OR32-05 Proteogenomics-based Development For Thyroid Cancer Molecular Classifier

Disclosure: Y. Kim: None. I. Wang: None. J. Jung: None. S. Cheon: None. S. Cho: None. D. Han: None. Y. Park: None. BackgroundCurrent leading-edgemolecular tests of thyroid nodule/cancer were based on DNA mutation/fusionprofile, and/or the expression profiles of mRNA and miRNA. They discriminated normalthyroid nodule from thyroid cancer with high performances. However, they could notsuggest any protein markers for the discrimination, and evaluate the risk ofcancer progression. Here, we deployed proteogenomics to discriminate normalthyroid from thyroid cancer, as well as capture the risk of cancer progression. MethodmRNAsequencing data of 86 normal thyroid (NT), 125 papillary thyroid cancer (PTC), 64follicular thyroid cancer (FTC) were obtained from our previous studies. Ofthese, 62 PTC samples harbored BRAF(V600E), and 24 FTC samplesharbored H/N/K RAS mutations, referred to as PTC-B and FTC-R, respectively. Freshfrozen tissues from 4 NT, 3 PTC-B, and 5 FTC-R were preparedfor tandem mass tag labeling, and followed by LC-MS/MS analysis. Differentiallyexpressed genes and differentially expressed proteins were determined betweenNT and PTC-B, as well as between NT and FTC-R, which were 86 genes and referredto as NPF (NT-PTC-FTC) genes. NPF genes were tested for their discriminativepower in our development cohort, and validated in TCGA-THPA, and new globalproteomics data including 72 FTC, and 76 PTC. Protein markerswith high discriminative performances validated with tissuemicroarray, which included formalin-fixed paraffin-embedded samples of 22 NT, 196PTC, and 76 FTC. NPF scores were calculated, and analyzed for the progressionof thyroid cancer in our development cohort and TCGA-THPA. ResultK-meansclustering using NPF genes identified 4 clusters; BRAF-like, RAS-like, Immune-rich,and NT-like subtypes. Samples with lymphoid thyroiditis were mainly involved inimmune-rich subtypes. We developed thyroid molecular classifier with NPF usingdecision tree model. The accuracy of this model to predict thyroid cancersubtypes was 0.92 in our development cohort. We validated our model with TCGA-THPAwhich predicted 97% of tumors with BRAFV600E and 84% of tumors with H/K/N RASmutation, and new global proteomics data which predicted 77.8% of FTC and 64.4%of PTC. MarkerA, MarkerB, and MarkerCshowed best discriminative performances in our classifier. In thevalidation by immunohistochemistry, MarkerA(high)- MarkerB(low)-MarkerC(high) indicated 70.6% of NT, MarkerA(low)- MarkerB(high)-MarkerC(low) indicated 97.2% of PTC, and MarkerB(low)-MarkerC(low) indicated 58.2% of FTC. Both in PTCand FTC, high NPF score was associated with intermediate to high risk based on ATArisk stratification in our development cohort and TCGA-THPA, and poorprogression-free survival in TCGA-THPA. ConclusionWedeveloped thyroid cancer molecular classifier which reflected molecularsubtypes of thyroid cancer, and their risk of progression using proteogenomics. Presentation: Sunday, June 18, 2023