Author ORCID Identifier

https://orcid.org/0000-0003-4579-3959

Date of Graduation

5-2022

Document Type

Dissertation (PhD)

Program Affiliation

Biostatistics, Bioinformatics and Systems Biology

Degree Name

Doctor of Philosophy (PhD)

Advisor/Committee Chair

Giulio F. Draetta, M.D. Ph.D.

Committee Member

P. Andrew Futreal, Ph.D.

Committee Member

William F. Symmans, M.D.

Committee Member

Paul Scheet, Ph.D.

Committee Member

Helen Piwnica-Worms, Ph.D.

Committee Member

Traver Hart, Ph.D.

Committee Member

Stacy Moulder, M.D.

Abstract

Tumors are highly heterogeneous and dynamic, continually adapting and evolving in response to their microenvironment as well as external perturbations. Multi-region (spatial) and single cell sequencing has enabled us to anatomize the heterogeneity further and provide evidence of its association with chemo and drug resistance. To investigate this further we took two different approaches to understand the chemo-resistance, and functional heterogeneity in Triple negative breast cancer (TNBC) and Pancreatic ductal carcinoma in situ (PDAC) from an evolutionary perspective.

The first approach was to leverage tumor profiling from an ongoing randomized clinical trial in triple-negative breast cancer (ARTEMIS) to assess mechanisms of chemo-resistance, and use longitudinally tracked patient biopsies to interrogate patterns of response and evolution under chemotherapy. We performed comprehensive molecular profiling using pre-treatment core needle biopsies from 300 TNBC patients treated neoadjuvantly to understand features associated with disease response. We identified six transcriptomic subtypes (ART-Types), each characterized by unique molecular and clinical features. Immune enriched and high proliferation subtypes had higher pCR rates while the low-immune subtype had the lowest pCR rate. Low pCR rates were associated with low T cells, B cells, myeloid cells, high neutrophils, EGFR amplification, and CDKN2A deletions. A machine-learning model integrating immune, transcriptomic and genomic features accurately predict pCR, with validation in a blinded cohort.

By tracking these tumors pre and post-chemotherapy (N=131), using RNASeq and whole-exome sequencing, we interrogated switching of ART-types, changes in transcriptomic phenotypes, and genomic evolution. Androgen response subtype was the most stable, not impacted by chemotherapy, while high proliferating subtypes switched the most. Switch to a low-IM subtype was associated with a poor prognosis. Orthogonally, an increase in immune infiltration and a decrease in tumor purity were associated with higher pCR rates. Similarly, low baseline purity and high tumor ploidy were also associated with lower pCR rates.

As part of a second strategy, we have developed a novel platform using single-cell lineage tracking (clonal replica tumors - CRTs), which enabled us to model and dissect functional heterogeneity in early passage PDAC patient-derived xenografts (PDX) models. These models were used to identify the fitness landscape of individual tumor lineages under three different drugs. High-throughput isolation and profiling of resistant and sensitive lineages were used to establish, and profile them using RNASeq and whole-exome sequencing. The transcriptomic signature developed by probing functional heterogeneity to inform on response to chemotherapy, within the PDX model was able to dissect TCGA pancreatic cancer cohort into groups with significantly different survival.

Keywords

Computational Biology, Machine Learning, Breast Cancer, TNBC, Cancer Genomics

Download

Share

COinS