PD-1 Status in CD8⁺ T Cells Associates with Survival and Anti-PD-1 Therapeutic Outcomes in Head and Neck Cancer

Immune checkpoint receptor (ICR) blockade has become a major focus of investigation in the field of cancer immunotherapy. Importantly, ICR blockade has shown beneficial results in the clinic for certain patient populations; however, biomarkers of response have not been clearly identified. Programmed death-1 (PD-1; ref. 1) and cytotoxic T-lymphocyte antigen-4 (CTLA-4) are two of the main clinical targets (2). While these have been increasingly investigated, less is known about their expression in tumor-infiltrating lymphocytes (TIL) during anti-PD-1 therapeutic interventions. Given the promising data that have been reported in several malignancies (3–5), we investigated ICR levels on specific TIL subsets in patients with head and neck squamous cell carcinoma (HNC).

An increasingly important prognostic marker for HNC patients is HPV status, because the percentage of HPV⁺ oropharyngeal squamous cell carcinoma in North America has increased from 30% in the 1980s to 80% at present (6). Although it is known that HPV⁺ HNC has a better prognosis than HPV⁻ HNC, and better response to anti-PD-1–based immunotherapy (7), the mechanism(s) underlying these clinical differences remain elusive, as do differences in immune escape strategies (8). Because of these prognostic differences between HPV⁺ and HPV⁻ in HNC patients, the distinct immunologic features of the two groups should be compared in order to advance knowledge regarding tumor immune evasion. Therefore, we investigated differences in ICR expression at the mRNA and protein level and characterized immunologic properties of tumor-associated T lymphocytes, including CD8⁺ T effector cells, with respect to HPV status.

Promising results using PD-1 or PD-L1 blocking monoclonal antibodies (mAb) have emerged for advanced recurrent/metastatic HNC (7, 9), but still only a minority of patients (15%–20%) respond, despite elevated expression of PD-L1 in >50% of HNC patients (8). On the one hand, PD-1 ligation has been demonstrated to have a negative impact on T cells, and the blockade of this ligation results in improvement of their function (10). On the other hand, PD-1 positivity has been shown to represent antigen experienced, TA-specific T cells (11) and has been correlated with better clinical outcome (12). Additionally, other checkpoint receptors such as T-cell immunoglobulin-3 (Tim-3; refs. 13, 14), lymphocyte activation gene-3 (LAG-3) and B and T lymphocyte attenuator (BTLA) are under investigation. Tim-3 has been identified as a specific marker of fully differentiated IFN-γ producing CD4⁺ and CD8⁺ T cells (15). Its expression is regulated by the transcription factor T-bet (16) and negatively regulates Th1 and CD8⁺ cytotoxic T-cell responses (17). LAG-3 is upregulated on activated CD4⁺ and CD8⁺ T cells as well as in a subset of activated natural killer (NK) cells (18). Besides its capacity to bind to major histocompatibility complexes (MHC) class II and its role in T helper cell and regulatory T-cell (Treg) signaling, direct inhibitory effects of LAG-3 on CD8⁺ T effector cells have been demonstrated (19). BTLA negatively regulates T-cell activation by inhibiting T-cell proliferation and cytokine production. In contrast to other checkpoint receptors, BTLA is expressed on naïve T cells and only transiently upregulated upon TCR engagement (17). Thus, BTLA is downregulated on highly activated T cells (20).

Little is known about the role of checkpoint receptors in TIL from HPV⁺ versus HPV⁻ HNC. Interestingly, a prior report suggested that the presence of PD-1 T cells in HPV⁺ patients was associated with a beneficial effect on survival (12). Given that HPV⁺ patients demonstrate a much better outcome in the clinic (12), we hypothesized that intratumoral ICR expression and function might provide immunologic insight into this difference in prognosis. Thus, as opposed to solely analyzing the frequency of PD-1⁺ versus PD-1⁻ TIL, we investigated whether the intensity of PD-1 expression on TIL influences clinical outcome, providing a potential prognostic biomarker and for monitoring response to anti-PD-1–based immunotherapy.

RNAseq data from queried genes were downloaded from the UCSC cancer genomics browser (https://genome-cancer.ucsc.edu). The HNC gene expression profile from 500 HNC specimens was measured experimentally (21). The RSEM units to quantitate RNAseq expression data were described previously (22). Correlations from The Cancer Genome Atlas (TCGA) data were calculated using Pearson r test, and linear regression curve fits were graphed using GraphPad PRISM software v6.

Peripheral venous blood samples and tumors were obtained from HNC patients seen in the Department of Otorhinolaryngology at the University of Pittsburgh Medical Center. No active patient exclusion to clinical stage or HPV status was performed other than restriction to all consented, surgically treated patients with primary tumors and curative treatment intention. Specimen acquisition was continuously performed by the same physician during 2011–2013, the patients' HPV status was defined from surgical specimen by IHC determination of the p16 status (positive p16 expression was defined as strong and diffuse nuclear and cytoplasmic staining in at least 70% of the tumor cells). All subjects signed written informed consent approved by the Institutional Review Board of the University of Pittsburgh, patient studies were conducted in accordance with the Declaration of Helsinki.

Blood samples and tumor specimen were collected at the day of and prior to surgery and, therefore, at the beginning of the treatment. Blood samples from cancer patients (30–40 mL) were drawn into heparinized tubes and centrifuged on Ficoll–Hypaque gradients (GE Healthcare Bioscience). Peripheral blood mononuclear cells (PBMC) were recovered, washed in RPMI-1640 (Invitrogen), and immediately used for experiments or stored at −80 °C until further analysis. For TIL isolation, freshly isolated tumors from HNC patients were minced into small pieces in RPMI-1640 (Invitrogen), which then were transferred to a cell strainer (70 mm Nylon) and mechanically separated using a syringe plunge. The cells passing through the cell strainer were collected and subjected to Ficoll–Hypaque gradient centrifugation. After centrifugation, mononuclear cells were recovered and stored at −80 °C until flow cytometry analysis.

Quantitative (q) PCR was performed to investigate tumor specimen from 20 HPV⁺ and 20 HPV⁻ HNC patients. Acquisition of patient material was separately performed during 2007–2011, following the consistent standards as described in patients and specimens. RNA was extracted using the RNeasy Kit (Qiagen) following the manufacturer's protocol. RNA concentration was determined by spectrophotometer measurement (NanoVue Plus, GE Healthcare). Random hexamers and MultiScribe Reverse Transcriptase (Applied Biosystems) were used for first-strand cDNA synthesis according to the manufacturer's instructions. The samples were added to a 20-μL reaction using 2X Taqman Master Mix (Applied Biosystems) and analyzed on a StepOne Real-Time PCR System (Applied Biosystems) using the following PCR probes: Hs00169472_m1 for PD-1, Hs00158563_m1 for LAG-3, Hs00262170_m1 for TIM-3, Hs03044418_m1 for CTLA-4, Hs99999905_m1 for GAPDH as a reference gene (all Life Technologies).

For cell surface staining, PBMCs and TILs were washed twice in staining buffer (2% w/v fetal bovine serum) and stained for cell surface markers. Cells were incubated with relevant antibodies for 30 minutes at 4°C in the dark, washed twice and resuspended in staining buffer. Flow cytometry was performed using a CyAn flow cytometer (Dako) and Fortessa cytometry (Becton Dickinson), and data were analyzed using FlowJo software (TreeStar, Inc.). The acquisition and analysis gates were restricted to the lymphocyte gate based on characteristic properties of the cells in the forward and side scatter.

The following anti-human mAbs were used for staining: PD-1-APC (eBioscience), CD3-PerCP-Cy5.5, CTLA-4-PE, PD-1-PerCP Cy5.5 (all Biolegend), CCR7-FITC (R&D Systems Inc.) CD3-Alexa Fluor 405, CD4-PE-Texas Red, CD-8-PE-TR, CD45RA-PE-TR (all Life Technologies), CD3-APC-Cy7, CD8-APC, CD8-PE-Cy7 (all BD Biosciences), granzyme B FITC (clone GB11, Biolegend) including their respective isotypes, which served as negative controls for surface as well as intracellular staining. Antibodies were pre-titrated using activated as well as nonactivated PBMC to determine optimal staining dilutions. Viability was assessed by Zombie aqua (Biolegend).

Isolated T cells of healthy donors were stimulated for 5 days with CD3/CD28 beads. CD3⁺ CD8⁺ T cells were analyzed by flow cytometry for expression of PD-1 as demonstrated in Supplementary Fig. S1. Accordingly, TILs were analyzed by flow cytometry for CD8⁺ PD-1⁺ cells in reference to isotype controls. Ranging from highest detected PD-1 expression to the lower limit of PD-1⁺, the gated PD-1⁺ cells were divided into cells with high PD-1 (PD-1^high), intermediate (PD-1^int) and low (PD-1^low) expression levels.

Isolated TIL (n = 3 HNC patients) were stained for PD-1 expression of CD3⁺ CD8⁺ T cells and then compared for granzyme B positivity of PD-1^negative, PD-1^low, PD-1^int, and PD-1^high subgroups.

After TIL isolation from HNC tumors (n = 3), T cells were purified using EasySep Human T Cell Enrichment Kit (Stemcell Technologies). After staining, CD8⁺ T cells were sorted using MoFlo Astrios (Beckman Coulter) in PD-1–negative, PD-1 low–intermediate, and PD-1–high groups. Cells were rested overnight in human serum and compared for functional differences in an IFN-γ ELISPOT assay. Therefore, MultiScreen-IP filter plates (Millipore) were coated overnight with anti-human IFN-γ mAb 1-D1K (Mabtech; 10 μg/mL in PBS) at 4°C. After washing with PBS, plates were blocked for 1 hour at 37°C with 10% human serum in RPMI. Sorted cells were added to wells in duplicates (5 × 10³) and stimulated with anti-CD3/CD28 beads (bead: cell = 5:1). After incubation for 18 hours at 37°C, plates were washed with PBS/0.5% Tween 20, and incubated with biotinylated anti-IFN-γ mAb (Mabtech; 2 μg/mL in PBS/0.5% BSA) for 2 hours at 37°C. Plates were washed again and incubated with streptavidin–HRP (Mabtech, 1:500 in PBS/0.5% BSA) for 1 hour at 37°C. After washing, TMB substrate solution (Vector Laboratories, Inc.) was added. Color development was stopped by extensive washing in tap water after 5 minutes. Plates were dried and spots were counted using CTL ImmunoSpot Analyzer (CTL) and evaluated by CTL Professional Double Color Software.

All animal experiments were performed following institutional guidelines for animal experimentation. The murine in vivo HNC model was established as previously described by Hoover and colleagues (23). Briefly, C57BL/6 mice were obtained from The Jackson Laboratory (age 1–2 months). Immortalized, E6/E7 plus H-ras–transduced mouse tonsil epithelial cells (MTEC) were obtained from Hoover (23). After subcutaneous injection of 1 × 10⁶ cells into the neck of the mice, 33 mice with tumor growth were randomized and assigned into different treatment groups. The purpose was to compare the radiation therapy (RT) + anti-PD-1 mAb therapy group (n = 17) to the RT + isotype control (clone MOPC-21, mouse IgG1k, BioXcell) group (n = 16). Anti-PD-1 mAb (clone 4H2, IgG1 isotype) was obtained from Bristol-Myers Squibb (24). Anti-PD-1 therapy (3 mg/kg body weight) was administered on days 1, 4, 11, and 15. Radiotherapy was administered to induce PD-L1 expression (25) in the tumor microenvironment in order to enlighten potential protective effects of anti-PD-1 mAb therapy. In addition, RT increases tumor death, phagocytosis, and antigen presentation. Fractionated radiotherapy was administered for 5 consecutive days at 2 Gy/day during week 1 (D4-D8) and week 2 (D11-D15). Treatment response was analyzed by measurement of the tumor volume at intervals between 3 and 18 days. Additionally, mice spleen were collected and processed as described for lymphocyte isolation. PD-1^high and PD-1^int/low fractions were compared from CD3⁺ CD8⁺ PD-1⁺ cells and analyzed according to different treatment groups. Different clones for treatment (clone 4H2) and staining (BV421 conjugated anti-PD-1 Ab, IgG2a isotype, clone 29F.1A12) were utilized.

Differences in phenotypic counts by HPV status were calculated by the Wilcoxon signed rank test for paired data or the Wilcoxon test for independent groups. PD-1⁺ TILs were divided by flow cytometry into subsets of the proportions of cells classified as PD-1^high, PD-1^int, and PD-1^low. We investigated the impact of the proportions of PD-1^high and PD-1^low in a cohort of 56 HNC patients having curative resection and followed up for disease recurrence. We used Cox proportional hazards models and checked models for linearity and for adequacy of the proportional hazards assumptions. As a heuristic device Kaplan–Meier plots of time to recurrence were prepared by dividing proportions of PD-1^high and PD-1^low cells into tertiles. For the murine experiment, tumor volumes of day 18 were compared. By modeling the fixed and random coefficients of a polynomial regression model, we compared the slopes between the two treatment groups. Data represent observed mean with bootstrap 95% CI.

Approval for the study of specimens from informed and consented HNC patients was obtained by the institutional review board of University of Pittsburgh (#99-069). Study approval for animal experiments was obtained by the University of Pittsburgh (IACUC# IS00005860).

To investigate the quantitative expression of ICR in HNC patients, a cohort of 20 HPV⁺ and 20 HPV⁻ tumors was used for analysis by qRT-PCR. ICR expression was normalized to the average level of pharyngeal mucosa of healthy donors (n = 4, Fig. 1A). ICRs show highly variable expression for HPV⁺ and HPV⁻ tumors in qPCR in relation to reference tissue. In order to confirm and further investigate these results in a larger cohort, ICR expression data from TCGA was retrieved, and mRNA expression of the ICR shown in Fig. 1B–D was analyzed. We found that expression of all ICRs and CD8A, CD8B, and CD4 was higher in HPV⁺ tumors than in HPV⁻ ones. However, more importantly was the finding that PD-1 was the only ICR that was significantly higher in HPV⁺ tumors along with CD8 T-cell markers: CD8A and CD8B (Figs. 1E and F). In addition to comparing expression data from the two cohorts of patients, we correlated the expression of PD-1, CTLA-4, TIM-3, and LAG-3 with that of CD8A, given that this marker of CD8 T cells was highly upregulated in HNC tumors, especially in HPV⁺ samples. We found that all ICRs queried highly correlated with CD8A expression regardless HPV status. However, HPV⁺ tumors showed a higher correlation coefficient than HPV⁻ samples for all ICRs (Fig. 1G). Interestingly, PD-1 expression was highly correlated with that of CD8A in both HPV⁺ and HPV⁻ tumors, with CTLA-4 being the other ICR have a similar correlation coefficient in HPV⁺ tumors (Figs. 1G and H). Because mRNA expression of ICRs was significantly higher in tumors versus control mucosa, and because PD-1 was a highly expressed gene common to both HPV⁺ and HPV⁻ tumors and highly correlated with CD8A expression, we decided to further investigate protein expression of these ICRs by flow cytometry in different subsets of T cells in PBL and TIL from HNC patients.

In order to consider interindividual ICR expression of HPV⁺ and HPV⁻ HNC patients' cellular subsets, we analyzed CD8⁺ PBL for expression of PD-1 and CTLA-4, the best characterized ICR, in respect to traditional phenotypic markers (26). We investigated the proportions of native and memory T cells of HNC patients based on CCR7 (lymphoid tissue homing receptor) and CD45RA (a transmembrane tyrosine phosphatase) expression (Fig. 2A). As shown in Fig. 2B, the frequency of effector memory (T_EM: CCR7⁻CD45RA⁻) was significantly higher than naïve (CCR7⁺CD45RA⁺), central memory (T_CM: CCR7⁺CD45RA⁻), and terminal effectors (T_EMRA: CCR7⁻CD45RA⁺) in both HPV⁺ and HPV⁻ patients (P < 0.001). The frequency of T_EM was comparable between HPV⁺ and HPV⁻ patients. Approximately 20% of T_EM expressed PD-1, which was significantly higher than that expressed by naïve T cells in both HPV⁺ and HPV⁻ patients (Fig. 2C, P < 0.05 and P < 0.01 respectively). In HPV⁻ patients, PD-1⁺ cells in T_CM were significantly higher than in naïve T cells (P < 0.05). PD-1 single-positive (PD-1⁺) T cells from both HPV⁺ and HPV⁻ patients were significantly more frequent than double-positive (PD-1⁺CTLA-4⁺) T cells or CTLA-4 single-positive (CTLA-4⁺) T cells, indicating that PD-1 was more frequently expressed by T cells in peripheral blood than CTLA-4 (P < 0.001, Fig. 2D). However, in peripheral blood, the frequencies of PD-1⁺ T cells between HPV⁺ and HPV⁻ patients were similar. Taken together, these results indicate that PD-1 rather than CTLA-4 is frequently expressed by CD8⁺ effector T cells in both HPV⁺ and HPV⁻ patients. These results led us to investigate TILs by flow cytometry from these patient subgroups.

To evaluate the relevance of ICR in effector cells infiltrating HPV⁺ and HPV⁻ HNC, we compared CTLA-4 and PD-1 expression on CD8⁺ TIL isolated from both HPV⁺ and HPV⁻ patient tumors. As shown in Fig. 3A, the frequencies of CTLA-4⁺ and PD-1⁺ cells were notably higher in CD8⁺ TIL for HPV⁺ and HPV⁻ patients compared with those on CD8⁺ PBL. Most notably, the frequency of PD-1⁺ cells of CD8⁺ TIL was significantly higher in HPV⁺ patients than that in HPV⁻ patients (P < 0.006, Fig. 3B). PD-1 positivity was defined by isotype controls through flow cytometry as shown in Fig. 4A.

Others have suggested that higher numbers of intratumoral PD-1⁺ T cells may correlate with better prognosis in HPV⁺ HNC (12), and the role of HPV in HNC patients' response to anti-PD-1 therapy is controversial. Furthermore, Wherry has shown that the highest level of PD-1 expression reflects T-cell exhaustion status (27). Thus, based on positive PD-1 expression, CD8⁺ TIL were divided into three subgroups, from highest PD-1 expressing cells (PD-1^high), intermediate (PD-1^int) to lowest PD-1 positive cells (PD-1^low; Fig. 4A). Definition of subgroups was confirmed with PD-1⁺ subpopulations of CD3/CD28 bead-stimulated healthy donor CD8⁺ T cells (Supplementary Fig. S1). We previously showed that PD-1 expression level correlates with higher SHP-2 phosphatase, more potent dysfunction, and impaired Th-1 signaling (28). Thus, we investigated the activation status and functional differences in subsets of PD-1-negative, PD-1^low, PD-1^int, and PD-1^high T cells, taken directly from freshly excised HNC tumors. As a marker of activation, granzyme B expression in each of the subsets was analyzed, indicating the highest activation for PD-1^high subsets (Fig. 4B), enriched in HNC patient TIL (29). To analyze functional capabilities of these groups, sorted T-cell populations of negative, moderate (PD-1^low+PD-1^int), or high (PD-1^high) PD-1 expression were compared for their IFN-γ secretion capacity, after TCR stimulation using CD3/CD28 beads. ELISPOT analysis revealed significantly compromised IFN-γ secretion only in PD-1^high CD8⁺ T cells, as compared with robust IFN-γ secretion by PD-1–negative and moderate PD-1^(low/int) T cells (Fig. 4C). HPV⁺ and HPV⁻ patients were compared based on the proportions of PD-1^high and PD-1^low TIL subsets. The fraction of PD-1^high T-cells was greater in HPV⁻ patients (7.3% vs. 2.1% of PD-1^high cells, P = 0.0005 Fig. 4D). In contrast, the proportion of PD-1^low expressing T cells was higher in HPV⁺ patients than in HPV⁻ patients (65.7% vs. 53.0% P = 0.047, Fig. 4E).

Based on the higher frequency of PD-1^high T cells in HPV⁻ patients, we investigated the impact of the extent of PD-1 expression by CD8⁺ T cells on time to disease recurrence. A cohort of HNC patients (n = 56, 36 HPV⁻ and 20 HPV⁺) all of whom had curative therapy for locally advanced disease, were followed for disease recurrence. With a median follow-up 19 months, 40 patients remained free of disease and 16 recurrences were observed, yielding a 2-year probability of recurrence-free survival of 70%, which is typical for our institution (30). A proportional hazards regression analysis revealed that the proportion of PD-1^high CD8⁺ T cells potently increased the risk of disease recurrence in this cohort of HNC patients (hazard ratio = 2.15, 95% CI = 1.46–3.15, P < 0.0001). To validate this finding, we examined the effect of the proportion of PD-1^low cells upon recurrence and found the opposite impact. Indeed, greater fractions of PD-1^low CD8⁺ T cells were protective of recurrence (hazard ratio = 0.19, 95% CI = 0.07–0.49, P = .0006). Figure 5A–D illustrates these findings showing Kaplan–Meier curves arbitrarily divided into tertiles of the proportions of PD-1^high (Fig. 5A) and predicted log hazard ratio (Fig. 5B) and Kaplan–Meier curves arbitrarily divided into tertiles of the proportions of PD-1^low (Fig. 5C) and predicted log hazard ratio (Fig. 5D).

To investigate if PD-1 targeted Ab therapy influences PD-1 T-cell fractions and how therapeutic modalities influence PD-1 expression levels, we used a murine HNC model investigating treatment response to anti-PD-1 mAb or its isotype in combination with radiation therapy (RT). Figure 6A demonstrates the development of tumor volume for the different therapeutic groups. As expected, better treatment response (i.e., lowest tumor volumes at day 30) was observed in the combination therapy with anti-PD-1 mAb. Moreover, anti-PD-1 mAb therapy plus RT significantly reduced the fraction of PD-1^high CD8⁺ T cells and enhanced the frequency of PD-1^low cells (Figs. 6B and C), in association with the best treatment response. This beneficial effect reversed the enrichment of PD-1^high fractions that was observed in the RT + isotype control mAb group. This result demonstrates a valuable effect of anti-PD-1 mAb therapy, by dramatically reducing the poor prognostic subpopulation of PD-1^high TIL cells, and permitting potent antitumor activity to develop via PD-1^low/int, IFN-γ–positive T cells.

In peripheral blood, we observed that PD-1 is mainly expressed on effector T cells rather than by naïve T-cell populations. Despite an overall higher PD-1⁺ CD8⁺ T-cell frequency in the TIL of HPV⁺ HNC patients, HPV⁻ patients had larger fractions of PD-1^high CD8⁺ T cells. We demonstrate that these PD-1^high expressing CTL were highly activated (expressing the most granzyme B), but showed a severely dysfunctional phenotype, with compromised IFN-γ secretion capacity and negative prognostic impact in the clinic. In a murine PD-1 targeting immunotherapy model, we were able to observe that anti-PD-1 therapy overcomes the increase of PD-1^high expressing cells that could be observed in mice treated with RT alone, which drives PD-L1 levels within 24 hours of treatment (25). Furthermore, in contrast to PD-1^high CD8⁺ T cells, frequencies of PD-1^low/int T cells were associated with a better disease-free survival (DFS) in our cohort of HNC patients. These beneficial, activated T cells were significantly increased upon anti-PD-1 mAb therapy in the murine HNC model, correlating with clinical efficacy.

Clinical trials targeting ICR have shown substantial results for multiple cancer subsets including melanoma, renal cell carcinoma, non–small cell lung cancer (31, 32) and recently HNC (7). Nevertheless, the knowledge about ICR in HNC tumors and TIL is limited, but is of interest due to emerging immunotherapeutic targeting of these molecules in several cancer types (33). While cancer immunotherapy targeting ICR has emerged as a major therapeutic advance (34), biomarkers of response are inadequate and have so far focused on ligand expression (i.e., PD-L1) by tumor or immune cells. Here, we propose that response to anti-PD-1 mAb therapy may be influenced by levels of PD-1 on effector T cells in the circulation or the microenvironment. The effect we observe here may reflect salvage and protection of activated functional T cells during tumor response to anti-PD-1 therapy.

Upon ligation, PD-1 directly inhibits TCR-mediated effector functions and thereby causes T-cell dysfunction (35). Despite the immune-inhibitory pathways that are activated upon PD-1 engagement, reports of PD-1 positivity in TIL and its prognostic impact have been controversial to date. PD-1 expression reflects T-cell activation but means only that a T cell is receptive to an immunoinhibitory signal, not that the T cell is necessarily exhausted and dysfunctional. Our results demonstrate that part of this controversy might be attributed to the limitations of grouping PD-1 positive cells despite heterogeneous functional capacity. Indeed, we show that PD-1^high versus PD-1^low T cells possess distinct and important functional differences with significant prognostic importance.

In a recent meta-analysis of 29 studies investigating PD-1 expression and overall survival (OS) in patients with epithelial malignancies, PD-1 expression by TIL was associated with a shorter OS (36). In contrast, Badoual and colleagues reported PD-1-expressing tumor-infiltrating T cells as a favorable prognostic marker in HPV⁺ HNC (12). This conundrum intrigued us due to the potential for the level of PD-1 expression to more accurately mark different TIL subsets with unique behavior. Also, distinct etiologies of HNC such as HPV^+/− disease, with such dramatically different prognoses (37), provided an unparalleled window on this potential biomarker. HPV⁺ HNC represents a unique cancer subset (38), given its location in lymphoepithelial tissue with specific immunological properties (39), virally associated immune evasion properties (40) and yet better clinical prognosis than non–HPV-associated HNC (37). Therefore, we investigated immunological properties like ICR expression of both patient subgroups in antigen-specific cells from peripheral blood and tumor tissue. Our findings in CD8⁺ TIL confirm an overall increased number of PD-1⁺ T cells for HPV⁺ patients. Most importantly, when investigating the extent of PD-1 expression in PD-1⁺ cells, we observed significant differences between HPV⁺ and HPV⁻ patients. HPV⁺ patients had a significant lower fraction of the PD-1^high-expressing cells in comparison with the HPV⁻ patients. In contrast, the PD-1^low fraction was significantly elevated in HPV⁺ patients in comparison with HPV⁻ patients. In combination with the shown functional differences of PD-1^low and PD-1^high CD8⁺ TIL, these results might contribute to a better understanding of prognostic differences between HPV⁺ and HPV⁻ patients, and additionally, account for correlations of PD-1⁺ cells and favorable prognostics. Therefore, our analysis of DFS focused on the impact of fractions of PD-1^high and PD-1^low CD8⁺ TIL.

Here, we show that the patients with higher PD-1^high fractions had a significantly higher risk of recurrence (2.25-fold increased risk as this fraction of cells increased). On the contrary, high PD-1^low fractions were associated with improved DFS. These data support an important role for T-cell exhaustion in progression of disease, with the possibility that rescue of the PD-1^low/int cells represents the mechanism of clinical benefit as well as a more accurate biomarker of clinical response to anti-PD-1/PD-L1 targeting immunotherapy. In the setting of chronic viral infection, Blackburn and colleagues were able to identify a subset of exhausted T cells, which can be rescued through anti-PD-1 targeting therapy (10). The data from our murine HPV⁺ model support these observations in the case of cancer-associated T-cell exhaustion as demonstrated in treatment groups with anti-PD-1 treatment, which showed an increase of PD-1^low-expressing T-cell fractions. Moreover, radiotherapy treatment response was significantly increased through the addition of PD-1 targeting immunotherapy.

Our data show that the effector memory cell subset of HNC patients more highly expresses PD-1. As PD-1 has been shown to negatively regulate T-cell phenotype, proliferation, and instead induces apoptosis (13, 14, 28), the combination of our observations linking PD-1 and peripheral effector memory subsets in vivo supports the association of PD-1 with clonally expanded, tumor-reactive populations (11). In this situation, ligation of PD-1 by its ligands is likely, and even more so for the PD-1^high-expressing populations. This may explain the vulnerability and functional insufficiency of antigen-specific T-cell expansion in the tumor microenvironment. Indeed, Fourcade and colleagues report inverse correlations of expansion and PD-1 levels in a melanoma vaccine study (41). In our study, we observed a protective effect from these negative regulatory pathways in case of anti-PD-1 treatment, as demonstrated by the significant increase of the PD-1^low/int fractions in the murine anti-PD1 therapy model. The antitumor activity of the PD-1^low/int populations is also supported by the improved treatment response in this group. Additionally, analyzed CD8⁺ T cells were obtained from splenocytes, which represent circulating T cells. Therefore, in our point of view, the increase in PD-1^low fractions represents more likely a clonal expansion of PD-1⁺ antitumor effective T-cell populations, perhaps detectable in the circulation of anti-PD-1–treated patients.

By separating different extent of PD-1 expression levels in CD8⁺ TIL, we observed that cells of high PD-1expression levels possess a severely impaired IFN-γ secretion capacity, in contrast to PD-1^low and PD-1^int T cell fractions. This is supported by the fact that in freshly isolated TIL, the PD-1^neg groups secrete lower IFN-γ after stimulation in the ELISPOT assays, with increasing differences of IFN between PD-1^low/int and PD-1^high after 18 hours of stimulation during ELISPOT. These results are in congruence with reports of Flecken and colleagues, who showed reduced IFNγ secretion after expansion of tumor-associated antigen-specific CD8⁺ T-cells (42). This impairment did not resolve after Treg depletion and supports direct involvement of PD-1 ligation and thereby downregulation of functional cytokine production. A study of Zhang and colleagues identified alteration of IFNγ levels depending on PD-1 polymorphisms, guiding to a direct involvement of PD-1 receptor pathways on IFNγ production (43). Additionally, PD-1 blockade has been associated with an increased IFNγ production and elevation of IFNγ–dependent associated chemokines (44). These relationships may contribute to difficulties in separating reasons for—and effects of—both PD-1 and PD-L1 expression. PD-1–positive cells have been associated with high TA specificity. The release of cytokines like IFNγ of these TA-specific cells in the tumor microenvironment has been shown to induce PD-L1 with negative impact on function and cytokine production of PD-1⁺ cells. Therefore, a more sophisticated view of different PD-1 expression levels is needed.

In their analysis of tissue-specific differences in PD-1 expression of T cells, Blackburn and colleagues describe that their ability to degranulate is dependent on the PD-1 expression levels, independent of anatomical location and thereby PD-L1 expression (45). This is supported by our findings of higher granzyme B positivity for PD-1^high-expressing TIL. Therefore, the consideration of different PD-1 expression levels exemplifies the two disparate interpretations of PD-1 expression as a marker of activated, competent tumor reactive T cells, on the one hand, and PD-1^high expression as a marker of exhausted, dysfunctional cells with a negative influence in the tumor microenvironment on the other hand.

Our results are supported by the fact that anti-PD-1 therapy has been reported to be effective in HPV⁻ than HPV⁺ patients (46, 47). Thus, it is possible that higher frequencies of PD-1^high T cells from HPV⁻ patients may contribute to worse prognosis, as well as to a good response to anti-PD-1 immunotherapy. In consequence, we demonstrate that PD-1 expression in CD8⁺ TIL of HNC represents (over-) activation and—at highest expression levels—exhaustion of effector CD8⁺ T cells that is associated with better clinical outcome. Thereby, we conclude two essential consequences for immunotherapy: first, low/intermediate but positive PD-1 expression levels are favorable in terms of tumor-specific functionality and prognosis, and, second, these data support that PD-1 blockade may allow the expansion of TA-specific T cells, keeping their PD-1 expression at lower levels with enhanced antitumor response and reduced susceptibility to PD-L1 ligation, indicating a possible predictive role of the level of PD-1 expression on TIL. Whether this effect is due to active restoration of proliferation of PD-1^high cells or enhanced proliferation and responses of PD-1^low cells has to be investigated. Further studies must validate our finding using anti-PD-1 mAb–treated cancer patients, including HNC. Baseline PD-1^high versus PD-1^low levels should be measured and correlated with responder status.

G.J. Freeman has ownership interest (including patents) in Roche, Merck, Bristol Myers Squibb, Novartis, EMD-Serono, AstraZeneca, Boehringer-Ingelheim, Roche, and Novartis and is a consultant/advisory board member for Roche, Novartis, Eli Lilly, Bristol-Myers Squibb, Seattle Genetics, Bethyl Laboratories, Surface Oncology, and Novartis. R.L. Ferris is a consultant/advisory board member for Astra-Zeneca/MedImmune, Bristol-Myers Squibb, Lilly, Merck, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B.A. Kansy, H.-B. Jie, Y. Lei, R.L. Ferris

Development of methodology: B.A. Kansy, R.M. Srivastava, H.-B. Jie, D.A. Clump, R.L. Ferris

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.A. Kansy, R.M. Srivastava, H.-B. Jie, G. Shayan, Y. Lei, J. Moy, J. Li, N.C. Schmitt, D.A. Clump, R.L. Ferris

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.A. Kansy, F. Concha-Benavente, R.M. Srivastava, H.-B. Jie, G. Shayan, J. Moy, J. Li S. Lang, G.J. Freeman, W.E. Gooding, R.L. Ferris

Writing, review, and/or revision of the manuscript: B.A. Kansy, F. Concha-Benavente, H.-B. Jie, J. Moy, S. Brandau, S. Lang, N.C. Schmitt, G.J. Freeman, W.E. Gooding, D.A. Clump, R.L. Ferris

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Moy, J. Li, R.L. Ferris

Study supervision: R.L. Ferris

We would like to thank Bratislav Janjic and Michael Meyers from the Flow Cytometry Facility for excellent technical assistance.

This work was supported by National Institute of Health grants R01 CA206517, DE019727, P50 CA097190, T32 CA060397 (R. L. Ferris), the University of Pittsburgh Cancer Institute award P30 CA047904 (R. L. Ferris), P50CA101942 (G. J. Freeman), DE024173 (Y. Lei), and the IFORES program of University of Duisburg-Essen (B. A. Kansy).

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.