Response-Adaptive Surgical Timing in neoadjuvant immunotherapy demonstrates enhanced pathologic treatment response in Head and Neck Squamous Cell Carcinoma

Abstract

Purpose:

We evaluated whether IDO-inhibitor BMS986205 (IDOi) + PD-1 inhibitor nivolumab enhanced T-cell activity and augmented immune-mediated antitumor responses in untreated, resectable HNSCC. We employed response-adaptive surgical timing to identify responders to immunotherapy and enhance their response.

Patients and Methods:

Patients with HNSCC were 3:1 randomized to receive nivolumab with or without BMS986205 PO daily (NCT03854032). In the combination arm, BMS986205 was initiated 7 days prior to nivolumab. Patients were stratified by HPV status. Response-adaptive surgical timing involved response assessment by radiographic criteria 4 weeks after nivolumab in both arms. Non-responders underwent surgical resection, while responders received 4 more weeks of randomized therapy before surgery. Biomarker analysis utilized pathologic treatment response (pTR) and RNA sequencing.

Results:

Forty-two patients were enrolled, and the addition of IDOi to nivolumab did not result in greater rate of radiographic response (p=0.909). Treatment was well-tolerated with only 2 (5%) patients experiencing grade 3 immune-related adverse events. The addition of IDO-inhibitor augmented rates of pTR in patients with high baseline IDO RNA expression (p<0.05). Response-adaptive surgical timing demonstrated reliability in differentiating pathologic responders versus non-responders (p=0.009). A pretreatment NK cell signature, PD-L1 status, and IFN-g expression in the HPV^– cohort correlated with response. HPV⁺ cohort found B-cell and CAF signatures predictive of response/non-response.

Conclusions:

Response-adaptive surgical timing enhanced treatment response. IDO-inhibitor BMS986205 augmented pTR in patients with high IDO1-expression in baseline samples, indicating a need for identifying and targeting resistant nodes to immunotherapy. HPV-status-dependent signatures predicting response to immunotherapy in HNSCC warrant further study.

Introduction:

Immune checkpoint inhibitors (ICIs) have been recognized as a standard of care method to treat recurrent or metastatic head and neck squamous cell carcinoma (HNSCC) by eliciting effective immune responses. ICI use in the neoadjuvant setting has been shown to generate pathologic responses by our group and others, with ongoing investigation underway to delineate the effects of single versus multiple cycles of treatment on pathologic response and survival outcomes (1–6). Evaluation of biologic specimens in the context of these neoadjuvant trials both before and after treatment has the potential to not only delineate effective novel drug combinations but also to identify novel biomarkers predictive of response (7). The culmination of these neoadjuvant trials will untangle the amazing and often over simplified immune-tumor interactions. Understanding the pre-existing tumor-resident T cells that have cytotoxic potential at baseline and during treatment or the peripheral blood immunoprofiling that leads to understanding primed and exhausted systems is critical to addressing resistance (6,8)

The impressive depth and duration of response of T cell-directed immunotherapies are contrasted by the fact that only a subset of recurrent/metastatic patients experience long-lasting remissions after single modality immune checkpoint therapy (response rates of approximately 20%) (9,10). This discrepancy has reinforced interest in understanding how both systemic and local factors in the tumor microenvironment (TME) such as anergic T cells, Th2/M2 phenotype, and inhibitory costimulatory molecules, may contribute to suboptimal responses to ICIs. Combinatorial therapeutic approaches to augment the immune and clinical responses seen with PD-1 inhibitors are being investigated in multiple tumor types. A better understanding of how to predict responders and identification of novel drug combinations to increase this response rate may provide the key to extending anti-PD-1 benefits.

Indoleamine 2,3-dioxygenase (IDO1) activity promotes immune tolerance by inhibiting T-cell function through local depletion of tryptophan and generation of inhibitory kynurenine pathway metabolites (11). IDO1 expression is associated with a decrease in tumor infiltrating immune cells, an increase in the proportion of regulatory T-cells (Treg), tumor progression, and increased programmed death receptor ligand 1 (PD-L1) expression (12,13). These findings suggest that IDO1 is an important regulator of the immunosuppressive mechanisms responsible for tumor escape from host immune surveillance. In a phase 1/2 multi-solid tumor trial, inhibition of IDO and anti-PD-1 with epacadostat and pembrolizumab, respectively, was found to be well-tolerated in 8 HNSCC patients, but without report of efficacy (14). A melanoma phase III trial with epacadostat did not reach its endpoint and this trial was conceived to look for a signal in the treatment of HNSCC with BMS986205, an IDO selective inhibitor (15).

Based on these observations, blocking the pathway that produces the IDO enzyme continues to be an approach of interest to amplify T-cell directed immunotherapies in the neoadjuvant setting. To further understand the complex effects of immunomodulators on the TME and potential clinical implications, we sought to assess the impact of neoadjuvant nivolumab with and without the IDO-inhibitor, BMS986205, on previously untreated, resectable HNSCC and interrogate pre- and post-treatment samples for markers of response or resistance to anti-PD-1 therapy. To demonstrate an enrichment of response to ICI treatment, we present a novel clinical trial design with response-adaptive surgical timing.

Methods

Study population and design

We conducted a two-arm randomized, open-label trial (NCT03854032) (Figure 1A) at two sites: Thomas Jefferson University, Philadelphia, PA, and The Ohio State University, Columbus, OH. Eligibility included male or female subjects, age ≥18 years, with untreated, resectable oral cavity, hypopharynx, larynx, or oropharynx HNSCC of any stage (AJCC 8th Edition) with the exception that HPV⁺ oropharynx required 1 node positive for eligibility. Researchers and Adequate organ function (as determined by hematologic and biochemical parameters) was required. Subjects could not have a history of an autoimmune disorder, use of systemic steroids ≥ the equivalent of 10 mg of prednisone daily within 14 days of initiation of therapy, human immunodeficiency virus, hepatitis B/C, concurrent malignancies, or G6PD Deficiency. The representativeness of the study population is included in Supplementary Table 1.

Figure 1. Trial schematic and waterfall plots of pathologic treatment response and radiographic response rates.

A: Trial Schematic. B: Pictorial depiction of patients included and excluded for radiographic and pathologic analyses. C and D: Degree of pathologic treatment response (pTR) at the primary tumor site (C) and overall pTR (combined analysis of both primary tumor and lymph nodes [when applicable]) (D). E: Per RECIST criteria, degree of overall radiographic response (combined analysis of both primary tumor and lymph nodes [when applicable]).

Upon enrollment, subjects underwent a biopsy of the primary site as well as blood sample collection. Computed tomography (CT) or magnetic resonance imaging (MRI) was performed within 28 days of enrollment. Subjects were randomized 3:1 to receive nivolumab + BMS986205 (Bristol Myers Squibb, New York, NY) or nivolumab (Bristol Myers Squibb, New York, NY) alone with stratification by HPV status. Subjects in arm A received BMS986205 100 mg PO once a day (qday) beginning 1 week prior to nivolumab 480 mg IV. Subjects in arm B received 480 mg IV of nivolumab alone. Both groups were evaluated at week 5. We introduced response-adaptive surgical timing as a novel trial design. This was carried out by analyzing radiographic response at both the primary site and any involved lymph nodes. Subjects with at least a 10% reduction in volume at either the primary or lymph nodes with no evidence of progression continued to a second cycle followed by surgery. Patients with stable disease or progression proceeded to surgery at the 5-week evaluation without additional study treatment. We hypothesized that the addition of an IDO inhibitor to PD-1 blockade would increase radiographic treatment response. The primary endpoint was the proportion of patients proceeding to a second cycle of treatment.

The chosen cutoff of 10% reduction on radiographic analysis was informed by the results from two of our previous publications. (2,16) We found that utilizing volumetric analysis (i.e. considering all three axes) increased the detection of change over a single short axis measurement, as directed by RECIST v1.1. Furthermore, this threshold allows for some increase in volume (0–9%) due to immune infiltrate while capping any growth over 10% as progression.

Arm A was made up of 31 nivolumab + BMS986205 subjects. Arm B included 11 nivolumab alone subjects from this trial plus an additional prospective 19 subjects from the nivolumab alone cohort in a contemporaneous neoadjuvant clinical trial (NCT03238365) for a total of 30 subjects (Figure 1B) (2). Patients from this contemporaneous neoadjuvant trial (NCT03238365) were dosed at 240 mg of nivolumab every 2 weeks whereas the equivalent dosing for this present trial (NCT03854032) were dosed at 480 mg of nivolumab every 4 weeks. Our protocol outlined this prospective inclusion of these concomitant trial patients. The specimens from this companion trial were not analyzed for secondary and exploratory endpoints in the following results as they did not go on for the second dose if radiographic response was identified.

Both subject inclusion in this neoadjuvant trial as well as adjuvant treatment recommendations were based on recommendations by a multidisciplinary tumor board. At the time of surgical resection, biopsies of the primary tumor, involved lymph nodes, and blood were obtained and processed for correlative analysis. Patients were followed for 3 months after surgery and evaluated for surgical complications and adverse events (AEs). All AEs were documented prospectively utilizing the Common Terminology Criteria for Adverse Events (CTCAE) version 5. This clinical trial was approved by Thomas Jefferson University and Ohio State University Internal Review Board and all patients signed informed consent. This trial was conducted in accordance with the Declaration of Helsinki and the guidelines for Good Clinical Practice.

Assessment of pathologic treatment response:

Pathologic specimens were independently graded by two pathologists as outlined in our previous publication (2). The entire mass or surgical specimen (in the event of a partial/complete response) was sectioned to calculate the pTR. Criteria for response included evidence of macrophage reaction, multinucleated giant cells and granulomas, fibrosis adjacent to residual tumor nests. In complete responders where no more viable tumor was identified, alteration of the normal tissue architecture by fibrosis and macrophage reaction was used to identify areas of previously viable malignancy. Tumors with pTR of ≥20% were defined as responders (R), while those with pTR <20% were defined as minimal or non-responders (NR) at the primary site. At the time of the trial design in 2017, we had limited data to create our cut points for the prospective aspect of this trial. We chose an ordinal data strategy to separate out our minor responders from major responders as others have to-date. To calculate overall pTR, we used an average of the percentages (e.g. 0% pTR at the primary, 80% pTR at lymph node (LN) 1, and 60% pTR at LN2 was considered overall 47% pTR.) We elected not to factor in volume of tumor to prevent greater emphasis on deposits of larger tumor which biologically did not correlate with pTR.

Clinical staging was determined at time of enrollment utilizing the American Joint Committee on Cancer criteria (AJCC 8^th edition) using clinical exam plus CT and FDG-PET/CT scans. Correlation of imaging to pathologic response has previously been published (2).

Next-generation sequencing and bioinformatic analysis

Methods for next-generation sequencing (NGS), quality control, whole exome sequencing (WES) alignment, variant calling, calculation of tumor mutational burden (TMB), whole transcriptome analysis, HPV-status calling, and statistical approaches for molecular correlatives are further described in Supplementary Methods.

Statistical Methods:

Patients were randomized 3:1 based on random permuted blocks within strata defined by location of tumor (oropharyngeal [HPV⁺] vs non-oropharyngeal [HPV^–]). Primary endpoint was a comparison between arm A (nivolumab + IDO Inhibitor BMS986205) and arm B (nivolumab alone) whereby significance would be met if arm A had 35% greater number of patients qualifying for a second dose of treatment as compared to arm B based on radiographic response after 4 weeks (e.g., 75% radiographic response rate in arm A, 40% radiographic response rate in arm B). Secondary endpoints were rates of pTR at the primary site and regional lymph nodes, rates of radiographic response (by this trial’s volumetric analysis and per RECIST criteria), safety, and tolerability of treatment regimen.

We prospectively included 19 additional subjects from the nivolumab alone cohort in a contemporaneous neoadjuvant clinical trial (NCT03238365) to achieve 80% power with two-sided significance level of 5% analysis for a total of 30 subjects (nivolumab alone) eligible for primary endpoint analysis (2). The specimens from this companion trial were not analyzed for secondary and exploratory endpoints (Figure 1B).

Data Availability Statement

The clinical data generated in this study are available upon request from the corresponding author. The processed data from this study are available at NCBI GEO (GSE281729).

Results:

Between June 2019 and January 2022, 42 patients were enrolled and completed treatment at two institutions. These 42 patients were combined with the 19 nivolumab alone patients from the companion trial (NCT03854032) referenced in the methods to allow for 61 patients analyzed for the primary endpoint. Demographic and tumor characteristics were similar between those receiving nivolumab + BMS986205 (n=31) and nivolumab alone (n=30) (Table 1). For analysis of pathologic response, 4 patients could not be assessed due to withdrawing from the trial prior to surgery and proceeding with non-operative treatment due to disease progression (n=2), patient preference (n=1), and adverse events (n=1); this resulted in 38 evaluable surgical specimens. Trial schematic and waterfall plots of pathologic and radiographic responses are presented in Figure 1 and Supplemental Figure 1A-C. The addition of IDO-inhibitor to nivolumab did not result in a greater rate of radiographic response that qualified for continuation of treatment (two-tailed p=0.710): 13 (42%) in arm A qualified for additional treatment versus 14 (47%) in arm B (inclusive of control subjects from the trial NCT03854032).

Table 1.

Patient Characteristics

Characteristic	Nivolumab Only (n = 30)	Nivolumab + IDO-inhibitor (n = 31)	All (n = 61)
Age, years, median (range)	60.5 (44–86)	62.5 (45–84)	61 (44–86)
Sex, n
Male	27	23	50
Female	3	8	11
Primary site, n (%)
Oral Cavity/nasal cavity	13 (43)	11 (35)	24 (39)
Oropharynx	15 (50)	18 (58)	33 (54)
Larynx/hypopharynx	2 (7)	2 (6)	4 (7)
HPV status, n (%)
Positive	15 (50)	15 (48)	30 (49)
Negative	15 (50)	16 (52)	31 (51)
cT stage (8th edition), n (%)
T1	8 (27)	10 (32)	18 (30)
T2	12 (40)	13 (42)	25 (41)
T3	4 (13)	4 (13)	8 (13)
T4	6 (20)	4 (13)	10 (16)
cN stage (8th edition), n (%)
N0	10 (33)	9 (29)	19 (31)
N1	14 (47)	17 (55)	31 (51)
N2 (HPV+ Oropharynx)	2 (7)	1 (3)	3 (5)
N2a	1 (3)	1 (3)	2 (3)
N2b	2 (7)	1 (3)	3 (5)
N2c	0 (0)	2 (6)	2 (3)
N3	1 (3)	0 (0)	1 (2)
Smoking Status, n (%)
Never	10 (33)	12 (39)	22 (36)
Former	15 (50)	11 (35)	16 (26)
Current	5 (17)	8 (26)	23 (38)
Alcohol Use, n (%)
None/light	23 (77)	25 (81)	48 (79)
Moderate/heavy	7 (23)	6 (19)	13 (21)

Response-adaptive surgical timing demonstrated reliability in differentiating pathologic responders versus non-responders

In the 38 evaluable patients for pathologic treatment response (pTR) who were enrolled on this clinical trial protocol, response adaptive surgical timing was a reliable predictor for patients who would also respond pathologically, with 14/17 (82%) of radiographic responders also demonstrating overall pathologic response (>20% pTR) as compared to only 9/21 (43%) of radiographic non-responders demonstrating overall pathologic response (Fischer’s exact p=0.020) (Figure 1C,D). The one non-radiographic responder who received one dose of cycle of treatment was found to have a profound pathological response demonstrated by pCR at primary site and 65% response at 1 lymph node, but 0% response at 2 other lymph nodes. This discordant response may or may not have gone on to complete pathologic response had a second dose of anti-PD1 been given. The remainder of patients with a discordant pathologic response between primary site and lymph node demonstrated an overall response from 20–85% and had gone on for 2 doses of ICI.

Of the total 38 evaluable patients for pTR, 23 (61%) had demonstrated overall pTR (defined as ≥20% pathologic response averaged between response at primary site and lymph node site(s) [if applicable]). Three (8%) patients demonstrated pathologic complete response (pCR) and 20 (53%) were partial pathologic responders (pR; defined as 20–99% pTR). Six patients (16%) were minor pathologic responders (pMinR; defined as 1–19% pTR) and 9 (24%) were non-responders with a pTR of 0%. At the primary site alone, 7 (18%) were pCR, 8 (21%) were pR, 12 (32%) were pMinR and 11 (29%) were pNR. Three radiographic responders (19%) had overall pCR compared to 0 (0%) radiographic non-responders. In Arm A, 11 (41%) patients demonstrated pTR at the primary site, compared to 4 (36%) in Arm B (p=1.00). There were likewise similar rates of overall pTR between the two treatment arms (A: 17 [63%], B: 6 [55%], p=0.721). HPV⁺ status was associated with radiographic response by both this trial’s volumetric analysis (Fisher’s exact p=0.004, Figure 1C-D) and per RECIST criteria (Fisher’s exact p<0.05, Figure 1E). In radiographic responders, mean overall pTR was 58.5% compared to 15.9% in radiographic non-responders (p<0.001). HPV status was not associated with pathologic response (Fisher’s exact p=0.325).

Nivolumab + IDO inhibitor BMS986205 combination is well tolerated and tumors with increased IDO RNA expression at baseline had a significantly greater pathologic response.

Nivolumab with and without BMS986205 was well-tolerated with the exception of 2 subjects experiencing grade 3 immune-related adverse events (IRAEs) in the nivolumab + BMS986205 arm. No patients experienced grade 4–5 treatment-related AEs. IRAEs and non-IRAEs are described in Table 2. In the nivolumab + BMS986205 group, one patient experienced a grade 3 hepatic IRAE requiring hospitalization and delay in surgical treatment by 28.7 weeks. A different patient experienced grade 3 colitis (which occurred several months after surgery and therefore did not in surgical delay) which resolved with high dose steroids and did not recur after completion of neoadjuvant treatment. Eight (27%) patients in the nivolumab only and 11 (35%) in the IDO inhibitor group experienced an IRAE (p=0.58); 16 of 30 (53%) patients in the nivolumab only and 17 of 31 (55%) in the IDO inhibitor group experienced a non-IRAE (p=1.0). There were two appreciable cases of wound complications after surgery in the nivolumab only group.

Table 2.

Adverse Events

	All (n=61)	Nivolumab Alone (N=30)		Nivolumab + IDO-inhibitor (N = 31)		p-value
Adverse Events, n (%) Immune-related		Grade 1/2	Grade 3	Grade 1/2	Grade 3
Pulmonary	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	-
Dermatologic	14 (23)	8 (27)	0 (0)	6 (19)	0 (0)	0.497
Hepatic	3 (5)	0 (0)	0 (0)	2 (6)	1 (3)	0.238
Renal	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	-
Endocrine	5 (8)	3 (10)	0 (0)	2 (6)	0 (0)	1.000
Arthritis	1 (2)	0 (0)	0 (0)	1 (3)	0 (0)	1.000
Mucosal	2 (3)	1 (3)	0 (0)	1 (3)	0 (0)	1.000
Gastrointestinal (GI)	1 (2)	0 (0)	0 (0)	0 (0)	1 (3)	1.000
Adverse Event, n (%) Non-immune related
Headache	3 (5)	3 (10)	0 (0)	0 (0)	0 (0)	0.113
Fatigue	15 (25)	7 (23)	0 (0)	8 (26)	0 (0)	0.823
Arthralgia/Myalgia	8 (12)	5 (17)	0 (0)	3 (10)	0 (0)	0.473
Presyncope	5 (8)	2 (7)	0 (0)	3 (10)	0 (0)	1.000
Fever/Chills	3 (8)	1 (3)	0 (0)	2 (6)	0 (0)	1.000
Wound Complications	2 (3)	2 (7)	0 (0)	0 (0)	0 (0)	0.238
GI (nausea/vomiting/diarrhea/appetite change)	16 (26)	7 (23)	0 (0)	9 (29)	0 (0)	0.613
Pruritis	2 (5)	0 (0)	0 (0)	2 (6)	0 (0)	0.492

When we interrogated the cohort in arm A receiving the IDO inhibitor BMS986205, we noted a significantly higher RNA expression of IDO1 at baseline in those primary tumors that went on to pathologic response compared to non-responders (p<0.05; Supplement Figure 2A). Arm B receiving nivolumab alone did not have this same observed difference. In retrospective analysis the RNA expression of IDO1 at baseline trended towards predicting radiographic response in the nivolumab + BMS986205 but did not reach statistical significance (p=0.059). Interestingly, there was no difference in tryptophan kynurenine pathway signature scores between any of the patient groups (Supplement Figure 2B). Moreover, further analysis revealed a trend toward downregulation of this pathway in responders on nivolumab + BMS986205; however, on average for all other patients the tryptophan kynurenine pathway was upregulated on therapy (statistically non-significant) (Supplement Figure 2C).

Cohort landscape shows differences in HPV⁺ and HPV^– tumor microenvironment

Transcriptional profiles of tumor samples from the TJU cohort were compared to the TCGA HNSCC cohort(n=588) for additional quality control. Clustering analysis showed no technical batch effect in these two studies (Figure 2A). However, it revealed a clear separation of HPV⁺ and HPV^– samples from both cohorts. These differences in expression profiles can be partly explained by the different oncogenesis of these cancer subtypes. Moreover, these tumors were predominantly found in different locations, with the majority of TJU HPV⁺ and HPV^– tumors in the oropharynx (94%) and the oral cavity/hypopharynx (82%), respectively, as anticipated (Figure 2B). According to cell type deconvolution analysis, the TME of HPV⁺ and HPV^– samples was significantly different (Figure 2C). In HPV⁺ tumor procurement, we were careful to not include any non-cancerous surrounding lymphoid tissue that could skew resident tumor immune cell complement. While HPV⁺ samples were enriched with immune cells, including predominantly B cells, CD8+ and CD4+ T cells (Figure 2C-E), the TME of HPV^– samples was characterized by fibroblasts, endothelial cells, and neutrophils (Figure 2C,F,G). Further analysis of correlatives was performed separately on HPV⁺ and HPV^– sub-cohorts (using pooled data from both treatment arms), and the obtained results supported this distinction.

Figure 2. Differences in the TME of HPV^– and HPV⁺ HNSCCs.

A: Transcriptional profiles of TJU HNSCC and TCGA HNSCC (n = 588) cohorts with color-coding for HPV status. B: Tumor subsites for HPV⁺ and HPV^– TJU HNSCC samples. C: Cell type deconvolution landscape of TJU HNSCC cohort. Each sample is represented by a bar composed of major cell types predicted to be enriched in its tumor microenvironment (TME). The gray color shows tumor tissue plus all other unpredicted cell types. D-H: Percentages of various cell types that are differentially enriched in the TME of HPV⁺ and HPV^– samples (total n = 655). *p < 0.05**p < 0.01, ***p < 0.001.

HPV^– HNSCC have baseline immune gene expression profiles that correlate with overall pathologic response

Differential expression analysis comparing HPV^– pre-treatment samples of pathologic responders and non-responders to nivolumab therapy revealed a number of genes with log-fold change including Serping1, PDCD1LG2, IFIT3 and IFIT5. These genes expressed in our responders have the capacity to react favorably to ICI treatment; it is unknown, however, the overall effect of these upregulated genes beyond reporting this observation (Figure 3A). Gene expression of PDL2 was upregulated the most in responders at baseline (FDR<0.001), followed by upregulation of PDL1 and genes coding for interferon-induced proteins (FDR<0.05) (Figure 3B-D). Evaluation of known weakly associated biomarkers, including PDL1 and tumor mutational burden (TMB), revealed a significant association between PDL1 expression and response in the HPV^– cohort (p <0.05) (Figure 3D). However, we did not detect any correlation of TMB with overall response to nivolumab in this cohort (Figure 3E). Additional signature-based biomarkers showed significant differences in the HPV^- cohort between response and non-response including MHCI, MHCII, T-cell trafficking, NK cells, antitumor cytokines, checkpoint inhibitors, macrophages and dendritic cell trafficking (Supplement Figure 3A).

Figure 3. PDL1, PDL2, and interferon pathway as biomarkers of response to Immune Checkpoint Inhibition in HPV- HNSCC.

A: Volcano plot representing differential expression analysis comparing baseline samples of responders and non-responders in the HPV^– HNSCC samples (n = 17, 8 responders and 9 non-responders). Each gene is shown as a dot; the X axis shows log(Fold Change) of expression levels compared to non-responders; the Y axis shows −10log(False Discovery Rate). Statistically significant genes (FDR < 0.1) are colored blue. B-E: Comparison of PDL2 expression (B), IFNG signature score (C), PDL1 expression (D), and tumor mutational burden (TMB) (n=14 as three patients were excluded from whole exome sequencing due to <20% viable tumor cells in their sample)(E) values between baseline samples of responders (R) and non-responders (NR) in HPV^– samples (n = 17, except for TMB analysis n = 14 due to strict sample quality filtration for mutation calling). - ns, *p < 0.05, **p < 0.01, ***p < 0.001.

Cellular deconvolution was utilized with the Kassandra algorithm to compare the TME of baseline samples between responders and non-responders to nivolumab. HPV^– patients with an increased amount of immune cells in the TME were found to be likely responders to ICI therapy (Figure 4A). Cell populations that differed significantly included macrophages, NK cells and Tregs. Evaluation of the TME composition of post-treatment samples revealed significant enrichment of CD8+ T cells and NK cells compared to baseline (Figure 4B-C) (10). Without knowing the phenotype of these cells and their relation to epitope/TCR connection, it is not possible to ascertain the active cytotoxic effect of these cells in the environment. Interestingly, we found that the macrophages population increased in non-responders after treatment (p < 0.05, Figure 4D).

Figure 4. Cellular biomarkers of response to nivolumab.

A: Heatmap of TME deconvolution (median-scaled) representing major cell types in HPV^– HNSCC baseline samples (n = 17). Populations that significantly differ between responders and non-responders are marked by stars. B-D: CD8+ T cells (B), NK cells (C), and macrophages (D) content compared between baseline and post-treatment samples of responders (R) and non-responders (NR) in the HPV^– HNSCC cohort (n = 34). E: Heatmap of ​​​TME-associated expressional signature scores in baseline samples ​(n = 34) and segregated based on TME classification ​(Immune-enriched fibrotic, immune-enriched non-fibrotic, fibrotic and immune desert) demonstrating the immune-desert TME as associated with non-response.

In addition, primary tumor samples (n = 36) were classified into four TME subtypes using Bagaev et. al’s transcriptomic-based TME classification based on functional gene expression signatures (Fges): Immune-Enriched, Non-fibrotic (IE); Immune-Enriched, Fibrotic (E/F); Fibrotic (F); Immune Desert (D) (11). All 9 patients with immune-desert TMEs showed no primary site response (p = 0.03) and only one of these 9 patients showed overall response (p = 0.003, Figure 4E). In contrast, 4 out of 5 patients with an immune-enriched TME showed both primary site and overall response
(p=0.002, Figure 4E). Interestingly, patients with a fibrotic TME showed no primary site response but showed overall response at distant sites (Figure 4E). None of these associations were discovered in the HPV⁺ HNSCC cohort (Supplement Figure 3B).

HPV⁺ pathologic responders demonstrate B cell maturation and unique anti-inflammatory CAF subtype differentiation during treatment.

HPV⁺ tumors originating from tonsil tissue, by definition, would not demonstrate any immune desert TME subtypes. As expected, analysis of individual gene expression sourced from signature-based biomarkers demonstrated no correlation with response to ICI in this cohort (Supplement Figure 3B). However, unbiased longitudinal post- versus pre-treatment gene set enrichment analysis (GSEA) revealed cell type, bioprocess, and pathway signatures in the HPV⁺ group predicting response (17). First, in pathologic responders, we observed a shift from an immature pro-B gene set enriched in the pre-treatment samples (Normal Enrichment Analysis [NES]=−3.19, False Discovery Rate [FDR]=0.00) toward a mature B cell enrichment with downstream antigen-activated second messengers, post-treatment (NES=2.14, FDR=0.00) (Figure 5A, left; Supplement Table 2A-B). This was observed even in responders that received only one dose of Nivo-IDO (Supplement Table 2C). This anticipated B-cell maturation acting as antigen presenting cells, thereby flooding the TME with proinflammatory Th1 polarizing cytokines and chemokines. Venn diagram analysis identified 58 genes unique to responders and 106 genes unique to non-responders of the genes at the leading edge of GSEA enrichment (Figure 5A, right top; for list of genes, see Supplement Table 3A). The 58 unique responder genes were analyzed functionally using STRING, which revealed significant enrichment pre-treatment for ‘cell cycle’ (FDR=9.78E-26) (Figure 5A, right top; for list of enrichments and FDR values, see Supplement Table 4A) (18). Venn diagram analysis of the second messenger gene set revealed 17 genes unique to non-responders and 16 genes unique to responders (Figure 5A, right bottom, enlarged in Supplement Figure 4A; Supplement Table 3B). STRING analysis of the unique genes in responders were significantly enriched post-treatment for ‘antigen activates B cell receptor signaling,’ ‘FCGR dependent phagocytosis,’ and ‘intracellular signaling by second messengers’ (FDR range: 1.05E-21 to 2.27E-05) (Figure 5A, right bottom; Supplement Table 4A). Conversely, non-responders demonstrated enrichment in the pro-B gene set post-treatment (NES=2.16, FDR=0.00), while enrichments for more mature B cell (BCR signaling) and antigen-activated second messenger gene sets manifested pre-treatment (pro-B: NES=–1.33, FDR=0.05, and second messenger: NES=–2.82, FDR=0.00) (Figure 5A, left).

Figure 5: Bulk RNAseq analysis in HPV-positive patients.

A: Schematic of B-cell maturation from pro-B to more mature B-Cell and downstream second messengers with post- vs pre-treatment GSEA at equivalent stages of maturation. Venn diagrams of non-responder (NR) and responder (R) leading edge genes of Pro-B (right top) and Second Messenger (right bottom) gene sets to identify genes unique to responders are displayed. STRING 12.0 was performed with the unique leading-edge genes in responders. Nodes of different colors demonstrate an enrichment with their associated FDRs. Genes associated with significant enrichment: cell cycle (teal) in the Pro-B gene set; antigen activates B cell receptor (red), FCGR dependent phagocytosis (blue), and intracellular signaling (green) in the second messenger gene set. B: GSEA of NR and R of the CAF2 gene set, post-treatment. Unique leading-edge genes of the NR and R CAF2 gene set were determined by using a Venn diagram. Leading edge genes were then analyzed by using STRING. CAF2 NR demonstrated enrichment of IL-4 and IL-13 signaling (red), and integrated stress response signaling (blue). In contrast, CAF2 Rs were enriched for collagen-containing extracellular matrix (green).

Supervised GSEA analysis of bulk RNA sequencing data via query with cancer associated fibroblast (CAF) gene sets (Supplement Table 2D-E) revealed a CAF2 gene set whose phenotype evolved during treatment. Post treatment, both responders and non-responders showed an enrichment of the CAF2 gene set (HPV⁺ responders NES=3.63, FDR =0.00; HPV⁺ non-responders NES=1.8, FDR=0.00, respectively) (Figure 5B). To further characterize these leading edges of enrichment a Venn diagram analysis was carried out (Figure 5B, bottom, enlarged in Supplement Figure 4B). This identified leading edge genes unique to responders (n=60 genes) and non-responders (n=29 genes). Thus, an evolving CAF2 subset was revealed. STRING network analysis of the CAF2 leading edge genes illustrates treatment-induced effects (Supplement Table 3C). Post-treatment, non-responders revealed significant cytokine signaling pathways of IL-4 and IL-13; IL-2; TGFβ; and IL-18 and integrated stress response (19). This is consistent with inflammatory CAF phenotype. The TME is a competitive space with cytokine expression in various levels that ultimately lead to a dominant signal that can drive biology. In this case, the pro-tumor cytokines of IL-4 and IL-13 have expression levels that may point to a contribution to the non-response. In contrast, the responder CAF2 gene set leading edge consisted of genes involved in tissue remodeling, fibrosis, and collagen-containing extracellular matrix (for FDRs, see Supplement Table 4B).

Discussion

Neoadjuvant nivolumab +/− IDO-inhibitor was well tolerated and demonstrated a wide variety of pTRs, ranging from 0–100% overall response rates. The addition of IDO-inhibitor was not sufficient to augment an ICI response to the level of statistical significance for our analysis with a primary outcome based on radiographic response between the two arms. However, analysis of pTR in those who received IDO-inhibitor BMS986205 showed a significant difference in pTR in patients with a high baseline level of IDO1 (Supplement Figure 2A). The result of this trial does not promote further investigation into IDO inhibitor as a universal strategy to improve anti-PD-1 therapy. The utility of IDO-inhibition, however, may find itself useful in a future state, whereby resistance to ICI therapy can be measured at the point of care and a therapeutic strategy tailored to attack this node of resistance.

We developed a novel form of neoadjuvant trial design called response-adaptive surgical timing. In this model, we utilized a patient-centered set of criteria based on observations of neoadjuvant immunotherapy trials conducted at our institution which facilitates biomarker discovery and utilizes 3-axis volumetric analysis of lymph node and primary to identify candidates who would benefit from additional neoadjuvant immunotherapy treatment (16). RECIST and iRECIST are limited in their consideration of discordant response. In comparison to RECIST, our criteria identified 44% of patients who qualified for a second cycle while RECIST identified 38% who qualified to proceed. Using our criteria, 81% of the patients identified as responders had evidence of pTR compared to 84% using RECIST criteria. Importantly, of the 23 pathologic responders in our cohort, only 9 (39%) were categorized as radiographic non-responders using our criteria compared to 11 (48%) using RECIST (Figure 1E). Thus, our trial’s more aggressive radiographic criteria for response identified a greater proportion of pathologic responders than RECIST, with no additional false positives. In trials where the interval of drug effect is short, such as noted here, this 3-axis approach allows for a sensitive measure of response or growth. We do not present data here that compares iRECIST to our tailored 3-dimensional analysis, but this warrants additional investigation.

The effects of single versus multiple doses of neoadjuvant ICI have yet to be established. Recent clinical trials demonstrate a wide range of patients achieving pTR even after 1 dose (1–6). The ongoing KEYNOTE-689 is currently investigating the effects of two doses of neoadjuvant pembrolizumab followed by surgical resection then standard of care and adjuvant pembrolizumab (20). By utilizing response-adaptive surgical timing in the presented study, radiographic responders were able to receive a second neoadjuvant dose of treatment with 82% of this cohort achieving pTR. Ultimately, however, this trial was not designed to identify if a second dose of treatment could convert a radiographic non-responder to a responder but was rather intended to enhance treatment to responders and limit treatment to non-responders.

Historically, anti-PD-1 immunotherapy was prescribed as a predominant T-cell mediated treatment. Our findings presented here, as well as those of others, implicate a role for tumor infiltrating B cells and tertiary lymphoid structures (21). We noted that HPV⁺ tumors had a significant greater number of B-cell population in the tumor perhaps due to the tumors arising from a lymphoid structure of the tonsil, yet their role in ICI treatment is not well defined (Figure 2D). We further characterized differences in B-cell phenotype that correlates with response to immunotherapy. Ruffin et al. found HPV⁺ patients have transcriptional signatures consistent with germinal centers with tumor infiltrating B cells and spatial organization consistent with TLS leading to favorable outcomes in overall treatment (15,17). Our pro-B cell population enriched in HPV⁺ pretreatment future responders are a naïve population that can give rise to an adaptive immune response during ICI treatment. GSEA and STRING analysis supports a shift to a more adaptive immune response with treatment in responders. STRING analysis of post-vs-pre-treatment groups alludes to a B-cell phenotype that is dynamic in the presence of ICI treatment. With the B-cell complement critical for long term immunity, identifying co-stimulatory surface interactions with other cells, or development of class II antigen processing, would help further elucidate the B-cell role. If these mechanisms are creating long-term memory, where does it reside and if we interrogate 5–10 years out from neoadjuvant ICI, is there evidence of this memory (22)?

This finding further substantiates possible downstream functions of the enriched B-cell second messenger system. Our non-responder HPV⁺ group demonstrated an enriched immature B-cell population post-treatment, which may explain a less adaptive and more fixed system like findings of BCR repertoire in the periphery with a dominant clone (21). Further investigation into this compartment is critical for understanding resistance to ICI neoadjuvant treatment in our HPV⁺ cohort with a baseline dominant B-cell population.

This trial presents data from 20 HPV⁺ HNSCC patients treated with nivolumab +/− an IDO-inhibitor, the largest cohort to-date. While there was no differential pathologic response observed in HPV⁺ versus HPV^– cohorts, our data demonstrates characteristics in the pre- and post-treatment TME that may predict for pTR in both HPV^– and HPV⁺ cohorts. First, we did not observe any significant differences in TMB in the patients, regardless of HPV status. Importantly, the mutational profile landscapes of our trial patients aligned with the TCGA cohorts with regard to HPV status (Figure 3A). Key differential biomarkers identified in pre-treatment samples include PDL-1, IFN-γ, NK cells, and macrophages. Furthermore, as indicated by a differential increase in the NK cell population post-treatment, response to nivolumab + IDO-inhibitor may be the result of NK cell reprogramming, an effect observed only in responders. In addition, macrophages play numerous roles in cancer evolution that are dependent on disease phenotype, including promotion of angiogenesis, induction of invasiveness and metastases, regulation of TME, and induction of therapeutic resistance. Of note, we observed an increase in both M1 and M2 macrophage signatures in pre-treatment responders. Macrophage reprogramming may contribute to resistance to the nivolumab + IDO-inhibition combination as evidenced by the differential increase in macrophage populations in non-responders. Unlike changes in a single biomarker, data from this trial point to an adaptive environment that can mount a response to ICI treatment.

Higher levels of baseline PD-1 expression in lymphoid tissues may result in different mechanisms of ICI resistance in HPV⁺ and HPV^– HNSCCs. CAFs in the TME are traditionally considered a rich source of epithelial-to-mesenchymal transition (EMT) and fibrosis that are pro-tumor and a leading cause of treatment resistance. Consistent with cell state transitions and phenotypic characteristics of certain cell types, CAFs can take on distinct subsets that can give rise to inflammatory (i)CAFs that have pro-tumor effects and can impact therapeutic responses (23). GSEA in our HPV⁺ cohort identified the enrichment of CAF2 gene set phenotype in our anti-PD-1 non-responders after treatment (Figure 5B). This enrichment is informative of the type of TME in HPV⁺ HNSCC non-response to ICI. Complementary treatment strategies should consider such phenotypic differences in CAFs. In contrast, the leading-edge of the same CAF2 gene set in responders had an evolution during treatment to a more myofibroblastic (myo)CAF (extracellular matrix) phenotype, which coincided with the ICI treatment of tumor cells. In the setting of tumor cell death in responders, a natural response by fibroblasts to clean up the area of carnage results in the findings that mark our pTR (fibrosis, extracellular matrix and granulation). Our findings point toward a possible unique immune resistance due to reprogramming of the TME via components of the CAF2 gene set, and on the other hand, a remodeled CAF that can aid in restoration after response to ICI treatment (23).

While pre-treatment PD-L1 expression was associated with pTR in our HPV^– cohort, PD-L1 expression as a predictor of pathologic response to and survival benefit from ICI treatment has demonstrated mixed results in previous studies (1,9,10,24). Interplayed with this PD-L1 interaction, increased pre-treatment IFN-γ expression was also present in responders, consistent with previous reports that IFN-γ is potentially the most potent inducer of PD-L1 (25). However, we did not detect any associated increase in EMT signature, inconsistent with previous reports of IFN-γ inducing EMT signature and thereby increasing a tumor’s metastatic potential (26). Further investigation is warranted since activation of EMT confers cancer cells the ability to trigger local immunosuppression and contributes to resistance to immunotherapy (27). Overall, our analysis pre- and post-treatment TME is critical to our understanding of nivolumab + IDO-inhibition in modulating the TME. Our future publication will outline the primed periphery that enables influx and efflux of cells and cytokines critical to this response and TME modulation.

IDO has been reported to play a key role in cancer immune evasion via the kynurenine pathway of tryptophan metabolism, thus sparking the clinical interest in IDO-inhibition as an adjunct to neo-adjuvant ICI (28). Previous studies have investigated this combination in melanoma, myelodysplastic syndrome, lung cancer, prostate cancer, and other oncologic entities with mixed results (29–33). One example described the use of IDO inhibitor epacadostat in combination with PD-L1 inhibitor pembrolizumab in a Phase III trial for unresectable or metastatic melanoma. The combination failed to demonstrate improvement in either overall or progression-free survival compared to pembrolizumab alone (9). Our trial is novel in its utilization of IDO inhibitor BMS986205 with ICI for treating HNSCC in the neoadjuvant setting. Subgroup analysis revealed that subjects with elevated baseline IDO1 levels had improved pathologic response to the combination nivolumab + IDO-inhibitor. This improvement in response was not observed in the nivolumab alone arm. While the results of this trial do not promote further investigation into IDO inhibitor as a universal strategy to improve anti-PD-1 therapy, this critical finding reveals a potentially predictive factor for patients who would most benefit from this combination neoadjuvant regimen.

Conclusion:

We applied a novel trial design utilizing response-adaptive surgical timing to our neoadjuvant PD-1 inhibitor nivolumab alone or in combination with the IDO-inhibitor BMS986205 clinical trial limiting drug delivery to pathologic non-responders and increasing treatment window for pathologic responders. ICI treatment response in HPV⁺ HNSCC may be driven by a B cell maturation mechanism, and a myofibroblastic CAF phenotype may be a mechanism leading to resistance. ICI response in the HPV^– HNSCC cohort had a differential expression of genes that correlated to response. IDO inhibitor BMS986205 may have pathologic treatment benefit in patients with high expression of IDO1 in their baseline tumor samples, pointing to a need for identifying nodes of resistance to ICI treatment and targeting those nodes with specific therapeutics.

Statement of translational relevance.

To explore why some patients with head and neck squamous cell carcinoma (HNSCC) respond well to neoadjuvant immune checkpoint inhibitors (ICIs) yet others show resistance to them. We applied a trial design utilizing response-adaptive surgical timing to our neoadjuvant PD-1 inhibitor nivolumab alone or combined with IDO-inhibitor BMS986205, limiting drug delivery to non-responders and increasing treatment window in responders. Responders to ICI treatment were identified and their responses enhanced through additional treatment prior to surgery, while ICI non-responders pivoted to surgical treatment earlier. Further, we evaluated biologic specimens both before and after treatment to identify effective novel drug combinations and potential biomarkers predictive of ICI response. ICI response differed between human papilloma virus negative (HPV^–) versus HPV positive (HPV⁺) HNSCC. These findings will inform future trials designed to selectively identify patients with strong response to ICI treatment and add evidence surrounding pathologic response to neoadjuvant doses prior to surgery.