Focal amplifications (FAs) resulting in copy number gain of short genomic regions, can mediate targeted therapy resistance. Understanding the structure, plasticity and vulnerability of FAs is critical for designing treatments that overcome such resistance. Here we developed a combined BRAF plus MEK inhibitor resistance melanoma model that bears high mutant BRAF amplifications through two modes of FAs: extrachromosomal double minutes (ecDNA/DMs) and intrachromosomal homogenously staining regions (HSRs), and investigated FA structure and dynamics in the context of drug resistance plasticity. We found that cells harboring BRAF FAs displayed mode switching between DMs and HSRs, from both de novo genetic changes and selection of pre-existing subpopulations. Plasticity is not exclusive to ecDNAs, as cells harboring HSRs also exhibit copy number and length changes through structural loss of amplicon repeats that allow them to respond to dose reduction and recover from drug addiction. DM and HSR mechanisms can couple with other BRAF genomic changes, such as kinase domain duplications and alternative splicing, to enhance therapy resistance. Amplicon plasticity is observed in other MAPK pathway genes, such as RAF1 and NRAS, and occurs in clinical cases of therapy resistance. We found that BRAF FA-induced dual MAPKi-resistant cells are more sensitive to pro-ferroptotic drugs, which extends the spectrum of melanoma resistance-derived ferroptosis sensitivity beyond cases of dedifferentiation.

Malignant transformation can result in melanoma cells that resemble different stages of their embryonic development. Our analysis of gene expression profiles from a large panel of human melanoma cell lines and patient tumors revealed that melanoma follows a two-dimensional differentiation trajectory that can be sub-classified into four progressive subtypes. This differentiation model is associated with subtype-specific sensitivity to iron dependent oxidative stress and cell death known as ferroptosis. Receptor tyrosine kinase mediated resistance to MAPK targeted therapies and activation of the inflammatory signaling associated with immune therapy involves transitions along this differentiation trajectory, which lead to increased sensitivity to ferroptosis. Therefore, ferroptosis-inducing drugs presents an orthogonal therapeutic approach to target the differentiation plasticity of melanoma cells to increase the efficacy of targeted and immune therapies. Melanoma cells are present at different differentiation states and have the ability to dedifferentiate under cellular stress. This has important therapeutic implications as dedifferentiation contributes to intrinsic and acquired resistance to MAPK pathway inhibitors, and occurs as a response to inflammatory signaling during immunotherapy. Therefore, targeting dedifferentiation is a logical approach to strengthen these current therapeutic strategies. Here we categorize melanoma differentiation as four distinct stepwise stages and identify a heightened sensitivity to ferroptosis induction with the degree of differentiation. Our results further define tumor differentiation as an important parameter for patient stratification, and propose a new and highly orthogonal component to add to existing therapeutics, namely enhancing targeted signaling inhibition and immune therapies by synthetic lethal induction of ferroptosis.

For precision oncology to become a reality, clinicians need access to tumor tissue in a real-time, repeatable, and cost-effective manner. Circulating tumor cells (CTCs), cells of tumor origin that circulate in the blood, are one possible solution, with the potential to serve as a “liquid biopsy” for gastrointestinal (GI) tumors.

In this thesis, we demonstrate the development and validation of an assay for performing molecular analysis of CTCs as a biomarker for GI cancers. We used a microfluidic platform to develop GI tumor-specific CTC assays for pancreatic and liver cancer through incorporation of tissue-specific markers for CTC capture and identification. We then conducted prospective studies testing the utility of CTC enumeration as a biomarker in pancreatic (n=100) and liver (n=61) cancer. We found that the use of a tissue-specific assay resulted in higher CTC counts than in previous studies, that CTC counts correlated with increasing stage and that CTC enumeration may have diagnostic and prognostic utility as a biomarker in GI cancers.

Using a modified assay that allows for the complete isolation of the CTCs identified, we confirmed the tumor origin of the CTCs we identified by comparison of mutational status with the primary tumor using both Sanger and next generation sequencing methods.

After confirming their tumor origin, we developed a quality control assay that allowed us to determine how many CTCs we would need to isolate to reliably perform CTC molecular analysis. Using this metric, we modified our CTC sequencing protocol that resulted in reliable sequencing from as few as 3-5 CTCs as opposed to the 10-20 needed using the traditional CTC sequencing protocol. We then used our optimized CTC sequencing protocol in a prospective study to test the feasibility and potential utility of CTC molecular analysis as a biomarker in liver cancer.

The research described in this thesis serves two primary goals. The first is as a validated method that allows for the use of CTCs in liquid biopsy-type applications for both HCC and PDAC patients. The second is as a framework for validating CTC assays and definitions to ensure the reliability and reproducibility of CTC sequencing results.

Copy number alteration (CNA) profiling of human tumors has revealed recurrent patterns of DNA amplifications and deletions. These patterns are indicative of conserved selection pressures, but cannot be fully explained by known oncogenes and tumor suppressor genes. Using integrative analysis of CNA data from patient tumors and experimental systems, we report that principal component analysis-defined CNA signatures are predictive of glycolytic phenotypes, including FDG-avidity of patient tumors, and increased proliferation. The primary glycolysis-linked CNA signature is associated with p53 mutation and shows coordinate amplification of glycolytic genes and other cancer-linked metabolic enzymes including TIGAR and RPIA. In contrast, alternative signatures involve both different mechanisms of tumor suppression loss (eg, MDM2 amplification) and different glycolysis enzyme isoforms. Furthermore, a cross-species CNA comparison identified 21 conserved CNA regions, containing 13 enzymes in the glycolysis and pentose phosphate pathways in addition to known cancer driving genes. In validation experiments, exogenous expression of hexokinase and enolase enzymes resulted in reduced propensities for amplifications at the corresponding endogenous loci. Our findings support metabolic stress as a selective pressure underlying the recurrent CNAs observed in human tumors, and further cast genomic instability as an enabling event in tumorigenesis and metabolic evolution.

There is a critical need for more effective systemic therapies for the treatment of soft tissue sarcoma and, specifically, undifferentiated sarcoma. Immunotherapy has shown signs of efficacy for the treatment of soft tissue sarcoma, particularly in undifferentiated sarcoma, though selecting the patients who will benefit from immunotherapy remains difficult and unclear. Further studies are needed to characterize the immune landscape of soft tissue sarcoma and to develop strategies to identify patients that are likely to benefit from immunotherapy.

In this study, I investigated the immunologic heterogeneity and identify transcriptomic and genomic correlates of immune cell infiltration in undifferentiated sarcoma. In doing so, I determined the immune cell landscape, the optimal high-throughput tools, and the transcriptomic and genomic changes associated with high and low immune cell infiltration in soft tissue sarcoma.

This study synthesized many datasets and data types for a comprehensive analysis of the immune landscape in soft tissue sarcoma. I first characterized the immune cell landscape in soft tissue sarcoma using flow cytometry data from fresh operative samples (n=105) of multiple soft tissue sarcoma subtypes. I then generated a tissue microarray with matched RNA sequencing data from 60 samples of untreated undifferentiated sarcoma to determine the optimal method of in-silico immune deconvolution, which allows for the expansion of this analysis to other next generation sequencing data. Finally, I synthesized multiple publicly available datasets containing next generation sequencing data (RNA sequencing and whole exome sequencing) from undifferentiated sarcoma samples. This data is combined with data from the aforementioned samples for a total of 193 samples and is used to determine the transcriptomic and genomic correlates of immune cell infiltration in undifferentiated sarcoma.

In the analysis of the flow cytometry data, I found that undifferentiated sarcoma tumors are characterized by a myeloid predominance and a relative abundance of suppressor cells, such as Treg cells and CD11b cells. I additionally found that the immune composition of peripheral blood was associated with intratumoral leukocyte infiltration, and specifically that myeloid-predominant tumor and lymphocyte-predominant blood are mutually exclusive. I then determined the optimal in-silico immune deconvolution tool in undifferentiated sarcoma by determining the correlation between mIF and in-silico immune deconvolution scores. Based on these findings, I suggest the following practices when applying in-silico immune deconvolution tools to undifferentiated sarcoma: (1) Use TIMER to define overall immune cell infiltration. (2) Use MCP counter to define monocyte infiltration or use CIBERSORTx, EPIC, quanTIseq, TIMER, or xCell to define macrophage infiltration. (3) Use caution when using in-silico immune deconvolution tools to define CD8+ T cell infiltration. CIBERSORTx most accurately defines CD8+ T cell immune infiltration, however, there are still many instances when tumors with high CD8+ T cell infiltration will be missed using this technique. (4) Avoid applying in-silico immune deconvolution results to define B cell or CD4+ T cell immune infiltration. Finally, I found that increased copy number changes were associated with low immune cell infiltration in undifferentiated sarcoma. These findings were suggested in both transcriptomic and genomic analyses. Interestingly, this association between CNA and immune invasion were unique to the UPS and DDLPS subtypes of STS, but it was not seen in other subtypes of STS. The mechanisms underlying this association are not clear and warrant further study.

These insights provide necessary information to understand which patients may benefit from immunotherapy and guide future studies to further the treatment of soft tissue sarcoma. These studies provide the groundwork for further investigation in this study of immune cell infiltration in soft tissue sarcoma and provide insights into how we may be able to improve outcomes in this rare and devastating disease. The mechanisms underlying these findings remain unclear and warrant further investigation. A deeper understanding of the drivers of immune cell infiltration, the unique tumor microenvironment in soft tissue sarcoma, and role that chromosomal instability plays in soft tissue sarcoma will hopefully ultimately lead to insights to new, and much-needed, treatments for this disease.

Focal amplifications (FAs) are important contributors to genomic instability within cancer cells and can influence their response to targeted therapies. FAs can take on two different forms: double minutes (DMs), which are circular segments of extrachromosomal DNA (ecDNA), or homogeneously staining regions (HSRs), which are intrachromosomal regions of amplification. DMs, or ecDNAs, are particularly known for enrichment of oncogenes and their contribution to cell plasticity, especially in conjunction with BRD4 which promotes overexpression of the oncogenes. However, while FAs confer advantages upon the cells, there are also weaknesses that are introduced due to the increased instability of the genome. This project aims to identify weak points in cells with FAs through large-scale computational analysis of drug screens. The area under the dose response curve (AUC) was chosen as a measure of sensitivity for this analysis since a smaller AUC indicates that the sample responded more quickly to the drug. We performed linear regressions for hundreds of drugs to identify those with a smaller AUC value when treating samples with FA compared to samples without FA. The targets associated with these drugs were then compared to differential expression analysis and network correlation analysis modules to identify if they were differentially expressed or part of a larger network specific to cells with DMs and HSRs. We found several targets that appeared to be consistently associated with our criteria for sensitivity including AKT1, BRD2, BRD3, BRD4, DNMT3A, FLT3, KIT, and PDGFRB. These targets appeared even when removing samples with known driver genes and reducing tissue-specific effects on the linear regressions. A subset of these targets has been demonstrated to be associated with DNA repair and genomic instability in cancer cells from previous studies. Most importantly, AKT1 and FLT3 were found to be effective targets for joint inhibition with BRD4 in previous research, which introduces potential for targeting ecDNA and DM hubs. Transcriptome (RNA-seq) data found that numerous genes scoring in the drug sensitivity analysis were differentially regulated. Network analysis of co-expressed gene modules identified by network correlation analysis showed interactions between identified drug targets and differentially expressed genes and support the existence of gene interactions that were not apparent from the sensitivity analysis alone. These results provide more insight into potential targeted therapies for cells with FA and the potential weak points within the cells.

There is a growing interest in metabolomic profiling of immune cells, as there is accumulating evidence that metabolism influences immune cell function and fate decisions. The non-adherent nature of these cells presents certain challenges not factored into extraction protocols developed for adherent cells. Here, we present a method developed for metabolite extraction from human T-cells that emphasizes simplicity and economy. We apply this method to profile the metabolic changes effected by in-vitro programmed cell death 1 receptor (PDCD1/PD-1) immune checkpoint engagement by liquid chromatography and mass spectrometry. We demonstrate that we can efficiently confirm the known reduction in aerobic glycolysis and glutaminolysis that accompanies PD-1 signaling and extend this by showing a block in de novo nucleoside phosphate biosynthesis.

Alternative splicing (AS) is an elaborately regulated molecular process incorporating different exons, expanding the proteome diversity from a single gene. AS is involved in virtually all biological processes and diseases, in which the AS process is often dysregulated. Dysregulated AS in cancer is so prevalent that it is now considered a hallmark of cancer. This dysregulation introduces a novel modality to target cancer through splicing modulation. Recent developments in RNA-Sequencing technologies have enabled transcriptome-wide quantification of AS changes in biological processes and diseases, requiring the development of computational methods for deciphering the molecular regulatory mechanisms for the observed AS changes in biological processes, diseases, and treatments.AS process is regulated by multitudes of RNA binding proteins (RBPs), each playing unique roles in the splicing processes, where molecular changes in RBPs causes AS changes in biological processes. Thus, the mechanism of AS changes can be attributed to molecular changes in RBPs, but determining the causal RBPs, based on the observed AS changes, has been challenging due to the complexity of AS regulation. We present Splicing Profile Analysis using RBP KD/KO Signatures (SPARKS), a computational framework to infer causal RBPs by quantifying the similarity in AS between the study of interest and the perturbation of RBPs from a large-scale RBP perturbation database. We verified that splicing factor perturbation leads to unique AS changes, which SPARKS leverages to identify causal RBPs. We demonstrated the efficacy of SPARKS in identifying the causal RBPs in RBP perturbation experiments and biological processes with known regulatory mechanisms in AS. We applied SPARKS to the AS changes in AML patients with the recurrent somatic mutation in U2AF1 and uncovered that the mutant U2AF1 and PTBP1 are involved in the AS changes. We also discovered the candidate RBPs responsible for pan-cancer AS changes upon PRMT5 inhibition therapy, EFTUD2, PCBP2, and PRPF8. Integrating the molecular changes on the candidate RBPs, we generated a potential model for the mechanism of action for PRMT5 inhibition in splicing. In summary, this work illustrates a computational analysis framework to decipher the molecular mechanisms of alternative splicing regulation in biology, disease, and therapy.

Macrophages exposed to immune stimuli reprogram their epigenomes to alter subsequent functions. Here, we have dissected the roles of interferon regulatory factors (IRFs) in innate immune de novo enhancer formation. We found that upon transient endotoxin exposure, such enhancers may remain poised for days. Endotoxin-activated IRF3 induces only a small number of de novo enhancers directly, but it is indirectly required for the formation of a large number of enhancers via type I interferon-induced ISGF3. However, ISGF3 is unable to trigger enhancer formation by itself – it must cooperate with NFκB-induced IRF1. In type II IFN, IRF1 induces enhancer formation by cooperating with GAF. We found that IRF1 is particularly required in locations that show less chromatin accessibility in naïve macrophages, and a fine timecourse revealed that IRF1 is required for the initial opening of chromatin before ISGF3 extends it and recruits enzymes to deposit H3K4me1 marks. Our results reveal an IRF1-mediated combinatorial logic gate to provide innate immune memory formation in cells exposed to pathogen or activated IFNγ-secreting T cells, but not bystander macrophages that benefit from the transient anti-viral message of type I interferon.