Abstract 4700: Quantification of intratumoral heterogeneity in individual luminal A breast cancers from whole transcriptome data through semi-supervised learning

Autor: Amit Sethi, Peter H. Gann, Neeraj Kumar, Yash Dharmamer
Rok vydání: 2019
Předmět:
Zdroj: Cancer Research. 79:4700-4700
ISSN: 1538-7445
0008-5472
DOI: 10.1158/1538-7445.am2019-4700
Popis: Introduction PAM50 gene profiling assigns individual breast cancers (BCas) to one of five intrinsic subtypes. However, individual tumors vary in their adherence to the assigned subtype, and a subset of these may show admixture with another subtype. We developed a novel metric to quantify this intratumoral heterogeneity using whole transcriptome expression data of individual BCas and studied specific pairwise admixtures for Luminal A (LumA) cases. Methods We obtained whole transcriptome expression data and PAM50 labels of 1081 and 1980 BCa cases from TCGA and METABRIC cohorts, respectively. We combined the two datasets, by identifying a common gene set. METABRIC data was pre-processed to normalize and log-transform expression levels and remove batch effects. For each case, the combined cohort had expression data of 11,379 genes, and probabilities of its adherence to each of the four major subtypes were computed using semi-supervised non-negative matrix factorization. The four probabilities obtained for each case were then combined using the proposed Aggregated Differential Sum (ADS) metric. For each subtype, the cases were subsequently stratified into pure, neither and admixed categories by computing ADS tertiles. For every case, we did comparative analysis of the ADS terms to identify the specific alternate molecular subtype for pairwise admixture analysis. Results Using the proposed ADS criteria, pure and admixed LumA cases spanned tertile 1 (T1) and 3 (T3), respectively. Compared to admixed cases, pure LumA cases were younger and had favorable clinical characteristics including a surrogate LumA profile (ER/PR+ and HER2-), smaller tumor size, lower stage, less prevalence of TP53 mutations, and high likelihood of PIK3CA and CBFB mutations. Kaplan-Meier curves and Cox proportional hazard analysis showed that the admixed cases had worse overall survival and higher mortality risk compared to their pure counterparts. Pairwise admixture analysis revealed that 59% (n=226), 27% (103) and 15% (55) LumA cases in T3 were admixed with Luminal B, Basal and HER2 subtype, respectively. LumA cases admixed with HER2 showed worst tumor characteristics, survival outcomes and highest mortality risk followed by, in order, the admixture with Luminal B and Basal subtypes. Conclusion Our results demonstrate that stratification of LumA BCas by subtype purity using ADS identifies the admixed cases which usually have worse tumor characteristics and survival outcomes. LumA cases were predominantly admixed with Luminal B subtype followed by Basal and HER2 subtypes. Eventually, we will use the pure cases to train computer vision classifiers for morphological identification of subtype admixture in routine H&E images. This would be a cost-effective complement to molecular subtyping that could take advantage of the spatial information available in whole slide images. Citation Format: Neeraj Kumar, Yash Dharmamer, Amit Sethi, Peter Gann. Quantification of intratumoral heterogeneity in individual luminal A breast cancers from whole transcriptome data through semi-supervised learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 4700.
Databáze: OpenAIRE
Externí odkaz: