Pan-cancer transcriptomic analyses have revealed that both HPV+ and HPV- HNSCC are among the most highly infiltrated tumors, and that immune infiltration is higher among HPV+ than HPV- HNSCC.27 Furthermore, based on these transcriptional analyses, HNSCC exhibits the highest median CD56^dim NK cell infiltration of any major tumor type.27 HPV+ HNSCC is more highly infiltrated with NK cells than HPV- HNSCC.28 Increased infiltration of NK cells, specifically CD56^dim, is associated with improved disease-free and overall survival independent of HPV status.27–31 Peripheral NK cell activity also correlates with fewer pharyngeal cancer metastases.30 Compared with circulating NK cells, HNSCC-infiltrating NK cells more frequently express the inhibitory receptor NKG2A and less frequently express the inhibitory KIR (KIR3DL1 and KIR2DL1/2/3).32 33 Although the dominant NK cell subset in HNSCC tumors is mature CD56 dim NK cells, compared with other major tumor types, HNSCC is enriched in immature, cytokine-producing CD56^bright, CD16^dim/negative NK cells lacking expression of CD57.28 32 34 Basal ADCC activity in HNSCC tumors varies considerably.35

NKG2D is an activating transmembrane receptor expressed on NK cells and on a subset of T cells.36 The ligands for NKG2D, including MICA/B, ULBP1/2/3, and Rae-1, are preferentially overexpressed on transformed cells and render them susceptible to NK-mediated tumor elimination.15 Tumors are capable of shedding soluble NKG2D ligands via proteolytic cleavage, which can paradoxically render NK cells less cytotoxic, possibly by non-functional occupation of the NKG2D receptor binding site.37 Levels of soluble NKG2D ligands (MICA/B and ULBP1/2/3) in the plasma of patients with HNSCC are significantly elevated compared with healthy controls, presumably due to shedding from the primary tumor after proteolytic cleavage.34 Primary healthy human NK cells incubated with plasma derived from patients with HNSCC were significantly less efficient at killing target cells in vitro; this effect was reversed by depletion of soluble NKG2D ligands prior to incubation, which suggests that NKG2D ligand shedding may be an important mechanism of HNSCC immune evasion.34

Defects in HLA expression in HNSCC are a common event.38 HLA class I expression is lower in HPV+ than HPV- oropharyngeal HNSCC.39 Interestingly, low HLA class I expression is associated with favorable prognosis in HPV+ HNSCC, but a poor prognosis in HPV- HNSCC.39 40 This may reflect the observation that HPV- HNSCC are relatively less infiltrated with NK cells compared with HPV+ HNSCC,28 41 and therefore may be protected from the consequences of HLA loss.

HPV+ and HPV- HNSCC primary tumors share a similar mutational and copy-number alteration burden; however, their mutational profiles, mechanism of malignant transformation, and clinical behavior are markedly distinct.42 43 HPV-associated tumorigenesis occurs via expression of the viral proteins E6 and E7, which inactivate p53 and retinoblastoma protein (RB), respectively. High-risk HPV infection also leads to upregulation of the cyclin-dependent kinase inhibitor p16, which is widely used as a surrogate marker for HPV positivity in diagnostic testing.44 HPV+ HNSCC most commonly arises in the oropharyngeal tonsillar tissue and is associated with a more favorable prognosis than HPV- HNSCC, even when controlling for tumor stage, smoking status, and other potential confounders.3 Recent large-scale tumor sequencing initiatives such as The Cancer Genome Atlas and MSK-IMPACT revealed that HPV- and HPV+ HNSCC harbor distinct mutational profiles.42 43 Overall, HPV- HNSCC mimicked the mutational landscape of lung and esophageal squamous cell carcinoma, which share tobacco smoke as a common risk factor.42 The tumor suppressor gene TP53 is mutated in the majority of HPV- HNSCC primary, locoregionally advanced, and recurrent/metastatic tumors, and it is independently associated with a poor prognosis.42 43 45 TP53 is rarely mutated in HPV+ HNSCC.45 Even when controlling for HPV and smoking status, TP53 mutations in HNSCC correlated with reduced NK cell infiltration and markers of NK cytolytic activity.46 The authors observed HLA downregulation in specimens with TP53 mutations, which may explain why these tumors were more immunosuppressed, even though they bore a higher mutational—and therefore neoantigen—burden.46

The Epstein-Barr virus belongs to the herpesvirus family and is an etiologic agent of nasopharyngeal carcinoma (NPC). The immune landscape of NPC and associated NK cell characterization is not as well-defined as non-nasopharyngeal HNSCC. NPC are characteristically well-infiltrated, ‘immune-hot’ tumors. Like oropharyngeal carcinoma, NK cell infiltration is correlated with longer overall survival and progression-free survival.31 Pretreatment blood sampling demonstrated NPC tended to have lower expression of the NCRs NKp30 and NKp46 compared with healthy donors.47 It is still unknown, however, how highly NKG2A, HLA-E, and other important ligands or receptors are expressed in NPC.

Over 50% of the US population is latently infected with cytomegalovirus (CMV), and some countries approach near 100% CMV seroprevalence.48 Although infection is not a risk factor for HNSCC, CMV exposure profoundly shapes the repertoire of NK cells (and T cells) within an individual. CMV infection serves as a useful case study of how environmental exposures imprint on NK and T cell diversity, with important consequences for immunotherapy. For instance, CMV infection may lead to stable increases in numbers of adaptive or ‘memory-like’ NK cells that express the activating isoform of NKG2A: NKG2C.49–51 Like NKG2A, NKG2C similarly recognizes HLA-E, but preferentially binds.51 Expansion of NKG2C+ NK cells in response to CMV infection is dependent on sequence variation in the viral UL40-derived peptides, which form a complex with HLA-E.50 This effect can be highly significant—as much as 60% of CD56dim NK cells in some CMV-seropositive individuals are composed of clonally expanded, educated NKG2C+ NK cells.51 NKG2A expression on NK cells mutually exclusive of NKG2C expression. CMV-expanded NKG2C+ NK cell populations have subsequently been found to bear epigenetic changes that alter inhibitory receptor expression, cytokine production and responsiveness, proliferative capacity, and signaling protein expression.52 53

Although immunotherapy has heralded a new era in the treatment of HNSCC, rates of response to treatment have remained low. As a result, research has begun to focus on how combination regimens may synergize to improve responses by unlocking the activity of other effector cells. NK cells are at the nexus of these investigations. Given our understanding of the complex interrelationships between NK cells and other critically important antitumor effector cells, it is worth considering how NK cell function and dysfunction may contribute to the success or failure of immunotherapy in head and neck cancer.

Most previously published studies have provided evidence suggesting Programmed cell death protein 1 (PD-1) is not induced on human NK cells. However, several groups have shown PD-1 expression on NK cells in certain clinical settings,70 71 including head and neck cancer.72 To date, the data are inconclusive, and PD-1 expression on NK cells is outside the scope of this review. This topic has been discussed extensively by Judge et al. 73 Regardless, the observation that depletion of NK cells eliminated the efficacy of PD-1/PD-L1 blockade in tumor-bearing mice suggests that PD-1-based immunotherapy for HNSCC may benefit from proper NK cell function.74 This is possibly by preventing immunosuppressive interactions between NK cells and other PD-1-expressing cells within the tumor microenvironment (TME), such as myeloid-derived suppressor cells (MDSCs).75 However, there are no data suggesting direct interactions between NK cells and PD-1-expressing MDSCs. Strikingly, high levels of pretreatment NK cell gene expression signatures in tumors predicted longer progression-free survival for patients with solid tumors—including HNSCC—after anti-PD-1 therapy, more than any other immune signature, including PD-1 expression and CD8+ T cell signatures.76 It is worth noting, however, that the NK cell gene signature in Prat et al was minimally defined by the expression of three genes (SPN, XCL2, NCR1). Further analyses are required, as these genes are not unique to NK cells. Tumors are capable of responding to IFN-gamma produced by NK cells, which leads to increased PD-L1 expression.77 Intratumoral NK cells are also indirectly responsible for PD-1 responsiveness by producing FLT3L, CCL5, and XCL1, which collectively enrich cDC1 in the TME that themselves closely correlate with PD-1 inhibitor responsiveness.22 23

NKG2A is an immunoreceptor tyrosine-based inhibition motif (ITIM)-bearing receptor that forms a heterodimer with CD94. On binding of NKG2A with its ligand HLA-E, the tyrosine phosphatase SHP-1 is recruited, which ultimately leads to downstream inhibition of NK-mediated cytotoxicity, including ADCC.78 NKG2A is stably expressed on approximately half of circulating and tumor-filtrating NK cells, and it was enriched on HNSCC-infiltrating CD8+ T cells.33 HLA-E expression protects tumor cells from NK-mediated killing.79 80 HLA-E is enriched on a broad range of carcinomas, including HNSCC, and is associated with poor prognosis.81–83 Expression of soluble HLA-E by tumors has been observed in ex vivo studies of human melanoma cells and in cell lines derived from various tumor types, but has not yet been directly described in HNSCC.84

In 2018, André et al reported the results of an ongoing trial investigating a humanized IgG4 anti-NKG2A monoclonal antibody, monalizumab, for R/M-HNSCC. In combination with cetuximab, monalizumab treatment led to an objective response rate (ORR) of 27.5% in patients with HNSCC, per Response Evaluation Criteria in Solid Tumors (RECIST) criteria.33 Although of course direct comparisons cannot be made without a randomized controlled trial, monalizumab-cetuximab compared favorably to single-agent cetuximab, which has a historical ORR of 13%.85 In vitro studies revealed that monalizumab enhanced NK cell and CD8+ T cell effector functions, and this effect synergized with both durvalumab and cetuximab separately.33 Monalizumab monotherapy is ineffective at promoting ADCC in vitro but synergizes with cetuximab to enhance ADCC. This suggests that monalizumab provides the additional benefit of potentiating cetuximab-induced NK-mediated ADCC. To date, no results have been published on monalizumab in combination with PD-(L)1 axis blockade in HNSCC, but this combination is actively being studied in microsatellite-stable colorectal cancer (NCT02671435), and will be included as an expansion cohort in NCT02643550. It remains to be seen how HLA genetics may stratify responders and non-responders, and whether HLA-E expression levels during NK cell education at steady-state predict the magnitude of response to monalizumab. Furthermore, stable expansion of NKG2C+ cells, such as occurs after CMV infection, may modulate the response to monalizumab, as NKG2C:HLA-E binding leads to NK cell activation rather than inhibition.83 Thus, we advocate that future studies of anti-NKG2A treatment substratify patients according to CMV-seropositivity, HLA-E expression and HLA-B genotype.

The observation that allogeneic transfer of NK cells lacking inhibitory KIR for recipient HLA prevents relapse in patients with some hematological malignancies has generated interest in KIR blockade for cancer treatment.86–88 To that end, the anti-KIR2D monoclonal antibody lirilumab (IPH-2101) was developed to limit inhibitory KIR2DL1, KIR2DL2, and KIR2DL3 signaling and improve NK antitumor response.89 Lirilumab is capable of binding to and blocking KIR2DL1, KIR2DL2, and KIR2DL3 expressed on approximately half of peripheral NK cells, blocking their inhibitory interaction with HLA-C, in addition to some off target blockade of KIR2DS1 and KIR2DL2.89 Lirilumab showed considerable efficacy in preclinical mouse studies, where it was capable of clearing leukemic blasts after just a single injection.89 Lirilumab was studied in combination with nivolumab for patients with platinum-refractory R/M-HNSCC (NCT01714739). Early results of this phase I/II study showed an encouraging improvement in objective response rate, although the trial was abandoned in November 2017 due to lack of evidence for benefit provided by the combination regimen. Lirilumab is now being studied in the neoadjuvant setting alongside nivolumab for HNSCC (NCT03341936).

Why have the promising preclinical results of KIR blockade not yet been realized in the clinic? Separate analyses in patients with smoldering multiple myeloma have shown that NK cells derived from patients receiving lirilumab unexpectedly showed NK cell hyporesponsiveness.90 Lirilumab treatment induced significant contraction of KIR2DL-positive NK cells by triggering trogocytosis of lirilumab-bound KIR protein.90 This observation raises special concern for KIR2D+ NK cells that express no other self-reactive inhibitory KIR, as loss of KIR2D signaling may compromise the education of this subset of NK cells and induce functional hyporesponsiveness.

The poliovirus receptor (PVR), or CD155, is the ligand for TIGIT, an inhibitory receptor expressed on NK cells and subsets of T and Treg cells.91 TIGIT signaling prevents NK-mediated cellular lysis and is associated with depressed capacity for cytokine production and degranulation capacity.92 TIGIT competes with the DNAM-1 receptor (CD226), which similarly binds CD155 but with lower affinity than TIGIT and provides an activating rather than inhibitory signal to NK cells93; DNAM-1 is enhanced on educated NK cells.94 Thus, TIGIT and DNAM-1 provide opposing regulatory signals to NK cells, and this opposition has important functional implications for tumor control. In melanoma, NK cells rely on DNAM-1:CD155 signaling to prevent metastasis.95 Antibody blockade of TIGIT in vitro and in mice reduces NK cell exhaustion, inhibits tumor growth, and enhances NK-mediated proinflammatory cytokine production.96 97 TIGIT blockade monotherapy was insufficient to induce tumor regression, but TIGIT/PD-1 co-blockade produced durable antitumor responses and prolonged survival compared with anti-PD-1 monotherapy in tumor-bearing mice.98 Based on these promising preclinical results, several anti-TIGIT mAbs have been developed and now are being investigated in phase I/II trials either as monotherapy or in combination with other checkpoint inhibitors for patients with advanced malignancies (table 1).99 The results of the ongoing phase III SKYSCRAPER trials in the setting of advanced lung cancer may be the first indication of the true clinical efficacy of anti-TIGIT agents (NCT04294810, NCT04256421).

The importance of TIGIT in head and neck cancer is poorly understood. CD155 is highly expressed on HNSCC tumor cells and intratumoral myeloid cells compared with healthy mucosa, and high expression independently associated with worse overall survival.100 Bulk expression analysis of HNSCC tumors shows that, compared with HPV- tumors, HPV+ tumors exhibit significantly higher expression of TIGIT, although it could not be assessed whether increased TIGIT expression was selective to NK cells, T cells, or both.41 Notably, in HPV+ HNSCC, high bulk TIGIT expression correlated with improved survival, whereas in HPV- HNSCC, TIGIT expression did not significantly correlate with either improved or worsened survival.41 TIGIT blockade delayed tumor growth and prevented T cell exhaustion in a mouse model of HNSCC.100 Taken together, these findings warrant further investigation into NK cell PVR-receptor expression dynamics and TIGIT blockade for HNSCC.

Several early-phase clinical trials are now underway investigating the use of T cell immunoglobulin mucin-3 (TIM3) checkpoint inhibition for the treatment of advanced solid malignancies, including HNSCC (NCT02608268, NCT03744468). Although the clinical efficacy of these agents has yet to be determined, there is promising evidence that TIM3 plays a role in the progression of head and neck cancer. TIM3 is an inhibitory receptor that is upregulated on the NK cell surface after activation via the Fc receptor.101 102 The major ligand for TIM3 is Galectin-9. Although the precise functional status of TIM3+ NK cells remains controversial,102 previous studies in melanoma have shown that TIM3 inhibition can reverse NK cell exhaustion.103 Bulk expression analyses have shown that TIM3 is upregulated in HNSCC tumors compared with healthy or dysplastic mucosa and correlated with increased likelihood of a lymph node metastasis and disease recurrence.72 104 Further, anti-TIM3 suppressed tumor growth in a mouse model of HNSCC.104

Similarly, several large trials have now begun using anti-lymphocyte activating gene 3 (LAG3) monoclonal antibodies for the treatment of advanced solid tumors (NCT01968109, NCT03311412). LAG-3 is an inhibitory receptor expressed on the surface on some NK cells and binds to the major histocompatibility complex II (MHCII). LAG-3 is highly expressed on tumor-infiltrating lymphocytes (TILs) in HNSCC—especially in recurrent or metastatic disease—and correlates with worse overall survival.105 LAG-3 blockade in a mouse model of HNSCC restricted tumor growth.105 Still, the extent to which the beneficial effects of LAG-3 blockade can be attributed to NK cells is unknown, as LAG-3 expression is also observed on adaptive lymphocytes.