All the schemes of this study were examined and approved by the Ethics Committee of Nanjing Stomatology Hospital, Medical School of Nanjing University (No.2019NL-009(KS)). Informed consent was provided by the patients for the use of their tissues and data. The study was carried out in accordance with the Declaration of Helsinki. From 2014 to 2017, 105 primary OSCC patients were enrolled. The inclusion and exclusion criteria of patients were the same as those of our previous studies (Zhu et al., 2020). None of the patients received preoperative chemotherapy, radiotherapy, or other cancer-related treatments. Patients with history of systemic illness or missing survival data were excluded. These patients with primary tumor were diagnosed by hematoxylin and eosin staining by two experienced pathologists. These patients were followed up for 2–60 months, and the median was 38 months. Paraffin-embedded OSCC tissue slices were obtained from the pathology department and used for IHC study. 104 blood samples from OSCC patients were obtained for flow cytometry assay before any related treatments.

The protocol of IHC of formalin-fixed paraffin-embedded sections and scoring details of IHC was performed as previously described (Zhao et al., 2019). Anti-ICOSLG (ab257321, Abcam, Waltham, MA, USA) were used at a dilution ratio of 1:400, and the serial sections were incubated with primary antibodies such as anti-CD4 (ZSGB-BIO, ZM-0418), anti-CD8 (ZSGB-BIO, ZA-0508), anti-CD19 (ZSGB-BIO, ZM-0038), anti-CD56 (ZSGB-BIO, ZM-0057), anti-CD68 (ZSGB-BIO, ZM-0464), and anti-Foxp3 (ab253297, Abcam). We used PBS to replace the primary antibody as negative control. Due to insufficient tissues and individual differences in OSCC samples, certain regions or cell types, such as CAFs and TILs, could not be detected in the IHC staining.

Protein expression was evaluated according to stain intensity and the percentage of positive cells. The intensity of staining was graded as 1 = weak staining, 2 = moderate staining and 3 = strong staining. The percentage of stained cells was graded as 0 = 0–5%, 1 = 6–25%, 2 = 26–50%, 3 = 51–75% and 4 = 75–100%. The final score was obtained by multiplying the two scores. The expression levels of ICOSLG in TCs and CAFs, TILs were defined as “low” when it is lower than the median value and as “high” when it is equal to or greater than the median. The IHC staining results of ICOSLG were evaluated by two senior pathologists who did not know the patient’s data, and the median values were calculated for further analysis.

Fresh whole blood from patients was collected with EDTA tube (ethylenediamine tetraacetic acid tube, BD Vacutainer). On average, 1.0 × 107 cells were isolated from 5 mL of whole blood. The collected whole blood was 2× diluted with Hanks’ Balanced Salt solution (HBSS, Gibco, Rockville, MD, USA) for loading on Ficoll-paque (Pharmacia, Uppsala, Sweden). The blood-loaded sample on Ficoll solution is centrifuged at 2000 rpm for 20 min (Acceleration/Break = lowest/zero), and middle layer was collected as PBMC. The collected cells were enumerated and stored in liquid nitrogen tank until us. All study participants provided informed consent, and was approved by the ethical committee of Nanjing Stomatology Hospital, Medical School of Nanjing University.

For the cell subtypes of PBMC analysis, cells were collected and washed with PBS twice and then suspended in 200 μL PBS. For enumeration of mature human T (CD3⁺) cells, helper/inducer T (CD3⁺ CD4⁺) cells, cytotoxic T (CD3⁺ CD8⁺) cells, B (CD19⁺) cells, and NK (CD3⁻CD16⁺and/orCD56⁺) lymphocytes, CD3⁻FITC/CD8⁻PE/CD45⁻PerCP/CD4⁻APC reagent, BD Multitest CD3⁻FITC/CD16⁻PE⁺ CD56⁻PE/CD45⁻PerCP/CD19⁻APC reagent were used according to the manufacturer’s instructions, respectively (Cat No.340503, BD Multitest™), then quantified by flow cytometry on a FACS Calibur instrument.

The cBioPortal for Cancer Genomics (http://cBioPortal.org) is a website for exploration of multi-dimensional cancer genomics data, providing readily understandable gene expression event (Gao et al., 2013). We used cBioPortal to analyze the correlation between ICOSLG and specific lymphocyte subset markers as well as specific immune checkpoint molecules in HNSCC. Co-expression was calculated based on the cBioPortal’s online instructions.

Tumor Immune Single-cell Hub 2 (Tisch2, http://tisch.comp-genomics.org) is a scRNA-seq database focusing on tumor microenvironment (TME). TISCH2 provides detailed cell-type annotation at the single-cell level, enabling the exploration of TME across different cancer types. We used TISCH2 to evaluate the difference in ICOSLG between tumor cells and normal cells in different tumors. In addition, according to the online description of Tisch2, we also evaluated the correlation between ICOSLG and specific immune infiltrating cell subsets at the transcriptional level.

SPSS 18.0 and GraphPad Prism 8.0 software packages were used for data analysis and graphic processing. Pearson’s chi-square test, Fisher’s exact test and Chi-square test were used to compare clinicopathological features. The Mann–Whitney U test was used to compare the two groups. Survival analysis includes overall survival (OS), metastasis-free survival (MFS) and disease-free survival (DFS), which were evaluated by Kaplan–Meier and log-rank test. Further multivariate analysis was carried out by Cox proportional hazards regression model to determine the independent risk factors, adjusted hazard ratio (HR) and 95% confidence interval (CI) of OSCC. Co-expression between ICOSLG and immune cell markers and immune checkpoint molecules was investigated by Pearson correlation analysis. The partial Spearman’s correlation analysis was used to analyze the association between ICOSLG and markers of specific immune infiltrating cell subset at transcription level. All statistical tests were two-sided, and p < 0.05 was considered to be significant.

ICOSLG was expressed in the cell membrane and cytoplasm of tumor cells (TCs), cancer-associated fibroblasts (CAFs) and tumor infiltrating lymphocytes (TILs) in 105 patients with OSCC. The typical low expression and high expression of ICOSLG IHC staining were shown in Figure 2A. In OSCC, the IHC score of TILs was generally higher than that of TCs. (Figure 2B). In the TISCH2 database, single-cell sequencing analysis showed that the mRNA expression of ICOSLG was significantly higher in TILs in liver hepatocellular carcinoma (Figure 2C), colorectal cancer (Figure 2D), and head and neck squamous cell carcinoma (Figure 2E).

We then analyzed the relationship between the expression of ICOSLG and clinicopathological features of OSCC patients (Table 1). The results showed that in OSCC patients, the expression of ICOSLG was not significantly correlated with gender, age, and differentiation, but the high expression of ICOSLG in TCs (ICOSLGTCs) and TILs (ICOSLGTILs) was associated with higher risk of lymph node metastasis and advanced TNM stage. IHC results showed that high expression of ICOSLG in TCs (ICOSLGTCs) was associated with higher risk of lymph node metastasis (Figure 3A) and advanced TNM stage (Figure 3B), while ICOSLG in TILs (ICOSLGTILs) was significantly associated with advanced TNM stage (Figure 3C) and T stage (Figure 3D).We analyzed the Tisch2 database and found that in patients with head and neck squamous cell carcinoma, the high expression of ICOSLGTCs was significantly positively correlated with advanced TNM stage, while the high expression of ICOSLGTILs, especially CD4⁺ T cells, was negatively correlated with advanced TNM stage (Figure 3E). The high expression of ICOSLG on CD4⁺ T cells is positively correlated with advanced TNM stage in colorectal cancer (CRC) (Figure 3F).

In order to confirm the prognostic value of ICOSLG for OSCC, we used Kaplan-Meier survival rate to analyze the survival rate of patients with oral squamous cell carcinoma included in this study. The results showed that patients with increased ICOSLG expression in TCs had shorter overall survival (OS r = 3.184), metastasis-free survival (MFS r = 2.981) and disease-free survival (DFS r = 3.073) (Figures 4A–C). In addition, OSCC patients with more ICOSLG in TILs had shorter OS (r = 2.664), MFS (r = 2.700) and DFS (r = 2.747) (Figures 4G–I), which was not observed in ICOSLG in CAFs (ICOSLGCAFs) (Figures 4D–F).

We used univariate and multivariate Cox regression to analyze the prognostic value of clinicopathological features. The results showed that gender, age, TNM stage, T stage, differentiation and ICOSLG in CAFs (ICOSLGCAFs) had no significant predictive value for OS and MFS (all p > 0.05). Lymph node metastasis and high expression of ICOSLG in TCs and TILs were significantly different in OS and MFS, but not independent prognostic indicators of oral squamous cell carcinoma (Figure 4J).

To explore whether the immune cell subsets of tumor tissues with high expression of ICOSLGTCs in OSCC patients changed, we determined the ICOSLG, CD4, CD8, CD19, CD68 and Foxp3 level of tumor centers and infiltration fronts in serial sections (Figures 5A,B). Next, we compared the proportion of CD4, CD8, CD19, CD68 and Foxp3 positive cells in lymphocytes of patients with high and low expression of ICOSLGTCs, as well as the proportion of CD4, CD8 positive cells and Fopx3+ cells, and divided them into invasive Frontier (Figure 5C) and tumor center (Figure 5D) for analysis. CD4⁺ cells and CD19⁺ cells in patients with high expression of ICOSLGTCs at the invasive Frontier showed a decreasing trend, while Foxp3+ cells showed an increasing trend. However, CD4⁺ cells and CD8⁺ cells in patients with high expression of ICOSLGTCs at the tumor center showed a decreasing trend.

We also analyzed the ICOSLGTILs-regulated in the intra-group correlation comparison of CD4, CD8, CD19, CD68, Foxp3, CD8/Foxp3, ICOSLGTCs, ICOSLGCAFs and ICOSLGTILs, we found that ICOSLGTILs were negatively correlated with CD4 and CD19 in the invasive Frontier (Figure 5E), while ICOSLGTILs were positively correlated with Foxp3 and ICOSLGTCs. In the tumor center (Figure 5F), ICOSLGTCs and ICOSLGTILs were negatively correlated with CD4 and CD8. Immune checkpoint proteins play a crucial role in the negative regulation of cellular immunity. Therefore, we used cBioPortal to further analyze the correlation between ICOSLG and immune checkpoint molecules, and analyzed the correlation between immune checkpoints (Figure 5G). We found that ICOSLG was positively correlated with inducible T cell co-stimulator (ICOS r = 0.403), TACTILE (CD96 r = 0.444), programmed death-ligand 1 (CD274 r = 0.124), colony-stimulating factor 1 receptor (CSF1R r = 0.409), cytotoxic T lymphocyte antigen 4 (CTLA4 r = 0.356), hepatitis A virus cellular receptor 2 (HAVCR2 r = 0.404), indoleamine 2,3-dioxygenase 1(IDO1 r = 0.224), interleukin 10 (IL10 r = 0.247), programmed cell death 1 (PDCD1 r = 0.400), and these immune checkpoints were also positively correlated.

IHC serial sections of OSCC patients showed a correlation between ICOSLG and the number of CD4⁺, CD8⁺, CD19⁺, Foxp3+ cells in situ. Therefore, we used flow cytometry to analyze the proportion of peripheral blood T, B and NK cells between the low ICOSLG group and the high ICOSLG group, and the strategy of gating lymphocytes was shown in Figure 6A. We found that the absolute counts of CD3⁺ T cells, CD3⁺ CD4⁺ T cells and CD3⁺ CD8⁺ T cells in ICOSLGTCs high expression samples were significantly lower (Figure 6E). However, there was no difference in the percentage of lymphocyte subsets in ICOSLGTCs (Figure 6B) and the frequency and number of lymphocyte subsets in ICOSLGCAFs (Figures 6C, F), ICOSLGTILs (Figures 6D, G) between the low and high subgroups of ICOSLG.