The use of machine learning to discover regulatory networks controlling biological systems.

Biological systems are composed of a vast web of multiscale molecular interactors and interactions. High-throughput technologies, both bulk and single cell, now allow for investigation of the properties and quantities of these interactors. Computational algorithms and machine learning methods then provide the tools to derive meaningful insights from the resulting data sets. One such approach is graphical network modeling, which provides a computational framework to explicitly model the molecular interactions within and between the cells comprising biological systems. These graphical networks aim to describe a putative chain of cause and effect between interacting molecules. This feature allows for determination of key molecules in a biological process, accelerated generation of mechanistic hypotheses, and simulation of experimental outcomes. We review the computational concepts and applications of graphical network models across molecular scales for both intracellular and intercellular regulatory biology, examples of successful applications, and the future directions needed to overcome current limitations.

Matrix factorization and transfer learning uncover regulatory biology across multiple single-cell ATAC-seq data sets.

While the methods available for single-cell ATAC-seq analysis are well optimized for clustering cell types, the question of how to integrate multiple scATAC-seq data sets and/or sequencing modalities is still open. We present an analysis framework that enables such integration across scATAC-seq data sets by applying the CoGAPS Matrix Factorization algorithm and the projectR transfer learning program to identify common regulatory patterns across scATAC-seq data sets. We additionally integrate our analysis with scRNA-seq data to identify orthogonal evidence for transcriptional regulators predicted by scATAC-seq analysis. Using publicly available scATAC-seq data, we find patterns that accurately characterize cell types both within and across data sets. Furthermore, we demonstrate that these patterns are both consistent with current biological understanding and reflective of novel regulatory biology.

Transcriptomic forecasting with neural ordinary differential equations.

Single-cell transcriptomics technologies can uncover changes in the molecular states that underlie cellular phenotypes. However, understanding the dynamic cellular processes requires extending from inferring trajectories from snapshots of cellular states to estimating temporal changes in cellular gene expression. To address this challenge, we have developed a neural ordinary differential-equation-based method, RNAForecaster, for predicting gene expression states in single cells for multiple future time steps in an embedding-independent manner. We demonstrate that RNAForecaster can accurately predict future expression states in simulated single-cell transcriptomic data with cellular tracking over time. We then show that by using metabolic labeling single-cell RNA sequencing (scRNA-seq) data from constitutively dividing cells, RNAForecaster accurately recapitulates many of the expected changes in gene expression during progression through the cell cycle over a 3-day period. Thus, RNAForecaster enables short-term estimation of future expression states in biological systems from high-throughput datasets with temporal information.

Antibody dependent cell-mediated cytotoxicity selection pressure induces diverse mechanisms of resistance.

Targeted monoclonal antibody therapy has emerged as a powerful therapeutic strategy for cancer. However, only a minority of patients have durable responses and the development of resistance remains a major clinical obstacle. Antibody-dependent cell-mediated cytotoxicity (ADCC) represents a crucial therapeutic mechanism of action; however, few studies have explored ADCC resistance. Using multiple in vitro models of ADCC selection pressure, we have uncovered both shared and distinct resistance mechanisms. Persistent ADCC selection pressure yielded ADCC-resistant cells that are characterized by a loss of NK cell conjugation and this shared resistance phenotype is associated with cell-line dependent modulation of cell surface proteins that contribute to immune synapse formation and NK cell function. We employed single-cell RNA sequencing and proteomic screens to interrogate molecular mechanisms of resistance. We demonstrate that ADCC resistance involves upregulation of interferon/STAT1 and DNA damage response signaling as well as activation of the immunoproteasome. Here, we identify pathways that modulate ADCC sensitivity and report strategies to enhance ADCC-mediated elimination of cancer cells. ADCC resistance could not be reversed with combinatorial treatment approaches. Hence, our findings indicate that tumor cells utilize multiple strategies to inhibit NK cell mediated-ADCC. Future research and development of NK cell-based immunotherapies must incorporate plans to address or potentially prevent the induction of resistance.

Interrogation of T Cell-Enriched Tumors Reveals Prognostic and Immunotherapeutic Implications of Polyamine Metabolism.

Metabolic features of the tumor microenvironment (TME) antagonize anti-tumor immunity. We hypothesized that T cell infiltrated tumors with a known antigen should exhibit superior clinical outcomes, though some fare worse given unfavorable metabolic features leveraging T cell-infiltrated (Thi), human papillomavirus-related (HPV+) head and neck squamous cell carcinomas (HNSC) to test this hypothesis. Expression of 2,520 metabolic genes were analyzed among Thi HPV+ HNSCs stratified by high-risk molecular subtype. RNAseq data from The Cancer Genome Atlas (TCGA; 10 cancer types), single cell RNAseq data, and an immunotherapy-treated melanoma cohort were used to test the association between metabolic gene expression and clinical outcomes and contribution of tumor versus stromal cells to metabolic gene expression. Polyamine (PA) metabolism genes were overexpressed in high-risk, Thi HPV+ HNSCs. Genes involved in PA biosynthesis and transport were associated with T cell infiltration, recurrent or persistent cancer, overall survival status, primary site, molecular subtype, and MYC genomic alterations. PA biogenesis gene sets were associated with tumor intrinsic features while myeloid cells in HPV+ HNSCs were enriched in PA catabolism, regulatory, transport, putrescine, and spermidine gene set expression. PA gene set expression also correlated with IFNγ or cytotoxic T cell ssGSEA scores across TCGA tumor types. PA transport ssGSEA scores were associated with poor survival whereas putrescine ssGSEA scores portended better survival for several tumor types. Thi melanomas enriched in PA synthesis or combined gene set expression exhibited worse anti-PD-1 responses. These data address hurdles to anti-tumor immunity warranting further investigation of divergent polyamine metabolism in the TME.

Evaluating the impact of age on immune checkpoint therapy biomarkers.

Both tumors and aging alter the immune landscape of tissues. These interactions may play an important role in tumor progression among elderly patients and may suggest considerations for patient care. We leverage large-scale genomic and clinical databases to perform comprehensive comparative analysis of molecular and cellular markers of immune checkpoint blockade (ICB) response with patient age. These analyses demonstrate that aging is associated with increased tumor mutational burden, increased expression and decreased promoter methylation of immune checkpoint genes, and increased interferon gamma signaling in older patients in many cancer types studied, all of which are expected to promote ICB efficacy. Concurrently, we observe age-related alterations that might be expected to reduce ICB efficacy, such as decreases in T cell receptor diversity. Altogether, these changes suggest the capacity for robust ICB response in many older patients, which may warrant large-scale prospective study on ICB therapies among patients of advanced age.

Multi-omic profiling of lung and liver tumor microenvironments of metastatic pancreatic cancer reveals site-specific immune regulatory pathways.

BACKGROUND: The majority of pancreatic ductal adenocarcinomas (PDAC) are diagnosed at the metastatic stage, and standard therapies have limited activity with a dismal 5-year survival rate of only 8%. The liver and lung are the most common sites of PDAC metastasis, and each have been differentially associated with prognoses and responses to systemic therapies. A deeper understanding of the molecular and cellular landscape within the tumor microenvironment (TME) metastasis at these different sites is critical to informing future therapeutic strategies against metastatic PDAC. RESULTS: By leveraging combined mass cytometry, immunohistochemistry, and RNA sequencing, we identify key regulatory pathways that distinguish the liver and lung TMEs in a preclinical mouse model of metastatic PDAC. We demonstrate that the lung TME generally exhibits higher levels of immune infiltration, immune activation, and pro-immune signaling pathways, whereas multiple immune-suppressive pathways are emphasized in the liver TME. We then perform further validation of these preclinical findings in paired human lung and liver metastatic samples using immunohistochemistry from PDAC rapid autopsy specimens. Finally, in silico validation with transfer learning between our mouse model and TCGA datasets further demonstrates that many of the site-associated features are detectable even in the context of different primary tumors. CONCLUSIONS: Determining the distinctive immune-suppressive features in multiple liver and lung TME datasets provides further insight into the tissue specificity of molecular and cellular pathways, suggesting a potential mechanism underlying the discordant clinical responses that are often observed in metastatic diseases.

Chromatin structure regulates cancer-specific alternative splicing events in primary HPV-related oropharyngeal squamous cell carcinoma.

Human papillomavirus-related oropharyngeal squamous cell carcinoma (HPV+ OPSCC) represents a unique disease entity within head and neck cancer with rising incidence. Previous work has shown that alternative splicing events (ASEs) are prevalent in HPV+ OPSCC, but further validation is needed to understand the regulation of this process and its role in these tumours. In this study, eleven ASEs (GIT2, CTNNB1, MKNK2, MRPL33, SIPA1L3, SNHG6, SYCP2, TPRG1, ZHX2, ZNF331, and ELOVL1) were selected for validation from 109 previously published candidate ASEs to elucidate the post-transcriptional mechanisms of oncogenesis in HPV+ disease. In vitro qRT-PCR confirmed differential expression of 9 of 11 ASE candidates, and in silico analysis within the TCGA cohort confirmed 8 of 11 candidates. Six ASEs (MRPL33, SIPA1L3, SNHG6, TPRG1, ZHX2, and ELOVL1) showed significant differential expression across both methods. Further evaluation of chromatin modification revealed that ASEs strongly correlated with cancer-specific distribution of acetylated lysine 27 of histone 3 (H3K27ac). Subsequent epigenetic treatment of HPV+ HNSCC cell lines (UM-SCC-047 and UPCI-SCC-090) with JQ1 not only induced downregulation of cancer-specific ASE isoforms, but also growth inhibition in both cell lines. The UPCI-SCC-090 cell line, with greater ASE expression, also showed more significant growth inhibition after JQ1 treatment. This study confirms several novel cancer-specific ASEs in HPV+OPSCC and provides evidence for the role of chromatin modifications in regulation of alternative splicing in HPV+OPSCC. This highlights the role of epigenetic changes in the oncogenesis of HPV+OPSCC, which represents a unique, unexplored target for therapeutics that can alter the global post-transcriptional landscape.