Characterization of the tumor immune microenvironment of sinonasal squamous cell carcinoma

Abstract

Background:

Treatment and prognosis of sinonasal squamous cell carcinoma (SNSCC) have not significantly improved despite improvements in radical therapy. Characterization of the tumor immune microenvironment (TiME) may identify patient subgroups associated with disease recurrence, and provide new biomarkers for improved patient stratification and treatment.

Methods:

The tumor immune microenvironment was quantitatively evaluated by multiplex immunohistochemistry (mIHC) in archived tissue sections from 38 patients with SNSCC, and were assessed for differences between recurrent (n=20) and non-recurrent groups (n=18). Hierarchical clustering analyses were performed to identify phenotypic TiME subgroups within the cohort, and used to compare survival outcomes.

Results:

Our mIHC analysis revealed increased T cell populations and decreased myeloid cell populations in SNSCC patients without recurrent disease, as compared to patients with recurrent disease. Within T cell subsets, there was a significantly higher percentage of granzyme B⁺, T-bet⁺, Eomes⁺ T cells, as well as higher proliferation of CD8⁺ T cells within the non-recurrent group relative to the recurrent group. Furthermore, immune cell complexity profiles of SNSCC revealed hyper- and hypo-T cell inflamed, myeloid-inflamed, B cell-inflamed and broadly hypo-inflamed subtypes not previously identified by gene expression analyses. Our study revealed that presence of either hyper- or hypo-T cell inflamed TiME subtypes were associated with increased survival outcomes as compared to broadly hypo-inflamed TiME subtypes (p=0.035 and 0.0376, respectively).

Conclusions:

The TiME of SNSCC reveals distinct subtypes which may correlate with recurrence and survival outcomes.

INTRODUCTION

Sinonasal malignancies make up 5% of head and neck cancers, with squamous cell carcinoma as the most common histologic subtype. The overall incidence of sinonasal squamous cell carcinoma (SNSCC) ranges from 35% to 58%, and SNSCC accounts for 60% to 75% of malignant neoplasms of the paranasal sinuses.¹ Although overall incidence of SNSCC has been declining over the past 30 years, treatment and prognosis have not improved significantly. Despite improvements in surgery, radiotherapy, and systemic therapy, the 5-year survival rate of patients with SNSCC remains poor, and has been reported to range from 30-50%, with an overall recurrence rate of 31-56%.^2,3 In addition, SNSCC tends to invade local structures including the orbits, oral cavity, and skull base, leading to significant functional and aesthetic morbidity following currently available treatment options.

Despite radical therapy, SNSCC has a tendency to recur, and may recur even after a 5-year disease-free period.^4,5 Several factors are known to contribute to high rates of recurrence, including late stage at presentation, anatomic considerations that can make achieving negative margins difficult and aggressive underlying histology.^6,7 Immune evasion as a mechanism of recurrence and metastatic progression of SNSCC has yet to be fully elucidated. Tumor cells in SNSCC may evade immune-mediated clearance or control by numerous mechanisms, including expression of inhibitory immune checkpoint proteins, release of pro-tumorigenic cytokines, and recruitment of T cell suppressive myeloid cells in the tumor immune microenvironment (TiME).^8,9

Critically, the TiME of SNSCC has not been deeply audited with regards to immune cell composition or functional status. Herein, we aim to fill these gaps by employing a multiplex immunohistochemistry (mIHC) platform enabling spatially-oriented quantification of leukocyte complexity and effector status in the TiME from surgical resection material. The mIHC platform has been successfully employed for deep characterization of the TiME of oropharyngeal head and neck squamous cell carcinoma¹⁴, and has been shown to be technically and economically equivalent to standard IHC.¹⁵

Using the mIHC platform, we have quantified the numbers, location and effector status of immune cells infiltrating primary and recurrent SNSCC. This work provides the first description of the TiME in SNSCC and represents the first step towards identifying potential biomarkers of disease recurrence. A more precise understanding of the TiME in SNSCC will also improve our ability to stratify and monitor patients and allow us to focus our efforts on more aggressive therapies and closer surveillance for patients at a higher risk of recurrence. Understanding the immune cell repertoire in SNSCC may also more effectively guide revision of immunotherapeutic strategies to enhance anti-tumor immune activity and result in better patient care.

METHODS

Clinical samples

Formalin-fixed paraffin embedded (FFPE) samples of previously resected SNSCC (N=38) were obtained from the Oregon Health and Science University (OHSU) Knight Cancer Institute Biolibrary. The sample population was identified using electronic chart review in October, 2019 from a heterogeneous adult (≥ 18 years of age) patient population presenting to the Division of Head and Neck Surgery in the Department of Otolaryngology-Head and Neck Surgery at OHSU. All cases were diagnosed between June, 1996 and October, 2017 with sinonasal squamous cell carcinoma (SNSCC) following cytopathology review. Patients with a history of immunotherapy, immunocompromise, or other immune altering pathologies or therapies were excluded. Cases were ascertained through ICD histology code review and staged according to the eighth edition AJCC/UIC TNM classification as shown in Table 1. Patients elected curative treatment including radiation therapy, chemotherapy, and/or surgical resection at the primary site by a fellowship trained Head and Neck surgeon at OHSU. Samples were all from primary tumor resection material, and matched by overall AJCC stage at diagnosis. Specimens were divided into those with or without recurrence per chart review, with recurrence defined as any subsequent presentation of local, regional, or distant spread of SNSCC. A total of 162 patients were diagnosed with SNSCC at OHSU during the study period. Out of this cohort, 104 patients with pathology samples available for processing were identified. Of the final cohort of 104 patients, we randomly selected 40 patients (20 with recurrent disease and 20 with non-recurrent disease) as a representative sample. Two samples were excluded from the non-recurrent group due to poor quality of the tissues. The Institutional Review Board at OHSU provided ethical oversight and approval for data collection (eIRB #19903).

Table 1:

Patient characteristics (n=38)

Characteristic:	N(%) Overall	N (%) Recurrent	N(%) No Recurrence
Age at SNSCC dx	62.5 (13.1)*	63.4 (11.8)*	60.7 (13.0)*
Female	17 (44.7)	8 (40.0)	9 (50.0)
Male	21 (55.3)	12 (60.0)	9 (50.0)
Race: White/Caucasian	35 (92.1)	17 (85.0)	18 (100.0)
Race: African American	2 (5.3)	2 (10.0)	0 (0)
Tobacco use	25 (65.8)	13 (65.0)	12 (66.7)
Alcohol (ETOH) use	18 (47.4)	11 (55.0)	7 (38.9)
Overall Anatomic Stage at dx:
Stage-1	3 (7.9)	2 (10.0)	1 (5.6)
Stage-2	3 (7.9)	2 (10.0)	1 (5.6)
Stage-3	9 (23.7)	4 (20.0)	5 (27.9)
Stage-4	23 (60.5)	12 (60.0)	11 (61.1)
Tumor location:
Nasal cavity	17 (44.7)	10 (50.0)	7 (38.9)
Maxillary sinus	18 (47.4)	8 (40.0)	10 (55.6)
Ethmoid sinus	3 (7.9)	2 (10.0)	1 (5.6)
Margin involvement of the primary site:
Positive (+) margin status	12 (31.6)	7 (35.0)	5 (27.8)
Negative (−) margin status	26 (68.4)	13 (65.0)	13 (72.2)
Perineural invasion	11 (29.0)	6 (30.0)	5 (27.8)
Treatment type:
Surgery alone	11 (29.0)	6 (30.0)	5 (27.8)
Surgery + radiation	15 (39.5)	9 (45.0)	6 (33.3)
Surgery + CRT	11 (29.0)	5 (25.0)	6 (33.3)
Surgery + chemotherapy	1 (2.6)	0 (0)	1 (5.6)
Recurrence:
No recurrence	18 (47.4)	----	----
Local	6 15.8)	----	----
Regional LN	6 (15.8)	----	----
Distant metastasis	8 (21.1)	----	----

^*Average (SD)

Multiplex Immunohistochemistry and Image Acquisition

The multiplex immunohistochemistry (mIHC) technique utilizes sequential immunohistochemical staining of FFPE tissue sections with immune lineage epitope-specific antibodies for immunodetection, and was performed as previously described^14,15 using the panel shown in Supplemental Table 1. Briefly, slides are baked and deparaffinized, followed by incubation with a primary antibody, then a F(ab’) fragment-specific secondary antibody-labeled polymer conjugated to horseradish peroxidase. Following detection, slides are then visualized using an alcohol-soluble peroxidase substrate 3-amino-9-ethylcarbazole (AEC), followed by whole-slide digital scanning. Secondary antibodies are either then inactivated and another primary produced in a different species is applied as described above, or antibodies are stripped in heated citrate buffer (pH 6.0). This allows for multiple primary antibodies produced by distinct species to be stained and detected before heat stripping, allowing greater utilization of FFPE tissue. Our mIHC protocol has been shown to be technically equivalent to standard IHC and flow cytometry.^14,15 Altogether, FFPE tissues were stained with a “discovery” panel of 23 validated antibodies for identification of 13 of distinct immune cell types and various functional states within (Figure 1).

Figure 1.

Leukocyte lineages and functionality biomarkers revealed with the ‘Discovery-23’ antibody panel.

Image Processing and Analysis

We applied a computational image processing pipeline to the acquired images in order to quantitatively analyze matched samples of recurrent and non-recurrent FFPE tissue specimens from SNSCC patients. Image processing and analysis was performed using ImageJ Version 1.48, CellProfiler Version 2.2.0, Matlab R2019b, and FCS Express 7 Image Cytometry RUO, analogous to the previously reported 12-plex mIHC image analytic workflow encompassing image processing, visualization, and quantification.¹⁵ For image processing, image co-registration was performed using the detectSURFfeatures algorithm from the Computer Vision Toolbox in Matlab. Key points were identified from the final hematoxylin image and used as target points to register each AEC stained image from all cycles and rounds such that single-cell measurements could be associated across images, as previously reported.^14,15,17,18 Following image co-registration, AEC signal was extracted using an Red Green Blue to Cyan Magenta Yellow Key (RGB to CMYK) conversion which utilizes a maximum gray component replacement to subtract the lowest brightness level from all channels. AEC chromogen signal is separated into the ‘Y’ channel and the middle 90% of pixels are rescaled to 0-255¹⁹. Hematoxylin images were used to segment nuclei for identification of single-cells. Single cell mean intensity signal measurements for every stained marker was performed using CellProfiler. All pixel intensity and shape-size measurements were saved as comma separated version files and imported in FCS Express 7 Image Cytometry. In image cytometry analysis, cell lineages were quantitatively evaluated based on the gating strategy we have reported previously to identify leukocyte lineages (Figure 1). Gating thresholds for quantitative identification were determined based on data in negative control subsets. Immune cell numbers were normalized as either cell density (cell count/mm² tissue area assessed) or percentages of total CD45⁺ cells, and subjected to unsupervised hierarchical clustering based on Ward's minimum variance method, using "pheatmap" from "R".

Statistics

Two-way ANOVA, Kruskal-Wallis test, Chi-squared test, and Wilcoxon signed-rank tests were used to determine statistically significant differences. P values were adjusted for multiple comparisons using Benjamini-Hochberg false discovery rate adjustments. Statistical calculations were performed by Graphpad Prism, version 9.1.0, and R software, version 4.0.3 (http://r-project.org/). An unsupervised hierarchical clustering was performed with Ward’s minimum variance method (pheatmap from R).²⁰ All P values < 0.05 were considered statistically significant.

RESULTS

Immune microenvironment of recurrent SNSCC exhibits myeloid predominance

The TiME of SNSCC has not been deeply audited with regards to immune cell composition or functional status. In order to investigate the TiME of SNSCC, we identified a total of 38 subjects with SNSCC to be included in this study (Table 1); 20 patients had recurrent disease, and 18 had no recurrence. Patients were followed for 5 years to monitor for signs of clinical recurrence. There were no significant differences between groups with respect to age, tumor location, margin status, or overall stage. mIHC and image cytometry analysis were performed to quantitatively identify differences in the TiME of recurrent and non-recurrent SNSCC samples.

Significantly increased density of CD45⁺CD3⁺CD8⁺ T-cells (p=0.021) and CD45⁺CD3⁺CD8⁻ helper T cells (p=0.0079) were noted in the non-recurrent group relative to the recurrent group (see Figure 3B for gating strategy). Additionally, there were significantly more CD45⁺CD3⁻CD20⁻CD66b⁺ granulocytes (p=0.025) and mononuclear phagocytes (monocytes, macrophages and dendritic cells (p<0.0001)– see Figure 1 for complete list of biomarkers) within the recurrent group relative to the non-recurrent group. No significant differences were noted within CD20⁺ B cells between groups (p=0.107) (Figure 2). These findings indicate that a myeloid predominant TiME may play a role in recurrence of SNSCC, whereas a T cell-enriched TiME may support tumor control.

Figure 3.

Significant differences within T cell subsets with relatively more granzyme⁺, Tbet⁺ and EOMES⁺, as well as Ki67⁺ in the non-recurrent group than recurrent group. No significant differences within PD1⁺, Treg, Th1 or Tfh populations between groups.

(A) Immune cell percentages were quantified as a percentage of total CD45⁺ cells. Horizontal bars and vertical lines show mean and standard error of the mean (SEM) respectively. Statistical differences were determined via two-way ANOVA using two-stage step-up method of Benjamini, Krieger, and Yekutieli with false discovery rate (FDR) adjustments, with *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

(B) Image cytometry-based cell population analyses for the lymphoid antibody panels to identify specific T cell population and functional status. The antibodies used for identification of cell lineages are shown in Figure 1. Gating thresholds for quantitative identification were determined based on data in negative controls. The x and y axes are shown on a logarithmic scale.

(C&D) Representative images of 23-biomarker mIHC to visualize immune cell phenotypes in a single FFPE section of human SNSCC with recurrence (C) and without recurrence (D). Chromogenic signal was extracted for each marker, pseudocolored and overlaid in ImageJ to simultaneously visualize nuclei, CD45+ immune cells, T cells, B cells, myeloid cells, and tumor cells. Scale bar (500mm) and colors are shown.

Figure 2.

Significant differences with more T cells (CD3⁺, CD8⁺ and CD8⁻) in the non-recurrent group relative to recurrent group. Significant differences with more myeloid cells (CD66B⁺ and total monocytes/macrophages/dendritic cells) in recurrent group relative to non-recurrent group.

(A) Immune cell percentages were quantified as a percentage of total CD45⁺ cells. Horizontal bars and vertical lines show mean and standard error of the mean (SEM) respectively. Statistical differences were determined via two-way ANOVA using two-stage step-up method of Benjamini, Krieger, and Yekutieli with false discovery rate (FDR) adjustments, with *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

(B) Image cytometry-based cell population analyses for the lymphoid and myeloid antibody panels. The antibodies used for identification of cell lineages are shown in Figure 1. Gating thresholds for qualitative identification were determined based on data in negative controls. The x and y axes are shown on a logarithmic scale.

(C&D) Representative images of 23-biomarker mIHC to visualize immune cell phenotypes in a single FFPE section of human SNSCC with recurrence (C) and without recurrence (D). Chromogenic signal was extracted for each marker, pseudocolored and overlaid in ImageJ to simultaneously visualize nuclei, CD45⁺ immune cells, T cells, B cells, myeloid cells, and tumor cells. Scale bar (500mm) and colors are shown.

Comparison of T cell functional status differences in TiME

In order to further analyze T cells within the TiME of SNSCC, we evaluated the functional status T cell populations by assessing granzyme B, Tbet, Eomes, and Ki67 (Figure 3). The T-box transcription factor T-bet controls genes crucial for regulating T helper type 1(Th1) cell lineage commitment in CD4⁺ T helper cells and effector function in CD8⁺ T cells, and Eomes regulates memory function in CD8⁺ T cells. Additionally, T-bet and Eomes have been implicated in anticancer responses, with higher expression of T-bet shown to be associated with a more favorable outcome in colorectal cancer patients.²¹ We evaluated cytotoxic activity of T cells and proliferation by staining for granzyme B and Ki67 respectively. There was a significantly higher percentage of granzyme B⁺, T-bet⁺, Eomes⁺ T cells (p=0.004), as well as higher proliferation of CD8⁺ T cells (p=0.003) as noted by a higher percentage of Ki67⁺CD8⁺ T cells within the non-recurrent group relative to the recurrent group. No statistically significant differences were noted between percentages of FoxP3⁺ T-regulatory cells (p=0.296), CD3⁺CD8⁻Tbet⁺ Th1 cells (p=0.323), CD3⁺CD8⁺PD1⁺ T cells (p=0.146) or CD3⁺CD8⁻PD1⁺ T-follicular helper cell populations (p=0.139) between groups. Of note, CD8⁺ T cells were frequently found within tumor foci in non-recurrent patients (Figure 3). Altogether, these results indicate the presence of a predominant type-1 anti-tumor immune response within T cells of the non-recurrent group that was lacking in the recurrent group.

Comparison of myeloid cell subsets

We found that myeloid cells were increased in recurrent SNSCC (Figure 2). In order to investigate whether a particular myeloid cell subset correlated with SNSCC recurrence, we analyzed several myeloid cell biomarkers within the TiME of our patient cohort. We measured the levels of the macrophage markers CD169 and CD163, dendritic cell markers CD11c cells expressing DCLAMP, monocytes expressing CCR2, and functional markers HLA-II and PD-L1 among CD45⁺CD3⁻CD20⁻CD66b⁻ CD68⁺ mononuclear phagocytes. We observed a significantly higher percentage of CD45⁺CD3⁻CD20⁻CD66b⁻CD68⁺CD169⁺ macrophages (p<0.0001), and CD45⁺CD3⁻CD20⁻CD66b⁻CD68⁺HLAII⁻ macrophages (p=0.0038) in the recurrent group than in the non-recurrent group (Figure 4). These results indicate that HLAII⁻ macrophages and CD169⁺ macrophages may possess immune suppressive activities within the TiME of SNSCC.

Figure 4.

Significant difference within CD169⁺ myeloid cells and HLA Class II⁻ macrophages with relatively higher percentage in the recurrent group relative to non-recurrent. No significant differences within other myeloid cell subsets.

(A) Immune cell percentages were quantified as a percentage of total CD45+ cells. Horizontal bars and vertical lines show mean and standard error of the mean (SEM) respectively. Statistical differences were determined via two-way ANOVA using two-stage step-up method of Benjamini, Krieger, and Yekutieli with false discovery rate (FDR) adjustments, with *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

(B) Image cytometry-based cell population analyses for the myeloid antibody panels to identify individual myeloid cell populations. The antibodies used for identification of cell lineages are shown in Figure 1. Gating thresholds for quantitative identification were determined based on data in negative controls. The x and y axes are shown on a logarithmic scale.

(C&D) Representative images of 23-biomarker mIHC to visualize immune cell phenotypes in a single FFPE section of human SNSCC with recurrence (C) and without recurrence (D). Chromogenic signal was extracted for each marker, pseudocolored and overlaid in ImageJ to simultaneously visualize nuclei, CD45+ immune cells, T cells, B cells, myeloid cells, and tumor cells. Scale bar (500mm) and colors are shown.

Assessment of B cells and tumor cells

B cells are emerging players in HNSCC, serving as antigen-presenting cells and activating T cells, and their presence in the TiME has been linked to improved patient survival in multiple cancer entities including oral squamous cell carcinoma.^22,23 To investigate whether SNSCC-infiltrating B cells may correlate with recurrence, we analyzed CD20⁺ B cell abundance and phenotypes in our patient cohort. The fraction of CD20⁺ B cells within the total CD45⁺ immune infiltrate was not different between groups (Figure 2). When we queried for expression of activation markers PD-L1 and HLA-II, atypical B cell marker CD11c²⁴ and proliferation marker Ki67 on B cells, we did not observe statistically significant differences between recurrent and non-recurrent patients (p=0.233) (Figure 5). Similarly, we did not find statistically significant differences in the fraction of CD45⁻PanCK⁺ squamous carcinoma cells within total segmented cells (p=0.672), or in the expression of PD-L1 (p=0.807), HLA-I (p=0.946) and Ki67 (p=0.250) on CD45–PanCK⁺ squamous carcinoma cells (Figure 5).

Figure 5.

No significant differences in proportions of B cell subsets or tumor cell subsets between recurrent and non-recurrent groups.

(A) Immune cell percentages were quantified as a percentage of total CD45⁺ cells. Horizontal bars and vertical lines show mean and standard error of the mean (SEM) respectively. Statistical differences were determined via two-way ANOVA using two-stage step-up method of Benjamini, Krieger, and Yekutieli with false discovery rate (FDR) adjustments, with *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

(B) Image cytometry-based cell population analyses for the lymphoid antibody panels and to identify individual B cell and epithelial tumor cell population subsets. The antibodies used for identification of cell lineages are shown in Figure 1. Gating thresholds for qualitative identification were determined based on data in negative controls. The x and y axes are shown on a logarithmic scale.

(C&D) Representative images of 23-biomarker mIHC to visualize immune cell phenotypes in a single FFPE section of human SNSCC with recurrence (C) and without recurrence (D). Chromogenic signal was extracted for each marker, pseudocolored and overlaid in ImageJ to simultaneously visualize nuclei, CD45+ immune cells, T cells, B cells, myeloid cells, and epithelial tumor cells. Scale bar (500mm) and colors are shown.

Survival implications of the TiME of SNSCC

In order to identify distinct TiME subgroups based on immune complexity profiles, we performed an unsupervised hierarchical clustering analysis based on tumor and immune cell fractions, as identified by image cytometry gating strategies (Figure 1). Although the number of patients was limited, the clustering analysis revealed the presence of five distinct sub-groups: hyper T cell-inflamed, hypo T cell-inflamed, myeloid-inflamed, B cell-inflamed and broadly hypo-inflamed (Figure 6). Based on these TiME subgroups, we performed survival analyses, and found a significant increase in overall survival of the hyper T cell-inflamed subtype as compared to the broadly hypo-inflamed subtype (p=0.035). Survival differences between either hyper or hypo T cell-inflamed and myeloid-inflamed groups (p=0.482 and 0.512 respectively) did not reach statistical difference, although trends were clearly appreciated (Figure 6). The B cell-inflamed group had too few patients for comparison but seemed to be associated with early-stage disease. These findings indicate a survival benefit for patients with increased T cell infiltration, and that mIHC of SNSCC tissues represents a viable option to keep a closer surveillance for patients at a higher risk of recurrence.

Figure 6.

Hierarchical clustering analysis reveals five distinct subgroups: hyper and hypo T cell-inflamed, myeloid-inflamed, B cell-inflamed, and broadly hypo-inflamed, and improved survival outcomes in hyper and hypo T cell-inflamed subgroups relative to hypo-inflamed subgroups. No significant survival differences seen with myeloid subgroups.

(A) Heat map indicating scaled immune cell densities (fraction of CD45⁺ or total cells) according to color scale with a dendrogram of unsupervised hierarchical clustering depicting (from top to bottom) hyper T cell-inflamed, myeloid inflamed (split into two groups), B cell-inflamed, hypo T cell-inflamed, and hypo-inflamed.

(B) Kaplan-Meier analysis of overall survival of SNSCC patients stratified by subgroups. Statistical significance was determined by log-rank test.

DISCUSSION

The management of SNSCC remains challenging and survival outcomes have not improved significantly despite progress in therapeutic approaches. Immunotherapy with inhibitors of immune checkpoint molecules, including nivolumab and pembrolizumab, has been recently approved for treatment of patients with metastatic or unresectable recurrent HNSCC (KEYNOTE-048 and CheckMate-141 trials). Still, only a portion of patients respond, while many fail to respond and/or develop resistance.^10-13 A more detailed understanding of the immunologic basis for recurrence of SNSCC will enable identification of biomarkers for earlier detection of recurrence and discovery of potential targets for therapeutic intervention. In this study, we investigated tumor immune characteristics of archival SNSCC tumors from patients with and without recurrent disease using a multiplex IHC platform optimized for assessment of immune complexity and analysis of cell phenotypes. The mIHC approach allowed for quantitative assessment of immune infiltrates based on sequential IHC and image cytometry. Our analysis revealed significant differences in TiME of recurrent and non-recurrent SNSCC which further correlate with significant survival differences.

In particular, our study uncovered increased T cell populations and decreased myeloid cell populations in SNSCC patients without recurrent disease, as compared to patients with recurrent disease (Figures 1, 2). Within T cell subsets, there was a significantly higher percentage of cytotoxic and effector/memory T cells, as well as higher proliferation of CD8⁺ T cells within the non-recurrent group relative to the recurrent group (Figure 2). These observations are in line with the known role of CD8⁺ T cells in anti-tumor immune responses.²⁵ Interestingly, CD8⁺ T cells are believed to be the target of checkpoint-based immunotherapy, and thus our findings support the use of T cell modulatory agents in subsets of SNSCC patients.

Immune cell complexity profiles of SNSCC revealed distinct signature-based subtypes not previously identified by gene expression analyses.^7,26,27 The ability to stratify treatment naïve patients according to the TiME will be important in order to improve our ability to monitor patients by focusing our efforts on more aggressive therapies and closer surveillance for those patients at a higher risk of recurrence. Hierarchical clustering analysis revealed that presence of even low levels of T cell tumor infiltrates was associated with increased survival outcomes, as compared to broadly hypo-inflamed TiME subtypes (p<0.05). The presence of myeloid cells was also a poor prognostic factor, with a 50% survival chance within 3 years from diagnosis. Interestingly, B cell-inflamed TiME seemed to associate with early-stage disease, as previously reported for other HNSCC types^28,29 (Figure 6), which warrants further investigations on the spatial organization of TLS in SNSCC. Altogether, our results substantiate prior reports of the important role of T cell-mediated antitumor immunity.^15,21,30,31

Although we observed significantly increased myeloid cells, and specifically, increased macrophage and neutrophils within the TiME of recurrent SNSCC relative to non-recurrent SNSCC, we did not observe a significant survival difference between myeloid inflamed and either hyper- or hypo-T cell inflamed subgroups. Prior reports within head and neck squamous cell carcinoma have found decreased survival and increased disease aggressiveness within myeloid-inflamed TiMEs.^15,17 Within our myeloid-inflamed subgroup, although this did not reach significance, there was a trend towards decreased survival relative to hyper- or hypo T cell inflamed subgroups. This result may largely be limited by smaller sample sizes in our analysis. Overall, these results suggest a myeloid-inflamed TiME may be associated with recurrence in SNSCC, but further evaluation may be needed to clarify its association with survival.

This study represents one of the first characterizations of the TiME of SNSCC. However, there are inherent limitations to this study. Due to the relative rarity of SNSCC we chose a retrospective analysis to power the study. In order to match stage groups between recurrent and non-recurrent disease, and due to a limited number of early stage recurrent SNSCC at our institution, there were fewer early stage SNSCC patients represented in this cohort. Despite these limitations, our study provides an initial characterization of the TiME of SNSCC and reveals previously unreported phenotypes with significant clinical and survival implications.

In conclusion, our findings support the tenant that immune contexture can be effectively used as a metric to predict clinical outcomes and responses to therapy, and will enable improved patient surveillance. A better understanding of immune contexture will provide the basis for identifying additional targets for immunotherapeutic interventions.