High CD3 and ICOS and low TIM-3 expression predict favourable survival in resected oesophageal squamous cell carcinoma

With the increasing oncological potential of immunotherapy, several immune checkpoint modulators are being investigated. The value of immune markers, including programmed cell death ligand-1, programmed cell death-1 (PD-1), inducible co-stimulator (ICOS), lymphocyte activation gene-3, T-cell immunoglobulin, and mucin-dominant containing-3 (TIM-3), is not well known. Using tissue microarrays of 396 patients who underwent surgery for oesophageal squamous cell carcinoma (ESCC), infiltrated T-cell subsets (CD3, CD8, and Foxp3) and checkpoint protein expression were scored. With a median follow-up of 24.8 months, CD3+ TIL subsets (50.0%) had longer median recurrence-free survival (RFS, 55.0 vs 21.4 months) and overall survival (OS, 77.7 vs 35.8 months). Patients with high ICOS expression (46.5%) had longer median RFS (53.9 vs 25.3 months) and OS (88.8 vs 36.9 months). For PD-1, RFS (hazard ratio [HR] 0.67) and OS (HR 0.66) were significantly longer in the high-expression group (45.2%). In the multivariate analysis, high TIM-3 expression (50.8%) had a significant relationship with shorter RFS (HR = 1.52) and OS (HR = 1.60). High CD3+ TIL and T-cell ICOS expression were associated with favourable prognosis, whereas high TIM-3 expression suggested a poor prognosis. Our findings may confer new insights to improve ESCC outcomes beyond the application of PD-1 blockade.

treatment. Clinical trials with pembrolizumab and nivolumab are ongoing and include almost every tumour type from solid tumours to haematologic malignancies and from neoadjuvant and adjuvant to the palliative setting 7 .
Here, we investigated the distribution and frequency of CD3 + /CD8 + T-cells, regulatory T-cells, and other immune checkpoints, including PD-L1, PD-1, ICOS, LAG-3, and TIM-3, in surgically resected ESCC and also determined their prognostic value in identifying potential therapeutic targets.

Results
Baseline characteristics. Overall, 407 patients who underwent surgical resection for ESCC were identified.

Prognostic impact of PD-L1 expression on tumour and immune cells. PD-L1 expression on
tumour and immune cells with various cut-offs (1%, 5%, 10%, 25%, and 50%) were evaluated among 396 samples and analysed in relation to the survival data (Fig. 3). The median PD-L1 expression rate was 0% in both tumour and immune cells, while the corresponding means were 5.6% and 8.0%, respectively. Even after incorporating various cut-off points in tumour and immune cells, our data indicated no significant correlation between PD-L1 expression and RFS and OS. Multivariate Cox analysis showed similar results (Tables 2 and 3). Median RFS and OS of patients with PD-L1 expression in ≥50% immune cells (n = 21, 5.3%) were numerically greater than those in patients with PD-L1 expression in <50% of cells (median RFS: 111.1 vs 30.7 months, P = 0.355, median OS: 111.1 vs 42.3 months, P = 0.367). However, this difference was not statistically significant.

Relationships among expression of various immune markers.
Overall, the median expression levels of FoxP3, ICOS, PD-1, and PD-L1 were 0, and any expression from these markers was considered as a "higher expression". Moreover, the expression of one immune marker has close associations with other markers (Supplementary Table 10). After performing Spearman's rank correlation analysis, CD8 + TIL expression was related with LAG-3, PD-1, TIM-3, and PD-L1 expression in immune cells (P < 0.001, respectively). Two immune markers, CD3 and ICOS, that displayed a prognostic role in this study were linked (r = 0.128, P = 0.011). However, no significant association was present between CD8 and ICOS expression or between ICOS and LAG-3 expression. into four subgroups based on CD3 + TIL positivity and ICOS expression, and the same method was applied for CD3 with PD-1 and CD3 with TIM-3, as these were proposed to have a prognostic impact in the current analysis. The Kaplan-Meier analysis revealed significant differences in RFS among these four types in each subgroup ( Supplementary Fig. 4). Consistent with the strong prognostic characteristics of ICOS and PD-1, CD3 + subgroups with positive ICOS or PD-1 expression showed the longest RFS (median: 88.8 and 58.0 months, respectively) compared with subgroups that were CD3 + /ICOS-or CD3 + /PD-1-(median: 18.6 and 17.6 months, respectively). Furthermore, consistent with the poor prognostic profile of TIM-3, subgroups that were CD3 + /TIM-3 − demonstrated the longest RFS (median: 59.7 vs 3.5 months in CD3 − /TIM-3 − ). However, these correlations with RFS did not translate into OS.

Discussion
ESCC is one of the most aggressive cancers. Definitive surgical therapy, including oesophagectomy, is considered a standard treatment for resectable diseases; however, >50% of patients undergo local recurrence and/or distant metastasis 15,16 . Even with recent improvements in sequencing techniques and precision medicine, molecularly targeted therapy has a limited role. Furthermore, traditional cytotoxic agents have not demonstrated progress in the last decade. Therefore, ESCC prognosis is dismal, with a high unmet need. In other cancer types such as lung cancer, immune checkpoint inhibitors, represented by PD-1/PD-L1 blockade, have recently shown promising efficacy in treating advanced/metastatic ESCC 17,18 . In a clinical trial, the objective response rate of 17% and a median OS of 10.8 months appear to be encouraging considering the heavily treated characteristics in the study population 18 . Despite the impressive outcomes of PD-1/PD-L1 inhibitory cancer immunotherapy, efficacy is suboptimal, and most patients with many tumour types do not show a response 19 . Thus, the focus has shifted to targeting alternative immune checkpoints receptors.
Combining two or more anti-tumour drugs for improving cancer treatment is not a new concept in oncology, and it is also valid in the immuno-oncology era 20 . Combination therapy involving PD-1/PD-L1 inhibitors with other immune checkpoint modulators has been considered an important strategy to overcome primary and secondary resistances to PD-1/PD-L1 blockade 21 . Several immune modulators are being investigated in pre-clinical models and clinical trials; however, our understanding of these checkpoints is incomplete, especially in ESCC. Furthermore, the expression level cut-offs, associations with clinicopathological features, and prognostic impact of these various checkpoints have not been established for ESCC. The PD-L1 expression level has a critical role in T-cell regulation, but the precise significance varies among studies and cancer types 22 . We comprehensively investigated the prognostic impact of immune cell infiltration and various immune checkpoints, including PD-L1, PD-1, ICOS, LAG-3, and TIM-3, in resected ESCC. Additionally, the significance of clinical factors such as sex, age, smoking status, and TNM stage was explored. The most significant prognostic factor was the TNM stage (Table 2 and Supplementary Fig. 5). Given that the number of stage IV subjects was small, it may not have a significant impact on OS and mitigate the influence of immune markers.
Several studies report that TILs have a positive prognostic effect across various cancer types [23][24][25] . TILs have an important role in anticancer immune responses and the cancer immunity cycle 26 . However, TILs are divided into several subgroups based on cell surface antigen expression. Each subset has a distinct interactive function between the tumour and tumour microenvironments and sometimes the roles differ across cancer types 27 . CD3 is a transmembrane protein that is exclusively expressed in the T-cell pedigree and thus is considered to represent a mature T-cell marker and whole tumour-infiltrating T-cells 28 . CD8 T-cells generally represent cytotoxic T-cells, and Foxp3 is the distinct transcription marker of regulatory T-cells 29 . Among different lymphocyte types, CD3-positive TILs are most significantly associated with survival across various tumours 27 . Consistent with previous literature on total TILs or stromal TILs in ESCC, in the present ESCC study, high CD3 + TIL density showed good prognostic impact, whereas CD8 + TILs and Foxp3 + TILs did not [30][31][32] . We attempted to investigate stromal and intra-tumoural CD3 + or CD8 + TILs independently; however, TIL abundance made it unfeasible. One meta-analysis showed that CD8 + TILs were associated with good prognosis; however, it included a heterogeneous population of patients with oesophageal adenocarcinoma and squamous cell carcinoma 33 .
To our knowledge, we are the first to report ICOS, LAG-3, and TIM-3 expressions and their clinical implication in a large ESCC cohort. As expected, ICOS, LAG-3, and TIM-3 expressions were related to PD-1 expression on TILs, possibly indicating the feedback nature of the immune system (Supplementary Table 10).
ICOS is a surface antigen in T-cells, and its expression is low in naïve T-cells. Once ICOS is stimulated by both the T-cell receptor and CD28 signals, it is upregulated in activated T-cells 34 . Co-stimulation by ICOS and ICOS ligand confers an anticancer response, but ICOS signalling also engages regulatory T-cell activity induction 35   www.nature.com/scientificreports www.nature.com/scientificreports/ across various tumour types (e.g., melanoma, colorectal cancer, breast cancer, gastric cancer, and renal cell carcinoma), and these clinical findings support the dual role of ICOS in carcinogenesis 8 . In our study, high ICOS expression in immune cells was significantly associated with longer RFS and OS in the Kaplan-Meier method and multivariate Cox regression model ( Fig. 2 and Table 2). Our findings suggest that ICOS-expressing immune cells include tumour neoantigen-specific T-cells and are associated with favourable prognosis, at least in ESCC. However, this should be validated in an independent patient cohort and in clinical trials with ICOS-targeted therapies.
Despite the existence of several LAG-3-related studies, data regarding the prognostic impact of LAG-3 in most cancer types are limited 9 . Two reports on NSCLC have contradictory results 36,37 . Moreover, one exploratory study www.nature.com/scientificreports www.nature.com/scientificreports/ showed that LAG-3 expression was associated with good prognosis in resected ESCC; however, this is inconsistent with the current data 38 . Accordingly, the role of LAG-3 in cancer and ESCC in particular remains elusive.
As a marker of CD8 + T-cell dysfunction or exhaustion, TIM-3 is occasionally co-expressed with PD-1 in CD8 + T-cells in several tumours 19 . Moreover, TIM-3 is expressed in natural killer cells, regulatory T-cells, dendritic cells, B-cells, macrophages, and other myeloid cells. TIM-3 expression inhibited anti-tumour immunity by T-cell exhaustion, suppressed immune response from the status of innate immunity, and exhibited pro-tumour activity by promoting myeloid-derived suppressor cells 9 . Consistent with the previously known immune regulatory role of TIM-3 in oncogenesis, TIM-3 expression is associated with poor prognosis in several cancers 39,40 . The high probability of recurrence or death was also related to high TIM-3 expression in the present study, which is the first to report the prognostic impact of TIM-3 in ESCC. In several preclinical studies, blocking TIM-3 enhanced cancer immunity by T-cell proliferation and cytokine production, as well as combination with PD-1 blockade, also demonstrated a remarkable synergy in these models 10 . Additionally, many clinical trials focusing on targeting TIM-3 are ongoing globally 41 . One important issue regarding blocking TIM-3 might be related to their expression levels in various tumour-infiltrating immune cells. Most patients (92%, 364/396, data not shown) in the present study expressed TIM-3 in T-cells. Thus, based on the high frequency of TIM-3 expression and its demonstrated poor prognostic value, TIM-3 may be a potential target candidate in ESCC. In conjunction with the current role of PD-1 blocking agents in ESCC, the use of TIM-3-targeted therapy in combination with PD-1 blockade may be a promising strategy. Although more studies are needed, our study may provide some evidence supporting the approach to target TIM-3 in ESCC.
Together with advances in PD-1/PD-L1 blockade, the relationship of PD-1/PD-L1 expression and its prognostic impact in various tumour types are gaining increased attention. Although the Kaplan-Meier method revealed that high PD-1 expression in TILs was associated with better RFS and OS, in Cox regression analyses, patients with high PD-1 expression had increased risk of relapse and death compared with those with low PD-1 expression. Considering the high association with PD-1 and other immune checkpoints, the overlap and diverse interaction between various immune checkpoints may attenuate the prognostic impact of PD-1 in ESCC (Supplementary Table 10). www.nature.com/scientificreports www.nature.com/scientificreports/ Regarding PD-L1, 25.5% (101/396) of patients showed PD-L1 expression in ≥1% of tumour cells and in 47.7% (189/396) of immune cells (Supplementary Fig. 3). We used various cut-offs of PD-L1 expression in tumour or immune cells to determine associations with prognosis, but we did not identify any meaningful associations with recurrence or survival. Because blocking PD-1/PD-L1 has gained much attention, several studies have investigated the prognostic role of PD-L1 expression in ESCC. However, these studies presented inconsistent results 30 . Some studies reported that PD-L1 expression was associated with poor prognosis, but another study indicated it was not associated with poor prognosis. A meta-analysis showed a positive but statistically non-significant trend in prognosis with PD-L1 expression 42 . In our study, there were no clinically significant relationships between survival and PD-L1 expression in tumour or immune cells. These conflicting results might derive from differences in the patient population, staining method and antibodies, and definition of PD-L1 positivity. Thus, heterogeneity in the prognostic role of PD-L1 has been observed in several cancer types.
Our study has some limitations. First, it was retrospective in nature. Second, it did not include the localisation information of each marker and therefore did not provide data on the spatial distribution of immune cells. Furthermore, since most patients underwent oesophagectomy when the neoadjuvant concurrent chemoradiation data was published and incorporated into clinical practice, none of the patients received preoperative treatment 43 . Additionally, no patient underwent immunotherapy; thus, this study did not provide any data regarding the response to immune checkpoint modulators.
In conclusion, we demonstrated that a high frequency of CD3 + TILs and high ICOS expression on immune cells are related to a favourable prognosis in ESCC. Furthermore, TIM-3 expression in immune cells was associated with poor prognosis in uni-and multivariate analyses. Our findings may help improve ESCC outcomes beyond the application of PD-1 blockade. These discoveries have important implications for using antibody therapies in addition to the PD-1/PD-L1 signalling pathway, such as TIM-3, in ESCC.

Methods
Patients and tissue samples. The study population comprised ESCC patients who underwent radical surgical resection at Severance Hospital and Samsung Medical Center, Seoul, Korea between 1996 and 2012. The inclusion criteria were (1) surgically resected ESCC with a curative aim and (2) availability of tumour tissue. Clinicopathologic data were collected and reviewed. Tumours were re-classified based on the 7 th edition of the AJCC TNM cancer staging system. Twelve subjects underwent oesophagectomy and were subsequently identified as having stage IV. The study was approved by the Institutional Review Board (IRB) of Severance Hospital and was conducted according to the ethical principles, guidelines, and relevant regulations. The IRB approved our research, and as the study was a retrospective review of subjects, the requirement for informed consent was waived.
Tissue microarray preparation and IHC. ESCC specimens were histologically reviewed by two experienced pathologists (Y.C. and S.S.). Haematoxylin-and eosin-stained sections from formalin-fixed paraffin-embedded tissues were reviewed to identify the invasive carcinoma area. The most densely viable carcinoma areas were chosen as the representative areas (core -3.0 mm). Tissue microarrays were constructed from these samples and assessed by immunohistochemical staining for PD-L1, PD-1, CD3, CD8, Foxp3 (forkhead box P3), ICOS, TIM-3, and LAG-3.
IHC scoring. Two well-experienced pathologists manually scored each IHC in the tissue microarray samples.
PD-L1 expression on tumour and immune cells was analysed separately. Tumour cells showing any intensity of membranous and/or cytoplasmic staining were defined as having positive staining, and the proportion of PD-L1-positive tumour cells was estimated as the percentage of total tumour cells in the whole section. The pre-specified cut-off values at 1%, 5%, 10%, 10%, 25%, and 50% were established. The same method was applied to assess PD-L1 expression in immune cells. The expression was defined as high density for CD3, CD8, ICOS, Foxp3, LAG-3, and PD-1 when the expression level was above the median value per four high-power fields (4HPFs).

Statistical analysis.
We investigated the association between various immune markers and clinical data using the chi-squared test or Fisher's exact test. Survival was analysed by the Kaplan-Meier method with log-rank for categorical variables. Uni-and multivariate analyses for predicting recurrence and survival were performed using the Cox regression method for the following variables: age, sex, smoking status, stage, histology, location, and each immune marker status. RFS was defined from the time of surgery to initial relapse or death. OS was defined as the time from the surgery until death from any cause or the most recent follow-up. The cases for OS included cases that were still alive and cases whose causes of death were not related to oesophageal cancer. Living patients were censored at the time of the last follow-up. Spearman's rank correlation test was used to assess the relationship among various immune markers. All statistical analyses were performed using SPSS version 23.0 (SPSS, Chicago, IL). P < 0.05 was considered statistically significant by the Kaplan-Meier method.