Shandong Cancer Hospital and Institute Jinan, Shandong
L. Li1, N. Liu1, G. Wang1, and S. Yuan1,2; 1Department of Radiation Oncology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, Shandong, China, 2Department of Radiation Oncology, The First Affiliated Hospital of USTC, Division of Life Sciences and Medicine, University of Science and Technology of China, Hefei, Anhui, China, Hefei, China
Purpose/Objective(s): Realization of gene function is mainly achieved by encoded proteins’ expression information. The aim of this study was to analyze the correlation between proteomic signatures of baseline tumor samples and response of radical chemoradiotherapy in locally advanced esophageal squamous cell cancer (ESCC). Materials/
Methods: Proteomic data of 180 patients with locally advanced ESCC who underwent radical chemoradiotherapy were analyzed. Differentially expressed protein analysis and pathway enrichment analysis were used to identified the biological process associated with chemoradiotherapy response. For the response prediction, a random forest recursive feature elimination (RF-RFE) approach was used for feature selection and modeling. Results: Totally 6,770 proteins were detected in our study, among which 218 proteins where differentially expressed in responders and non-responders. Compared with non-responders, 117 proteins were upregulated and 101 proteins were downregulated in the responders. The upregulated proteins were enriched immune related pathways, while the downregulated proteins were enriched in metabolism related pathways, consistent with gene set enrichment analysis (GSEA) results. Based on the differentially expressed proteins, we built a response assessment RF-RFE model with 22 features, with AUROC at 0.83 in the train set and 0.86 at test set. We further classify patients by combining efficacy and toxicity into two groups, including responders with low toxicity (Good) and non-responders or high toxicity (Poor). Interestingly, we obtained a RF-RFE model with higher performance, with AUROC at 0.97 in the train set and 0.95 at test set. Among features used for Good-Poor assessment, two proteins (Q16401_PSMD5, Q15257_PTPA) were associated with patient PFS. Combinatorial analysis of PSMD5 and PTPA revealed two distinct risk groups, patients in the low-risk group experienced markedly prolonged PFS (median PFS = 13.47 vs 22.83 months, p=0.0083) compared with those in the high-risk group. The model showed good performance with AUC of 0.67, 0.78 and 0.67 in predicting 2 years, 3 years and 4 years’ PFS. Conclusion: This study shows associations between proteomic characteristics and the response to radical chemoradiotherapy, and obtains a RF-RFE model with high performance in locally advanced ESCC.
