Immune Score Predicts Outcomes of Gastric Cancer Patients Treated with Adjuvant Chemoradiotherapy

Background Substantial evidence has demonstrated that tumor-infiltrating lymphocytes (TILs) are correlated with patient prognosis. The TIL-based immune score (IS) affects prognosis in various cancers, but its prognostic impact in gastric cancer (GC) patients treated with adjuvant chemoradiotherapy remains unclear. Methods A total of 101 GC patients who received chemoradiotherapy after gastrectomy were retrospectively analyzed in this study. Immunohistochemistry staining for CD3+ and CD8+ T-cell counts in both tumor center (CT) and invasive margin (IM) regions was built into the IS. Patients were then divided into three groups based on their differential IS levels. The correlation between IS and clinical parameters was analyzed. The prognostic impact of IS and clinical parameters was evaluated using Kaplan–Meier analysis and Cox proportional hazard regression analysis. Receiver operating characteristic (ROC) curves were plotted to compare the area under the curve (AUC) of IS with other clinical parameters. Nomograms for disease-free survival (DFS) and overall survival (OS) prediction were constructed based on the identified parameters. Results Finally, 20 (19.8%), 57 (56.4%), and 24 (23.8%) GC patients were identified with low, intermediate, and high IS levels, respectively. GC patients with higher IS levels exhibited better DFS (p < 0.001) and OS (p < 0.001). IS was an independent prognostic factor for both DFS (p < 0.001) and OS (p < 0.001) in multivariate analysis. IS presented a better predictive ability than the traditional pathological tumor-node-metastasis (pTNM) staging system (AUC: 0.801 vs. 0.677 and 0.800 vs. 0.660, respectively) with respect to both DFS and OS. The C-index of the nomograms for DFS and OS prediction was 0.737 and 0.774, respectively. Conclusions IS is a strong predictive factor for both DFS and OS in GC patients treated with adjuvant chemoradiotherapy, which may complement the traditional pTNM staging system.


Introduction
Gastric cancer (GC) is the fifth most common malignant tumor with the third highest mortality rate worldwide [1]. Despite the innovations of modern screening methods, emerging new drugs, and multidisciplinary management, the prognosis of GC patients remains poor [2][3][4]. e American Joint Committee on Cancer/tumor-node-metastasis (AJCC/TNM) staging system is now widely implemented for GC treatment in clinical practice [5]. However, due to tumor heterogeneity, patients in the same TNM stage may still have different clinical outcomes. Since the prognostic information provided by the TNM staging system is insufficient, many studies have turned to other markers, such as tumor biomarkers, molecular features, or genetic signatures [6,7]. However, only the intrinsic properties of tumors are considered in these measures, regardless of the tumor microenvironment (TME). e TME is where the host immune response occurs. Its components are quite complex and include various immune cells, fibroblasts, and tumor vasculature [8,9]. Tumor-infiltrating lymphocytes (TILs), key components in the TME, have now been shown to be strongly correlated with patient prognosis in various cancers [10][11][12][13]. To quantify TIL infiltration, the construction of immune score (IS) was attempted. Galon et al. proposed the "Immunoscore which combines CD3+ and CD8+ T-cell densities from the tumor center and invasive margin [14][15][16]. e international validation of consensus Immunoscore ® held by Pages et al.
Research studies on the IS for gastric cancer were attempted in several ways. Jiang et al. selected 5 (CD3 invasivemargin(IM) , CD3 centeroftumor(CT) , CD 8 IM , CD 45ROCT, and CD 66b IM ) of 27 immune features using the LASSO model to build the IS system [26]. Zhang et al. combined intratumoral and stromal TILs in the scoring system [27]. Recently, Yun et al. adopted the same protocol as Pages et al. to establish the IS system in stage II/III GC patients following 5-FU-based adjuvant chemotherapy [28]. Despite the fact that the exact scoring methods used in different studies were not the same, IS based on different immune cell densities in different locations was found to have potential prognostic value in gastric cancer. However, most research studies have been restricted to GC patients who received only adjuvant chemotherapy, while patients who received adjuvant chemoradiotherapy were excluded. Whether the IS system is predictive of prognosis in these patients is still unknown.
In this study, we developed an IS system using immunohistochemistry in GC patients who received adjuvant chemoradiotherapy after curative gastrectomy, aiming to explore the prognostic significance of IS.

Patients.
We retrospectively reviewed patients with resectable gastric cancer who underwent curative surgery followed by chemoradiation at our institute from 2006 to 2015. Criteria for inclusion were as follows: (1) histologically confirmed gastric adenocarcinoma, (2) underwent curative gastrectomy with D2 lymph node dissection, (3) received no preoperative treatment, (4) received adjuvant chemoradiotherapy, (5) adequate formalin-fixed and paraffin-embedded (FFPE) tissue blocks could be collected, and (6) complete clinical information. Patients diagnosed with metastatic lesions were excluded. Patients without adequate FFPE tissue blocks were also excluded. All these GC patients received curative D2 gastrectomy with R0 resection followed by adjuvant chemoradiation. e total dose was 45 Gy delivered in 25 fractions, 5 days per week. Patients received continuous fluorouracil infusion (225 mg/m 2 ) or oral capecitabine (625 mg/m 2 ) twice daily concurrently with radiotherapy from days 1 to 5 every week. Fluorouracil-based chemotherapy was performed in 1-2 cycles before and 4-6 cycles after radiotherapy. Patient clinicopathological information was retrieved, such as age, sex, Eastern Cooperative Oncology Group (ECOG) performance score, tumor location, lesion size, gastrectomy type, postoperative serum lactate dehydrogenase (LDH) level, carcinoembryonic antigen (CEA) level, histologic grade, lymphatic and vascular invasion (LVI), perineural invasion (PNI), Borrmann classification, and pT and pN staging. Borrmann classification includes type I (polypoid type), type II (ulcerated type with demarcated margin), type III (ulcerated type with peripheral infiltration), and type IV (diffuse infiltrating type) [29]. e pathologic stage was classified based on the 8th edition of the American Joint Committee on Cancer (AJCC). All patients were followed up for at least 3 years, with recurrence and death information records. Disease-free survival (DFS) was defined as the length of the period of survival without recurrence after surgery. Overall survival (OS) was calculated from the date of surgery to the date of death. is study was approved by the Ethics Committee of Fudan University Shanghai Cancer Center. e written informed consents were obtained from all patients included in this study.

Immunohistochemistry.
For each patient, one tumor tissue block with both tumor and invasive margins was chosen and made into 4 μm-thick paraffin-embedded tissue sections. Two tissue sections were dewaxed with xylene and rehydrated with decreasing concentrations of gradient alcohol. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 15 min. Antigen retrieval was performed in the steamer for 20 min with citrate antigen retrieval solution (pH 6.0). Nonspecific antigens were blocked with 10% normal goat serum for 1 hour at 37°C. en, the tissue sections were incubated with anti-CD3 antibody (ab5960, goat multiclonal, 1 : 200, Abcam, Cambridge, UK) and anti-CD8 antibody (#70306, mouse monoclonal, 1 : 500, Cell Signalling Technology, Danvers, MA, USA) in the wet box overnight at 4°C. Subsequently, the secondary antibody (GK5005, GTVision ™ Detection System/Mo&Rb, Gene Tech, Shanghai, China) was incubated for 30 min at 37°C, and diaminobenzidine (DAB) was used for chromogen reaction. After all of the above steps, all sections were counterstained with hematoxylin-eosin solution and dehydrated with increasing xylene and alcohol gradient solutions.

Immune Score Construction.
Finally, stained tissue sections were examined under the optical microscope. All stained sections were independently viewed by two observers who were blinded to the clinical information. We identified the tumor center (CT) and invasive margin (defined as the peritumoral area with a width of 0.6 mm, IM) under 200× magnification and chose 3 representative fields from each. e number of CD3+ or CD8+ T cells in each of the 3 fields was quantified by manual counting using ImageJ software v1.8.0 (Rawak Software Inc., Stuttgart, Germany). e cell density for each region is represented using the mean cell count of the 3 fields at 200× magnification. en, we established the IS based on the proposal raised by Pages et al. For each case, the CD3+ and CD8+ T-cell densities were compared to the entire group and converted into percentiles. e mean of the four percentiles (two from CT and two from IM) was then translated into the immune score. Based on the three-category immune score classification, 0∼25% was classified as low, 25%∼70% was classified as intermediate, and 70∼100% was classified as high.

2
Journal of Oncology

Statistical Analysis.
e association between markers was analyzed using Spearman's correlation test. r > 0.8, 0.5 ≤ r < 0.8, and r < 0.5 were classified as high, moderate, and low correlations, respectively. e correlation between IS level and clinicopathological factors was investigated using Pearson's chi-square test or Fisher's exact test. To compare survival among groups, Kaplan-Meier curves were developed. Prognostic factors for OS and DFS were identified using univariate and multivariate Cox regression analyses. Variables identified with p value <0.1 in univariate analysis were included in the further multivariate analysis. e prognostic nomograms were constructed based on those identified factors, and their predictive performance was assessed by the bootstrap-corrected Harrell's concordance index (C-index) and calibration curves. To compare the predictive ability among parameters, a receiver operating characteristic (ROC) curve was generated, and the area under the curve (AUC) value was calculated. All of the above analyses were performed using IBM SPSS 25.0 (IBM Inc., Armonk, NY, USA), GraphPad Prism 9 (GraphPad Software Inc., San Diego, CA, USA), and R software v4.0.5 (R Project for Statistical Computing, Vienna, Austria). e survival package, rms package, and time ROC package were used in R software. All p values are all two-sided. A p value <0.05 was considered statistically significant.

TILs and Immune Score.
e CD3+ and CD8+ T-cell counts in each region were quantified from stained tissue sections (representative pictures are shown in  Table 1). Both CD3+ and CD8+ cell counts in the CT region were slightly higher than those in the IM region. e density of CD3+ cells in the CT region was moderately correlated with that in the IM region (r � 0.680, p < 0.001).
e same results were observed in CD8+ cells (r � 0.681, p < 0.001, Supplementary Figure 1(A-D)). e immune score was constructed as previously defined based on the mean percentiles obtained. e numbers of patients classified as IS-low, IS-int, and IS-high were 20 (19.8%), 57 (56.4%), and 24 (23.8%), respectively. e association between IS and clinical parameters was explored, and patients with low IS had higher pT stage (p � 0.006), higher pN stage (p � 0.023), higher pTNM stage (p � 0.009), and larger lesions (p � 0.029) and were more likely to be LVI positive (p � 0.020). Although not statistically significant, patients with low IS tended to have a lower differentiation grade (p � 0.092) ( Table 1).

IS and Prognosis.
Next, the survival curves were drawnfor IS discrimination. A higher IS level indicated better DFS (p < 0.001) and OS (p < 0.001). e median DFS and median OS for IS-high group patients were not reached. e median DFS for the IS-low and IS-intermediate (IS-int) groups was 17.0 months (standard deviation, SD ± 3.1) and 40.7 months (SD ± 12.6), respectively. e median OS for the IS-low and IS-int groups was 20.4 months (standard deviation, SD ± 0.5) and 49.4 months (SD ± 9.1), respectively (Figures 2(a) and 2(b)).

Subgroup Analysis Based on Stages.
When stratified by stage, IS was a strong prognostic factor in stage III GC patients (p < 0.001 for DFS and p < 0.001 for OS) but not in stage II GC patients (p � 0.398 for DFS and p � 0.397 for OS) (Figures 2(c) and 2(d)). Among all patients, stage III GC patients with low IS levels had the worst survival. Stage III GC patients with high IS levels had a similar prognosis to stage II GC patients (p � 0.718 for DFS; p � 0.934 for OS).

e Predictive Ability of Parameters and Prognostic Nomogram Construction.
e area under the curve (AUC) was calculated to assess the predictive ability of the identified parameters from multivariate Cox regression analysis (Figures 3(a) and 3(d)). Using 3 years as a cutoff, Journal of Oncology the predictive ability of IS (AUC 0.801 and AUC 0.800, respectively) was significantly higher than that of pTNM staging (AUC 0.677 and AUC 0.660, respectively) for both DFS and OS.
e AUC was 0.601 for age in OS prediction. ough the AUC value changed with time, the predictive ability of IS was always higher than that of the pTNM staging system. e nomograms for DFS and OS prediction were constructed using the identified parameters (Figures 4 and 5). e internal validation was performed using the bootstrap method. e bootstrap-

Discussion
Gastric cancer is a highly heterogeneous cancer with a poor prognosis. Patients with the same AJCC/TNM staging can exhibit quite different clinical outcomes. erefore, the prognostic information provided by AJCC/TNM staging is not sufficient. We need more information to guide treatment strategies in clinical practice. e tumor microenvironment, where the host immune response against tumors occurs, is composed of various components, such as immune cells, fibroblasts, and epithelial cells. Substantial studies have shown that infiltrating lymphocytes are closely correlated with patient prognosis [30,31]. e immune score, which serves as a quantitative evaluation method for TILs, has been widely discussed. Based on the density and position of CD3+ and CD8+ cells, Galon   Journal of Oncology immune scoring system in colorectal cancer [12]. Recently, the consensus Immunoscore ® of colon cancer was validated in a large international, prospective, multicenter clinical trial, which showed that the IS was an independent prognostic factor for recurrence [17]. In gastric cancer, previous studies have also demonstrated the prognostic impact of immune cells, such as CD3+, CD8+, Foxp3, and CD45RO cells [32][33][34][35]. However, few studies have reported the immune score in GC, and the construction methods differed in those published essays. What is more, no study has reported an association between IS and GC patients who have received adjuvant chemoradiotherapy. In our study, we obtained CD3+ and CD8+ cell densities from different locations using immunohistochemistry. CD3 is a specific marker for T cells, and CD8 is a specific marker for cytotoxic T cells. ey were the most frequently used markers in previous studies, and research studies in gastric cancer have confirmed their role in prognosis prediction. Combining the CD3+ and CD8+ cell densities together, the TIL infiltration level is well reflected. Counts of these two markers in two different locations were obtained from stained tumor sections. A moderate correlation was found between the markers. A consensus about immune score construction has not yet been reached in gastric cancer. In some studies, a cutoff value, usually the median [22,36], is given for each marker to distinguish low infiltrating levels (score 0) from high infiltrating levels (score 1). By combining 2 markers in 2 regions together, patients obtain scores of 0 to 4. However, the median may not be the best cut off value, and much information could be lost in this way. In this study, we established the IS system using the consensus Immunoscore ® modified by Pages et al. e exact whole immune infiltration status is revealed by converted percentiles from the immune cell counts. Based on the IS levels, we classified patients into three groups, namely, IS-low (20,19.8%), IS-intermediate (57, 56.4%), and IS-high (24, 23.8%). e number of patients in each group was relatively balanced. rough analysis, we found that lower IS was associated with higher TNM stage, larger lesions, LVI-positive state, and lower differentiation grade. It seemed that a lower IS level was associated with a heavier tumor burden, which is generally consistent with previous studies. In this study, we explored the prognostic impact of IS. Kaplan-Meier curve analysis showed that IS was predictive for DFS and OS at p < 0.001. IS remained an independent prognostic factor in multivariate Cox regression analysis. A higher IS level indicated better DFS and OS, which is also concordant with previous studies. In addition, the IS score was a better predictive parameter than the TNM staging system. IS can be used to complement the TNM staging system and help to identify patients who can benefit from adjuvant chemoradiotherapy. Prognostic nomograms were constructed using the identified parameters from the Cox regression analysis. Both of them had a good predictive performance with C-index >0.7, and the corresponding calibration curves showed favorable results.
Subgroup analysis was also performed. When stratified by stage, we found a prognostic impact of IS in stage III but not stage II GC patients. is result may be caused by the small sample size of only 22 stage II GC patients. Among them, there were only 14 patients with intermediate IS levels, 8 patients with high IS levels, and no patients with low IS levels. To determine the exact prognostic effect of IS, we need to include more stage II GC patients in the future. Among stage III GC patients, those with high IS levels exhibited the best survival, which was no worse than that of the stage II GC patients (p � 0.718 for DFS, p � 0.934 for OS). Although no stage II GC patients with low IS level were included here and this result still needs to be validated in larger samples, it may suggest that stage III GC patients with high IS exhibit distinct biological behaviors from others.
is may help guide stratified treatment to avoid possible overtreatment.
In summary, IS was demonstrated as a strong prognostic indicator for GC patients in our study. Higher IS level, which meant higher TIL infiltration, was correlated with a better prognosis. However, how to use the immune score to guide the treatment remained unclear. Research studies on colon cancer found that those with high IS levels benefited from more cycles of adjuvant chemotherapy while those with low IS levels did not [17,18]. is may suggest a need for more aggressive treatment in GC patients with high IS levels, which should be explored in the future. Besides, since the abscopal effect showed the great potential of radiation on the activation of the immune system, the combination of radiation and immunotherapy has been a hot issue nowadays [37,38]. e numbers of studies about the impact of TIL infiltration on prognosis in immunotherapy-treated patients were performed [39,40]. We should expect the potential of TIL-based IS in guiding the immunotherapy and even its combination with radiotherapy.
ere are still limitations to this study. First, this was a single-center retrospective analysis with limited evidence for clinical application. Second, the relatively small sample size may lead to bias.
ird, the stained cells were counted manually, and this human error cannot be overlooked. Fourth, we may try to include more immune markers such as CD45RO, Foxp3, and PD-L1 to improve the IS. In addition, gene signatures like HER-2 and MSS state may also be included to explore its correlation with the tumor microenvironment. A multicenter prospectively designed study is required to further confirm the feasibility and reproducibility of these findings.

Conclusions
We established an immune scoring system in GC patients who received post-surgical chemoradiation in our study. Patients with high immune scores were more likely to have lower pT stages, lower pN stages, smaller lesions, and LVI-negative states. e immune score was found to be a powerful prognostic factor for both DFS and OS, which may complement the TNM system to better identify GC patients who may benefit from adjuvant chemoradiotherapy.

Data Availability
e data used to support the findings of this study are included within the article. Journal of Oncology 9