Abstract C066: Molecular subtyping of NCI PDMR models using gene expression profiles

Abstract Background: Patient-derived xenografts (PDX) have proven invaluable research and pre-clinical tools for studying cancer. NCI has been building an extensive patient-derived model repository (PDMR) of diverse histologies, many of which have been utilized for pre-clinical drug studies. In this study, we have used gene expression profiles from RNA-Seq and mutation data from whole exome sequencing (WES) to identify the molecular subtypes of PDX models from 4 major cancer types: breast, ovarian, colon, and cholangiocarcinoma. We also report the key differentially expressed genes and pathways and highlight the key mutations associated with each subtype. Materials and methods: Gene expression data from originator patient and PDX specimens were obtained using R tximport and DESeq2 packages based on RNA-Seq data. ComBat-Seq was used to computationally alleviate batch effects between originators and PDX samples. The gene expression values were log2-transformed when being visualized via heatmap. The R genefu package was used for PAM50 classification of breast cancer models. For ovarian cancer and cholangiocarcinoma subtyping, the PAM algorithm was used on TCGA data to train the classification model. The R CMScaller package was used to predict colon cancer subtypes. Results: We characterized the molecular subtypes of 256 PDX models from breast (n=58), ovarian (n=11), colon (n=174), and cholangiocarcinoma (n=13) tumor histologies. First, we used Least Absolute Shrinkage and Selection Operator (LASSO) to build a histology classifier and identified differential expression from many histology-specific genes such as GATA3 for breast cancer and SOX17 for ovarian cancer. This confirmed that our PDX models resemble the original patient specimens. Second, by using either a pre-trained molecular subtype classifier or a newly trained model from TCGA data, we were able to assign many molecular subtypes of clinical relevance and rediscover the gene and pathway-level molecular signatures associated with prognosis and treatment response. Finally, we observed the expected mutation profiles for the identified subtypes in the WES data, such as a significant increase for the TP53 mutations and decrease of PIK3CA mutations in Basal subtypes compared to other breast cancer subtypes. Conclusions: The NCI PDMR provides an extensive resource for preclinical studies of molecular subtypes of cancer, as demonstrated by our ability to carry out subtype-specific analysis based on gene expression and mutational profiles. We were able to characterize, for example, the basal subtype of breast cancer and the mesenchymal subtype of colon cancer, both of which remain among the most challenging subtypes to treat. The presented molecular subtypes here could not only guide the future direction for translational projects, but also serve as a reference point for other high-impact research topics. Citation Format: I-Fan Wu, Li Chen, Shahanawaz Jiwani, Cindy R. Timme, Yvonne A. Evrard, Biswajit Das, Kelly Benauer, Gonzalo Salguero. Corado, Ruth Thornton, Irene Song, Poorva Juneja, Thomas Forbes, Dianne Newton, Melinda G. Hollingshead, Chris A. Karlovich, James H. Doroshow. Molecular subtyping of NCI PDMR models using gene expression profiles [abstract]. In: Proceedings of the AACR-NCI-EORTC International Conference on Molecular Targets and Cancer Therapeutics; 2025 Oct 22-26; Boston, MA. Philadelphia (PA): AACR; Mol Cancer Ther 2025;24(10 Suppl):Abstract nr C066.