Formalin-fixed paraffin-embedded tumor samples of patients were obtained from 266 surgical resection specimens of OSCC spanning 10 years of time (2005-2014) from the archives of the Department of Pathology at the University of Iowa Hospitals and Clinics. Cases were chosen selectively to ensure patients with no history of radiation or chemotherapy and to ensure a mixture of patients regarding recurrent status, node metastases, margin status, age, and smoking status. After excluding cases with unavailable tissue blocks, 146 cases were included in the current study. TMAs were constructed using 3-6 morphologically representative areas (tumor and stroma). Sections (4μm) were obtained from the TMAs on poly-L-lysine-coated glass slides. Routine hematoxylin-and eosin (H&E) sections were reviewed to confirm the original diagnosis. Subject clinicopathological characteristics considered were age, sex, smoking history, tumor site, T stage, N stage, differentiation, and presence of perineural invasion, lymphovascular invasion, bone invasion and local recurrence. Clinicopathological characteristics were obtained from medical records where tumor microscopic features and TNM stage had been previously evaluated by board certified pathologists. The TNM status was based on the American Joint Committee on Cancer 7 (AJCC7).

Antigen retrieval was performed on freshly cut sections in a decloaking chamber for 5 min at 125°C in TRIS buffer (pH 9.0). Endogenous peroxidase was blocked by incubation with 3% peroxide at room temperature for 8 min. IHC was performed with the following antibodies: EGFR (H11, Dako) at 1:200 dilution, CD3 (Dako A0452) at 1:200 dilution, CD4 (Novocastra NCL-L-CD4-368) at 1:100 dilution, and CD8 (Dako M7103) at 1:100 dilution. Bound antibody was detected using the HRP-DAB Cell & Tissue Staining Kit. All slides were counterstained with hematoxylin.

EGFR immunostaining was evaluated by semiquantitative scoring (score 0-3) where 0 represents staining in <10% neoplastic cells and 1, 2, and 3 representing weak, moderate, and strong staining in >10% neoplastic cells according to Gamboa-Domingez and colleagues (56). Immunostaining scores of 3 and 2 were designated as EGFR-positive (EGFR+) and scores of 1 and 0 were designated as EGFR-negative (EGFR-). CD3+, CD8+, CD4+ and FoxP3+ TILs were evaluated based on the density of positive inflammatory cells (score 0-3): 0 represents none or very few positive cells; 1 represents single isolated positive cell or small aggregates of positive cells 2-4 cells; 2 represents discrete nodules or aggregates of positive cells more than 2-4 cells; 3 represents bands or continuous aggregates of positive cells. For CD3, the immunostaining scores of 3 and 2 were designated as high CD3 (CD3+), while scores of 1 and 0 were designated as low CD3+ (CD3-). Due to low number of specimens that received scores of 3 (n=2) and 2 (n=11) for CD8, the immunostaining scores of 3, 2 and 1 were designated as CD8+, while a score of 0 were designated as CD8-. For CD4, immunostaining scores of 3 and 2 were designated as high CD4 (CD4+), while scores of 1 and 0 were designated as low CD4 (CD4-). Immunoexpression of EGFR, CD3, CD4 and CD8 was scored by 2 pathologists. In the event of disagreement, a consensus score was given. Each tissue core received an individual score, and the average score was calculated from all the tissue cores that were from the same case and used as a final score. Very few HPV+ OSCC cases were present in the patient cohort and were excluded from the study.

Power for a sample size of 146 cases was estimated at 95.2% calculating from previous work (57). The association between expression scores and patient clinicopathological characteristics were analyzed by Chi-square test. Survival outcomes differences were plotted using the Kaplan-Meier method while estimates for the group hazard ratios were obtained using Cox proportional hazards (PH) modeling. Overall survival (OS) is defined as the length of time (in months) from the date of diagnosis that the patients remain alive. Progression free survival (PFS) is defined as the time from diagnosis to disease progression or death (in months) from any cause. All testing was performed on the univariate level and unadjusted for multiple comparisons. Differences between survival curves were compared using the log-rank test. A p-value below 0.05 was considered statistically significant. GraphPad Prism 8.1.2 was utilized for data analysis.

Baseline characteristics for the OSCC patients are summarized in Table 1 . Fifty eight percent of the patients were male and 42% were female. Female patients were diagnosed at an older average age then male patients (66 versus 58 years respectively, p=0.002, Table 1 ), although there were no significant differences in survival outcomes (OS or PFS) between sexes ( Supplemental Figures 1A, B ). Patients 60 years and older demonstrated significantly worse OS (but not PFS) compared to patients under 60 years ( Figures 1A, B ). Active smokers comprised 44% of patients, with 35% having never smoked, 9% which quit smoking for less than 10 years, and 12% which quit smoking for more than 10 years ( Table 1 ). There was no difference in OS between active smokers, never smokers and patients that quit (p=0.2, Figure 1C ), however significant differences were observed in PFS where never smokers had improved PFS compares to active and patients that quit smoking (p=0.04, Figure 1D ). Of the oral cavity disease sites, tongue was the most common disease location (38%, Table 1 ) and there was a trend toward differences in OS (p=0.07) but not PFS (p=0.38) by tumor site ( Supplemental Figures 1C, D ). Patients with T3/T4 tumors represented 41% of the patient cohort ( Table 1 ) and showed a trend toward worse OS compared to T1 and T2 tumors (p=0.06) but no differences in PFS ( Supplemental Figures 1E, F ). The majority of patients presented with no lymph node metastasis (N0, 51%, Table 1 ) and demonstrated significantly more favorable OS (but not PFS) compared to patients that presented with N1 or N2/N3 disease ( Figures 2A, B ). Patients with poorly differentiated tumors represented 25% of the patient cohort ( Table 1 ) and were associated with a trend toward worse OS (p=0.08, Figure 2C ) and significantly worse PFS (p=0.04, Figure 2D ) compared to well and moderately differentiated tumors. Perineural (PNI), lymphovascular (LVI) and bone invasion (BI) were observed in 50%, 37% and 30% of tumors respectively ( Table 1 ). Patients presenting with PNI demonstrated significantly worse OS (p=0.004, Figure 2E ) and PFS (p=0.04, Figure 2F ) compared to patients with no PNI. Patients presenting with LVI demonstrated significantly worse OS (p=0.004, Figure 2G ) and a trend toward worse PFS (p=0.06, Figure 2H ) compared to patients with no LVI. BI status did not affect OS nor PFS in the patient cohort ( Supplemental Figures 1G, H ). The average follow-up time was 107 ± 37 months. Of the 146 cases, 48 patients survived 5 years after diagnoses.

The prognostic value of EGFR expression in the OSCC cases represented in the TMA was initially evaluated. Examples of EGFR expression scores are shown in Figure 3A . EGFR+ expression represents tumors with strong (score of 3) and moderate (score of 2) expression, while EGFR- expression represents low (score of 1) and no (score of 0) expression. Patient clinicopathological characteristics based on EGFR status (EGFR+ vs EGFR-) are shown in Table 1 where there were no significant associations observed. There were also no differences observed in OS (p=0.14) according to EGFR+ and EGFR- expression ( Figure 3B ). However significant differences were observed in PFS (p=0.01) with higher EGFR expression (EGFR+) being associated with worse PFS ( Figure 3C ). Similar results with OS and PFS were obtained when the prognostic value of each EGFR expression score (3, 2, 1 and 0) was previously evaluated shown here (58). However, the combined (3 + 2 [EGFR+], 1 + 0 [EGFR-]) expression scores were utilized for the remainder of the study to maintain sufficient case numbers to further analyze T cell subsets in these EGFR expression groups. These results support prior reports that EGFR expression is a strong predictor for PFS but not OS in OSCC patients (27, 59).

The prognostic value of TIL expression was initially assessed by pan-T cell (CD3) expression. Images of no [0], low [1+], moderate [2+], and strong [3+] CD3 expression scores are shown in Figure 4A . CD3 expression scores were combined in a manner identical to EGFR above where combined 3 and 2 scores represent CD3+ and combined 1 and 0 scores represented CD3-. CD3+ expression was associated with tongue tumors compared to other disease sites (p=0.018), and also associated with cases presenting with no lymph node metastasis (N0) compared to N1 and N2/3 cases (p=0.008) ( Table 2 ). Additionally, CD3+ OSCCs were associated with favorable OS compared to CD3- tumors (p=0.01, Figure 4B ). CD3 expression did not impact PFS (p=0.19, Figure 4C ). These results suggest that CD3+ TILs may suppress lymph node metastasis and thus survival outcomes. The prognostic value of CD3 was next evaluated in EGFR+ and EGFR- OSCC patients. There was no difference in CD3 expression between EGFR+ and EGFR- tumors ( Figure 5A ). However, EGFR+/CD3+ expression was most frequently observed in younger males (p=0.01) and associated with less lymph node metastasis (p=0.046) compared to EGFR+/CD3- patients ( Table 3 ). Patients with CD3+ OSCCs were associated with significantly higher OS (p=0.02, cox ph = 0.0097) compared to CD3- tumors, but only in EGFR+ patients ( Figure 5B ) and not EGFR- patients ( Figure 5C ). There were no differences in PFS observed with any of the combined EGFR/CD3 expression scores ( Supplemental Figures 2A, B ).

In order to determine what type of TIL represented the CD3 expression observed in the EGFR+/CD3+ patients, CD4 and CD8 expression was assessed in each OSCC case. CD4 expression scores and images are shown in Figure 6A and were combined as carried out for EGFR and CD3 where 3 and 2 scores represented CD4+ and combined 1 and 0 scores represented CD4-. Patient clinicopathological characteristics based on CD4 expression are shown in Table 2 . CD4+ expression was associated with tongue tumors, no lymph node metastasis (N0, p=0.019) and no LVI (p=0.006, Table 2 ). CD4- expression were associated with floor of mouth tumors (p=0.004, Table 2 ). CD4- OSCCs were associated with significantly lower OS compared to CD4+ OSCCs (p=0.048, Figure 6B ) and there was no impact on PFS ( Supplemental Figure 3A ). When combined EGFR and CD4 expression was analyzed, it was found that EGFR+/CD4+ expression was associated with never smokers (p=0.03), tongue tumor disease site (p=0.03), T1 tumors (p=0.02), N0 tumors (p=0.002), no LVI (p=0.02) and no BI (p=0.01) ( Table 4 ). There was no difference in CD4 expression between EGFR+ and EGFR- tumors (p=0.69, Figure 6C ) and there were no differences observed in OS ( Figures 6D, E ) or PFS ( Supplemental Figures 3B, C ) with any of the combined EGFR/CD4 expression scores. Patients that were CD3+ also were more likely to be CD4+, but EGFR status had no influence on this observation ( Figure 6F ). These results suggest that combined EGFR+/CD4+ expression is associated with favorable clinicopathological features for survival.

To assess CD8 expression, scores of 0-3 were again assigned as shown in Figure 7A . However, due to the low number of patients with CD8 scores of 3 (n=2) and 2 (n=11), we used scores of 3, 2, and 1 to represent CD8+ tumors (n=90) and a score of 0 to represent CD8- tumors (n=49) in order to have sufficient numbers for group comparisons and further subset analyses. Patient clinicopathological characteristics based on CD8 expression are included in Table 2 and there was no significant association between CD8 expression and any of the characteristics analyzed. However, CD8+ OSCCs were associated with significantly more favorable OS compared to CD8- OSCCs (p=0.02, Figure 7B ) and there was no impact on PFS ( Supplemental Figure 3D ). When combined EGFR and CD8 expression was analyzed, EGFR+/CD8+ was associated with tumors in male patients but no other associations with clinicopathological characteristics were observed ( Table 5 ). There were no differences in CD8 expression between EGFR+ and EGFR- tumors (p=0.66, Figure 7C ). There were also no differences observed in OS ( Figures 7D, E ) or PFS ( Supplemental Figures 3E, F ) with any of the combined EGFR/CD8 expression scores. Notably, EGFR-/CD8+ expression showed a trend (p=0.07) toward association with favorable OS compared with EGFR-/CD8- expression when analyzed by log-rank test ( Figure 7E ), but showed significance (p=0.01) when the Gehan-Breslow-Wilcoxon test was used which gives more weight to deaths that occur at early time points. As shown with CD4+ ( Figure 6F ), patients that were CD3+ were also more likely to be CD8+ but EGFR status had no influence on this observation ( Figure 7F ).

Altogether, these results demonstrate that combined EGFR+/CD3+ expression identified a subset of OSCC patients with favorable prognosis compared to other biomarker profiles which may be attributed to a lack of lymph node metastasis.