Abstract

e14661 Background: Response of neoadjuvant immunotherapy can vary among patients in locally advanced ESCC, with the lack of a promising biomarker which need fully capture the heterogeneity of TME. Now, based on final results of the SEEK-01 study (2023 ASCO, P#4047), we have, for the first time, established a precise correlation from baseline TME to pathological response, baseline LN status, and 1-year EFS. Methods: In SEEK-01 study, 22 locally advanced ESCC patients (cT2-4NxM0) were enrolled. Following neoadjuvant chemotherapy (paclitaxel-albumin+carboplatin) combined with tislelizumab, patients underwent minimal invasive esophagectomy (MIE). Fresh tumor tissues were obtained via endoscopy prior to neoadjuvant therapy for scRNAseq analysis. Pathological response was evaluated using the irRVT score, leading to classification into pCR (irRVT=0%), MPR (0<irRVT<10%), and non-responder (irRVT>60%). Baseline LN status was determined based on pathological diagnosis after surgery. We identified subclusters of tumor-associated myeloid cells and stroma cells that significantly predicted pathological response, baseline LN metastasis and recurrence within 1 year. Prediction model of efficacy for neoadjuvant immunotherapy was developed using baseline expression ratios of differential subclusters. Results: Among 22 enrolled patients, 18 completed neoadjuvant therapy and MIE. 4 achieved pCR, 2 got MPR and 6 were non-responder. 12 of 18 proved baseline LN metastasis. 5 of 18 suffered recurrence within 1 year. Myeloid cells, cancer-associated fibroblast (Caf) and endothelium (Endo) shew heterogeneity that can predict pathological response, baseline LN metastasis or recurrence within 1 year. Upgrade of CXCL9 + tumor-associated-macrophage (TAM), LAMP3 + mregDC3 was a strong predictor of pCR, while the upgrade of STAT1 + TAM, C1q + TAM, IL1b + TAM, IL1b + Neutrophil, C1q + Neutrophil was a solid biomarker of non-responder. Besides, presence of PTGS1 + Caf, RGS5 + Caf, S100B + Caf, RND1 + Endo and S100 + Endo highly correlated with baseline LN metastasis and recurrence within 1 year, which was better than PET-CT. On this basis, we constructed predictive models for pathological response. With further validated by public scRNAseq and TCGA database, it showed excellent predictive ability and robustness. Conclusions: The final results of the SEEK-01 study demonstrate the differential baseline infiltration of myeloid cells that can predict pathological response of neoadjuvant immunotherapy. The characteristic changes in Caf and Endo in TME can predict occult LN metastasis and 1-year EFS. A predictive model for efficacy of neoadjuvant immunotherapy in locally advanced ESCC was developed to facilitate individualized and precise neoadjuvant therapy before treatment. Clinical trial information: NCT05807542 .