Identification of the Cell-Intrinsic and -Extrinsic Pathways Downstream of EGFR and IFNγ That Induce PD-L1 Expression in Head and Neck Cancer

Cancer immunoediting implies that lymphocytes successfully suppress tumor growth (1). However, tumor cell immune escape can eventually occur by downregulating HLA class I antigen processing (2) or by providing inhibitory signals (3, 4). Programmed death-1 (PD-1) is an immune checkpoint receptor expressed by tumor infiltrating lymphocytes (TIL), which limits the function of activated T lymphocytes (3, 5). Its cognate ligand, programmed death ligand-1 (PD-L1) is expressed in many types of cancers (5). Indeed, trials targeting the PD-L1/PD-1 pathway with blocking antibodies show encouraging results where tumor PD-L1 expression enriches for clinical responders (6–10). The incidence of HPV⁺ tumors is rapidly increasing (11) and these tumors are more responsive to oncologic therapy, which may be in part immune mediated (12–14). A previous report suggested that PD-L1 expression contributed to immune resistance in HPV⁺ HNC (15). In contrast, PD-1⁺ CD8⁺ T cells with an activated phenotype may be a favorable prognostic biomarker in HPV⁺ patients (16). Therefore, the stimuli and signaling pathways that induce PD-L1 expression in HNC may permit more effective therapeutic approaches.

PD-L1 expression in tumor cells may be regulated by two major mechanisms. First, an “extrinsic” mechanism where an antitumor cellular immune response driven by natural killer cells (NK) and CD8⁺ TIL produce IFNγ, which in turn may induce PD-L1 expression on tumor cells. Second, an “intrinsic” mechanism may exist in which constitutive oncogenic signaling pathways within the tumor cell lead to PD-L1 overexpression. In glioblastoma, PTEN deletion promotes PI3K–AKT-mediated PD-L1 overexpression (17), while EGFR-mutant lung cancer cells have been associated with PD-L1 overexpression (18, 19). In contrast to lung cancer, in the setting of HNC, EGFR mutations are extremely rare, whereas wild-type EGFR is overexpressed in approximately 80% to 90% of tumors (20). We hypothesized that targeting signaling molecules involved in PD-L1 expression in HNC cells might synergize with current anti-EGFR antibody targeted immunotherapies such as cetuximab, that are known to activate NK cells and cytotoxic T lymphocytes (CTL; ref. 21). However, because activated NK and T cells secrete IFNγ, a known stimulus for PD-L1 expression, understanding the complex signaling pathways regulating PD-L1 is crucial. Because extrinsic (IFNγ-mediated) and intrinsic (EGFR-mediated) mechanisms may cooperate to promote PD-L1 upregulation, we investigated signaling pathways that mediate both IFNγ and EGFR induced PD-L1 upregulation in tumor cells. These findings have relevance for immunotherapeutic combinations with cetuximab, which can both block EGFR signaling and stimulate IFNγ secretion via activation of NK cells and CTL (21–23).

HPV⁻ HNC cell lines (JHU020, JHU022, JHU029, PCI13) and HPV⁺ (SCC2, SCC47, SCC90, 93VU) were cultured in IMDM complete media (Mediatech). JHU020, JHU022, and JHU029 were a gift from Dr. James Rocco in January of 2006 (Ohio State University). SCC90 and PCI13 were isolated from patients treated at the University of Pittsburgh through the explant/culture method, authenticated and validated using STTR profiling and HLA genotyping (24, 25). SCC2 and SCC47 were a kind gift from Dr. Thomas Carey in December of 2005 (University of Michigan) and 93VU a gift from Dr. Henning Bier in October of 2013 (Technische Universitat Munchen, Germany). HNC lines were tested every 6 months and were free of Mycoplasma infection.

The anti-PD-L1 mAb (clone 405.9A11) was previously validated (26) by Dr. Gordon J. Freeman (Dana-Farber Cancer Institute, Boston, MA), anti-pJAK2 (Y1007 and Y1008; Abcam) were used for IHC staining. The anti-PD-L1-PE (BD Pharmingen) anti-HLA-ABC-FITC mAb (clone w6/32, eBioscience), IgG1-PE isotype control, pSTAT1 Tyr701-PE, STAT1-PE and STAT3-PE (BD Biosciences), phospho-AKT (Thre308)–PE and primary anti-p44/42 MAPK(Erk1/2) and phospho-p44/42 MAPK(Erk1/2) (Thr202/Tyr204) and secondary anti-rabbit-PE antibodies (Cell Signaling Technology). WB antibodies included rabbit anti-human JAK2, pJAK2 (y1007/1008), total AKT, pAKT, pERK and mouse anti-human β-actin (Cell Signaling Technology). IFNγ (R&D Systems) was used at 10 IU/mL. IFNα2a (PBL Interferon Source) was used at 1,000 IU/mL. The JAK2 inhibitor BMS-911543 was characterized previously (25), provided by Bristol-Myers Squibb and used at 10 μmol/L. JAK1/3 inhibitor (ZM39923; Tocris Bioscience) was characterized previously (27, 28) and was used at 10 μmol/L. Wortmannin (Cell Signaling Technology) was used at 1 μmol/L. PI3Kα110 subunit inhibitor (BYL-719) was used at 1 μmol/L and MEK1/2 inhibitor (PD0325901) was used at 5 μmol/L (Tocris Bioscience).

Cell viability was determined by Zombie Aqua staining (Biolegend; ref. 29), then incubated with fluorophore-conjugated antibodies at 1:10 dilution for 15 minutes at 4°C, then washed twice and resuspended in 2% PFA solution until analysis. Intracellular flow cytometry was performed (30) and analyzed using an LSR Fortessa cytometer (BD Biosciences), and FlowJo version 10 software. Median fluorescence intensity (MFI) fold change was calculated by normalizing after subtracting the isotype control MFI.

Western blotting was performed as described previously (2) using the indicated antibodies.

Cells were transfected as described elsewhere (2). After 48 hours of transfection, cells were incubated ±IFNγ (10 IU/mL) or EGF (10 ng/mL) for 48 hours, then harvested and analyzed by flow cytometry. siRNA STAT1: 5-CUACGAACAUGACCCUAUTT-3(s) and 5-AUAGGGUCAUGUUCGUAGGTG-3(as) siRNA STAT3: 5-GCCUCAAGAUUGACCUAGATT-3(s) and 5-UCUAGGUCAAUCUUGAGGCCT-3(as) and siRNA nontargeting 5-AGUACAGCAAACGAUACG Gtt-3 control: (s) and 5-CCGUAUCGUUUGCUGUACUtt-3(as).

qPCR was performed as described previously (31). PCR probes for PD-L1 (Hs01125301_m1) and GUSB (Hs99999908_m1; Applied Biosystems for TaqMan Gene Expression Assay). GUSB was amplified as an internal control. Relative expression of the target genes to GUSB control gene was calculated using the ΔC_T method: relative expression = 2^−ΔC_T, where ΔC_T = C_T (target gene) − C_T (GUSB).

Cells were serum starved for 18 hours prior to incubation with IFNγ (10 IU/mL) for 30 minutes at 37°C or sequentially with cetuximab (10 μg/mL) for 30 minutes at 37°C. Then chromatin immunoprecipitation (ChIP) assay was performed as described previously (2) using the Ez-ChIP kit (Millipore). Purified DNA was used in quantitative RT-PCR using the EpiTect ChIP qPCR (Qiagen) SYBR-Green Master Mix method using a primer for PD-L1 promoter NM_014143.2 (−)16Kb. qPCR amplification data were normalized and analyzed as a percentage of input (32) and expressed as relative enrichment to percentage of input.

The University of Pittsburgh IRB #99-069 approved the use of clinical samples and written informed consent was obtained. Slides were deparaffinized and rehydrated. Antigen retrieval was performed using Diva Retrieval (Biocare Medical) and a decloaking chamber at 124°C, 3 minutes, and cooled for 10 minutes. Slides were placed on an Autostainer Plus (Dako) using a TBST rinse buffer (Dako) and stained using: 3% H2O2 (ThermoFisher Scientific) for 5 minutes, CAS Block (Invitrogen) for 10 minutes, the primary antibody for PD-L1 (clone 405.9A11; ref. 33) and the pJAK2 (Y1007-1008; clone E132) were used per instructions. The secondary consisted of Envision Dual Link + (Dako) polymer for 30 minutes, rinsed, then a TBST holding rinse was applied for 5 minutes. The substrate used was 3,3, Diaminobenzidine + (Dako) for 7 minutes and counterstained with hematoxylin. PD-L1 and pJAK2 staining were quantified by positive pixel count v9 algorithm (Aperio). A head and neck pathologist blinded to clinical patient data examined tumor sections. Scoring was determined by the percentage of tumor stained for PD-L1 or pJAK2, respectively. Tumors with <5% tumor cell–positive staining were considered negative.

Cytotoxicity was determined by a ⁵¹Cr release assay as described before (34). Tumor targets and NK cells were coincubated with cetuximab or IgG1 isotype (10 μg/mL) for 4 hours. Controls for spontaneous and maximal lysis were included. Specific lysis = (experimental lysis − spontaneous lysis)/(experimental lysis − maximal lysis) × 100.

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 (35). The RSEM units to quantitate RNAseq expression data were described previously (36). 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.

Software was accessed via the University of Pittsburgh license. Path Explorer tool available in IPA Ingenuity was utilized for associating EGFR and IFNγ pathways with PD-L1 expression.

IHC staining of tumor specimens (n = 134) revealed that 59.7% of HNC patients express detectable PD-L1 on the tumor cell surface, as determined by >5% positive tumor cells (Fig. 1A). Furthermore, when segregated by HPV status (n = 127, 63 HPV⁻ and 64 HPV⁺), we noted that PD-L1 expression was more frequent in HPV⁺ specimens (70% vs. 43.3%, respectively, Fig. 1B) and the percentage of PD-L1 expression was also significantly higher in in HPV⁺ tumors (Fig. 1C). Interestingly, PD-L1 expression was more intense on the cell membrane than in the cytoplasm and was heterogeneously expressed within the microenvironment, generally forming clusters of PD-L1⁺ tumor cells with a higher intensity at the cluster periphery (Fig. 1D and E). To study the stimuli and pathways by which PD-L1 is upregulated in vitro, we analyzed a panel of HPV⁺ and HPV⁻ HNC cell lines for PD-L1 expression, which was expressed variably (Fig. 1F) similar to patient tumors by IHC.

We then analyzed PD-L1 expression in a large cohort of HNC specimens for which gene expression TCGA repository data were available (35). Because PD-L1 expression has been linked with that of CD8 and IFNγ (15, 23), we investigated the Th1 mRNA expression profile of HPV⁺ versus HPV⁻ specimens. Pooled data from 88 HNC specimens were plotted using a heat map, segregated by HPV status (Fig. 2A, red boxes depict higher expression in HPV⁺ tumors). A Th1-type expression profile (PD-1, CD8A, CD8B, IFNG, and JAK2) was significantly higher in HPV⁺ than HPV⁻ tumors (Fig. 2B), suggesting that activated immune effector cells readily infiltrate HPV⁺ tumors, which may be important for PD-L1 induction due to this source of IFNγ. Importantly, JAK2 expression (but not JAK1) was also higher in HPV⁺ tumors (Fig. 2B). Therefore, JAK2 was associated with a Th1 profile and with PD-L1 expression, particularly in HPV⁺ tumors.

Given that JAK2 is a common signaling molecule downstream of the EGFR and IFNγ pathways, we found that PD-L1 and JAK2 mRNA expression were highly correlated (n = 500) and persisted when the cohort was segregated by HPV status (Fig. 2C). In order to further assess the relationship between JAK2 and PD-L1 in vivo at the protein level, we determined phospho-JAK2 and PD-L1 using IHC from adjacent sections of HNC specimens (n = 23). Corroborating our previous findings, PD-L1 was predominantly expressed on the tumor cell membrane while phospho-JAK2 exhibited strong nuclear staining, with occasional weak-moderate cytoplasmic staining. PD-L1–positive tumor islands were found to be strongly positive for phospho-JAK2 (Fig. 2D). Furthermore, we also found a significant correlation between EGFR and PD-L1 expression, which was somewhat weaker in HPV⁻ tumors (Fig. 2E). Likewise, PD-L1 expression was highly correlated with a Th1-type expression profile (IFNγ, CD8A, and PD-1) regardless of HPV status (Fig. 2F and Table 1). In addition, correlation of EGFR and CD8 or JAK2 was only significant in HPV⁺ tumors, given that their expression level was higher than in HPV⁻ tumors. However, this finding did not preclude the fact that JAK2 could also be important for PD-L1 expression in HPV⁻ tumors given that they were strongly correlated regardless of HPV status.

Because PD-L1 expression strongly correlated with a Th1-type expression profile in the tumor microenvironment, we hypothesized that PD-L1 may depend on STAT1 activation, a known Th1-type transcription factor. Indeed, STAT1 emerged as one of the highly predicted transcription factors binding to the PD-L1 promoter region and common to EGFR and IFNγ pathways when utilizing previously validated software for transcription factor binding prediction (MATCH) and pathway exploration (Ingenuity IPA; ref. 37). Given that previous reports presented STAT3, PI3K, and MAPK as possibly involved in PD-L1 expression in other tumor types, we included these in our investigation. We pooled RNAseq data collected from 46 paired specimens of tumor versus matched normal mucosa and found that STAT1 (but not STAT3, PIK3CA, or MAPK1) was significantly upregulated in tumor tissue (Fig. 3A). Furthermore, STAT1 expression highly correlated with that of PD-L1 in TCGA (n = 500), which was preserved when segregated by HPV status (Fig. 3B). Concordant with TCGA data, we found that STAT1 protein was widely expressed in HNC tumor tissues, and that PD-L1–positive tumor islands were also strongly positive for total STAT1 staining (Fig. 3C, circled areas highlight colocalization, 100× inset). Interestingly, STAT1 expression also showed strong correlation with that of EGFR (Fig. 3D). As expected, STAT1 tumor expression was also strongly correlated with that of IFNγ (Fig. 3E). Notably, STAT3 and PI3K pathway components (AKT1, TORC1, or 4EBP1) showed no correlation with PD-L1 in HPV⁺ tumors and only weakly in the HPV⁻ HNC. Likewise, MAPK1 was not correlated with PD-L1 expression in either cohort (Table 2). Overall, our findings suggest that the JAK2/STAT1 pathway may serve as an important common mediator for both EGFR- and IFNγ-mediated PD-L1 expression in HNC tumors, regardless of HPV status.

Based on our TCGA analysis and previous reports linking IFNγ with PD-L1 expression at the mRNA level, we investigated the signaling pathway by which IFNγ upregulates PD-L1 expression in vitro. Indeed, a panel of HPV⁺ and HPV⁻ HNC cell lines upregulated PD-L1 expression after IFNγ treatment (Fig. 4A). Given that IFNγ-mediated PD-L1 upregulation was linked with PI3K pathway activation (17), we tested whether wortmannin (pan-PI3K inhibitor) or BYL-719 (PI3Kα110 subunit specific inhibitor) could prevent IFNγ-mediated PD-L1 upregulation. PI3K pathway inhibition did not induce PD-L1 downregulation, under conditions in which these inhibitors effectively prevented AKT phosphorylation (Supplementary Fig. S1A–S1D). Because IFNγ signals via JAK1 and JAK2, we utilized a clinical grade, selective JAK2 inhibitor BMS-911345 (JAK2i), which was previously characterized (38), finding an abrogation of IFNγ-mediated PD-L1 upregulation in all cell lines tested, both at the mRNA and protein level (Fig. 4A and B; Supplementary Fig. S1E and S1F). Interestingly, specific JAK1/3 inhibition (JAK1/3i) did not show a significant downregulation of IFNγ-mediated PD-L1 protein expression (Fig. 4C). We then used IFNα, which signals via JAK1 and TYK2, but not JAK2. IFNα treatment did not upregulate PD-L1 expression (Fig. 4D) but still induced pSTAT1 upregulation, although to a lesser extent than IFNγ, in all cell lines tested. Moreover, JAK2 inhibition did not affect HLA-ABC upregulation (Supplementary Fig. S2A and S2B), which suggests that the kinetics of IFNα-induced pSTAT1 binding to the PD-L1 promoter differ for HLA-ABC.

In order to determine whether the IFNγ-mediated PD-L1 upregulation was solely STAT1 dependent, we silenced each transcription factor using siRNA technology (80%–90% knockdown efficiency; Supplementary Fig. S3A). STAT1 but not STAT3 knockdown potently impaired IFNγ-mediated upregulation of PD-L1 (Fig. 4E). Moreover, ChIP assays documented that pSTAT1 but not pSTAT3 binds to the promoter region of PD-L1 after IFNγ treatment (Fig. 4F). Interestingly, cetuximab-mediated EGFR blockade downregulated IFNγ-induced pSTAT1 binding to the PD-L1 promoter and significantly downregulated the IFNγ-mediated PD-L1 upregulation at the mRNA and protein level, respectively (Fig. 4G and H and Supplementary Fig. S3B).

Because EGFR strongly correlated with PD-L1 expression in TCGA specimens and a previous report showed that EGFR activating mutations induce PD-L1 in lung cancers (18), we hypothesized that wild type EGFR, overexpressed in 80% to 90% of HNC, may promote PD-L1 upregulation. EGF treatment induced upregulation of PD-L1 protein in 7 of 8 HNC lines studied, though to a lesser extent than that induced by IFNγ (Fig. 5A). This effect was also seen at the mRNA level (Fig. 5B). Although EGFR activates multiple downstream pathways, including the PI3K, MAPK, and JAK/STAT pathway, TCGA analysis yielded weak if any correlation between PD-L1 and PIK3CA or MAPK1 (Table 2). However, a strong correlation was observed with that of JAK2 and STAT1 (Figs. 2C and 3B, respectively). Given that JAK2 serves as a common signaling molecule for both IFNγ and EGFR pathways, we investigated whether EGF-mediated PD-L1 upregulation was JAK2 and/or STAT1 dependent. Indeed, basal expression of PD-L1 in HNC cell lines was downregulated by JAK2 but not JAK1/3 inhibition (Fig. 5C). Furthermore, EGF induced JAK2 phosphorylation (Supplementary Fig. S4A) and upregulation of basal PD-L1 expression (Fig. 5D). Additionally, specific JAK2, but not JAK1/3, inhibition prevented EGF induced PD-L1 upregulation (Fig. 5D and Supplementary Fig. S4B). Interestingly, the EGF-mediated PD-L1 upregulation was higher in cell lines with a higher EGFR expression (JHU029 and JHU022 vs. 93VU and SCC90). Likewise, JAK2 inhibition more strongly downregulated basal and EGF-mediated PD-L1 expression in the EGFR^high cell lines (Fig. 5D; JHU022 and JHU029).

Because EGFR activates PI3K and MAPK pathways, we tested whether these mediated PD-L1 upregulation after EGFR stimulation. We found that neither wortmannin-mediated PI3K inhibition nor MEK1/2-mediated MAPK inhibition prevented EGF-induced PD-L1 upregulation (Supplementary Fig. S4C and S4D). However, these inhibitors effectively suppressed AKT and ERK phosphorylation, respectively (Supplementary Fig. S4E and S4F). In light of this result and the positive correlation found between EGFR and STAT1, we hypothesized that EGF may be activating STAT1 phosphorylation, mediated by JAK2. Indeed, EGF induced STAT1 (tyrosine701) phosphorylation reaching its maximum peak at 24 hours, while total STAT1 levels remained stable (Fig. 5E). Furthermore, siRNA-targeted STAT1 knockdown efficiently suppressed total STAT1 levels, (Supplementary Fig. S5) as well as significantly abrogating EGF-induced PD-L1 upregulation (Fig. 5F).

Because JAK2 represents a key player in PD-L1 upregulation in both EGFR (intrinsic) and IFNγ (extrinsic) pathways in vitro, we tested whether JAK2 inhibition enhanced NK-mediated killing via antibody-dependent cell cytotoxicity (ADCC; ref. 21) against PD-L1⁺ HNC cells. When NK cells were cocultured with HNC targets and cetuximab, activated NK cells upregulated tumor PD-L1 expression in an IFNγ-dependent fashion (Fig. 6A, open bars). As a control, the EGFR specific mAb panitumumab (IgG2 isotype), which does not bind to CD16 on NK cells, did not induce PD-L1 upregulation, most likely because of a lack of NK cell activation and IFNγ secretion (39). Importantly, the IFNγ-mediated PD-L1 upregulation on HNC cells was prevented when they were pretreated with the JAK2 inhibitor (left panel, closed bars), but not with a JAK1/3 specific inhibitor (right panel, closed bars). We therefore tested the hypothesis that NK cells would more efficiently lyse JAK2 inhibitor pretreated tumor cells, in the setting of reduced PD-L1 expression. Indeed, NK cells showed approximately 25% higher specific lysis of HNC cells pretreated with the JAK2 inhibitor (Fig. 6B). Overall, these results confirm that JAK2 is an important regulator of PD-L1 expression in HNC tumor cells, and its inhibition reverses PD-L1–mediated tumor cell escape from cetuximab-mediated ADCC.

Given the known importance of HPV infection in the etiology of HNC, several studies have sought to correlate PD-L1 expression with HPV status. Recent reports have shown higher PD-L1 expression in HPV⁺ tumors (15, 16, 40, 41). Lyford-Pike and colleagues (15) reported only 29% PD-L1 positivity among a small cohort (n = 9) of HPV⁻ HNC patients, while Malm and colleagues reported 80% (41), making the association of PD-L1 expression with HPV status controversial. In our large series of 134 patients, the majority (approximately 60%) was PD-L1⁺. Most importantly, we found HPV⁺ tumors to be more frequently PD-L1 positive (70%, n = 64) and have a significant higher percent area and intensity of PD-L1. Importantly, a previous study showed that PD-L1 colocalized with CD3 in 56% of tumors while 44% showed a diffuse pattern with no colocalization noted (41), raising the question of how PD-L1 is regulated in those tumors. Thus, PD-L1 expression could be “extrinsically” induced by IFNγ secreting TILs (where colocalization was found), particularly in HPV⁺ tumors and “intrinsically” induced via endogenous EGFR signaling (where no colocalization was found), particularly in HPV⁻ tumors.

Because HPV⁺ tumors showed higher PD-L1 protein expression in vivo and to extend findings reported previously, we took advantage of the RNAseq expression data available in TCGA (n = 500). HPV⁺ tumors have significantly higher expression of a Th1-type profile including CD8A, PD-1, IFNG, and JAK2 (Fig. 2A and B). Our findings are concordant with those of a previous report where HPV⁺ HNC tumors showed more PD-1⁺ CD8⁺ T-cell infiltration, which correlated with favorable clinical outcome (16). These data suggest that PD-1 expressing cells are biologically relevant and may play a crucial role in HPV⁺ disease and PD-L1 induction. Importantly, we report that PD-L1 expression highly correlated with that of JAK2 at the mRNA (TCGA) and protein level in vivo (IHC). Additionally, we found that pJAK2 staining was significantly higher in HPV⁺ than HPV⁻ specimens (data not shown). Interestingly, PD-L1 was also strongly correlated with a Th1-type profile regardless of HPV status, suggesting that the PD-L1/PD-1 axis represents an important mechanism of immune evasion in both HPV-negative and -positive tumors, such that HPV⁻ tumors may rely more on a tumor-intrinsic (EGFR-driven) PD-L1 expression, while HPV⁺ tumors rely more on a tumor extrinsic IFNγ-mediated Th1-like response.

STAT1 was upregulated in HNC tumors when compared with paired autologous normal mucosa and that was significantly correlated with PD-L1 expression regardless of HPV status. Notably, components of other signaling pathways such as PI3K and MAPK, which have been previously associated with PD-L1 expression in other types of cancer (17, 28), did not show significant correlation with PD-L1 expression in HPV⁺ tumors or induce PD-L1. Indeed, the unique biology, mutational landscape, and predominant signaling pathways in HNC may explain the discrepancy in PD-L1 expression. It has been recently reported that PTEN loss-of-function mutations are frequent in glioblastoma (31.9% of specimens, TCGA data; ref. 42). Furthermore, PD-L1 was upregulated after PTEN loss/PI3K activation in glioblastoma lines, suggesting that gliomas may rely more on this signaling pathway (17). Likewise, in the setting of non–small cell lung cancer (NSCLC), PD-L1 protein was upregulated after EGFR/RAS/MAPK pathway–activating mutations. Indeed, HRAS and EGFR mutations in NSCLC are far more frequent than in HNC (2%–5%; refs. 43, 44). Therefore, mutant EGFR may induce a stronger MAPK pathway activation than wild-type EGFR. On the other hand, PTEN or PIK3CA mutations are rather infrequent in HNC, 7% and 8%, respectively (44). Hence, in the setting of HNC, the intrinsic oncogenic signaling mostly depends on overexpressed wild-type EGFR stimulation and presents as a unique feature of this type of cancer, in which the JAK/STAT3 oncogenic pathway is best characterized (45). Importantly, in our series, STAT3 showed no significant correlation with PD-L1 expression in HPV⁺ tumors and only a weak correlation in HPV⁻ tumors, most likely because of higher EGFR expression in these tumors versus HPV⁺ ones.

Concordant with other types of cancer, IFNγ induced PD-L1 upregulation in all of the HNC cell lines tested in our study; however, its upregulation was not PI3K dependent as reported for glioma, lymphoma, or lung cancer (28, 46, 47). We believe this is the first report in which specific JAK2 inhibition completely abrogates IFNγ-mediated PD-L1 upregulation at the mRNA and protein level. Interestingly, IFNα, which does not signal via JAK2, did not upregulate PD-L1 expression, confirming the specific role of JAK2 upregulating PD-L1. Likewise, we should emphasize that IFNα not only induces STAT1 phosphorylation but also STAT2, and complexes with IRF9 in the transcription factor assembly cascade, forming the ISGF3 transcription complex, where IRF9 is the main DNA binding domain (48). In contrast, IFNγ mainly induces the formation of pSTAT1 dimers that directly bind to the promoter region of the target gene, which explains why IFNα may not upregulate PD-L1, although it still upregulates pSTAT1. In addition, and corroborating our TCGA findings, in vitro knockdown experiments show that STAT1 but not STAT3 mediates IFNγ-induced PD-L1 upregulation, further supported by ChIP assay, providing additional evidence that pSTAT1 but not pSTAT3 binds to the PD-L1 promoter region as early as 30 minutes after IFNγ treatment. Our findings complement that of a previous report performed in lung cancer that shows IRF-1 binding to the PD-L1 promoter after IFNγ treatment (49). Interestingly, we are the first to report that cetuximab-mediated EGFR blockade significantly downregulates IFNγ-induced PD-L1 expression, suggesting cross-talk between the IFNγ and EGFR pathways in regulating PD-L1 expression mediated through STAT1 modulation.

EGFR and PD-L1 showed a higher correlation in HPV⁺ tumors that corresponds to the higher correlation seen with CD8A, JAK2, and STAT1 as well. These otherwise counterintuitive results may be explained by the fact that PD-L1 expression might be more dependent on the strength of EGFR/JAK2 pathway activation than EGFR higher expression in HPV⁻ tumors, which may induce an increased STAT1 activation and induction of PD-L1 expression. Alternatively, other immune cells infiltrating the tumor microenvironment may also express PD-L1, such as dendritic cells, macrophages, monocytes, B cells, as well as tumor-associated fibroblasts and stromal cells (26, 50). PD-L1 expression on these cells may confound the strength of correlation with that of EGFR, given that the expression of the latter is mostly on tumor cells, given that the TCGA RNAseq values represent unfractionated tumor.

Thus, JAK2/STAT1 signaling is a major common regulator for PD-L1 transcription driven by IFNγ and EGFR pathways. Because EGFR mutations are very rare in HNC (2%; ref. 43), EGFR pathway overactivation, rather than activating mutations, is more important for PD-L1 upregulation in HNC. We are the first to report that that wild-type EGFR pathway induces PD-L1 upregulation at the mRNA and protein level, and that specific JAK2 inhibition significantly downregulated baseline and EGF-induced PD-L1 upregulation, though not completely, suggesting that other alternative pathways also contribute to PD-L1 expression in HNC. Noteworthy, JAK2 inhibition more effectively downregulated basal and EGF-mediated PD-L1 expression in EGFR^high-expressing cell lines (JHU029 and JHU022). In addition, we are the first to report that EGFR stimulation induces phosphorylation of STAT1, which in turn mediates PD-L1 upregulation, because its silencing completely abrogated this effect, EGFR and JAK2 inhibition may synergize downregulating the “intrinsic” PD-L1 expression. Most important, however, is the speculation of potential added benefit of JAK2 inhibition with the simultaneous blockade of the “extrinsic” IFNγ-mediated PD-L1 upregulation, which seems to be more important in HPV⁺ tumors. An added benefit of using combined JAK2 inhibition and cetuximab-mediated EGFR blockade would be that the JAK2-mediated downregulation of PD-L1 expression on tumor cells would ultimately enhance the effector properties of PD-1⁺ NK cells activated in the tumor microenvironment. Indeed, we show that JAK2 inhibition in tumor targets enhances cetuximab-mediated ADCC to a significant extent, therefore reversing PD-L1-mediated immunoescape of tumor cells to NK cell killing. Moreover, JAK2 inhibition might synergize with mAbs targeting PD-1 and/or CTLA-4 by enhancing ADCC of PD-1⁺CTLA4⁺ lymphoid or myeloid suppressor immune infiltrates in the tumor microenvironment as well as reducing PD-L1 expression.

G.J. Freeman has ownership interest (including patents) in Merck, Bristol-Myers Squibb, Roche, EMD Serono, Amplimmune, Boehringer Ingelheim, and Novartis and is a consultant/advisory board member for Novartis, Roche, Bristol-Myers Squibb, Eli Lilly, and Surface Oncology. R.L Ferris reports receiving commercial research grant from Bristol-Myers Squibb and AZ/Meddimune and is a consultant/advisory board member for Merck, Celgene, Bristol-Myers Squibb, and AZ/Meddimune. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Concha-Benavente, R.L. Ferris

Development of methodology: F. Concha-Benavente, R.M. Srivastava, G.J. Freeman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Concha-Benavente, S. Trivedi, Y. Lei

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Concha-Benavente, R.M. Srivastava, Y. Lei, U. Chandran, R.R. Seethala, R.L. Ferris

Writing, review, and/or revision of the manuscript: F. Concha-Benavente, R.M. Srivastava, S. Trivedi, R.L. Ferris

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Concha-Benavente, G.J. Freeman, R.L. Ferris

Study supervision: R.L. Ferris

This study was supported by National Institute of Health grants 1R01 CA 206517-01, R01 DE019727, P50 CA097190, CA110249 and University of Pittsburgh Cancer Center Support Grant P30CA047904, P50CA101942 (G.J. Freeman), and DE024173 (Yu Lei).

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.