A study for prediction model of clinical response to programmed death-1 inhibitors in advanced gastric cancer

Abstract

Immune checkpoint inhibitors are an emerging treatment option for patients with advanced gastric cancer. The identification of predictive biomarkers or models is necessary for the selection of patients who might benefit the most from immunotherapy. This dissertation aims to discover a novel and clinically meaningful biomarker or predictive model for clinical response to PD-1 inhibitors in advanced gastric cancer. In chapter 3.2, histopathologic features were examined as a clinical predictive marker for clinical response. Seven histological features (signet ring cell, fibrous stroma, myxoid stroma, tumor-infiltrating lymphocytes [TILs], necrosis, tertiary lymphoid follicles, and ulceration) detected in surgical resected tissues (N = 44) were used to train the model. The presence of signet ring cell became an optimal decision model for pathology alone (AUC = 0.78). In chapter 3.3, gene expression data of RNA sequencing were analyzed as a clinical predictive marker for clinical response. Analysis of differentially expressed genes for prediction of genomics markers showed that C-X-C motif chemokine ligand 11 (CXCL11) was high in responders (P < 0.001). In chapter 3.4, immunohistochemistry (IHC) was performed to verify its potential as a biomarker. IHC revealed that CXCL11 was mainly expressed in the cytoplasm, and locally in the cell membrane. The expression of CXCL11 was associated with responsiveness (P = 0.003). In chapter 3.5, the response prediction model was trained by integrating the results of analysis of pathological factors and RNA sequencing. When trained with the C5.0 decision tree model, the categorical level of the expression of CXCL11, a single variable, was shown to be the best model (AUC = 0.812). The AUC of the random forest trained model was 0.944. Survival rate analysis revealed that the C5.0 trained model (log-rank P = 0.01 for progression-free survival; log-rank P = 0.012 for overall survival) and the random forest trained model (log-rank P < 0.001 for progression-free survival; log-rank P = 0.001 for overall survival) predicted prognosis more accurately than the PD-L1 test (log-rank P = 0.031 for progression-free survival; log-rank P = 0.107 for overall survival). Through four sub-chapters, two predictive models (signet ring cell and expression level of CXCL11 gene) were suggested: 1) The presence of signet ring cell can be a biomarker for predicting non-responsiveness with only histopathology. 2) The expression level of the CXCL11 gene can be a predictive model for responsiveness. These prediction model can be applied to the clinic and used to predict patient outcomes.