Pembrolizumab and cabozantinib in recurrent metastatic head and neck squamous cell carcinoma: a phase 2 trial

Abstract

Anti-programmed cell death protein 1 (PD-1) therapy is a standard of care in recurrent metastatic head and neck squamous cell carcinoma (RMHNSCC). Vascular endothelial growth factor inhibitors, including tyrosine kinase inhibitors, have immunomodulatory properties and have offered promising results when combined with anti-PD-1 agents. We conducted a phase 2, multicenter, single-arm trial of pembrolizumab and cabozantinib in patients with RMHNSCC who had Response Evaluation Criteria in Solid Tumors v.1.1 measurable disease and no contraindications to either agent. We assessed the primary end points of tolerability and overall response rate to the combination with secondary end points of progression-free survival and overall survival and performed correlative studies with PDL-1 and combined positive score, CD8⁺ T cell infiltration and tumor mutational burden. A total of 50 patients were screened and 36 were enrolled with 33 evaluable for response. The primary end point was met, with 17 out of 33 patients having a partial response (52%) and 13 (39%) stable disease with an overall clinical benefit rate of 91%. Median and 1-year overall survival were 22.3 months (95% confidence interval (CI) = 11.7–32.9) and 68.4% (95 % CI = 45.1%–83.5%), respectively. Median and 1-year progression-free survival were 14.6 months (95% CI = 8.2–19.6) and 54% (95% CI = 31.5%–72%), respectively. Grade 3 or higher treatment-related adverse events included increased aspartate aminotransferase (n = 2, 5.6%). In 16 patients (44.4%), the dose of cabozantinib was reduced to 20 mg daily. The overall response rate correlated positively with baseline CD8⁺ T cell infiltration. There was no observed correlation between tumor mutational burden and clinical outcome. Pembrolizumab and cabozantinib were well tolerated and showed promising clinical activity in patients with RMHNSCC. Further investigation of similar combinations are needed in RMHNSCC. The trial is registered at ClinicalTrials.gov under registration no. NCT03468218.

The incidence rates of head and neck squamous cell carcinoma (HNSCC) in the United States have significantly increased in the past several decades with roughly 67,000 cases diagnosed yearly¹. Most patients with recurrent metastatic HNSCC (RMHNSCC) have limited survival and limited therapeutic options^1,2. Harnessing the immune system for antitumor effect in HNSCC is the subject of intensive research^3,4. Immunotherapy with the anti-programmed cell death protein 1 (PD-1) agents nivolumab and pembrolizumab emerged as a treatment option in patients with recurrent or metastatic disease and showed more satisfactory therapeutic effects and lower toxicity compared to cytotoxic chemotherapy^5–8. Yet despite these encouraging results, RMHNSCC is largely a fatal disease with a meager response to therapy and a need for new and effective therapeutic strategies¹. Because vascular endothelial growth factor (VEGF) inhibitors, including tyrosine kinase inhibitors (TKIs), have immunomodulatory properties, combining these agents with immune checkpoint inhibitors has emerged as a promising strategy to advance the immunotherapy paradigm in different malignancies, including head and neck cancer⁹.

Cabozantinib is a widely used multikinase TKI that targets VEGFR-1, 2 and 3, MET, as well as TAM receptor kinases^10,11. It has gained approval in multiple tumor types, including as a single agent in advanced renal cell carcinoma¹², hepatocellular carcinoma¹³, in combination with nivolumab as a first-line therapy in renal cell carcinoma after previous treatment with sorafenib¹⁴, and after first-line multi-TKI therapy in metastatic radioiodine-refractory differentiated thyroid cancer¹⁵. VEGF inhibitors are known to enhance effector T cell tumor infiltration and function, while decreasing the proportion of regulatory T cells and myeloid-derived suppressor cells¹⁶. Additionally, MET and TAM signaling, which is cotargeted by cabozantinib, contributes to tumor progression^17,18. Other immune modulatory effects of cabozantinib include inhibition of AXL¹⁹ and MER signaling, which in turn polarizes macrophages into an immunosuppressive M2 phenotype reversed by cabozantinib^20,21. Given these immunomodulatory properties, the combination of cabozantinib with immune checkpoint inhibitors could be expected to mount a more effective immune response even in noninflamed or ‘cold’ tumors.

Results

Patient characteristics

Between March 2019 and June 2022, a total of 50 patients were screened and 36 were enrolled on the study. The patient characteristics of the 36 enrolled patients are summarized in Table 1. Fourteen patients were deemed ineligible at screening. The reasons for screen failures are summarized in Extended Data Table 1. Three patients were removed from the study before having evaluable disease for the following reasons: moving to another state; interest in palliative therapy; and organ failure. At the time of the analysis, 29 patients had discontinued therapy, 16 of whom had disease progression, with 9 patients discontinuing as a result of adverse events (AEs). A total of 14 patients were still in follow-up, with 7 patients continuing treatment (Fig. 1). Of note, 24 patients were treated for distant metastatic disease (M1) at presentation. A total of 9 patients failed cisplatin with radiotherapy (enrolled under 6 months), 1 patient failed cetuximab with radiotherapy within 6 months, 2 patients had no previous treatment at presentation and 24 patients were enrolled more than 6 months after completion of previous therapy.

Table 1 |.

Characteristics of the patients treated

Variable	n=36a
Race
Asian	5 (14%)
Black or African American	4 (11%)
White	27 (75%)
Ethnicity
Hispanic or Latino	1 (2.8%)
Non-Hispanic	34 (94%)
Unknown	1 (2.8%)
Sex
Male	30 (83%)
Female	6 (17%)
Age at enrollment, years	62 (54–67)
Smoking status
Current	9 (25%)
Former	8 (22%)
Never	19 (53%)
Distant metastasis	35 (97%)
Stage at initial diagnosis
0	1 (2.8%)
I	4 (11.1%)
II	8 (22.2%)
III	6 (16.7%)
IV	16 (44.4%)
Unknown	1 (2.8%)
Previous treatment
Radiation	33 (91.6%)
Platinum-based chemotherapy	26 (72.2%)
No chemotherapy	2 (5.6%)
Cetuximab	3 (8.3%)
Primary disease site
Oropharynx	22 (61%)
Oral cavity	2 (6%)
Hypopharynx	2 (6%)
Larynx	4 (11%)
Nasopharynx	6 (16%)
Dose reduction	16 (44%)
HPV/P16 status
Negative	12 (33%)
Positive	17 (47%)
Not applicable/unknown	7 (19%)
PD-L1-CPS
<20	17 (50%)
≥20	17 (50%)
TMB
<6.71	8 (50%)
≥6.71	8 (50%)
ECOG status
0	18 (50%)
1	18 (50%)

^an (%); median (IQR).

Fig. 1 |. CONSORT diagram.

Study CONSORT diagram.

Toxicity analysis

A total of 18 (50%) patients had grade 3 or higher AEs and 6 (17%) had grade 3 or higher treatment-related AEs (TRAEs). Toxicities including all-grade AEs, grade 3 or higher toxicities and grade 3 or higher TRAEs are summarized in Table 2. Fatigue (n = 16, 44.4%), diarrhea (n = 12, 33.3%) and hypothyroidism (n = 12, 33.3%) were the most common AEs (all grades). Grade 3 or higher TRAEs included increased aspartate aminotransferase (AST) (n = 2, 5.6%), hyponatremia (n = 1, 2.8%) and increased liver function tests (n = 1, 2.8%), as well as oral mucositis. Nine patients (25%) discontinued all therapy as a result of AEs. Cabozantinib dose reduction was allowed for any grade 2 or higher intolerable AEs deemed related to treatment. In 16 patients (44.4%), the dose of cabozantinib was reduced to 20 mg per day. The most common reasons for dose reduction are described in Extended Data Table 2 and included oral mucositis, hand and foot skin reaction, and diarrhea. No grade 5 TRAEs were noted.

Table 2 |.

Frequency of AEs: all grades, grade 3 or higher and treatment-related grade 3 or higher

Most common AEs with any-grade toxicity	n (%)=36
Fatigue	16 (44.4)
Diarrhea	12 (33.3)
Hypothyroidism	12 (33.3)
Constipation	11 (30.6)
Dry mouth	10 (27.8)
Anorexia	9 (25.0)
Headache	9 (25.0)
Hypertension	9 (25.0)
Hyponatremia	9 (25.0)
Oral mucositis	9 (25.0)
Hypomagnesemia	8 (22.2)
Pain	8 (22.2)
Dysphagia	7 (19.4)
Hyperglycemia	7 (19.4)
Palmar-plantar (hand-foot) erythrodysesthesia syndrome	7 (19.4)
Sore throat	7 (19.4)
Abdominal pain	6 (16.7)
Back pain	6 (16.7)
Hypophosphatemia	6 (16.7)
Nausea	6 (16.7)
Oral pain	6 (16.7)
Maculopapular rash	6 (16.7)
Decreased white blood celis	6 (16.7)
Anemia	5 (13.9)
Cough	5 (13.9)
Dysgeusia	5 (13.9)
Respiratory, thoracic and mediastinal disorders—Other, specify	5 (13.9)
Increased alanine aminotransferase	4 (11.1)
Increased AST	4 (11.1)
Dizziness	4 (11.1)
Dyspnea	4 (11.1)
Hypotension	4 (11.1)
Investigations—Other, specify	4 (11.1)
Decreased platelet count	4 (11.1)
Vomiting	4 (11.1)
Dehydration	3 (8.3)
Increased gamma-glutamyl transferase	3 (8.3)
General disorders and administration site conditions— Other, specify	3 (8.3)
Hematuria	3 (8.3)
Hepatobiliary disorders—Other, specify	3 (8.3)
Hypoalbuminemia	3 (8.3)
Hypocalcemia	3 (8.3)
Hypokalemia	3 (8.3)
Increased lipase	3 (8.3)
Pain in extremity	3 (8.3)
Paresthesia	3 (8.3)
Acneiform rash	3 (8.3)
Increased serum amylase	3 (8.3)
Skin and subcutaneous tissue disorders—Other, specify	3 (8.3)
Weight loss	3 (8.3)
Increased alkaline phosphatase	2 (5.6)
Anxiety	2 (5.6)
Arthralgia	2 (5.6)
Aspiration	2 (5.6)
Increased blood bilirubin	2 (5.6)
Dry skin	2 (5.6)
Dyspepsia	2 (5.6)
Ear pain	2 (5.6)
Epistaxis	2 (5.6)
Infections and infestations—Other, specify	2 (5.6)
Leukocytosis	2 (5.6)
Decreased lymphocyte count	2 (5.6)
Oral dysesthesia	2 (5.6)
Pancreatitis	2 (5.6)
Sinusitis	2 (5.6)
Stomach pain	2 (5.6)
All AEs with grade 3 and 4 toxicity	n (%)=36
Overall	18 (50%)
Dysphagia	3 (8.3)
Hypertension	3 (8.3)
Increased AST	2 (5.6)
Back pain	2 (5.6)
Hypotension	2 (5.6)
Oral mucositis	2 (5.6)
Increased alanine aminotransferase	1 (2.8)
Increased alkaline phosphatase	1 (2.8)
Anemia	1 (2.8)
Anorexia	1 (2.8)
Aspiration	1 (2.8)
Increased blood bilirubin	1 (2.8)
Diarrhea	1 (2.8)
Dyspnea	1 (2.8)
Increased gamma-glutamyl transferase	1 (2.8)
Headache	1 (2.8)
Hepatobiliary disorders—Other, specify	1 (2.8)
Hyperglycemia	1 (2.8)
Hyponatremia	1 (2.8)
Hypophosphatemia	1 (2.8)
Hypoxia	1 (2.8)
Increased lipase	1 (2.8)
Myocardial infarction	1 (2.8)
Nausea	1 (2.8)
Gastric obstruction	1 (2.8)
Oral pain	1 (2.8)
Pain	1 (2.8)
Pancreatitis	1 (2.8)
Decreased platelet count	1 (2.8)
Sore throat	1 (2.8)
Vomiting	1 (2.8)
All AEs with grade 3 and 4 toxicity attributed to treatment	n (%)=36
Overall	6 (17%)
Increased AST	2 (5.6)
Increased alanine aminotransferase	1 (2.8)
Increased alkaline phosphatase	1 (2.8)
Increased blood bilirubin	1 (2.8)
Increased gamma-glutamyl transferase	1 (2.8)
Hypertension	1 (2.8)
Hyponatremia	1 (2.8)
Increased lipase	1 (2.8)
Oral mucositis	1 (2.8)

Thirty-six patients were accounted for in the AE count. Patients may have more than one AE reported.

Efficacy analysis

Of 33 patients who could be evaluated, the best reported responses were partial response (PR) in 17 patients (52%), stable disease (SD) in 13 patients (39%) and progressive disease (PD) in 3 patients (9%) with an overall response rate (ORR) of 52% and an overall clinical benefit (PR + SD) observed in 30 patients (91%). Responses were seen in both human papillomavirus (HPV)-related and unrelated disease (Fig. 2a). Responses were in general early and some were deep and durable (Fig. 2b).

Fig. 2 |. Response assessments.

a, Waterfall diagram describing the percentage change in tumor size for 33 evaluable patients. (Patients with missing response or percentage tumor size change were excluded.) The red star represent patients with a high median TMB ≥ 6.71; responses noted by HPV or p16 status and programmed cell death 1 ligand 1 (PD-L1)-CPS < 20 or ≥ 20. b, Spider plot curve for 33 patients evaluable for response with highlighted tumor HPV or p16 status. (The ORR was calculated with 95% CI using a one-sided binomial distribution; the biomarker correlation with the ORR was estimated by chi-squared test or logistic regression model.).

With a median follow-up of 10.6 months, the median progression- free survival (PFS) for the 36 treated patients was 14.6 months (95% CI = 8.2–19.6) and the 1-year PFS was 54% (95% CI = 31.5%–72%) (Fig. 3a). The median overall survival (OS) was 22.3 months (95% CI = 11.7–32.9) and the 1-year OS was 68.4% (95% CI = 45.1%–83.5%) (Fig. 3b). The median time to response was 4.17 months (95% CI = 2.367–6.016) and the median duration of response had not been reached by the time of follow-up (68.9% had persistent response at 12 months) (Fig. 3c). The initial T, N or M stage did not correlate with PFS or OS (Extended Data Fig. 1). It is noteworthy that responders to therapy had a significantly higher PFS (median PFS of 19.4 versus 8.8 months; P = 0.0278) compared to nonresponders (Extended Data Fig. 2a). However, this did not reach significance for OS (median OS of 17.0 versus 16 months; P = 0.347) (Extended Data Fig. 2b). When excluding patients with nasopharyngeal carcinoma, we observed a median PFS of 11.7 months and OS of 17.8 months. No survival correlation was observed with smoking; former or current smokers had a median OS of 17.8 months compared to 22.3 months for never smokers (P = 0.442) (Extended Data Fig. 2c). In addition, there was no correlation between having grade 3 or higher AEs and PFS (P = 0.960) or OS (P = 0.560).

Fig. 3 |. Survival assessments.

a, PFS for the 36 treated patients. The median PFS was 14.6 months (95% CI = 8.2–19.6); the 1-year PFS was 54% (95% CI = 31.5%–72%). b, OS of the 36 treated patients. The median OS was 22.3 months (95% CI = 11.7– 32.9) and the 1-year OS was 68.4% (95% CI = 45.1%–83.5%). c, Swimmer plot for patients who responded to therapy. Each bar represents one patient. Patients with arrows represent continued response. A durable responder is a patient with confirmed response for at least 6 months.

Using the Cox proportional-hazards model, univariate association with OS was significant for Eastern Cooperative Oncology Group (ECOG) performance status (0 versus 1; P = 0.045). There was no association with sex, tumor HPV status, primary tumor site, T, N or M stage, or tumor combined positive score (CPS) (Fig. 4a) despite a trend favoring a CPS ≥ 20. Patients with a CPS less than 20 had a median OS of 14.6 months (95% CI = 8.2–NE) and a 1-year OS of 54.9% (95% CI = 24.5%–31.4%), whereas patients with a CPS ≥ 20 had a median OS of 32.9 months (95% CI = 6.9–32.9) and a 1-year OS of 83.6% (95% CI = 48.0%–95.7%) (Fig. 4b). CD8⁺ T cell tumor infiltration was significantly correlated with clinical response (P = 0.0152) in 26 patients (Fig. 4c). CD8⁺ T cell infiltration was not correlated with HPV status in responders and nonresponders (P = 0.969). Improved OS was seen in the entire intention-to-treat population, including those patients who had dose reductions. Patients with dose reductions did not appear to have a worse OS (hazard ratio (HR) = 0.5; P = 0.312).

Fig. 4 |. Correlative assessments.

a, Univariate Cox proportional-hazards association with OS by patient and clinical characteristics at baseline. Data are presented as the HR and 95% CI by potential confounders. HR < 1 represents a lower hazard; HR > 1 represents a higher hazard. OP, oropharynx carcinoma. b, OS of 36 patients treated with PD-L1-CPS. CPS < 20, median OS of 14.6 months (95% CI = 8.2–NE); 1-year OS of 54.9% (95% CI = 24.5%–31.4%). CPS ≥ 20, median OS of 32.9 months (95% CI = 6.9–32.9); 1-year OS of 83.6% (95% CI = 48.0%–95.7%). Association of OS with CPS was assessed using the Kaplan–Meier method (log-rank test); the significance level was set at P < 0.05 (two-tailed). c, Preexisting CD8⁺ T cell tumor infiltration (25 patients) with representative IHC. Data are presented as mean values ± s.d. The length of the error bars is the s.d. Mean number of CD8⁺ positive cells per field in the PD + SD and PR groups is 52.57 and 137.8, respectively. An unpaired two-tailed t-test was used. WCI, Winship Cancer Institute.

TMB results

We determined the TMB in 16 patients with sufficient tumors available for whole-exome sequencing (WES). The TMBs for these patients are illustrated in Extended Data Fig. 3 and summarized in Extended Data Table 3. The median TMB for these patients was 6.71 per megabase (Mb) (range: 0.39–10.66 per Mb), which is similar to the previously published TMB of HNSCC²². Only one patient had a TMB > 10 per Mb). There was no correlation between TMB level and response to therapy (P = 0.614). There was no correlation between TMB and PFS (P = 0.343). A lower TMB was correlated with improved OS (P = 0.008) (Extended Data Fig. 4). Note the limitation of not having normal DNA from all patients, which could have resulted in the TMB estimation not being an accurate reflection of patients’ actual TMB.

Discussion

This phase 2 trial of pembrolizumab and cabozantinib in RMHNSCC revealed encouraging clinical activity in this disease and met its primary end point of improvement in ORR. These findings build on previous cabozantinib-based combinations with PD-1 inhibitors leading to notable improvement in outcome in patients with other tumors, such as renal cell carcinomas²³.

The observed PFS and OS suggest that the combination of cabozantinib and pembrolizumab improved PFS (and OS) over historical single-agent data. Focusing on patients with biomarker-positive disease (CPS ≥ 20) receiving the single-agent pembrolizumab in KEYNOTE-048, the median PFS was notably 3.4 months and the median OS was 14.9 months. The results observed with pembrolizumab and cabozantinib represent a notable improvement compared to historical controls with single-agent pembrolizumab and, therefore, open the door for further exploration of this combination in RMHNSCC. Whether the observed enhanced clinical activity is related predominantly to VEGF inhibition is a question that could be further clarified once results from the ongoing trials with other VEGF inhibitor TKIs are available (LEAP 9 and 10; NCT04428151 and NCT04199104). Cabozantinib exerts anti-VEGF and immunomodulatory effects that may contribute further to the observed anti-tumor activity. This feature of cabozantinib may be exerting a greater role in tumors such as HNSCC where immune infiltration is a prominent characteristic²⁴. The relative contributions of each of these mechanisms (VEGF versus MET and TAM inhibition) to the enhanced clinical activity of these combinations should be further investigated. Tumor-inherent transcriptomic and epigenetic factors can also influence the activity of immune modulatory agents such as cabozantinib, as suggested in previous studies²⁵. Such an analysis would be important to identify potential predictive tumor characteristics for the observed clinical benefit of the cabozantinib and immune checkpoint inhibitor combinations in RMHNSCC.

Although the OS in our patient cohort did not correlate with PD-L1-CPS, the observed trend favoring higher CPS suggests that a relevant response to immune PD-1 blockade is a necessary condition for the activity of cabozantinib and PD-1 inhibitors³. Our TMB data did not show a clear correlation between a higher TMB and response. The correlation between a lower TMB and improved OS suggest an activity of the combination despite a low TMB, a question worthy of further evaluation in similar combinations. The data must however be interpreted with caution given the small sample of tumors with available TMB, the lack of normal DNA from all patients and the fact that only one patient had a truly high TMB (>10 per Mb), in addition to the single-arm design. Further investigation of TMB as a predictive biomarker of PD-1 inhibitors and multi-TKI combinations is warranted given the paucity of data in this domain.

Our observed correlation between overall response and PFS is another departure from historic data using single-agent immune checkpoint inhibitors as these did not appear to significantly impact PFS. In addition to being supportive of further investigation in a randomized setting, our findings raise the question of whether PFS is an important end point for combination PD-1 and TKI trials unlike single-agent PD-1 inhibitors. Of interest is the lack of clear influence on disease outcome in patients with nasopharyngeal carcinoma, which is suggestive of a clinical activity in this disease. Noted limitations of our results are the lack of a comparative arm of single-agent PD-1 inhibitor, the predominance of archival tissues as a source of biomarkers and the limited number of patients with TMB data. In addition, discrepancy in the proportion of HPV-related oropharyngeal squamous cell carcinoma, high CPS and other biological tumor characteristics could have influenced the clinical results when compared to historical controls.

Although dose reductions of cabozantinib were common, these did not appear to reduce clinical benefit for patients who continued on the 20-mg dose. Of importance, cabozantinib dose reductions mitigated the noted treatment-related toxicities, which did not overlap between the two agents. These observations point to a potential advantage of such combinations over immunotherapeutic doublets, which often have overlapping toxicity profiles. Further possible advantages to the cabozantinib and pembrolizumab combination include the lack of a needed specific immune biomarker tumor profile, which is often the case with new immune targets such as TIGIT and LAG3 inhibitors^26,27. Despite these findings, it is worth mentioning that 14 patients were screen failures mostly because of contraindications to receiving VEGF inhibitors; 9 patients (25%) discontinued therapy as a result of AEs, stressing that the use of VEGF-TKI is not applicable in all patients with head and neck malignancies as corroborated by previous reports²⁸. Of note is a higher rate of AE-related treatment discontinuation of 25%, compared to a 12% rate in the KEYNOTE-048 trial.

Based on these encouraging results, we believe that the combination of pembrolizumab and cabozantinib is highly active, well tolerated overall and warrants further investigation in patients with RMHNSCC.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41591-023-02275-x.

Methods

Study design

This was an open label, single-arm, multicenter, phase 2 study of pembrolizumab and cabozantinib in patients with RMHNSCC who had not received previous immune checkpoint inhibitors. The study was approved by the institutional review boards at Emory University and the Moffitt Cancer Center and was conducted in compliance with ethical standards and good clinical practice. After institutional review board approval, the trial enrolled patients at the Winship Cancer Institute of Emory University (leading site) and Moffitt Cancer Center; enrollment started in March 2019 and was completed in June 2022 (NCT03468218). Written informed consent was received from all participants in accordance with the Declaration of Helsinki v.2013. The first patient was enrolled on 5 March 2019 and the last patient on 16 March 2022. The Emory Winship Cancer Institute clinical trials office served as the coordinating center for the study. The sponsor reviewed responses, toxicity and patient information on a weekly basis with participating sites. The Emory Winship Cancer Institute data safety and monitoring committee reviewed study information on a regular basis. All study participant information was kept confidential by assigning a random number to each study participant. All data were kept confidential as per institutional guidelines and policies.

Patients were enrolled if they had a documented diagnosis of RMHNSCC of the oral cavity, oropharynx, paranasal sinuses, hypopharynx, nasopharynx or larynx. Patients with squamous cell carcinoma of unknown primary origin in cervical lymph nodes were included only if their HPV status was positive. To be eligible, patients needed to have recurrent or metastatic disease that was deemed inoperable. Patients who had received previous systemic therapy in the definitive or metastatic setting were required to have documentation of disease progression after previous therapy. A maximum of one previous radiotherapy regimen (curative or palliative intent) to the head and neck was allowed. If the radiation was combined with chemotherapy, a minimum of 4 months must have elapsed between the end of radiotherapy and registration. If radiation was given alone, a minimum of 8 weeks must have elapsed between the end of radiotherapy and registration. A minimum of 3 weeks must have elapsed between previous radiation to other anatomical areas and registration. Previously irradiated sites must have healed with no sequelae from radiation that would predispose to fistula formation. Patients were required to be 18 years of age or older with a life expectancy greater than 3 months and have an ECOG performance status of 0 or 1. Sexually active participants (men and women) needed to agree to use medically accepted methods of contraception. Patients were enrolled regardless of sex or gender, which was determined by patient self-reporting. Because HNSCC predominantly affects male patients, which was consistent with our study enrollment (83% male), an analysis based on sex or gender was not performed.

Participants were excluded if they had experienced any significant gastrointestinal bleeding, had radiographic evidence of cavitating lung metastatic lesions, tumor invasion or encasement of any major blood vessel, invasion of the gastrointestinal tract, evidence of endotracheal or endobronchial invasion, were receiving concomitant anticoagulation or were unable to swallow intact tablets of cabozantinib. Participants did not receive any compensation while on the study.

The study primary end point was evaluation of the ORR and evaluation of safety and tolerability of the combination. Secondary objectives included estimation of the PFS and OS in addition to performing tissue and blood correlative studies. OS was defined as the length of time from start of treatment to death. PFS was defined as the length of time from start of treatment to disease progression (including death). Time to response and duration of response were defined as the time from start of treatment to the first objective tumor response and from first occurrence of response to disease progression or death.

Treatment plan

Pembrolizumab was administered intravenously at a dose of 200 mg every 21 days and cabozantinib was administered orally at a dose of 40 mg once daily. A run-in phase of cabozantinib at 40 mg per day was performed for the first six patients, beginning in cycle 1. If 2 or more dose-limiting toxicities were observed during cycle 1 in the first six patients, the dose of cabozantinib would be decreased to 20 mg per day for the remainder of the study. As only one patient developed a dose-limiting toxicity (grade 3 hyponatremia) among the first six treated patients, accrual continued with 40 mg per day of cabozantinib.

Cabozantinib was supplied in 20-mg tablets by Exelixis. Laboratory testing (chemistry, hematology tests, urinalysis and urine protein) was performed every 3 weeks for the first 9 weeks followed by assessments every 6 weeks. AE seriousness, severity grade and relationship to study treatment were assessed by investigators with each cycle of therapy. Severity and grade were defined by the National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events v.4.0. Tumors were assessed by Response Evaluation Criteria in Solid Tumors v.1.1 response criteria using computed tomography, magnetic resonance imaging or positron emission tomography computed tomography scan at a frequency of every 9 weeks and were confirmed by the principal investigator and by radiology staff at each participating institution.

Biomarkers

Tumor samples were collected from patients through archival tissue or biopsies performed from at least one site of disease before the first dose administration of therapy.

Tumor PD-L1 protein expression and CPS were analyzed in baseline biopsies or surgical specimens. Baseline tumor samples were analyzed by immunohistochemistry (IHC) for CD8⁺ T cell infiltration. Blood samples were collected at screening, at 9 weeks and at 6 months on treatment, as well as at the end of treatment or disease progression, whichever was sooner.

Tumor CPS

PD-L1 (CPS) expression was assessed at a central laboratory in the Emory University Department of Pathology using the PD-L1 IHC 22C3 pharmDx assay (Agilent) and characterized by CPS as described by Emancipator et al.²⁹.

Intratumoral CD8⁺ T cell infiltration

Clinical samples were collected with informed consent of the individuals or their guardians under Health Insurance Portability and Accountability Act guidelines. Paraffin-embedded human specimens were cut into 3-μm sections and mounted on slides. IHC was performed as described elsewhere^30,31. Briefly, tissue sections were deparaffinized with xylene, rehydrated through a graded alcohol series and incubated in 3% hydrogen peroxide. Sections were placed in 10 mM sodium citrate buffer (pH 6.0) at subboiling temperatures for 10 min and incubated with 10% normal goat serum, followed by incubation with a primary monoclonal antibody against human CD8α (clone C8/144B, 1:300, catalog no. 90257, Cell Signaling Technology). Immunoreactivity was visualized using the DAB Detection Kit (Vector Laboratories). Slides were counterstained with hematoxylin, dehydrated and mounted; images were captured using a charge-coupled device camera (Olympus). Quantitation of CD8⁺ cells in the tumor sections was counted in ten random fields at a low power magnification of ×200. The average number of positive cells per reported field was calculated based on the results provided by three investigators who were blinded to pathological information. The majority of available tissue consisted of archival specimens.

WES and TMB calculation

For WES, tumor DNA samples from 16 formalin-fixed, paraffin-embedded (FFPE) tumors and germline DNA samples from four buffy coats were extracted and sequenced at Admera Health. Briefly, isolated DNA extraction from FFPE samples was performed using Mag-Bind FFPE DNA/RNA (OMEGA Bio-Tec) according to the manufacturer’s recommendation. Extracted genomic DNA was then quantified with Qubit 2.0 DNA HS Assay (Thermo Fisher Scientific) and quality-assessed using the TapeStation Genomic DNA Assay (Agilent Technologies). Library preparation was performed using the KAPA Hyper Prep Kit (Roche) according to the manufacturer’s recommendations. Exome capture was performed with the xGen Exome Research Panel v.2.0 (Integrated DNA Technologies). Library quality and quantity were assessed with Qubit 2.0 DNA HS Assay, TapeStation High Sensitivity D1000 Assay (Agilent Technologies) and QuantStudio 5 System (Applied Biosystems). Illumina 8-nt dual indices were used. Equimolar pooling of libraries was performed based on quality control values and sequenced on an Illumina NovaSeq S4 with a read length configuration of 150 M paired-end reads. Tumor FFPE samples were targeted for 156 M paired-end reads (78 M in each direction) and samples originating from buffy coat were targeted for 52 M paired-end reads (26 M in each direction). Because not all tumor samples had paired germline DNA samples, the four germline DNA samples were concatenated and merged as a combined normal for somatic variant calling. Paired-end FASTQ files were aligned to the human reference genome GRCh38 using the Burrows–Wheeler Aligner. Next, PCR duplicates were removed and somatic variants (indels and single-nucleotide variants) were called using the Illumina Manta v.1.6.0³² and Strelka2 v.1.6.0³³ software. The resulting variant call format files were annotated to include allele frequency from the Exome Aggregation Consortium. Alignment, variant calling and annotation workflow were integrated in several containerized Snakemake workflow that is publicly available (https://github.com/HuntsmanCancerInstitute/Workflows). The output variant call format files were analyzed in R v.4.2.2 using vcfR v.1.13.0³⁴; to retain somatic gene coding mutations, the following filtering steps were used: (1) coverages at tumor variant base fewer than 100 reads were excluded; (2) tumor variant allele frequencies less than 5% were excluded; (3) variants found in the Exome Aggregation Consortium database of more than 5% were discarded; (4) variants present in the normal of more than 0.1% were discarded; (5) only somatic variants considered as missense, frameshift, start-lost, stop-gained or stop-lost were retained. After these variant filtering steps, the TMB for each sample was calculated by dividing the total mutation number with the length of the human exon (38 Mb). The WES data was submitted to the Sequence Read Archive (SRA) under BioProject ID PRJNA903894.

Statistical analysis

The ORR for this patient population is reported as 18% with pembrolizumab as a single agent^6,35. We estimated that the ORR would improve to 35% with the combined treatment. Patients were excluded from the statistical analysis if they did not meet the protocol-specified eligibility criteria. Sample size was estimated to be 34 by a one-arm design with a null hypothesis of ORR ≤ 0.15 versus its one-sided alternative. If the number of responses was 8 or fewer out of 34, the intervention would be deemed not promising. Based on an interim futility analysis established in the protocol, accrual continued after 20 patients were evaluable for response (if 4 or fewer responses were observed, the trial would stop earlier and futility was claimed based on a Bayesian optimal phase 2 design with one interim analysis)³⁶. Because ten out of the first 20 accrued patients had a documented response, the statistical boundaries for futility were not crossed and accrual proceeded as planned. This design yielded a one-sided type 1 error of 0.05 and a power of 83.5% when the true response rate is 35%. The related calculation was done at https://trialdesign.org/one-page-shell.html#BOP2.

The ORR was calculated with a 95% CI using a binomial distribution. The median PFS and OS were estimated by Kaplan–Meier method along with 95% CI³⁷. The ability of biomarkers to predict ORR was estimated using a chi-squared test or logistic regression model³⁸; association with PFS and OS was assessed using the Kaplan–Meier method, log-rank test and Cox model³⁹. All data analysis was carried out in SAS v.9.4 (SAS Institute). The Oncore v.2021R3 software was used for data collection.

Extended Data

Extended Data Fig. 1 |. Survival outcomes by T and N stage.

Overall survival by T (a) and N (b) stage (p=0.891, p=0.550 respectively), and progression-free survival by T (c) and N (d) stage (p=0.599, p=0.732 respectively).

Extended Data Fig. 2 |. Survival outcomes by response and smoking history.

PFS (a) and OS (b) in responders versus non-responders (p=0.028 and p=0.3347 respectively). (c) OS by smoking history (former or current smokers versus never smokers) (p=0.442).

Extended Data Fig. 3 |. Tumor mutational burdens for HNSCC patients.

Distribution of the TMB (Mut/MB) computed for the 16 HNSCC patients with sufficient tumor sample. The red line indicates the median of TMB for these patients, which is 6.71/MB.

Extended Data Fig. 4 |. Survival outcomes by tumor mutation burden.

(a) Overall survival and (b) Progression-free survival by tumor mutation burden (TMB).

Extended Data Table 1 |.

Number and reasons for screen failures

Screen Failure Reason	n
Cavitating pulmonary lesions	4
Inability to swallow cabozantinib tablets	2
Baseline grade >1 AE	2
Patient decision to withdraw and seek alternative therapy	1
Evidence of Gl bleeding within 6 months	1
Significant recent illness	1
Non-measurable disease	1
History of prior malignancy within 1 yr of enrollment	1
Patient on anti-coagulation	1

Extended Data Table 2 |.

Frequency and reasons for cabozantinib dose reductions

	N=36 n (%)
Cabozantinib dose reduction	16 (44.4)
Oral mucositis	4 (23.5)
Hand foot skin reaction	4 (23.5)
Diarrhea	2(11.7)
Physician’s discretion	2 (11.7)
Hyponatremia	1 (5.9)
Hypertension	1 (5.9)
Epistaxis	1 (5.9)
ALT / AST increase	1 (5.9)
Vomiting	1 (5.9)

Extended Data Table 3 |.

Mutation number and TMB for patients with available tumor and blood DNA samples

Sample	NumMut	TMB
WCI 4234–001	216	5.68421053
WCI 4234–002	346	9.10526316
WCI 4234–003	261	6.86842105
WCI 4234–004	248	6.52631579
WCI 4234–005	279	7.34210526
WCI 4234–006	329	8.65789474
WCI 4234–008	284	7.47368421
WCI 4234–009	201	5.28947368
WCI 4234–010	194	5.10526316
WCI 4234–012	275	7.23684211
WCI 4234–015	79	2.07894737
WCI 4234–016	15	0.39473684
WCI 4234-MFT-01	405	10.6578947
WCI 4234-MFT-01	249	6.55263158
WCI 4234-MFT-01	269	7.07894737
WCI 4234-MFT-01	232	6.10526316

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author upon written request and clarification from the requesting party as to how the data will be used. Requests for data sharing will be responded to within 2 weeks. The WES data has been submitted to the SRA. The full study protocol is available as Supplementary Information.