Genomic sequence context differs between germline and somatic structural variants allowing for their differentiation in tumor samples without paired normals.

There is currently no method to distinguish between germline and somatic structural variants (SVs) in tumor samples that lack a matched normal sample. In this study, we analyzed several features of germline and somatic SVs from a cohort of 974 patients from The Cancer Genome Atlas (TCGA). We identified a total of 21 features that differed significantly between germline and somatic SVs. Several of the germline SV features were associated with each other, as were several of the somatic SV features. We also found that these associations differed between the germline and somatic classes, for example, we found that somatic inversions were more likely to be longer events than their germline counterparts. Using these features we trained a support vector machine (SVM) classifier on 555,849 TCGA SVs to computationally distinguish germline from somatic SVs in the absence of a matched normal. This classifier had an ROC curve AUC of 0.984 when tested on an independent test set of 277,925 TCGA SVs. In this dataset, we achieved a positive predictive value (PPV) of 0.81 for an SV called somatic by the classifier being truly somatic. We further tested the classifier on a separate set of 7,623 SVs from pediatric high-grade gliomas (pHGG). In this non-TCGA cohort, our classifier achieved a PPV of 0.828, showing robust performance across datasets.

Double-strand break repair-associated intragenic deletions and tandem duplications suggest the architecture of the repair replication fork.

Double-strand break (DSB) repair is associated with a 1000-fold increase in mutations compared to normal replication of the same sequences. In budding yeast, repair of an HO endonuclease-induced DSB at the MATα locus can be repaired by using a homologous, heterochromatic HMR::Kl-URA3 donor harboring a transcriptionally silenced URA3 gene, resulting in a MAT::URA3 (Ura+) repair product where URA3 is expressed. Repair-associated ura3- mutations can be selected by resistance to 5-fluoroorotic acid (FOA). Using this system, we find that a major class of mutations are -1 deletions, almost always in homonucleotide runs, but there are few +1 insertions. In contrast, +1 and -1 insertions in homonucleotide runs are nearly equal among spontaneous mutations. Approximately 10% of repair-associated mutations are interchromosomal template switches (ICTS), even though the K. lactis URA3 sequence embedded in HMR is only 72% identical with S. cerevisiae ura3-52 sequences on a different chromosome. ICTS events begin and end in regions of short microhomology, averaging 7 bp. Long microhomologies are favored, but some ICTS junctions are as short as 2 bp. Both repair-associated intragenic deletions (IDs) and tandem duplications (TDs) are recovered, with junctions sharing short stretches of, on average, 6 bp of microhomology. Intragenic deletions are more than 5 times more frequent than TDs. IDs have a mean length of 60 bp, but, surprisingly there are almost no deletions shorter than 25 bp. In contrast, TDs average only 12 bp. The usage of microhomologies among intragenic deletions is not strongly influenced by the degree of adjacent homeology. Together, these data provide a picture of the structure of the repair replication fork. We suggest that IDs and TDs occur within the migrating D-loop in which DNA polymerase δ copies the template, where the 3' end of a partly copied new DNA strand can dissociate and anneal with a single-stranded region of microhomology that lies either in front or behind the 3' end, within the open structure of a migrating D-loop. Our data suggest that ~100 bp ahead of the polymerase is "open," but that part of the repair replication apparatus remains bound in the 25 bp ahead of the newly copied DNA, preventing annealing. In contrast, the template region behind the polymerase appears to be rapidly reannealed, limiting template switching to a very short region.

Collateral responses to classical cytotoxic chemotherapies are heterogeneous and sensitivities are sparse.

Chemotherapy resistance is a major obstacle to curing cancer patients. Combination drug regimens have shown promise as a method to overcome resistance; however, to date only some cancers have been cured with this method. Collateral sensitivity-the phenomenon whereby resistance to one drug is co-occurrent with sensitivity to a second drug-has been gaining traction as a promising new concept to guide rational design of combination regimens. Here we evolved over 100 subclones of the Eµ-Myc; p19ARF-/- cell line to be resistant to one of four classical chemotherapy agents: doxorubicin, vincristine, paclitaxel, and cisplatin. We then surveyed collateral responses to acquisition of resistance to these agents. Although numerous collateral sensitivities have been documented for antibiotics and targeted cancer therapies, we observed only one collateral sensitivity: half of cell lines that acquired resistance to paclitaxel also acquired a collateral sensitivity to verapamil. However, we found that the mechanism of this collateral sensitivity was unrelated to the mechanism of paclitaxel resistance. Interestingly, we observed heterogeneity in the phenotypic response to acquisition of resistance to most of the drugs we tested, most notably for paclitaxel, suggesting the existence of multiple different states of resistance. Surprisingly, this phenotypic heterogeneity in paclitaxel resistant cell lines was unrelated to transcriptomic heterogeneity among those cell lines. These features of phenotypic and transcriptomic heterogeneity must be taken into account in future studies of treated tumor subclones and in design of chemotherapy combinations.

Acquired resistance to PRMT5 inhibition induces concomitant collateral sensitivity to paclitaxel.

Epigenetic regulators play key roles in cancer and are increasingly being targeted for treatment. However, for many, little is known about mechanisms of resistance to the inhibition of these regulators. We have generated a model of resistance to inhibitors of protein arginine methyltransferase 5 (PRMT5). This study was conducted in KrasG12D;Tp53-null lung adenocarcinoma (LUAD) cell lines. Resistance to PRMT5 inhibitors (PRMT5i) arose rapidly, and barcoding experiments showed that this resulted from a drug-induced transcriptional state switch, not selection of a preexisting population. This resistant state is both stable and conserved across variants arising from distinct LUAD lines. Moreover, it brought with it vulnerabilities to other chemotherapeutics, especially the taxane paclitaxel. This paclitaxel sensitivity depended on the presence of stathmin 2 (STMN2), a microtubule regulator that is specifically expressed in the resistant state. Remarkably, STMN2 was also essential for resistance to PRMT5 inhibition. Thus, a single gene is required for both acquisition of resistance to PRMT5i and collateral sensitivity to paclitaxel in our LUAD cells. Accordingly, the combination of PRMT5i and paclitaxel yielded potent and synergistic killing of the murine LUAD cells. Importantly, the synergy between PRMT5i and paclitaxel also extended to human cancer cell lines. Finally, analysis of The Cancer Genome Atlas patient data showed that high STMN2 levels correlate with complete regression of tumors in response to taxane treatment. Collectively, this study reveals a recurring mechanism of PRMT5i resistance in LUAD and identifies collateral sensitivities that have potential clinical relevance.

Deoxycytidine Release from Pancreatic Stellate Cells Promotes Gemcitabine Resistance.

Pancreatic ductal adenocarcinoma (PDAC) is a leading cause of cancer deaths in the United States. The deoxynucleoside analogue gemcitabine is among the most effective therapies to treat PDAC, however, nearly all patients treated with gemcitabine either fail to respond or rapidly develop resistance. One hallmark of PDAC is a striking accumulation of stromal tissue surrounding the tumor, and this accumulation of stroma can contribute to therapy resistance. To better understand how stroma limits response to therapy, we investigated cell-extrinsic mechanisms of resistance to gemcitabine. Conditioned media from pancreatic stellate cells (PSC), as well as from other fibroblasts, protected PDAC cells from gemcitabine toxicity. The protective effect of PSC-conditioned media was mediated by secretion of deoxycytidine, but not other deoxynucleosides, through equilibrative nucleoside transporters. Deoxycytidine inhibited the processing of gemcitabine in PDAC cells, thus reducing the effect of gemcitabine and other nucleoside analogues on cancer cells. These results suggest that reducing deoxycytidine production in PSCs may increase the efficacy of nucleoside analog therapies. SIGNIFICANCE: This study provides important new insight into mechanisms that contribute to gemcitabine resistance in PDAC and suggests new avenues for improving gemcitabine efficacy.

Regulation of Cellular Heterogeneity and Rates of Symmetric and Asymmetric Divisions in Triple-Negative Breast Cancer.

Differentiation events contribute to phenotypic cellular heterogeneity within tumors and influence disease progression and response to therapy. Here, we dissect mechanisms controlling intratumoral heterogeneity within triple-negative basal-like breast cancers. Tumor cells expressing the cytokeratin K14 possess a differentiation state that is associated with that of normal luminal progenitors, and K14-negative cells are in a state closer to that of mature luminal cells. We show that cells can transition between these states through asymmetric divisions, which produce one K14+ and one K14- daughter cell, and that these asymmetric divisions contribute to the generation of cellular heterogeneity. We identified several regulators that control the proportion of K14+ cells in the population. EZH2 and Notch increase the numbers of K14+ cells and their rates of symmetric divisions, and FOXA1 has an opposing effect. Our findings demonstrate that asymmetric divisions generate differentiation transitions and heterogeneity, and identify pathways that control breast cancer cellular composition.

Identifying therapeutic targets by combining transcriptional data with ordinal clinical measurements.

The immense and growing repositories of transcriptional data may contain critical insights for developing new therapies. Current approaches to mining these data largely rely on binary classifications of disease vs. control, and are not able to incorporate measures of disease severity. We report an analytical approach to integrate ordinal clinical information with transcriptomics. We apply this method to public data for a large cohort of Huntington's disease patients and controls, identifying and prioritizing phenotype-associated genes. We verify the role of a high-ranked gene in dysregulation of sphingolipid metabolism in the disease and demonstrate that inhibiting the enzyme, sphingosine-1-phosphate lyase 1 (SPL), has neuroprotective effects in Huntington's disease models. Finally, we show that one consequence of inhibiting SPL is intracellular inhibition of histone deacetylases, thus linking our observations in sphingolipid metabolism to a well-characterized Huntington's disease pathway. Our approach is easily applied to any data with ordinal clinical measurements, and may deepen our understanding of disease processes.Identifying gene subsets affecting disease phenotypes from transcriptome data is challenge. Here, the authors develop a method that combines transcriptional data with disease ordinal clinical measurements to discover a sphingolipid metabolism regulator involving in Huntington's disease progression.

Hyper- and hypo- nutrition studies of the hepatic transcriptome and epigenome suggest that PPARα regulates anaerobic glycolysis.

Diet plays a crucial role in shaping human health and disease. Diets promoting obesity and insulin resistance can lead to severe metabolic diseases, while calorie-restricted (CR) diets can improve health and extend lifespan. In this work, we fed mice either a chow diet (CD), a 16 week high-fat diet (HFD), or a CR diet to compare and contrast the effects of these diets on mouse liver biology. We collected transcriptomic and epigenomic datasets from these mice using RNA-Seq and DNase-Seq. We found that both CR and HFD induce extensive transcriptional changes, in some cases altering the same genes in the same direction. We used our epigenomic data to infer transcriptional regulatory proteins bound near these genes that likely influence their expression levels. In particular, we found evidence for critical roles played by PPARα and RXRα. We used ChIP-Seq to profile the binding locations for these factors in HFD and CR livers. We found extensive binding of PPARα near genes involved in glycolysis/gluconeogenesis and uncovered a role for this factor in regulating anaerobic glycolysis. Overall, we generated extensive transcriptional and epigenomic datasets from livers of mice fed these diets and uncovered new functions and gene targets for PPARα.

Differential selective pressure alters rate of drug resistance acquisition in heterogeneous tumor populations.

Recent drug discovery and development efforts have created a large arsenal of targeted and chemotherapeutic drugs for precision medicine. However, drug resistance remains a major challenge as minor pre-existing resistant subpopulations are often found to be enriched at relapse. Current drug design has been heavily focused on initial efficacy, and we do not fully understand the effects of drug selective pressure on long-term drug resistance potential. Using a minimal two-population model, taking into account subpopulation proportions and growth/kill rates, we modeled long-term drug treatment and performed parameter sweeps to analyze the effects of each parameter on therapeutic efficacy. We found that drugs with the same overall initial kill may exert differential selective pressures, affecting long-term therapeutic outcome. We validated our conclusions experimentally using a preclinical model of Burkitt's lymphoma. Furthermore, we highlighted an intrinsic tradeoff between drug-imposed overall selective pressure and rate of adaptation. A principled approach in understanding the effects of distinct drug selective pressures on short-term and long-term tumor response enables better design of therapeutics that ultimately minimize relapse.

Network Modeling Identifies Patient-specific Pathways in Glioblastoma.

Glioblastoma is the most aggressive type of malignant human brain tumor. Molecular profiling experiments have revealed that these tumors are extremely heterogeneous. This heterogeneity is one of the principal challenges for developing targeted therapies. We hypothesize that despite the diverse molecular profiles, it might still be possible to identify common signaling changes that could be targeted in some or all tumors. Using a network modeling approach, we reconstruct the altered signaling pathways from tumor-specific phosphoproteomic data and known protein-protein interactions. We then develop a network-based strategy for identifying tumor specific proteins and pathways that were predicted by the models but not directly observed in the experiments. Among these hidden targets, we show that the ERK activator kinase1 (MEK1) displays increased phosphorylation in all tumors. By contrast, protein numb homolog (NUMB) is present only in the subset of the tumors that are the most invasive. Additionally, increased S100A4 is associated with only one of the tumors. Overall, our results demonstrate that despite the heterogeneity of the proteomic data, network models can identify common or tumor specific pathway-level changes. These results represent an important proof of principle that can improve the target selection process for tumor specific treatments.

Pathway-based network modeling finds hidden genes in shRNA screen for regulators of acute lymphoblastic leukemia.

Data integration stands to improve interpretation of RNAi screens which, as a result of off-target effects, typically yield numerous gene hits of which only a few validate. These off-target effects can result from seed matches to unintended gene targets (reagent-based) or cellular pathways, which can compensate for gene perturbations (biology-based). We focus on the biology-based effects and use network modeling tools to discover pathways de novo around RNAi hits. By looking at hits in a functional context, we can uncover novel biology not identified from any individual 'omics measurement. We leverage multiple 'omic measurements using the Simultaneous Analysis of Multiple Networks (SAMNet) computational framework to model a genome scale shRNA screen investigating Acute Lymphoblastic Leukemia (ALL) progression in vivo. Our network model is enriched for cellular processes associated with hematopoietic differentiation and homeostasis even though none of the individual 'omic sets showed this enrichment. The model identifies genes associated with the TGF-beta pathway and predicts a role in ALL progression for many genes without this functional annotation. We further experimentally validate the hidden genes - Wwp1, a ubiquitin ligase, and Hgs, a multi-vesicular body associated protein - for their role in ALL progression. Our ALL pathway model includes genes with roles in multiple types of leukemia and roles in hematological development. We identify a tumor suppressor role for Wwp1 in ALL progression. This work demonstrates that network integration approaches can compensate for off-target effects, and that these methods can uncover novel biology retroactively on existing screening data. We anticipate that this framework will be valuable to multiple functional genomic technologies - siRNA, shRNA, and CRISPR - generally, and will improve the utility of functional genomic studies.

Elucidating MicroRNA Regulatory Networks Using Transcriptional, Post-transcriptional, and Histone Modification Measurements.

MicroRNAs (miRNAs) regulate diverse biological processes by repressing mRNAs, but their modest effects on direct targets, together with their participation in larger regulatory networks, make it challenging to delineate miRNA-mediated effects. Here, we describe an approach to characterizing miRNA-regulatory networks by systematically profiling transcriptional, post-transcriptional and epigenetic activity in a pair of isogenic murine fibroblast cell lines with and without Dicer expression. By RNA sequencing (RNA-seq) and CLIP (crosslinking followed by immunoprecipitation) sequencing (CLIP-seq), we found that most of the changes induced by global miRNA loss occur at the level of transcription. We then introduced a network modeling approach that integrated these data with epigenetic data to identify specific miRNA-regulated transcription factors that explain the impact of miRNA perturbation on gene expression. In total, we demonstrate that combining multiple genome-wide datasets spanning diverse regulatory modes enables accurate delineation of the downstream miRNA-regulated transcriptional network and establishes a model for studying similar networks in other systems.

Extensive changes in DNA methylation are associated with expression of mutant huntingtin.

The earliest stages of Huntington disease are marked by changes in gene expression that are caused in an indirect and poorly understood manner by polyglutamine expansions in the huntingtin (HTT) protein. To explore the hypothesis that DNA methylation may be altered in cells expressing mutated HTT, we use reduced representation bisulfite sequencing (RRBS) to map sites of DNA methylation in cells carrying either wild-type or mutant HTT. We find that a large fraction of the genes that change in expression in the presence of mutant huntingtin demonstrate significant changes in DNA methylation. Regions with low CpG content, which have previously been shown to undergo methylation changes in response to neuronal activity, are disproportionately affected. On the basis of the sequence of regions that change in methylation, we identify AP-1 and SOX2 as transcriptional regulators associated with DNA methylation changes, and we confirm these hypotheses using genome-wide chromatin immunoprecipitation sequencing (ChIP-Seq). Our findings suggest new mechanisms for the effects of polyglutamine-expanded HTT. These results also raise important questions about the potential effects of changes in DNA methylation on neurogenesis and cognitive decline in patients with Huntington disease.