact of PD-L1 intratumoral IC staining on OS did not reach statistical significance in the full model (HR, 0.42; 95% CI, 0.16–1.11, P = 0.081), the impact became significant after backward elimination (HR, 0.39; 95% CI, 0.15–0.99, P = 0.048). CD8 was significantly associated with OS in the full model (HR, 0.37; 95% CI, 0.17–0.83, P = 0.015) and after backward elimination (HR, 0.43; 95% CI, 0.20–0.91, P = 0.027). Aside from intratumoral IC PD-L1 staining and CD8, only age (HR, 1.40 per 5 years increase; 95% CI, 1.16–1.70, P = 0.001) and ICON-S (HR, 1.82; 95% CI, 1.14–2.92, P = 0.013) remained significant after backward elimination.

High intratumoral IC PD-L1 (≥5%) and high CD8 ICs (≥30%) were able to identify a patient population with an excellent prognosis independent of stage. The presence of intratumoral IC PD-L1 staining ≥5% separated this group from the <5% group (intermediate prognosis), with a 3-year OS of 100%, 96% (95% CI, 88–100), and 100% for ICON-S stages 1, 2, and 3, respectively (Fig. 5A–C). Stage 4, which includes M1 disease, was excluded. Similarly, CD8⁺ infiltrate (≥30%) also identified a group with excellent prognosis, with a 3-year OS of 96% (95% CI, 91–100), 86% (95% CI, 76–97), and 100% for ICON-S stages 1, 2, and 3, respectively (Fig. 5D–F).

The significant morbidity associated with conventional therapy for advanced oropharyngeal cancer has led to interest in deintensification of therapy for patients with good prognostic HPV⁺ tumors, with the corollary requirement for accurate prognostic stratification in this population. Current anatomically based TNM-staging methodologies, designed for HPV-negative OPSCC, have well-documented limitations with respect to accurately discriminating prognostic groups for HPV⁺ OPSCC (6–8).

In this study, we quantitatively analyzed the prognostic significance of CD8⁺ TILs and PD-L1 staining of TCs and ICs in a large cohort of HPV⁺ OPSCC patients. Both CD8⁺ TILs and intratumoral PD-L1+ ICs have prognostic significance for OS. Intratumoral PD-L1⁺ ICs and CD8⁺ TILs may provide further prognostic stratification over existing anatomical TNM-based systems, including the ICON-S (UICC/AJCC 8th Edition) classification.

Our finding of frequent CD8⁺ lymphocyte infiltrates in HPV⁺ OPSCC and association with OS (for intratumoral or stromal CD8⁺ TILs ≥ 30%) is consistent with findings reported across a variety of tumor types. In the context of HNSCC, qualitative (23, 24) and quantitative differences (18, 25–29) have been described in IC infiltrates in HPV⁺ compared with HPV-negative tumors. HPV⁺ tumors have consistently been reported to have higher degrees of lymphoid infiltration with increased CD8⁺ lymphocytes but also increased FOXP3⁺ regulatory T cells and CD20⁺ B lymphocytes. Mandal and colleagues (30), in an analysis of transcriptome data from 280 HNSCCs from The Cancer Genome Atlas, found that HPV⁺ tumors have higher degrees of T cells and overall IC infiltration, as well as higher expression of immune activation markers, granzyme and perforin, compared with HPV-negative tumors. The presence of a CD8⁺ T-cell infiltrate correlates with improved prognosis in HPV⁺ tumors in multiple previous studies (25–29, 31), including a meta-analysis (18).

PD-L1, also known as B7-H1 or CD274, is the major ligand for PD-1, a receptor expressed on the surface of activated T cells that regulates T-cell proliferation and activation (32). PD-L1 may be expressed not only by TCs but also on ICs, including lymphocytes, dendritic cells, and macrophages (21, 32). Although we did not find a significant association of PD-L1 expression on TCs with prognosis, we demonstrated a strong association between PD-L1 expression on intratumoral ICs and outcome. We confirmed that expression of PD-L1 within TC nests was on nontumor ICs using multispectral immunofluorescence and dual-color IHC. Tumors with an intratumoral IC PD-L1 score of ≥5% (70/191 or 38% of our cohort) had markedly better OS. This predictive effect of intratumoral PD-L1 expression remained after adjustment for ICON-S stage and other known prognostic factors. In contrast, no prognostic significance was seen for PD-L1 staining of ICs in the peritumoral stroma (stromal ICs).

A key finding of our study is that immunophenotyping using quantification of CD8⁺ TILs and PD-L1 staining of intratumoral ICs can provide a powerful contribution to, and improvement of, the prognostic stratification of anatomically based staging methods. In our cohort of 191 patients with p16⁺ OPSCC, the UICC/AJCC 7th edition TNM staging failed to provide any discriminative prognostic stratification, as had been observed in other series (6–8). The ICON-S classification, which has been independently validated, did separate groups, particularly for stage I compared with stages II and III. However, when intratumoral PD-L1 IC scoring was added to the algorithm, it was able to separate stages I, II, and III into good and poor prognosis groups. The addition of PD-L1 tumoral IC staining (<5% or ≥5%) to TNM staging (ICON-S) improved the c-statistic, a measure that quantifies the capacity to discriminate patients with different survival times, from 0.69 (ICON-S) to a c-statistic of 0.74 (intratumoral IC PD-L1 and ICON-S). CD8⁺ TILs (<30% or ≥30%) also identified a group with good prognosis and improved the c-statistic to 0.72. Together, these data support the concept of adding immunophenotyping to anatomically based classifications for stratification of HPV⁺ OPSCC.

Similar findings with respect to PD-L1 staining of ICs were reported in a heterogeneous cohort of 402 surgically treated head and neck cancers, including tumors from the oral cavity, oropharynx, larynx, hypopharynx, and nasal cavity (20). Kim and colleagues found that PD-L1 expression in more than 5% of ICs is associated with improved recurrence-free survival and OS in resected HNSCC. As was the case in our study, no prognostic significance of TC PD-L1 expression was identified. In the subset of patients with p16⁺ OPSCC (n = 98), higher CD3⁺ and CD8⁺ cell infiltrates were more frequent than in patients with p16-negative tumors, and patients with PD-L1⁺ ICs had superior 5-year OS compared with patients with PD-L1–negative ICs (93.8% vs. 81.4%, P = 0.04). Although ICs were not defined according to intratumoral or stromal compartment, these data, using the same PD-L1 antibody (SP142 clone) and scoring cut-off, are in line with our data. Two studies, albeit in relatively small cohorts of HPV⁺ tumors, describe an association of PD-1, the receptor for PD-L1, and outcome. Badoual and colleagues (31) found a high degree of T-cell infiltration positive for PD-1 in HPV⁺ OPSCC and that higher levels of PD-1⁺ T cells correlated with improved prognosis. Balermpas and colleagues also demonstrated that intratumoral and invasive margin expression of PD-1 is associated with improved outcomes in OPSCC (33).

In addition to prognostic significance, the level of PD-L1⁺ ICs may predict response to PD-1/PD-L1 checkpoint inhibition. Badoual and colleagues demonstrate that the PD-1⁺ T cells in HPV⁺ OPSCC express activation markers and are not simply exhausted but functional after PD-1/PD-L1 blockade (31). Ferris and colleagues (34) describe an analysis of tumors from the CheckMate 141 phase III clinical trial, which shows that an increased abundance of tumor-associated ICs expressing PD-L1 is associated with longer OS and greater likelihood of response to the drug nivolumab (anti–PD-1). Similar findings have been reported in bladder cancer, where the presence of PD-L1 staining in >5% of tumor-infiltrating ICs is associated with increased response to the PD-L1 inhibitor atezolizumab (35) and in lung cancer where the effector T-cell (T_eff) gene signature is associated with PD-L1⁺ ICs and correlates with response to the PD-L1 inhibitor atezolizumab (22). Together, these data suggest that in addition to indicating good prognosis, PD-L1⁺ ICs indicate an immune microenvironment with an existing, though potentially attenuated, immune response that may be unleashed with the use of PD-1– or PD-L1–directed therapy.

Our study has several limitations. Although the study is based on a relatively large cohort, it is nonetheless a single retrospective analysis that needs independent validation in other cohorts of HPV⁺ OPSCC. We used p16 immunohistochemistry alone (rather than PCR or ISH) to determine HPV status, a method that is well accepted and specific when used with a cutoff of greater than or equal to 70% in oropharyngeal tumors (36). The PD-L1 antibody used in our analyses (clone SP142) is one of several antibodies available for assessment of PD-L1 in tumors, and whether similar data can be obtained using other antibodies is unknown at present. Although the SP142 clone performs differently to other PD-L1 antibodies for assessment of TC PD-L1 (37, 38), it is the only antibody currently FDA approved as a codependent diagnostic for the assessment of IC PD-L1 staining.

In summary, we have demonstrated that the assessment of the immune microenvironment of HPV⁺ OPSCC, through quantification of CD8⁺ TILs and PD-L1⁺ intratumoral ICs, provides prognostic information. This additional information improves the prognostic capacity of existing anatomic staging methods, including the new ICON-S classification. Ultimately, immunophenotyping of tumors may provide a useful clinical tool for prognostic stratification of HPV⁺ OPSCC to be used in addition to TNM-based staging to guide selection of therapy, including deintensification approaches.

No potential conflicts of interest were disclosed.

Conception and design: B. Solomon, R.J. Young, D. Urban, D. Rischin

Development of methodology: B. Solomon, R.J. Young, M. Kowanetz, D. Rischin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B. Solomon, R.J. Young, D. Urban, S. Hendry, A. Thai, C. Angel, A. Haddad, J. Corry, S. Fox

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B. Solomon, R.J. Young, M. Bressel, S. Hendry, M. Kowanetz, D. Rischin

Writing, review, and/or revision of the manuscript: B. Solomon, R.J. Young, M. Bressel, D. Urban, S. Hendry, C. Angel, A. Haddad, M. Kowanetz, T. Fua, J. Corry, S. Fox, D. Rischin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Solomon, R.J. Young, D. Urban, A. Thai, J. Corry

Study supervision: B. Solomon, D. Rischin

Other (interpretation of immunohistochemistry and advice of hitopathology): C. Angel

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.