Effects of tobacco smoking on the tumor immune microenvironment in head and neck squamous cell carcinoma

Abstract

Purpose:

Patients with head and neck squamous cell carcinoma (HNSCC) who actively smoke during treatment have worse survival compared to never-smokers and former-smokers. We hypothesize the poor prognosis in tobacco smokers with HNSCC is, at least in part, due to ongoing suppression of immune response. We characterized the tumor immune microenvironment (TIM) of HNSCC in a retrospective cohort of 177 current, former, and never smokers.

Experimental Design:

Tumor specimens were subjected to analysis of CD3, CD8, FOXP3, PD-1, PD-L1, and pancytokeratin by multiplex immunofluorescence, whole exome sequencing and RNA sequencing. Immune markers were measured in tumor core, tumor margin, and stroma.

Results:

Our data indicate that current smokers have significantly lower numbers of CD8⁺ cytotoxic T-cells and PD-L1⁺ cells in the TIM compared to never- and former-smokers. While tumor mutation burden and mutant-allele tumor heterogeneity score do not associate with smoking status, gene-set enrichment analyses reveal significant suppression of IFN-α and IFN-γ response pathways in current smokers. Gene expression of canonical IFN-response chemokines, CXCL9, CXCL10, and CXCL11, are lower in current smokers than in former smokers, suggesting a mechanism for the decreased immune cell migration to tumor sites.

Conclusion:

These results suggest active tobacco use in HNSCC has an immunosuppressive effect through inhibition of tumor infiltration of cytotoxic T-cells, likely as a result of suppression of IFN response pathways. Our study highlights the importance of understanding the interaction between smoking and TIM in light of emerging immune modulators for cancer management.

INTRODUCTION

It is estimated that there are 1.2 billion smokers over the age of 15 years globally (1). Tobacco smoking is a major cause of head and neck squamous cell carcinoma (HNSCC) which remains a significant cause of morbidity worldwide with approximately 400,000 new cases per year (2). It is well established that HNSCC patients with a significant tobacco smoking history have poorer prognosis compared to never-smokers (3,4). Furthermore, patients who actively smoke during treatment have worse survival compared to former-smokers, suggesting smoking has acute effects beyond the cumulative effects of genomic alterations associated with extensive smoking history on molecular progression of HNSCC from premalignant lesions to invasive disease (5–8).

Increasing evidence suggests the tumor immune microenvironment in HNSCC is immunosuppressive, and the tumor-mediated immunosuppression may play a pivotal role in HNSCC progression and treatment resistance (9–11). In addition, several studies have indicated the numbers of infiltrating immune cells reflecting antitumor immune response in HNSCC are predictive of overall survival (12–14). The importance of understanding the interaction between smoking and tumor immune microenvironment (TIM) is further highlighted by emergence of immune checkpoint inhibitors as a promising therapeutic option in management of HNSCC, as smokers are less likely to benefit from the anti-PD-1 checkpoint inhibitors (11,15–17). It is currently known that HNSCC evades immune surveillance by deregulating four key signaling steps for robust antitumor immunity: decreased TCR:HLA-peptide antigen interactions, increased immunosuppressive co-stimulatory (or co-inhibitory) signals, increased immunosuppressive type 2 cytokines, and decreased chemokines to recruit cellular immune populations into the TIM and amplify antitumor immunity (18). Each of these deregulated signals contributes to localized immunosuppression and unrestrained tumor growth in the TIM.

We hypothesize the poor prognosis in current smokers with HNSCC is, at least in part, due to ongoing suppression of immune response by active tobacco use. Our study reports that active tobacco use is associated with an immunosuppressive signature as observed by lower tumor infiltration of cytotoxic T-cells in concert with reduced expression levels of IFN response pathways. Our study highlights the importance of understanding the interaction between smoking and TIM in HNSCC.

MATERIALS AND METHODS

Patient selection and clinical data

HNSCC patients were identified through three IRB approved studies: Total Cancer Care (MCC#14690) with a tissue diagnosis of HNSCC and obtained written consent, Epidemiology of Head and Neck Cancer Study (MCC#17041, written consent waived), and Evaluation of The Tumor and Its Microenvironment in Head and Neck Cancer Patients (MCC#18754, written consent waived). Institutional IRB approval was obtained in accordance with the Department of Health and Human Services Federal Policy for the Protection of Human Subjects (US Common Rule). The study was initiated after the IRB approvals. From these studies, a cohort of 177 cases with human papillomavirus (HPV)-negative squamous cell carcinoma of the oral cavity, pharynx, and larynx seen at Moffitt Cancer Center between 2006 and 2016 were included (Supplementary Table 1 and 2). Clinical data were extracted from medical records and included patient demographics, smoking history, treatment history, and disease outcome. Specimens available for study were derived from primary tumors, second primary tumors, recurrent tumors or recurrent tumors from second primary tumors (Supplementary Table 1). The second primary tumors were separated from the original primary tumors by time (>10 years), tumor subsite, or histology. Expert review (JHP and CHC) of clinical and pathology records was performed to assess the origin of each of the specimens. Smoking history was classified into three categories: never-smokers having consumed less than 100 lifetime cigarettes, former-smokers who smoked more than 100 lifetime cigarettes and quit at least 12 months prior to the date of diagnosis, and current-smokers who smoked more than 100 cigarettes within the last 12 months.

Multiplex Immunofluorescence (mIF) staining

The mIF staining was performed using the autostainer BOND-RX (Leica Microsystem, Vista, CA), with Opal 7-color kit (PerkinElmer, Waltham, MA) and detected using the Bond Research Detection System 2 kit (Leica Biosystems, Buffalo Grove, IL). This kit uses individual tyramide signal amplification-conjugated fluorophores to detect the stained targets. Antigen retrieval was accomplished using Bond Epitope Retrieval 1 (pH 6.0) or 2 (pH 9.0) (Leica Biosystems, Buffalo Grove, IL). Archival formalin-fixed, paraffin-embedded (FFPE) tissue blocks were obtained from Moffitt Surgical Pathology and used to cut sequential 4-μm thick sections for mIF. Antibodies against the following were used for staining: CD3 (clone F7.2.38, dilution 1:100, Dako, Carpinteria, CA), PD-1 (clone D4W2J, dilution 1:200; Cell Signaling Technology, Danvers, MA), CD8 (clone C8/144B, dilution 1:100, Dako, Carpinteria, CA), PD-L1 (clone E1L3N, dilution 1:100; Cell Signaling Technology, Danvers, MA), FOXP3 (clone 236A/E7, dilution 1:200, Abcam, Cambridge, MA) and pancytokeratin (clone AE1/AE3, dilution 1:600, Dako, Carpinteria, CA) and DAPI for nuclei. Optimal dilutions were chosen based on the specific cell expression, background, uniformity, and pattern of the staining by an expert head and neck pathologist (JHP). Following the manufacturer’s instructions, Opal 7-colors kit (same order as antibodies) Opal520 (green), Opal 540 (pink), Opal 570 (yellow), Opal 620 (Orange), Opal 650 (red), and Opal 690 (Aqua blue) was used the dilution 1:150. The Opal 7-colors kit (Opal 520, Opal 540, Opal 570, Opal 620, Opal 650, and Opal 690) was used according to manufacturer’s instructions (PerkinElmer, Waltham, MA). A positive control was used for each protein as follows: human tonsil for CD3, CD8, PD-1, FOXP3, and pancytokeratin AE1/AE3 and human placenta for PD-L1. A negative control slide for autofluorescence was included and stained with primary and secondary antibody but omitting fluorochrome-tyramide.

Visualization and data analysis

Multiplex staining slides were imaged using the Vectra 3.0 spectral imaging system (PerkinElmer, Waltham, MA). Fluorescence intensity information was extracted using the fluorescence protocol at 10 nm λ from 420 nm to 720 nm. A similar method was used to build the spectral library using the InForm 2.2.1 image analysis software (PerkinElmer, Waltham, MA). For the multiplex immunofluorescence slides, each set of slides was scanned with the Vectra imaging system. After scanning the full image at low resolution, nine regions of interest (ROIs) from each slide were chosen by a pathologist (JHP) using Phenochart 1.0.4 (PerkinElmer, Waltham, MA), 3 in the tumor core (TC), 3 at the tumor margin (TM) and 3 outside in the adjacent stroma (S) area. The size of the ROIs was standardized at 1356×1012 pixels with a resolution of 0.5 μm/pixel for a total surface area of 0.343 mm². The TM area was chosen to cover the tumor/stroma interface such that approximately half of the image covered the tumor region and the other half covered the surrounding stromal region.

The data from Vectra were accessed by the InForm imaging software, and each individual staining channel was combined using the spectral library information to link each fluorochrome component with a mIF component. Immune cell populations were characterized and quantified using the cell segmentation and phenotype cell tool of the InForm image analysis software under pathologist supervision (JHP). The InForm cell recognition tool was trained on a tonsil sample for all markers, except for the use of placenta for PD-L1 staining. Cell-level data was collected in a single InForm batch run and summarized using the phenoptr package for R (19). Cells positive for CD8 or FOXP3 were also required to be positive for CD3 to be counted in the phenotyping process, but they will be referred to as CD8⁺ and FOXP3⁺ in the text. To reduce variability due to low cell counts, we summed cells for each phenotype across the three ROI replicates for each of the tissue classes (tumor core, tumor margin and stroma). To allow distinction between PD-L1⁺ lymphocytes and PD-L1⁺ tumor cells, we classified all PD-L1⁺ cells as either PCK⁺ or PCK⁻ using average intensity of PCK measurements. The number of marker-positive cells per ROI was used as our primary endpoint for all marker comparisons. All statistical analyses were performed using R Project for Statistical Computing (version 3.4.3) and p-values less than 0.05 were considered statistically significant.

DNA/RNA sequencing and data processing

Whole-exome and RNA-sequencing on all tumor and matched normal samples were performed at the HudsonAlpha Institute for Biotechnology. DNA was extracted from germline samples and fresh frozen tissues using the Qiagen QIASymphony method. Total RNA from tumor tissues was extracted using Qiagen RNAeasy plus mini kit. DNA and RNA from all FFPE samples were extracted using the Qiagen All prep FFPE DNA/RNA kit. Each exome was captured using Roche SeqCap EZ Exome v3.0 kit and sequenced on an Illumina HiSeq 4000 platform to a depth a ~100x for germline and ~300x for tumor tissues. RNA libraries were prepared using the Illumina TruSeq RNA Exome protocol and kit reagents and further sequenced on an Illumina HiSeq 4000 with 100 million total reads per sample (50 million paired reads). For DNA exome sequencing analysis, the trimmed and filtered reads were aligned to human genome (hs37d5) using BWA (0.7.7) (20) and further refined with Picard/GATK tool (21). Somatic mutation detection was then performed using software MuTect (1.1.4) (22) and Strelka (1.0.13) (23). The detected variants were annotated by software ANNOVAR (24). For RNAseq data processing, Cutadapt (1.16) was used to trim off adaptor containment sequences and low-quality bases at the ends. Trimmed and filtered reads were then aligned to the reference transcriptome (hs37d5) using STAR (2.5.3a) (25). Gene-level quantification was conducted using HTSeq-count (26). Raw count data was normalized and further analyzed using DESeq2 package (27). Gene expression values were log2(1+x) transformed for downstream analysis.

Tumor mutation burden (TMB) was calculated by counting mutations in each sample meeting the following criteria: observed as PASS in Strelka or in both Strelka and MuTect, located within the targeted regions, and absent in 1000 Genomes. Mutation counts were divided by target region size (63.6MB) giving mutations per megabase targeted. Tumor heterogeneity was quantified using MATH (Mutant-Allele Tumor Heterogeneity) score (28), which was calculated with R package Maftools (29) based on the mutation annotation format (MAF) files processed from the WES analysis pipeline. Gene set enrichment analysis (GSEA) was performed using the GSEA toolkit version 4.0 (30) against Hallmark of Cancer, Oncogenic Signatures, and Immune Signatures. False discovery rate (FDR) and p-value for the enriched pathways were estimated by performing 1000 gene-set permutations, and gene sets with FDR < 5% were considered as significant in this study. We also performed single sample GSEA (ssGSEA) analysis on the samples using the GenePattern workflow. CIBERSORT (http://cibersort.stanford.edu/) was used to estimate the tumor infiltrating immune cell populations of the samples (31). RNA-seq data were submitted to the CIBERSORT web server as mixture file, and 22 immune cell type (LM22) is selected as the signature gene file. Quantile normalization option was disabled as recommended by CIBERSORT for RNA-seq, and 1000 permutations were performed for estimating the immune cell populations (CIBERSORT Results for individual samples were provided in Supplementary Table 8).

RESULTS

Cohort and clinical parameters

To study the effects of smoking on the TIM, we selected a cohort consisting of 177 HNSCC cases with primary tumors and clinical information (Table 1). We elected to analyze only HPV-negative HNSCC because HPV-positive HNSCC is a biologically distinct disease with potentially different mechanisms of immune escape (32). One hundred (57%) of 177 patients were current-smokers compared to 22 (12%) never- or 55 (31%) former-smokers. There was a difference in average age at diagnosis and smoking status (p<0.001, Kruskal-Wallis test), which was attributed to a higher age for former-smokers than for current-smokers (p<0.001, Dunn post-test) (Supplementary Figure 1). There was no difference of smoking status with gender, Eastern Cooperative Oncology Group (ECOG) status, race, or disease stage (Table 1). The smoking distribution was unequal by primary tumor site (p=0.038, Fisher’s Exact test), which was mainly driven by a larger proportion of the larynx tumors in the current smoker group. We observed differences in overall survival (OS) based on smoking status, although these did not reach statistical significance (Supplementary Figure 2). The observed features and parameters in our cohort are largely in accordance with expectations from a retrospectively collected HNSCC cohort (3,33)

Table 1.

Patient Characteristics

	Smoking Status
	Current	Former	Never	Total	p-value*
Age (median, range)	59.5 (31–83)	68 (20–86)	62 (20–88)	62 (20–88)	<0.001
Gender					0.090
Male	76	36	12	124
Female	24	19	10	53
ECOG status					0.103
0	49	27	12	88
1	39	25	10	74
2	11	0	0	11
3	1	1	0	2
Not Available	0	2	0	2
Race					0.054
Caucasian	95	55	21	171
Black	5	0	0	5
Other	0	0	1	1
Disease Site					0.038
Oral Cavity	61	43	18	122
Larynx	31	8	2	41
Oropharynx	6	3	0	9
Hypopharynx	2	1	2	5
Disease Stage (AJCC 7th)					0.055
I	5	9	3	17
II	10	8	4	22
III	11	4	5	20
IV	48	21	6	75
Recurrent tumor	26	13	4	43
Tumor Origin					0.149
Primary tumor	67	37	13	117
Second primary tumor	7	5	5	17
Recurrent tumor	25	10	3	38
Recurrent second tumor	1	3	1	5
Total	100	55	22	177

^*Fisher’s Exact test for count data, except Age Kruskal-Wallis Test

Multiplex Immunofluoresence (mIF)

Of the full cohort of 177 cases, 107 had sufficient FFPE tumors for analysis by mIF (Supplementary Figure 3 and Supplementary Table 2 and 3) for expression of CD3, CD8, FOXP3, PD-1, PD-L1, and pancytokeratin (PCK). An overview of the 107 cases for mIF with associated clinical and molecular profiles is shown in Figure 1. The mIF data were obtained from three separate regions of interests (ROIs) selected within the tumor core (TC), tumor margin (TM), and surrounding stroma (S). These separate regions were chosen because several studies have demonstrated that host immune response is often concentrated at the tumor margins and that large differences in infiltrating immune cells may exist between tumor core, tumor margin and stromal areas of a tumor specimen (34,35). Of the cohort of 107 mIF cases, 80 had a sufficiently large specimen to allow the selection of all 3 regions of interest (ROIs) for each of the 3 compartments. TC was available for the complete 3 ROIs in 100 cases with the remaining 7 cases consisting of 2 TC ROIs. A full 3 ROIs were available for TM in 94 cases and for S in 93 cases (Supplementary Table 3). We distinguished the following phenotype classes: PCK⁺ (tumor cells), CD3⁺, CD3⁺CD8⁺, CD3⁺FOXP3⁺, PD-1⁺, and PD-L1⁺ cells. Examples of representative staining patterns are shown in Figure 2. The PD-L1⁺ cells were further separated into tumor (PCK⁺PD-L1⁺) and non-tumor (PCK⁻PD-L1⁺) cells based on PCK staining intensity using median as a cut-off. There were only 2 cases with detectable PD-1-positive cells in any compartment, which was too limited to incorporate in subsequent analyses. This result is similar to those obtained by Schneider et al. using a different antibody for PD-1 where PD-1 staining was not observed in tumor cells from 124 cases of HNSCC, while strong staining of infiltrating immune cells were mostly confined to oropharyngeal carcinomas (36). A separate study by Balermpas et al. also found that PD-1 expression is significantly more common in pharyngeal (~50%) than in oral cavity carcinoma (<20%) (37). Of the 107 cases with mIF measurements, only 8 tumors are from the oropharynx or hypopharynx which may explain the low number of cases with PD-1 staining in our cohort.

Figure 1.

Overview of main findings from mIF, RNAseq and WES analyses organized by smoking status. Listed in descending order are pack-years for each of the smoking groups, selected clinical parameters, mIF immune marker results, immune cell and IFN pathway measurements based on RNAseq and tumor mutation burden from WES. M-Male, F-Female; LX-Larynx, OC-Oral Cavity, HP-hypopharynx, OP-oropharynx; Stage 1 through 4 and R-recurrent tumor; blank cells indicate no RNA-seq or WES data except for Immune Marker intensities.

Figure 2.

Representative images of multiplex immunofluorescent staining. A) Tumor margin (TM) with a low lymphocyte density; B) Tumor core (TC) with a low lymphocyte infiltration; C) TM with a high lymphocyte density; D) TC with a high lymphocyte infiltration; E) PD-L1⁺ tumor cell and lymphocyte infiltration at YC; F) PD-L1⁺ tumor cell and low lymphocyte infiltration at TC; G) PD-L1⁻ tumor cells with high PD-L1⁺ immune cells at TM; H) Adjacent stroma with lymphocytes infiltration. All images at 20x magnification DAPI counter staining, green = CD3, yellow = CD8, red =FOXP3, aqua blue =pancytokeratin (PCK), orange = PDL1.

Recurrent tumors have a lower number of immune cells than primary tumors

To determine possible differences in the TIM between newly diagnosed and independent recurrent tumors, we compared each of the markers in TC, TM, and S for these groups. We found evidence of lower immune activity in TM and S where the numbers of CD3⁺ cells were significantly lower in recurrent tumors than in newly diagnosed primary tumors (p=0.005 and p=0.010, Wilcoxon rank test, respectively) suggesting recurrent tumors may have been selected for immune evasion allowing the recurrence (Supplementary Figures 4A and 4B). In TC, the numbers of PD-L1⁺ cells were generally lower in recurrent tumors than in newly diagnosed primary tumors (Supplementary Figure 4C). For the subclass of PCK⁻PD-L1⁺ cells, the difference was statistically significant (p=0.034, Wilcoxon rank test) while the difference for PCK⁺PD-L1⁺ cells were borderline significant (p=0.065, Wilcoxon rank test) (Supplementary Figure 4D and 4E). The cell numbers for the other markers in the three ROI types were not different between newly diagnosed and recurrent tumors (Supplementary Figures 4F–Q). Similar analyses were performed for the newly diagnosed primary versus second primary tumors, but no significant differences were observed. Although it is difficult to make a firm conclusion because our newly diagnosed and recurrent tumors were not matched pairs, we chose to use the full group of 134 primary tumors for subsequent analyses of the TIM excluding the 43 recurrent tumors based on our current data.

Tumor margin has higher immune cell numbers than tumor core or stroma

When we compared immune cell numbers across the 3 ROI types, higher immune activity was observed in TM compared to TC or S. For instance, TM had the highest numbers of CD3⁺ cells compared to TC and S (both p<0.001, Dunn post-test, Figure 3A). While absolute numbers were lower, a similar pattern was observed for both CD8⁺ and FOXP3⁺ cells (Figure 3B and 3C). Again, the difference could be attributed to higher numbers of CD8⁺ and FOXP3⁺ cells in TM compared to TC and S (both p<0.001, Dunn post-test). Using PCK as a marker for tumor cells, PCK⁺PD-L1⁺ cell counts were not different between TC and TM, but non-tumor PCK⁻PDL1⁺ counts were higher in TM than in TC (p=0.031, Wilcoxon rank test) (Figure 3D and 3E). These findings suggest that there is significant intra-tumor heterogeneity in immune cell infiltration.

Figure 3.

Numbers of marker-positive cells according to tumor core (TC), tumor margin (TM) and stroma (S) region of interest (ROI). A) Comparison of the numbers of CD3⁺ cells; B) Comparison of the numbers of CD8⁺ cells; C) Comparison of the number of FOXP3⁺ cells; D) Comparison of the number of PCK⁺PDL1⁺ tumor cells for TC and TM; E) Comparison of the number of PCK⁻PDL1⁺ non-tumor cells for TC and TM. Results of non-parametric test for differences are indicated in the figures.

Smoking suppresses cytotoxic immune response at the tumor margin

To determine the association between tobacco smoke exposure and the composition of immune cells in the TIM, we summarized cells classified as CD3⁺, CD8⁺, FOXP3⁺, PD-L1⁺ (both PCK⁺ tumor and PCK⁻ non-tumor cells) in TC, TM, and S from the newly diagnosed primary tumors and compared their numbers between current-, former-, and never-smokers. The numbers of CD8⁺ cells were significantly different in TM among the different smoking classes (p=0.038, Kruskal-Wallis test), but not for TC (p=0.235, Kruskal-Wallis test) (Figure 4A and 4B). Pairwise comparisons showed that TM had significantly lower numbers of CD8⁺ cells in current-smokers compared to never-smokers (p=0.009, Dunn post-test) while the difference with former-smokers almost reached significance (p=0.051, Dunn post-test). The numbers of PCK⁺PD-L1⁺ tumor cells in TM were also dependent on smoking status: current-smokers had significantly lower PCK⁺PD-L1⁺ tumor cells in TM (p=0.001, Kruskal-Wallis test) compared to former- or never-smokers, Figure 4C). PCK⁺PD-L1⁺ results were similar for TC, although they did not reach statistical significance (p=0.059, Kruskal-Wallis test, Figure 4D). The numbers of PCK⁻PD-L1⁺ cells were also significantly different with respect to smoking classes measured in TC and TM but did not reach significance in stroma (p=0.045, p=0.025, and p=0.097, Kruskal-Wallis test, respectively) (Figure 4E, 4F, and 4G). There was no significant difference between the smoking groups for CD3⁺ or FOXP3⁺ cells in any of the TC, TM, or S ROI types. Taken together, the absence of differences between former- and never-smokers suggests smoking might actively modulate the tumor and immune cell interaction at the tumor margin.

Figure 4.

Numbers of marker-positive cells according to smoking status categorized as never-, former- and current-smokers. A) Comparison of the number of CD8⁺ cells in tumor margin (TM); B) Comparison of the number of CD8⁺ cells in tumor core (TC); C) Comparison of the number of PCK⁺PDL1⁺ tumor cells in TM; D) Comparison of the number of PCK⁺PDL1⁺ tumor cells in TC; E) Comparison of the number of PCK⁻PDL1⁺ non-tumor cells in TM; F) Comparison of the number of PCK⁻PDL1⁺ non-tumor cells in TC; G) Comparison of the number of PCK⁻PDL1⁺ non-tumor cells in stroma. Results of non-parametric test for differences are indicated in the figures.

Tumor mutation burden (TMB) and tumor clonality by mutant-allele tumor heterogeneity (MATH) score are not associated with smoking status

High TMB is often used as a surrogate marker of immune response to tumors harboring neo-antigens due to high levels of somatic mutations (38). Furthermore, carcinogens in tobacco smoke are expected to cause permanent DNA damage that is reflected in TMB (39). Within the data set of 104 cases with available exome sequencing data, we observed a median TMB of 6.4 mutations/Mb (range: 1.5–41 Mb), which is similar to values reported in the literature (40). However, in this cohort, we did not find evidence for an association between smoking status and TMB (Supplementary Figure 5A). Another measure related to treatment resistance is the level of intra-tumor heterogeneity, which can be estimated based on differences among mutated loci as summarized by the MATH score (28). This measure also did not show any association with smoking status (Supplementary Figure 5B). Based on these findings, the presence of higher number of immune cells in the TM is suggested to be, at least in part, a result of active TIM modulation by smoking rather than simply driven smoking-related genomic changes.

Immune marker expression and overall survival.

Due to heterogeneity in the cohort and the quantitative nature of the immune marker measurements, we chose to evaluate effects on survival using Cox proportional hazards modeling. Starting with the 134 primary and second primary tumors in the cohort, we evaluated all possible clinical prognostic factors in a single Cox model that included age, sex, ECOG status (4 classes), smoking (3 classes), disease stage and node status (3 classes). The effects of age and nodal status were highly significant in this model (Supplementary Table 4A) while stage was not independently significant in this model. Thus, we chose age and nodal status as confounders in subsequent models for the immune markers. Because there was no significant difference in OS between N0 and N1 disease, we chose to collapse the nodal status to 2 classes for these models (N0 and N1 vs. N2). We evaluated OS according to immune markers in the 3 ROI types in the subset of 76 cases with primary tumors, full mIF data, and known nodal status. Using a Cox proportional hazards model that included age and node status, we found that higher numbers of CD8⁺ cells were associated with significantly better OS, especially in the TM (p < 0.002, Supplementary Table 4B). The numbers of cells positive for CD3, FoxP3, and PD-L1 were not associated with differences in OS in any of the 3 ROIs, although higher numbers of CD3⁺ cells in TM were associated with borderline significant better OS (p = 0.053, Supplementary Table 4C). These results confirm prior reports that a stronger immune response is associated with better survival (12–14).

Suppression of interferon response pathways in current smokers

We also obtained RNA sequencing data from 75 tumors to determine the differential gene expression. RNA sequencing data were available from 39 current-, 22 former-, and 14 never-smokers (Supplementary Table 5). After removal of the 17 recurrent tumors, we performed Gene Set Enrichment Analyses (GSEA) on the remaining 58 cases for Current/Never, Current/Former, and Former/Never smokers using Hallmark of Cancer, Oncogenic Signature, and Immune Signature (Supplementary Table 6 and Supplementary Figure 6) (30). As shown in Figure 5A, current- versus never-smokers showed several differential pathways mapped to Hallmark of Cancer, including lower interferon-α and-γ responses in current smokers compared to never smokers. Several smoking-associated pathways were higher in current smokers: oxidative phosphorylation and reactive oxygen species reflecting effects of active smoking. Similar comparisons using Oncogenic Signature and Immune Signature yielded similar results as the Hallmarks gene sets where the inflammation and immune related pathways were depleted in current smokers. The comparison between current- and former-smokers for Cancer Hallmarks yielded similar results with higher expression of oxidative phosphorylation, xenobiotic metabolism, and reactive oxygen species gene sets in current smokers, and lower expression of genes in the Interferon-α and -γ response pathways (Figure 5B).

Figure 5.

Gene enrichment analyses between smoking categories for Cancer Hallmarks. Indicated are the top 10 most differentially expressed pathways in Cancer Hallmarks gene sets in each direction. A) Current-smokers compared to Never-smokers and B) Current-smokers compared to Former-smokers. C) Differentially expressed chemokines and chemokine receptors in current-smokers compared to non-current smokers with a p-value below 0.05 are listed (unadjusted t-test of normalized expression values). D) Differentially expressed interferon signature in current-smokers compared to non-current smokers.

These results prompted further gene expression analysis of chemokines and their receptors in current- versus former/never-smokers. Former and never smokers were grouped together because they were indistinguishable in the mIF and RNAseq comparisons. All 58 primary tumor and second primary tumor cases with RNAseq data were used to test all expressed cytokines and cytokine receptors with a minimum of 50% differential expression as shown in Figure 5C. Chemokines are chemotactic cytokines that regulate immune-cell trafficking with importance for the TIM. The most differentially expressed chemokines included the canonical IFN-response CXCR3 chemokines, CXCL9, CXCL10, and CXCL11 which were expressed in lower levels in the current-smokers compared to the former-smokers (41). This was associated with a concurrent decrease in the expression levels of CCL5 that, along with CXCR3 chemokines, is produced in response to the STING/TBK1/IRF3 innate immune pathway (42,43) and is a known lymphocyte-recruiting chemokine (44). It was recently shown that tumor-derived CCL5 and immune-cell expressed CXCL9 are involved in recruitment of cytotoxic T-cells in response to IFN-γ (45). To further study the association of CXCR3 chemokines and CCL5 with immune cell infiltration, we performed correlation analyses of these markers in the 42 cases for which these data were available. A clear pattern emerged whereby higher expression of chemokines CXCL9, CXCL10, CXCL11, CCL5 and IFN-γ was correlated with higher CD3⁺ cell counts in TC and higher CD8⁺ cell counts in TM (Supplementary Table 7).

In previous studies, IFN-γ response genes have been related to a T-cell inflamed gene expression profile indicative of an inflamed TIM (46) that is predictive of response to PD-1 checkpoint inhibitor treatment across the KEYNOTE clinical datasets (47). We further investigated the 18-gene signature between current smokers (N=27) and former or never smokers (N=31) from primary or second primary tumors for which RNAseq data was available (Figure 5D). This analysis showed that all, except one, of the 18 genes had higher expression levels in tumors from non-current smokers compared to tumors from current smokers, again confirming higher activity of the IFN-γ signaling pathway in tumors from individuals who are not currently smoking. Taken together, our data is consistent with an immune suppressive effect of active smoking and suggests involvement of the IFN-α and -γ response pathways, altering the balance between anti-tumor T cells and tumor-promoting effects of CXCR2.

DISCUSSION

Tobacco smoking has been clearly established as a cause of many cancers including HNSCC (48). However, the pro-inflammatory and immunosuppressive effects of tobacco smoking within the disease site specific TIM has not been clearly delineated. In this study, we characterize the TIM in current-, former-, and never-smokers with HNSCC and observed ongoing immunosuppressive state that may result in differential response to immune modulating therapies and importance of smoking cessation during cancer treatment.

To date, most of the preclinical and clinical studies delineating the effects of tobacco smoking are conducted in chronic obstructive pulmonary disease and lung cancer models in context of chronic airway inflammation (48). Evaluation of the different cell types and immune microenvironment in the head and neck region is sorely lacking. These disease site specific data are particularly important in context of the cancer-immune set point, the threshold needed to mount effective cancer immunity (49). The cancer-immune set point is defined as the summation of all immune regulators that are inherent to each individual. In addition to the obvious immune regulators such as the T-cell receptor (TCR) signaling in CD8⁺ T cells, presence of neoantigens, host genetics, etc., there is mounting data that the TIM-driven factors may be important in mounting the effective immune response (50). It is clearly established that smoking can modulate the TIM by affecting the oral microbiome, oxygen deprivation, and biofilm formation (51). Furthermore, the degree of exposure and concentration of carcinogens contained in the tobacco smoking would also vary depending on the disease site (48). For example, Desrichard, et al. evaluated the mutational signatures associated with tobacco smoking in HNSCC and lung SCC (LSCC) from TCGA and correlated with expression profiles of immune infiltration, cytolytic activity, and interferon-gamma pathway signaling (11). The differential gene expression analyses indicated immunosuppression in smoking-high HNSCC involving lower expression of MHCII molecules, T-cell receptors, immunoregulatory molecules, cytotoxic effectors, and cytokines. In addition, pro-inflammatory profiles were also observed in smoking-high LSCC. Notably, the mutation and expression profiles of tobacco-related HNSCC and LSCC resemble each other very closely, suggesting immunosuppression in HNSCC is driven by interactions within the TIM (32). Therefore, comprehensive evaluation of the TIM specifically in HNSCC is required and extrapolating the data from other disease sites warrants caution.

We also found a temporal relationship to smoking exposure. Current-smokers had the lowest number of the immune cells compared to the former- and never-smokers. Considering we did not find any significant differences in the mutation profiles based on smoking status, this difference may be driven by differential gene or protein expression due to the recent exposure to the tobacco smoking. In addition, we found that the IFN-α and IFN-γ response pathways were significantly downregulated in current-smokers compared to never- and former-smokers. Further analyses of chemokine profiles revealed decreased expression of CXCL9, −10, and −11 in current-smokers. These chemokines are known to regulate immune cell migration, differentiation, and activation through recruitment of cytotoxic lymphocytes and natural killer cells in response to IFN-γ expression (52). Activation of the CXCL9, −10, −11/CXCR3 axis leads to migration of immune cells to their focal sites. Current data suggest that tumors with the IFN-γ signature and inflamed phenotype with a high number of TILs have higher response and survival benefits given anti- PD-1 checkpoint inhibitors (46,47). We observed that current-smokers had significantly lower numbers of PD-L1⁺ cells in the TC and TM compared to never- and former-smokers, which is consistent with data that patients with smoking-high HNSCC have a lower response rate given anti-PD-1 checkpoint inhibitors compared to smoking-low HNSCC (11). Using an in vivo model, Chheda, et al. demonstrated abrogation of anti-PD-1 inhibitor effects in CXCR3 knock out mice suggesting homing of T-cells to the tumor through the CXCL9, −10, −11/CXCR3 axis may be critical for the anti-PD-1 inhibitor efficacy (53). However, activation of CXCL9, −10, −11 signaling in tumor cells has shown to increase cell proliferation, angiogenesis, and metastasis in colon and breast cancer models (54,55). Understanding the interaction between the tumor and immune cells and therapeutic strategies to selectively inhibit the CXCL9, −10, −11/CXCR3 signaling in the tumors without dampening the anti-tumor immune cell migration to the tumor site will be a challenge in immunotherapy development for cancer treatments.

In summary, our results indicate that active tobacco use in HNSCC may have a significant immunosuppressive effect through suppression of T-cell chemotaxis. The importance of understanding the interaction between smoking and TIM is further highlighted by emergence of immune checkpoint inhibitors as a promising therapeutic option in management of HNSCC, and smokers are less likely to benefit from the anti-PD-1 checkpoint inhibitors (11,15–17). Our findings add to the importance of further research that will allow a more granular examination of the TIM and changes due to tobacco smoke exposure in HNSCC.

STATEMENT OF TRANSLATIONAL RELEVANCE.

Patients with head and neck squamous cell carcinoma (HNSCC) who actively smoke during treatment have worse survival compared to never-smokers and former-smokers. We hypothesized the poor prognosis in tobacco smokers with HNSCC is, at least in part, due to ongoing suppression of immune response, and understanding of deregulated signals due to smoking may provide insights to improve the clinical outcomes of HNSCC patients. We found current-smokers have lower numbers of cytotoxic T-cells in tumors and decreased expression of genes in the interferon α and γ response pathways compared to the former- and never-smokers. Our data suggest the canonical IFN-response CXCR3 chemokines, CXCL9, CXCL10, and CXCL11, contribute to the immunosuppressive tumor microenvironment.