This was a prospective, case-controlled pilot study to investigate inflammatory changes in saliva in patients with OSCC. The study has been approved by the Research Ethics Board (REB) of Sunnybrook Health Sciences Center (project identification number 223-2015) and University of Toronto (protocol number 32724).

The study population consisted of all patients referred for treatment of biopsy-proven OSCC at the Odette Cancer Center, Sunnybrook Health Sciences Center, Toronto, between September 2016 and October 2017. A total of 37 patients consented to participate in this study. A control group of 24 healthy patients with no history of oral cancer, as well as eight patients from the University of Toronto Faculty of Dentistry with moderate to severe periodontal disease and no history of oral cancer were included. The control patients were matched in age and sex with the OSCC patients. Table 1 shows the demographic information of the study and control populations.

The patients were recruited into four groups: (1) control patients without oral or systemic diseases, (2) Patients with a diagnosis of OSCC, (3) Patients with periodontal disease, and (4) Patients with dysplasia at resection. OSCC group: Inclusion criteria: biopsy proven OSCC referred for treatment at the Odette Cancer Center. Exclusion criteria: tumor location outside the oral cavity, non-squamous cell carcinoma, history of previous malignant disease, chemotherapy or head and neck radiation within the previous 5 years, presence of a second primary malignancy, immune disorders (such as Sjogren's syndrome and HIV infection), history of hepatitis, and severe systemic disease (ASA > 4). Patients with a biopsy confirmed OSCC that presented for surgery at the Odette cancer center but were found to have only premalignant lesions (dysplasia) and no invasive OSCC were included in the dysplasia group.

Control groups: healthy control patients with no reported oral disease or ongoing dental/periodontal treatment were included. Periodontal disease group: Periodontal disease patients from the Periodontal Disease Unit at the University of Toronto Faculty of Dentistry were only included if they had treatment appointments scheduled for diagnosis of moderate to severe periodontal disease, as recorded in their dental chart. Exclusion criteria included: history of previous malignant disease, chemotherapy or head and neck radiation within the previous 5 years, presence of a second primary malignancy, immune disorders (such as Sjogren's syndrome and HIV infection), history of hepatitis, and severe systemic disease (ASA > 4).

As a pilot project, our first objective was to test a protocol for collection and analysis of saliva samples. The collection protocol using 3 ml of a saline rinse was well-tolerated by all patients and all patients consented to the collection of saliva after the study and protocols were explained. The collection was carried during the pre-surgical consult and did not cause any disruption to standard treatment protocols. The filtered saliva supernatant was adequate for cytokine analysis as all samples passed initial quality control and were within the range of the Luminex test. The establishment of FMO and single stain controls required the use of a control saliva sample as blood controls showed different staining patterns compared to the saliva samples. To overcome this issue, a single donor collected multiple saliva samples that were combined and used to generate all FMO and single marker controls.

All patients were recruited during a pre-treatment oncology appointment at the Odette Cancer Center, Sunnybrook Hospital, Toronto. After informed consent was obtained, a short medical questionnaire and intraoral examination were performed. Saliva samples were obtained by rinsing the mouth for 30 s with 3 ml of saline. The samples were kept on ice and stabilized with 100 μl of protease inhibitor (EMD Millipore Protease Inhibitor Cocktail Set II, Calbiochem) until filtration using a 40 μm membrane filter. Time interval between collection and filtration of samples ranged from 20 to 150 min. The saliva samples after filtration were centrifuged (7 min at 3,000 rpm) in the Sunnybrook Hospital Clinical Pathology lab to obtain cell pellets and supernatants. The latter were frozen at −80°C until use. Cell pellets were re-suspended and fixed in 200 μl paraformaldehyde 4% and PBS and kept at 4°C until use for a duration ranging from 1 to 12 months (Figure 1). Control subjects and subjects with periodontitis provided saliva samples that were processed at the Faculty of Dentistry, University of Toronto. Inflammatory cell counts in blood were obtained from routine pre-surgical blood tests at the Odette Cancer Center, without change to standard of care or extra visits.

A 12-antibody panel (Supplemental Table 1) was optimized to determine the inflammatory cell profile in saliva. Prior to staining the cells, all samples were re-filtered with a 40 μm, 20 μm membrane filter, and then with a 10 μm membrane filter to remove debris. The cells were washed with PBS, and concentration was adjusted to a cell number of 1–5 × 10⁶ cells/ml with 300 μl ice cold FACS buffer. Two Hundred microliters of Fc receptor blocking solution (5 μl of TruStain FcXTM per million cells) was added to each sample (Fc blocking solution 1% BSA and 1% FBS diluted in PBS buffer) and incubated on ice for 10 min. Samples were centrifuged at 1,500 rpm for 5 min at 4°C. Supernatant were discarded. Primary labeled antibodies were added and incubated for 20 min at room temperature in the dark. Cells were washed three times by centrifugation at 1,500 rpm for 5 min and resuspended in 1,000 μl of ice cold FACS buffer and kept in the dark at 4°C until analysis. Vortex and a final filtration (10 μm filter) were done immediately before FCM to ensure even distribution of the cells in the samples and to minimize cells and particles aggregates.

Fluorescent-labeled saliva samples from the study and control patients (total 69 samples) were analyzed in a flow cytometer (BD Fortessa x20 5 laser) in three sessions. For each FCM session, the instrument was calibrated with rainbow beads (BD Bioscience), to ensure that the voltage applied to lasers/detectors always returned the same median fluorescence. Compensation beads stained with the 12 fluorescent antibodies were used to set voltage and adjust compensation matrix. Full minus one controls (FMOs) for each of the 12 stains, as well as one fully stained control and one unstained control were prepared, using saliva provided by a single healthy donor. The instruments were calibrated with the same parameters for each session. Supplemental Table 1 shows the laser configuration and antibodies/fluorophore used for detection of immune cells. For 38 of the 69 samples, 100,000 events were analyzed. For 30 of the 69 samples (10 cancer samples, seven perio samples, 13 healthy samples) the entire sample was analyzed in order to calculate the total number of immune cells in the samples. FlowJo 10.4.1 software was used for data analysis. Parameters that were studied included forward scatter (to analyze cell size) and side scatter (to analyze cell granularity and complexity), followed by fluorescent emission analysis of the 12 selected fluorescence-tagged antibodies (to sort immune cells subsets based on specific cell surface fluorescent-labeled antigens). Figure 1 shows the gating strategy that was developed to identify and quantitate each inflammatory cell type.

Assays were conducted in 96-well plates according to the immunoassay protocol for the Millipore Human Cytokine/Chemokine Magnetic Bead Assay Panel (HCYTMAG-60K-PX30 by EMD Millipore, USA) [including analytes EGF, G-CSF, GM-CSF, IFN-α2, IFN-γ, IL-1α, IL-1β, IL-1ra, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12 (p40), IL-12 (p70), IL-13, IL-15, IL-17, IP-10, MCP-1, MIP-1α, MIP-1β, TNF-α, TNF-β, VEGF, RANTES/CCL5, Eotaxin/CCL11] and Luminex? detection. The samples were prepared, and the data acquired and analyzed at the Princess Margaret Genomic Center (PMCC, Toronto, Canada) according to the manufacturer's instructions. Briefly, 25 μl magnetic beads, 25 μl assay buffer, and 25 μl of sample (1:2 diluted sample) were incubated overnight at 4°C with shaking. Beads were subsequently washed 2 X times and incubated with 25 μl of detection antibody for 1 h. Twenty-five microliters streptavidin phycoerythrin was added to the assay mixture for 30 min at room temperature. Beads were washed 2X and resuspended in 150 μl of Sheath Fluid. Assays were read with Luminex 100 Reader and data was analyzed using Bio plex Manager 6.0. Analyte concentrations are indicated in pg/ml units. From the 69 patients included in this cytokine analysis, 30 samples (17 cancer patients and 13 controls) were used as part of a previous publication [11].

A total of 25 cancer cases were available for analysis and 10–15 tissue slides were prepared from the resection specimens. Fluorescent immunohistochemistry (FIHC) was performed as described previously [11]. In brief, primary antibodies for PD-1, PD-L1, CD45, CD66b, CD3, CD4, and CD8 surface markers were used to assess the populations of neutrophils (CD45+ CD66b+), CD8+ T cells (CD3+CD8+), and CD4+ T cells (CD3+CD4+) in addition to PD-1 and PD-L1 distribution based on the colocalization of the two markers. Using Volocity software, all tissue samples were analyzed, and the average sum of pixel intensity was calculated for each slide.

Descriptive statistics, including means and standard deviations, were calculated for the outcome variables. These statistics were reported for the whole sample as well as separately by the four groups of interest. The Shapiro test of normality was run on the outcome variables and following this, non-parametric Kruskal-Wallis tests were used to examine whether there are significant bivariate associations between group membership and potential predictor variables. To explore differences among groups, multivariable logistic regression was applied to examine the relationship between group membership and predictors of interest. In the logistic regression model, group membership was dichotomized as Cancer vs. the others. Prior to modeling, multicollinearity among the predictors was examined using tolerance statistics (tolerance value <0.4). If multicollinearity was found to exist, then only one member of a correlated set of variables was retained for the final model. The model results were reported as odds ratios and their associated 95% confidence intervals. All analyses were carried out using SAS 9.4 (Cary, NC, USA). Correlation analyses and graphs were generated using GraphPad Prism 7.0 and IBM SPSS version 25. All figures were prepared using Adobe Illustrator CC 2019.

The demographics of the groups can be seen in Table 1. Twenty-eight [24] patients were treated with surgery and their tumors staged according to AJCC 7^th ed. based on the pathology report of the tumor resection; clinical staging was only used for two patients treated with primary chemoradiation and two patients did not have definitive treatment at the Odette Cancer Center (Table 2). Five patients had an initial biopsy of dysplasia suspicious for carcinoma, and showed only dysplasia (moderate to severe) in the resection specimen. These were included in the dysplasia group. There were no significant differences in the age distribution between control (61 ± 14), periodontal disease (51 ± 14), dysplasia (70 ± 13), or cancer patients (66 ± 13) (P = 0.07). There were more females in all groups except periodontal disease group (P = 0.049). Further analysis of dysplasia and cancer patients reveal that the most commonly involved site in all groups was tongue (n = 19, 51.3%) followed by gingiva (n = 9, 24.3%). 86.4% of lesions were painful and 40% of patients had abundant visible dental plaque. The presence of visible plaque and daily tooth brushing was significantly increased in periodontal disease patients compared to cancer patients (P = 0.0498 and P = 0.0333). There were no significant differences in alcohol consumption or smoking between groups and between cancer stages. Most dysplasia and cancer patients had a normal appearing remaining oral mucosa (45.9%, n = 17/37) while 18.9% (7/37) had other areas of leukoplakia (Table 2).

Thirty cytokines/chemokine were evaluated in the saliva samples and the results are reported as the fold change of the specific cytokine over control in a log2 scale (Figure 2). There was a significant increase in GMCSF (OSCC > Control p = 0.0057, OSCC > perio, p = 0.0080), IL-6 (OSCC > control p = 0.0001; OSCC > perio p = 0.0015), Rantes/CCL5 (OSCC > control p = 0.0079; OSCC > perio p = 0.0227) and MIP1α (CCL3) (OSCC > control p = 0.0034; OSCC > perio p = 0.0218) in OSCC patients compared to control and periodontal disease patients. TNFα (p = 0.0038), MCP1 (CCL2) (p = 0.0493), IL-1β (P = 0.0043), and IL-8 (p = 0.0035) were increased in OSCC compared to control only while MIP1β (CCL4) (p < 0.0365) was increased in OSCC patients compared to periodontal disease patients. IL-10 (p = 0.0364), IL-12p40 (p < 0.0364) were statistically significant but multiple comparison analysis did not show significant differences between groups. There were no significant differences between the dysplasia patients and OSCC patients.

Logistic regression analysis including the eight cytokines increased in cancer compared to control (GMCSF, IL-1β, IL-6, IL-8, Rantes/CCL5, MIP1α, TNFα, MCP1) was performed. The first logistic regression model including IL-6, IL-8, and TNFα showed that elevated IL-6 significantly increased the likelihood of having a diagnosis of cancer (p = 0.0040, odds ratio 3.991). Similar analysis including IL-1β, Rantes/CCL5, MCP1, GMCSF, TNFα, and IL-6 showed similar results with elevated IL-6 increasing the probability of having a cancer diagnosis (p < 0.0046) (Table 3).

We performed Pearson's correlation analysis between all cytokines in the control and cancer groups (Figures 3A,B). The dysplasia and periodontal disease group were excluded due to the limited number of cases available for analysis. We selected the eleven cytokines that were significantly increased in cancer (GMCSF, IL-6, IL-1β, Rantes/CCL5, MIP1α, MIP1β, TNFα, MCP1/CCL2, IL-8, IL-12p40, IL10) and selected all the positive Pearson's correlations with r > 0.8 and significance of p < 0.01. The selected cytokine correlations were analyzed using Cytoscape to create an interaction map in control (Figure 3C) and cancer (Figure 3D). The results reveal different networks in each group with IL-10 and Rantes/CCL5 clusters on control samples and GMCSF and IL-6 in cancer.

The most common inflammatory cell in saliva was neutrophils (78.61% ± 23.3) followed by lymphocytes (23.98% ± 29.59). The cancer group had the highest number of immune cells (CD45+) in saliva (99,240 ± 38,789) and was significantly higher than control patients (7,945 ± 5,790, P < 0.038) but not periodontal disease patients (83,180 ± 21,160). We also calculated the total number of identified cells (CD66b+, CD4+, CD8+, CD56+, CD68+, Siglec8+, CD138+) and OSCC patients showed significantly more cells (55,966 ± 22,151) compared to periodontal disease (32,550 ± 11,816) and control patients (6,355 ± 4,733). Cancer patients also had more neutrophils (p = 0.04) and lymphocytes (p = 0.05) compared to control patients. There were very few immune cells other than neutrophils and T lymphocytes in saliva, preventing further characterization of the cells (Table 4). The dysplasia group was not included in the total cell count analysis as only two samples were fully processed for cell counts. The differences between the individual cell subtypes were not statistically significant when normalized to the total number of cells identified except for decreased T lymphocytes in periodontal disease group compared to control and dysplasia (Table 5). There were no significant differences in the number of neutrophils according to the clinical appearance of the lesion, presence of plaque or tumor Stage (Supplemental Table 2).

We have analyzed complete blood counts from all cancer patients and found no significant difference in complete blood counts between stages of disease. There was no correlation between the salivary cell counts and blood counts for each patient and all the differential blood cell counts in OSCC were within normal limits (data not shown).

We investigated neutrophils (CD45+ CD66b+), CD8+ T cells (CD3+CD8+), and CD4+ T cells (CD3+CD4+) in addition to PD-1 and PD-L1 expression in 25 tissue samples that were available from the cancer cohort (Figure 4A). To normalize the number of samples in each group, we combined Stage 0–2 (low stage) and Stage 3–4 (high stage) patients. There was an increase in neutrophils (CD45+CD66b+) (P = 0.0262) in high Stage (III–IV) tumors compared to low stage (0-II) tumors (Figure 4B). There was also more PD-L1+ (P = 0.038) (Figure 4C) and CD45+ PD-1+ (P = 0.038) (Figure 4D) cells in high stage patients. No significant differences were found between the other cells tested including CD3+CD4+ (P = 0.219) and CD3+CD8+ (p = 0.33) T cells (Figures 4E,F).

The patient outcomes are summarized on Figure 5. We had up to 35 months for follow-up information and from the original cohort of cancer patients. Excluding dysplasia patients, eight patients had recurrences or died of disease (8/32, 25%). Stage IV patients had the worst outcome with 33.3% dying of disease at the end of the follow up period. Bivariate Cox models and Kaplan Meier analyses were used to determine potential predictors of diseased free survival (DFS). Age, gender, location, cytokines, and salivary immune cells were tested but no variables achieved statistical significance. Figure 5 shows Kaplan Meier curves for Stage (Log Rank test, p = 0.309) and presence of metastasis (Log Rank test, p = 0.0001).

The analysis of inflammatory cell profile in saliva by FCM is challenging, and we believe this is the first report of its kind. Saliva inherently contains debris and proteases that contribute to sample degradation and formation of aggregates. To minimize this, we used multiple filtrations prior to processing the samples. The use of formalin has the potential to alter antigen availability, therefore, compensation was done using fixed saliva from the same donor and the antibodies were tested in both fixed and fresh saliva. To determine the different cell populations and exclude debris, gating was carefully done using forward and side scatter and by applying multiple markers/gates for each cell. An important limitation of studying inflammatory cell profile in saliva by FCM is the number of cells available in each sample, particularly in control patients. By vigorously rinsing with saline, we were able to collect at least 100,000 inflammatory cells per sample which made FCM analysis feasible. To achieve reproducible results, we had to have access to filters, fixatives, protease inhibitors, ice, and a centrifuge adjacent to the collection area. This allowed us to minimize the time between collection, processing, and storage which was critical to achieve consistency.

Our results show an increased number of inflammatory cells and a distinct change in inflammatory mediators in the saliva of cancer and periodontal disease patients compared to control. The levels of salivary inflammatory mediators GMCSF, IL-6, Rantes/CCL5, MIP1α (CCL3), and MIP1β (CCL4) were elevated in OSCC compared to periodontal disease which supports that the inflammatory response in OSCC is specific and different from other causes of chronic inflammation such as periodontitis in the oral cavity. The changes in immune cells in saliva of OSCC patients were primarily due to an increase in total number of cells and not from specific shifts in different cell populations. This is potentially due to the fact the highly motile neutrophils would extravasate to saliva while most other inflammatory cells would stay attached to the stroma or connective tissue, therefore the changes in immune cells are expected to be seen in the tissue and not in saliva. The findings also raise the possibility that neutrophils in saliva are primarily caused by the presence of ulceration rather than a specific recruitment mechanism. Several findings challenge this hypothesis. First, there was no difference in the number of salivary neutrophils in lesions that were primarily ulcerated vs. non-homogeneous hyperkeratotic lesions or exophytic lesion. Second, there were no significant differences between the number of salivary neutrophils in different cancer Stages, suggesting that tumor size alone does not explain the observed neutrophilia. Finally, some patients with non-ulcerated lesions in periodontal disease showed similar levels of salivary neutrophils compared to cancer patients, suggesting that ulceration alone also does not explain the increase recruitment of neutrophils. Our data supports the conclusion that the increased number of immune cells in saliva of OSCC patients is due to the recruitment of inflammatory cells to the tumor microenvironment [11] which thus leads to more immune cells that come into direct contact with saliva. Further studies are needed to determine if other inflammatory, reactive, and ulcerated oral lesions also lead to salivary neutrophilia.

Our results suggest a markedly pro-inflammatory immune environment in OSCC. Signaling pathways downstream of growth factors and cytokines such as interleukin IL-1β, IL-6, TNFα, vascular endothelial growth factor (VEGF), and epidermal growth factor (EGF) [22, 25, 26] are complex and commonly rely on Nuclear factor kappa beta (NF-kB), and to a lesser extent signal transducer and activator of transcription 3 (STAT3). These pathways are also critical in carcinogenesis in head and neck cancer and may link the immune landscape of OSCC to tumor promotion [22, 27, 28]. Nuclear factor kappa beta and STAT3 hyperactivation in tumor cells and immune cells in the tumor microenvironment in turn promotes production of several pro-inflammatory cytokines including IL-1β, IL-6, and TNFα, which promote survival and proliferation of tumor cells [22, 24, 27]. In addition, both STAT3 and NF-kB interfere with p53 synthesis, and attenuate p53-mediated genomic surveillance [27]. This is also supported by our recent results that show that neutrophils and TNFα promotes the invasion of oral cancer cells in vitro [11]. Previous studies showed that an increased neutrophilic infiltration in OSCC tissue is associated with poor survival in advanced HNSCC [29–31]. Specifically, Wang et al. showed that tongue SCC associated with neutrophils infiltration displayed higher clinical stage, increased lymph node metastasis, and increased chance of tumor recurrence [29]. Our results support this pathway as higher stage OSCC showed more neutrophils. The pro-inflammatory response is in contrast with an increase in PD-1 and PDL-1 in high stage OSCC as this might be a specific immunomodulatory response from the tumor cells.