Study On Molecular Subtypes And Prognosis Of Esophageal Squamous Cell Carcinoma Based On Radiomics And Transcriptomics

BackgroundEsophageal squamous cell carcinoma(ESCC),the most common pathological type of esophageal cancer,is widely prevalent in East Asia.It is well known that accurate clinical staging of ESCC,especially lymph node staging,is of great significance for treatment decision-making and survival.In recent yeas,multiple studies have shown that for locally advanced ESCC patients with lymph node metastasis(LNM),neoadjuvant chemotherapy combined with immunotherapy has obtained great achievements,while a proportion of patients still fail to benefit,and the specific mechanism remains unclear.Therefore,accurately staging/typing of ESCC,and developing personalized therapeutic strategies based on the heterogeneity of tumor microenvironment(TME)have become the current research hotspot.Objective1.Based on PET/CT images of ESCC patients,multi-task deep learning algorithm was used to analyze the features containing tumor and highly correlated lymph node,and an automatic LNM prediction model was constructed.2.Molecular subtypes of ESCC were determined by RNA sequencing,and the TME of each subtype was investigated.Candidate biomarkers which may potentially predict the efficacy to neoadjuvant immunotherapy were identified to accurately make treatment regimens and predict the prognosis of patients.Methods1.Construction of LNM prediction model:A total of 689 ESCC cases with PET/CT were enrolled from three hospitals and divided into a training cohort and two external validation cohorts.Anatomic information from CT images was first obtained automatically via multi-organ segmentation using U-Net plus,and metabolic information from PET images was subsequently gained for predicting LNM using a gradient-based approach.an artificial intelligence-based computer-aided diagnosis system(AI-CAD)was developed and validated.2.Transcriptome sequencing:Transcriptomic profiling was applied in formalin-fixed and parrffin-embedded(FFPE)tissues from 103 ESCC patients,which were divided into retrospective cohort(66 treatment-naive patients)and prospective cohort(37 cases receiving neoadjuvant chemotherapy plus immunotherapy),and molecular subtypes of ESCC were determined.A prediction model was established following differentially expressed genes of each subtype.The subtype-specific TME characteristics were verified by a multiplex immunofluorescence assay.3.Screening of biomarkers:Univariant logistic regression was applied to identify candidate biomarkers which may potentially predict the efficacy to neoadjuvant immunotherapy and patients’ survival.Putative mechanisms mediating response to immunotherapy were evaluated by transcriptomic data.ResultsPart 1.18F-FDG PET/CT-based deep learning model predicted LNM in ESCC patientsWith the aid of AI-CAD,the human expert’s diagnostic performance was significantly improved[accuracy(95%CI):from 0.712(0.669-0.758)to 0.833(0.797-0.865),specificity(95%CI):from 0.697(0.636-0.753)to 0.891(0.851-0.928);P<0.001].Furthermore,AI-CAD prediction could also serve as an independent prognostic factor for progression-free survival and overall survival in-multiple clinical scenarios(multivariable-adjusted hazard ratios:1.572-2.597).Part 2.Transcriptional sequencing identified ESCC molecular subtypes and TME heterogeneityFour stratified molecular subtypes were identified by unsupervised clustering,which exhibited distinct biological features,including metabolism(C1),mesenchymal(C2),immune(C3)and G protein coupled receptor(GPCR)(C4).The C2 subtype,the most aggressive one,displayed an immune-unfavored microenvironment,represented by positive correlating with regulatory T cells,Helper 2 T cell as well as less infiltration of B cells,effector T cells and mast cells.By contrast,C3 subtype had a favorable survival due to the enrichment of IFN-y and B cell signatures.Part 3.Exploration of biomarkers for predicting the efficacy of immunotherapy in ESCCPLEK2 and IFI6 were highly expressed in C2 and correlated with immune resistance[PLEK2high,OR(95%CI):2.15(1.07-4.33),P=0.032;IFI6high,OR(95%CI):2.21(1.16-4.23),P=0.016]and poor prognosis,while overexpression of COL19A1 in C3 indicated good immune response and prognosis.ConclusionsThe AI-CAD successfully maximizes the preclinical justification for ’LNM in patients with ESCC.Locally advanced cases with high COL19A1 expression would be benefited from neoadjuvant immunotherapy,while PLEK2high or IFI6high ESCC displayed immune resistance.