Impact of intratumoural CD96 expression on clinical outcome and therapeutic benefit in gastric cancer

2.1 Study population

The research consisted of three independent patient cohorts. Cohort 1 included 496 patients from Zhongshan Hospital (ZSHS), Fudan University (Shanghai, China), the ZSHS cohort, with 59 patients excluded because of missing data, dot loss or because they suffered from metastatic diseases. The remaining 437 patients underwent radical gastrectomy and standard D2 lymphadenectomy between August 2007 and December 2008. All tissue samples from the ZSHS cohort were formalin-fixed and paraffin-embedded (FFPE). Patient clinicopathological characteristics, including age, gender, tumor size, tumor grade, Lauren's classification, tumor-node-metastasis (TNM) stage, and application of fluorouracil-based ACT were collected retrospectively. The T, N classification and TNM stage were assessed according to the 2010 International Union Against Cancer TNM staging system. Postoperative routine fluorouracil-based ACT was primarily given to patients with TNM II/III advanced tumors. No radiotherapy was administered to enrolled patients in ZSHS cohort. The endpoints of interest were overall survival (OS) and disease-free survival (DFS), computed from the date of gastrectomy to the date of death or disease recurrence, or the last follow-up. Subsequently, cohort 1 was randomly assigned into two independent data sets (discovery set, n = 219; validation set, n = 218). Our research had approval from the Clinical Research Ethics Committee of Zhongshan Hospital, Fudan University. All enrolled patients were informed of the usage of resected gastric tissue samples. Written consent was acquired from each patient. Cohort 2 was derived from the Cancer Genome Atlas (TCGA) with 412 GC patients in all, but 37 were excluded because of missing data or because they suffered from metastatic diseases. All patient characteristics and mRNA data were downloaded from https://xenabrowser.net/datapages/ on September 9, 2020. Cohort 3 was derived from ERP107734 published on the European Nucleotide Archive (ENA) with 61 patients treated with pembrolizumab. Due to the integrity of the mRNA expression matrix uploaded, we included 39 patients with certain characteristics in our research. An illustration of patients enrolled and study design is presented in Figure S1.

2.2 Immunohistochemistry

Immunohistochemistry (IHC) was applied to detect the expression of CD96 and the infiltration of immune cells on each TMA slide. The TMAs were constructed by Shanghai Outdo Biotech Co, Ltd. The protocol details of TMA construction and IHC staining have been described elsewhere.^{24, 25} Blinded to the clinical data, two pathologists evaluated the IHC score of CD96 independently according to the number of stained cells in three randomly selected 200× fields of view. The mean score of their evaluation was adopted. The cut-off value for the classification of high CD96 and low CD96 subgroups was the median value. On the basis of TCGA data, CIBERSORT was constructed to calculate the relative proportion of 22 immune cell types recognized as LM22. In addition, we selected CD8⁺ T cells, Treg cells (Foxp3⁺), and M2 macrophages (CD163⁺) as significant immune contexture in GC with high CD96⁺ cell infiltration. The associated antibodies are listed in Table S1.

2.3 Statistical analysis

SPSS 21.0 (SPSS Inc.) was applied for statistical analysis. The cut-off value for CD96 expression was the median value. Pearson's χ² test and Fisher's exact test were applied for categorical variables, and Student's t-test, the Mann Whitney U test, and one-way ANOVA test were applied for continuous variables. Kaplan–Meier curves, the log-rank test, and univariate and multivariate analyses were applied for survival outcomes. Receiver operating characteristic (ROC) analysis was used to compare the accuracy of the prediction of clinical outcome by the parameters. All analyses mentioned above were visualized by R (4.1.1). All statistical analyses were two-sided and p < 0.05 was regarded as statistically significant.

2.4 In silico analysis

The gene expression profile data in the TCGA cohort were used to quantify the infiltration of immune cells in tumor tissues by single-sample gene set enrichment analysis (ssGSEA)²⁶ in the R Bioconductor package Gene Set Variation Analysis (GSVA), and the infiltration of immune cells was obtained. The ssGSEA algorithm is a rank-based method that computes an enrichment score representing the degree to which genes in a particular gene set are coordinately up/downregulated in a single sample. The source and details of the gene signature are listed in Table S2.

3.1 CD96⁺ cells are enriched in GC tissues and associated with tumor progression

To discover the clinical roles of CD96 in the tumor progression of GC, clinical information for all GC patients enrolled in the ZSHS cohort was retrospectively collected and IHC staining of CD96 in gastric tissues is shown in Figure S2A. Interestingly, the intratumoral tissues showed significantly higher infiltration of CD96⁺ cells compared with peritumoral tissues (p < 0.001; Figure S2B), therefore the role of intratumoral CD96⁺ cells was the main focus of our following study. The clinicopathological characteristics of GC patients are listed in Table 1. Notably, the infiltration of CD96⁺ cells was significantly associated with tumor stages in GC and elevated in TNM stage III tumors, indicating that CD96⁺ cells might be correlated with the progression of GC (Figure S2C). Taken together, these findings suggest that CD96⁺ cells are enriched in GC tissues and associated with tumor progression.

3.2 High infiltration of CD96⁺ cells predicts poor prognosis in GC patients

To investigate the prognostic significance of CD96 in GC, clinical outcomes of GC patients with high CD96⁺ cells infiltration were compared with those of patients with low CD96⁺ cell infiltration. In both the discovery and validation sets, the high CD96⁺ cell subgroup patients had significantly inferior OS and DFS (p < 0.001, p < 0.001, p < 0.001, and p < 0.001; Figure 1A,B). Since CD96⁺ cell infiltration appears to be a potential prognosticator for GC, multivariate analysis based on clinicopathological characteristics was constructed to demonstrate whether CD96⁺ cells could serve as a potential independent prognostic factor for the survival outcomes. Notably, CD96⁺ cell infiltration was identified as an independent adverse prognostic factor for OS and DFS in both the discovery (hazard ratio [HR] 2.711, 95% confidence interval [CI] 1.788–4.111, p < 0.001 and HR 2.207, 95% CI 1.473–3.308, p < 0.001; Figure 1C) and validation (HR 2.848, 95% CI 1.841–4.408, p < 0.001 and HR 2.149, 95% CI 1.399–3.302, p < 0.001; Figure 1C) sets. Hence, high infiltration of CD96⁺ cells could predict inferior prognosis in GC patients.

3.3 CD96 indicates therapeutic responsiveness in GC patients

We investigated whether CD96⁺ cell infiltration could predict therapeutic responsiveness in GC patients. Since we found that receiving fluorouracil-based postoperative adjuvant chemotherapy could indicate superior overall survival for TNM stage II/III patients (p < 0.001; Figure 2A), we sought to explore the interaction between different CD96⁺ cell infiltration subgroups and the therapeutic responsiveness to ACT in stage II/III patients. Interestingly, ACT could lead to significantly better OS in low CD96⁺ cell subgroup patients, contrary to high CD96⁺ cell subgroup patients(p < 0.001 and p = 0.340, respectively; Figure 2A). This might indicate that low CD96⁺ cell subgroup patients received better survival benefits after ACT (p for interaction = 0.001; Figure 2A).

In addition to the ZSHS cohort, we enrolled a cohort of GC patients treated with pembrolizumab published as ERP107734 on the ENA to examine the potential value of CD96 in indicating immunotherapeutic responsiveness to PD-1 inhibition in GC. The baseline characteristics of GC patients are listed in Table S3.²⁷ Notably, response rates were significantly associated with CD96 mRNA expression (p = 0.002; Figure 2B), with objective response rates (ORRs) of 50.0% and 4.8% in the high CD96 subgroup and the low CD96 subgroup, respectively. Since the PD-L1 combined positive score (CPS) has been proven to be an indicator of response to pembrolizumab and is widely used as a clinical guideline,²⁸ we conducted a combined analysis of PD-L1 CPS and CD96 expression to more accurately identify patients who respond well to pembrolizumab. Interestingly, PD-L1-positive CD96 high patients showed the best responsiveness to pembrolizumab (Figure 2C). Additionally, several predictive indicators and characteristics of responsiveness to pembrolizumab in terms of immune signature, microsatellite instability, mutational load, mesenchymal subtype, and molecular subtype were grouped. Notably, the patients with better responsiveness showed an enrichment of biomarkers, including antitumor cytokines, checkpoint molecules, co-stimulatory ligands, co-stimulatory receptors, effector cells and effector cells traffic, Treg and Th2 traffic, and Treg signature, indicating a high-infiltration immune microenvironment (Figure 2C).

Accumulatively, our results suggest that CD96 could serve as a potential predictive biomarker for ACT and ICB treatment, as well as a promising therapeutic target for individualized precision medicine treatment of GC patients.

3.4 CD96 is associated with an immunosuppressive contexture characterized by an exhausted CD8⁺ T-cell phenotype in GC

Given that CD96 is regarded as an immune checkpoint,¹⁸ we next focused on the potential impact of CD96 on immune contexture. CIBERSORT was performed to investigate the association between CD96 mRNA expression and the typical immune cells infiltration in the TCGA cohort. Notably, high expression of CD96 mRNA expression was associated with high infiltration of memory B cells, CD8⁺ T cells, activated memory CD4⁺ T cells, follicular helper T cells, and M1 macrophages while low expression of CD96 mRNA was associated with resting memory CD4⁺ T cells, M0 macrophages, resting mast cells, and neutrophils (Figure 3A). To depict the functional states of immune cells more precisely, we found that CD96 mRNA expression was significantly associated with upregulation of both immune checkpoints and effector molecule mRNA level, which might indicate an inflamed but exhausted immune microenvironment (Figure 3A). To validate the results derived from the TCGA database, we further conducted IHC staining of immune cells, immune checkpoints, and effector molecules on the TMAs from the ZSHS cohort. Remarkably, high infiltration of CD96⁺ cells was significantly associated with high infiltration of CD8⁺ T cells, regulatory T cells, and M2 macrophages, and upregulation of interferon-γ, PD-L1, and lymphocyte-activation gene 3 IHC score (p = 0.006, p < 0.001, p < 0.001, p = 0.001, p < 0.001, p = 0.001; Figure 3B), which is consistent with the deduction that CD96 could potentially facilitate the immunosuppressive contexture. We further explored the association of these immune markers with the survival of GC patients and found that only PD-L1 expression significantly correlated with the poor prognosis in the ZSHS cohort (HR 1.619, 95% CI 1.213–2.162, p < 0.001; Figure S3). However, ROC analysis indicated that CD96 was associated with significantly higher prognostic accuracy compared with PD-L1 (CD96: AUC = 0.686; PD-L1: AUC = 0.583; Figure S4).

High infiltration of CD8⁺ T cells was regarded as a positive prognosticator in most cancer types.²⁹ However, recent studies have indicated that the induction of T-cell dysfunction in tumors with high infiltration of CD8⁺ T cells could correlate with tumor immune evasion.³⁰ To verify if CD8⁺ T cells displayed an exhausted phenotype, we further performed GSEA analysis. Notably, the exhausted CD8⁺ T cell signatures were significantly enriched in the high CD96 subgroup (Figure 3C). Furthermore, we combined CD96 and CD8 levels for survival analysis to discover the clinical significance of CD96-associated cytotoxic T-cell dysfunction. The analysis demonstrated that high CD8⁺ T cell infiltration could only identify prolonged OS in the high CD96⁺ cell subgroup, rather than the low CD96⁺ cell subgroup (Figure 3D). These results indicate that high CD96 expression is associated with immunosuppressive contexture characterized by exhausted a CD8⁺ T-cell phenotype, leading to poor prognosis.

3.5 Features of gene mutations and molecular subtypes based on CD96 expression in GC

Considering that progressive accumulation of gene alterations could facilitate tumorigenesis,³¹ we intended to profile detailed associations between CD96 and genomic features by delineating a holistic gene alteration landscape of GC from the TCGA cohort (Figure 4A). As a result, we identified the top 20 variant mutated genes, and AT-rich interaction domain 1A (ARID1A) and phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) gene alterations were frequently found in the high CD96 subgroup while tumor protein p53 (TP53) and spectrin repeat containing nuclear envelope protein 1 (SYNE1) were in the low CD96 subgroup (p = 0.003, p < 0.001, p < 0.001, p = 0.003, p = 0.029; Figure 4B).

Further research has revealed that molecular subtypes of GC have provided novel perspectives for patient stratification and individualized therapy.^{32, 33} Notably, CD96 mRNA expression was highest in Epstein–Barr virus (EBV)-positive subtype GC and lowest in chromosomal instable (CIN) subtype GC in the TCGA cohort (Figure 4C).

Recently, molecular targeted therapies, which target specific molecular deviations in signal transduction pathways and/or cancer proteins, including trastuzumab, bemarituzumab, savolitinib and zolbetuximab, have attracted growing attention in GC precision medicine. Thus, we wondered whether there was an association between CD96 mRNA expression and activation of the corresponding signaling pathways. Known as promising therapeutic targets in GC, human epidermal growth factor receptor 2, fibroblast growth factor receptor 2, epidermal growth factor receptor, MET, angiogenesis, homologous recombination repair signaling score, and CLDN18.2 mRNA expression were all elevated in the high CD96 subgroup (p < 0.001, p = 0.040, p < 0.001, p < 0.001, p < 0.001, p < 0.001, p = 0.018; Figure 4D). Taken together, these results suggest that CD96 is associated with unique genomic characteristics and specific molecular subtypes, and might guide the application of targeted therapies in GC.