Co-expression in tissue-specific gene networks links genes in cancer-susceptibility loci to known somatic driver genes.

BACKGROUND: The genetic background of cancer remains complex and challenging to integrate. Many somatic mutations within genes are known to cause and drive cancer, while genome-wide association studies (GWAS) of cancer have revealed many germline risk factors associated with cancer. However, the overlap between known somatic driver genes and positional candidate genes from GWAS loci is surprisingly small. We hypothesised that genes from multiple independent cancer GWAS loci should show tissue-specific co-regulation patterns that converge on cancer-specific driver genes. RESULTS: We studied recent well-powered GWAS of breast, prostate, colorectal and skin cancer by estimating co-expression between genes and subsequently prioritising genes that show significant co-expression with genes mapping within susceptibility loci from cancer GWAS. We observed that the prioritised genes were strongly enriched for cancer drivers defined by COSMIC, IntOGen and Dietlein et al. The enrichment of known cancer driver genes was most significant when using co-expression networks derived from non-cancer samples of the relevant tissue of origin. CONCLUSION: We show how genes within risk loci identified by cancer GWAS can be linked to known cancer driver genes through tissue-specific co-expression networks. This provides an important explanation for why seemingly unrelated sets of genes that harbour either germline risk factors or somatic mutations can eventually cause the same type of disease.

WEE1 inhibitor adavosertib in combination with carboplatin in advanced TP53 mutated ovarian cancer: A biomarker-enriched phase II study.

OBJECTIVE: In the first part of this phase II study (NCT01164995), the combination of carboplatin and adavosertib (AZD1775) was shown to be safe and effective in patients with TP53 mutated platinum-resistant ovarian cancer (PROC). Here, we present the results of an additional safety and efficacy cohort and explore predictive biomarkers for resistance and response to this combination treatment. METHODS: This is a phase II, open-label, non-randomized study. Patients with TP53 mutated PROC received carboplatin AUC 5 mg/ ml·min intravenously and adavosertib 225 mg BID orally for 2.5 days in a 21-day cycle. The primary objective is to determine the efficacy and safety of carboplatin and adavosertib. Secondary objectives include progression-free survival (PFS), changes in circulating tumor cells (CTC) and exploration of genomic alterations. RESULTS: Thirty-two patients with a median age of 63 years (39-77 years) were enrolled and received treatment. Twenty-nine patients were evaluable for efficacy. Bone marrow toxicity, nausea and vomiting were the most common adverse events. Twelve patients showed partial response (PR) as best response, resulting in an objective ORR of 41% in the evaluable patients (95% CI: 23%-61%). The median PFS was 5.6 months (95% CI: 3.8-10.3). In patients with tumors harboring CCNE1 amplification, treatment efficacy was slightly but not significantly better. CONCLUSIONS: Adavosertib 225 mg BID for 2.5 days and carboplatin AUC 5 could be safely combined and showed anti-tumor efficacy in patients with PROC. However, bone marrow toxicity remains a point of concern, since this is the most common reason for dose reductions and dose delays.

Predicting response to enzalutamide and abiraterone in metastatic prostate cancer using whole-omics machine learning.

Response to androgen receptor signaling inhibitors (ARSI) varies widely in metastatic castration resistant prostate cancer (mCRPC). To improve treatment guidance, biomarkers are needed. We use whole-genomics (WGS; n = 155) with matching whole-transcriptomics (WTS; n = 113) from biopsies of ARSI-treated mCRPC patients for unbiased discovery of biomarkers and development of machine learning-based prediction models. Tumor mutational burden (q < 0.001), structural variants (q < 0.05), tandem duplications (q < 0.05) and deletions (q < 0.05) are enriched in poor responders, coupled with distinct transcriptomic expression profiles. Validating various classification models predicting treatment duration with ARSI on our internal and external mCRPC cohort reveals two best-performing models, based on the combination of prior treatment information with either the four combined enriched genomic markers or with overall transcriptomic profiles. In conclusion, predictive models combining genomic, transcriptomic, and clinical data can predict response to ARSI in mCRPC patients and, with additional optimization and prospective validation, could improve treatment guidance.

The genome-wide mutational consequences of DNA hypomethylation.

DNA methylation is important for establishing and maintaining cell identity and for genomic stability. This is achieved by regulating the accessibility of regulatory and transcriptional elements and the compaction of subtelomeric, centromeric, and other inactive genomic regions. Carcinogenesis is accompanied by a global loss in DNA methylation, which facilitates the transformation of cells. Cancer hypomethylation may also cause genomic instability, for example through interference with the protective function of telomeres and centromeres. However, understanding the role(s) of hypomethylation in tumor evolution is incomplete because the precise mutational consequences of global hypomethylation have thus far not been systematically assessed. Here we made genome-wide inventories of all possible genetic variation that accumulates in single cells upon the long-term global hypomethylation by CRISPR interference-mediated conditional knockdown of DNMT1. Depletion of DNMT1 resulted in a genomewide reduction in DNA methylation. The degree of DNA methylation loss was similar to that observed in many cancer types. Hypomethylated cells showed reduced proliferation rates, increased transcription of genes, reactivation of the inactive X-chromosome and abnormal nuclear morphologies. Prolonged hypomethylation was accompanied by increased chromosomal instability. However, there was no increase in mutational burden, enrichment for certain mutational signatures or accumulation of structural variation to the genome. In conclusion, the primary consequence of hypomethylation is genomic instability, which in cancer leads to increased tumor heterogeneity and thereby fuels cancer evolution.

Integration of multiple lineage measurements from the same cell reconstructs parallel tumor evolution.

Organoid evolution models complemented with integrated single-cell sequencing technology provide a powerful platform to characterize intra-tumor heterogeneity (ITH) and tumor evolution. Here, we conduct a parallel evolution experiment to mimic the tumor evolution process by evolving a colon cancer organoid model over 100 generations, spanning 6 months in time. We use single-cell whole-genome sequencing (WGS) in combination with viral lineage tracing at 12 time points to simultaneously monitor clone size, CNV states, SNV states, and viral lineage barcodes for 1,641 single cells. We integrate these measurements to construct clonal evolution trees with high resolution. We characterize the order of events in which chromosomal aberrations occur and identify aberrations that recur multiple times within the same tumor sub-population. We observe recurrent sequential loss of chromosome 4 after loss of chromosome 18 in four unique tumor clones. SNVs and CNVs identified in our organoid experiments are also frequently reported in colorectal carcinoma samples, and out of 334 patients with chromosome 18 loss in a Memorial Sloan Kettering colorectal cancer cohort, 99 (29.6%) also harbor chromosome 4 loss. Our study reconstructs tumor evolution in a colon cancer organoid model at high resolution, demonstrating an approach to identify potentially clinically relevant genomic aberrations in tumor evolution.

Accurate detection of circulating tumor DNA using nanopore consensus sequencing.

Levels of circulating tumor DNA (ctDNA) in liquid biopsies may serve as a sensitive biomarker for real-time, minimally-invasive tumor diagnostics and monitoring. However, detecting ctDNA is challenging, as much fewer than 5% of the cell-free DNA in the blood typically originates from the tumor. To detect lowly abundant ctDNA molecules based on somatic variants, extremely sensitive sequencing methods are required. Here, we describe a new technique, CyclomicsSeq, which is based on Oxford Nanopore sequencing of concatenated copies of a single DNA molecule. Consensus calling of the DNA copies increased the base-calling accuracy ~60×, enabling accurate detection of TP53 mutations at frequencies down to 0.02%. We demonstrate that a TP53-specific CyclomicsSeq assay can be successfully used to monitor tumor burden during treatment for head-and-neck cancer patients. CyclomicsSeq can be applied to any genomic locus and offers an accurate diagnostic liquid biopsy approach that can be implemented in clinical workflows.

Predicting pathogenic non-coding SVs disrupting the 3D genome in 1646 whole cancer genomes using multiple instance learning.

Over the past years, large consortia have been established to fuel the sequencing of whole genomes of many cancer patients. Despite the increased abundance in tools to study the impact of SNVs, non-coding SVs have been largely ignored in these data. Here, we introduce svMIL2, an improved version of our Multiple Instance Learning-based method to study the effect of somatic non-coding SVs disrupting boundaries of TADs and CTCF loops in 1646 cancer genomes. We demonstrate that svMIL2 predicts pathogenic non-coding SVs with an average AUC of 0.86 across 12 cancer types, and identifies non-coding SVs affecting well-known driver genes. The disruption of active (super) enhancers in open chromatin regions appears to be a common mechanism by which non-coding SVs exert their pathogenicity. Finally, our results reveal that the contribution of pathogenic non-coding SVs as opposed to driver SNVs may highly vary between cancers, with notably high numbers of genes being disrupted by pathogenic non-coding SVs in ovarian and pancreatic cancer. Taken together, our machine learning method offers a potent way to prioritize putatively pathogenic non-coding SVs and leverage non-coding SVs to identify driver genes. Moreover, our analysis of 1646 cancer genomes demonstrates the importance of including non-coding SVs in cancer diagnostics.

Allele-specific expression of GATA2 due to epigenetic dysregulation in CEBPA double-mutant AML.

Transcriptional deregulation is a central event in the development of acute myeloid leukemia (AML). To identify potential disturbances in gene regulation, we conducted an unbiased screen of allele-specific expression (ASE) in 209 AML cases. The gene encoding GATA binding protein 2 (GATA2) displayed ASE more often than any other myeloid- or cancer-related gene. GATA2 ASE was strongly associated with CEBPA double mutations (DMs), with 95% of cases presenting GATA2 ASE. In CEBPA DM AML with GATA2 mutations, the mutated allele was preferentially expressed. We found that GATA2 ASE was a somatic event lost in complete remission, supporting the notion that it plays a role in CEBPA DM AML. Acquisition of GATA2 ASE involved silencing of 1 allele via promoter methylation and concurrent overactivation of the other allele, thereby preserving expression levels. Notably, promoter methylation was also lost in remission along with GATA2 ASE. In summary, we propose that GATA2 ASE is acquired by epigenetic mechanisms and is a prerequisite for the development of AML with CEBPA DMs. This finding constitutes a novel example of an epigenetic hit cooperating with a genetic hit in the pathogenesis of AML.

Cancer Type Classification in Liquid Biopsies Based on Sparse Mutational Profiles Enabled through Data Augmentation and Integration.

Identifying the cell of origin of cancer is important to guide treatment decisions. Machine learning approaches have been proposed to classify the cell of origin based on somatic mutation profiles from solid biopsies. However, solid biopsies can cause complications and certain tumors are not accessible. Liquid biopsies are promising alternatives but their somatic mutation profile is sparse and current machine learning models fail to perform in this setting. We propose an improved method to deal with sparsity in liquid biopsy data. Firstly, data augmentation is performed on sparse data to enhance model robustness. Secondly, we employ data integration to merge information from: (i) SNV density; (ii) SNVs in driver genes and (iii) trinucleotide motifs. Our adapted method achieves an average accuracy of 0.88 and 0.65 on data where only 70% and 2% of SNVs are retained, compared to 0.83 and 0.41 with the original model, respectively. The method and results presented here open the way for application of machine learning in the detection of the cell of origin of cancer from liquid biopsy data.

RAINFOREST: a random forest approach to predict treatment benefit in data from (failed) clinical drug trials.

MOTIVATION: When phase III clinical drug trials fail their endpoint, enormous resources are wasted. Moreover, even if a clinical trial demonstrates a significant benefit, the observed effects are often small and may not outweigh the side effects of the drug. Therefore, there is a great clinical need for methods to identify genetic markers that can identify subgroups of patients which are likely to benefit from treatment as this may (i) rescue failed clinical trials and/or (ii) identify subgroups of patients which benefit more than the population as a whole. When single genetic biomarkers cannot be found, machine learning approaches that find multivariate signatures are required. For single nucleotide polymorphism (SNP) profiles, this is extremely challenging owing to the high dimensionality of the data. Here, we introduce RAINFOREST (tReAtment benefIt prediction using raNdom FOREST), which can predict treatment benefit from patient SNP profiles obtained in a clinical trial setting. RESULTS: We demonstrate the performance of RAINFOREST on the CAIRO2 dataset, a phase III clinical trial which tested the addition of cetuximab treatment for metastatic colorectal cancer and concluded there was no benefit. However, we find that RAINFOREST is able to identify a subgroup comprising 27.7% of the patients that do benefit, with a hazard ratio of 0.69 (P = 0.04) in favor of cetuximab. The method is not specific to colorectal cancer and could aid in reanalysis of clinical trial data and provide a more personalized approach to cancer treatment, also when there is no clear link between a single variant and treatment benefit. AVAILABILITY AND IMPLEMENTATION: The R code used to produce the results in this paper can be found at github.com/jubels/RAINFOREST. A more configurable, user-friendly Python implementation of RAINFOREST is also provided. Due to restrictions based on privacy regulations and informed consent of participants, phenotype and genotype data of the CAIRO2 trial cannot be made freely available in a public repository. Data from this study can be obtained upon request. Requests should be directed toward Prof. Dr. H.J. Guchelaar (h.j.guchelaar@lumc.nl). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

svMIL: predicting the pathogenic effect of TAD boundary-disrupting somatic structural variants through multiple instance learning.

MOTIVATION: Despite the fact that structural variants (SVs) play an important role in cancer, methods to predict their effect, especially for SVs in non-coding regions, are lacking, leaving them often overlooked in the clinic. Non-coding SVs may disrupt the boundaries of Topologically Associated Domains (TADs), thereby affecting interactions between genes and regulatory elements such as enhancers. However, it is not known when such alterations are pathogenic. Although machine learning techniques are a promising solution to answer this question, representing the large number of interactions that an SV can disrupt in a single feature matrix is not trivial. RESULTS: We introduce svMIL: a method to predict pathogenic TAD boundary-disrupting SV effects based on multiple instance learning, which circumvents the need for a traditional feature matrix by grouping SVs into bags that can contain any number of disruptions. We demonstrate that svMIL can predict SV pathogenicity, measured through same-sample gene expression aberration, for various cancer types. In addition, our approach reveals that somatic pathogenic SVs alter different regulatory interactions than somatic non-pathogenic SVs and germline SVs. AVAILABILITY AND IMPLEMENTATION: All code for svMIL is publicly available on GitHub: https://github.com/UMCUGenetics/svMIL. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Gene Networks Constructed Through Simulated Treatment Learning can Predict Proteasome Inhibitor Benefit in Multiple Myeloma.

PURPOSE: Proteasome inhibitors are widely used in treating multiple myeloma, but can cause serious side effects and response varies among patients. It is, therefore, important to gain more insight into which patients will benefit from proteasome inhibitors. EXPERIMENTAL DESIGN: We introduce simulated treatment learned signatures (STLsig), a machine learning method to identify predictive gene expression signatures. STLsig uses genetically similar patients who have received an alternative treatment to model which patients will benefit more from proteasome inhibitors than from an alternative treatment. STLsig constructs gene networks by linking genes that are synergistic in their ability to predict benefit. RESULTS: In a dataset of 910 patients with multiple myeloma, STLsig identified two gene networks that together can predict benefit to the proteasome inhibitor, bortezomib. In class "benefit," we found an HR of 0.47 (P = 0.04) in favor of bortezomib, while in class "no benefit," the HR was 0.91 (P = 0.68). Importantly, we observed a similar performance (HR class benefit, 0.46; P = 0.04) in an independent patient cohort. Moreover, this signature also predicts benefit for the proteasome inhibitor, carfilzomib, indicating it is not specific to bortezomib. No equivalent signature can be found when the genes in the signature are excluded from the analysis, indicating that they are essential. Multiple genes in the signature are linked to working mechanisms of proteasome inhibitors or multiple myeloma disease progression. CONCLUSIONS: STLsig can identify gene signatures that could aid in treatment decisions for patients with multiple myeloma and provide insight into the biological mechanism behind treatment benefit.

MetaboShiny: interactive analysis and metabolite annotation of mass spectrometry-based metabolomics data.

Direct infusion untargeted mass spectrometry-based metabolomics allows for rapid insight into a sample's metabolic activity. However, analysis is often complicated by the large array of detected m/z values and the difficulty to prioritize important m/z and simultaneously annotate their putative identities. To address this challenge, we developed MetaboShiny, a novel R/RShiny-based metabolomics package featuring data analysis, database- and formula-prediction-based annotation and visualization. To demonstrate this, we reproduce and further explore a MetaboLights metabolomics bioinformatics study on lung cancer patient urine samples. MetaboShiny enables rapid and rigorous analysis and interpretation of direct infusion untargeted mass spectrometry-based metabolomics data.

A deep learning system accurately classifies primary and metastatic cancers using passenger mutation patterns.

In cancer, the primary tumour's organ of origin and histopathology are the strongest determinants of its clinical behaviour, but in 3% of cases a patient presents with a metastatic tumour and no obvious primary. Here, as part of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes (PCAWG) Consortium, we train a deep learning classifier to predict cancer type based on patterns of somatic passenger mutations detected in whole genome sequencing (WGS) of 2606 tumours representing 24 common cancer types produced by the PCAWG Consortium. Our classifier achieves an accuracy of 91% on held-out tumor samples and 88% and 83% respectively on independent primary and metastatic samples, roughly double the accuracy of trained pathologists when presented with a metastatic tumour without knowledge of the primary. Surprisingly, adding information on driver mutations reduced accuracy. Our results have clinical applicability, underscore how patterns of somatic passenger mutations encode the state of the cell of origin, and can inform future strategies to detect the source of circulating tumour DNA.

sv-callers: a highly portable parallel workflow for structural variant detection in whole-genome sequence data.

Structural variants (SVs) are an important class of genetic variation implicated in a wide array of genetic diseases including cancer. Despite the advances in whole genome sequencing, comprehensive and accurate detection of SVs in short-read data still poses some practical and computational challenges. We present sv-callers, a highly portable workflow that enables parallel execution of multiple SV detection tools, as well as provide users with example analyses of detected SV callsets in a Jupyter Notebook. This workflow supports easy deployment of software dependencies, configuration and addition of new analysis tools. Moreover, porting it to different computing systems requires minimal effort. Finally, we demonstrate the utility of the workflow by performing both somatic and germline SV analyses on different high-performance computing systems.

Molecular heterogeneity and early metastatic clone selection in testicular germ cell cancer development.

BACKGROUND: Testicular germ cell cancer (TGCC), being the most frequent malignancy in young Caucasian males, is initiated from an embryonic germ cell. This study determines intratumour heterogeneity to unravel tumour progression from initiation until metastasis. METHODS: In total, 42 purified samples of four treatment-resistant nonseminomatous (NS) TGCC were investigated, including the precursor germ cell neoplasia in situ (GCNIS) and metastatic specimens, using whole-genome and targeted sequencing. Their evolution was reconstructed. RESULTS: Intratumour molecular heterogeneity did not correspond to the supposed primary tumour histological evolution. Metastases after systemic treatment could be derived from cancer stem cells not identified in the primary cancer. GCNIS mostly lacked the molecular marks of the primary NS and comprised dominant clones that failed to progress. A BRCA-like mutational signature was observed without evidence for direct involvement of BRCA1 and BRCA2 genes. CONCLUSIONS: Our data strongly support the hypothesis that NS is initiated by whole-genome duplication, followed by chromosome copy number alterations in the cancer stem cell population, and accumulation of low numbers of somatic mutations, even in therapy-resistant cases. These observations of heterogeneity at all stages of tumourigenesis should be considered when treating patients with GCNIS-only disease, or with clinically overt NS.

A data-driven interactome of synergistic genes improves network-based cancer outcome prediction.

Robustly predicting outcome for cancer patients from gene expression is an important challenge on the road to better personalized treatment. Network-based outcome predictors (NOPs), which considers the cellular wiring diagram in the classification, hold much promise to improve performance, stability and interpretability of identified marker genes. Problematically, reports on the efficacy of NOPs are conflicting and for instance suggest that utilizing random networks performs on par to networks that describe biologically relevant interactions. In this paper we turn the prediction problem around: instead of using a given biological network in the NOP, we aim to identify the network of genes that truly improves outcome prediction. To this end, we propose SyNet, a gene network constructed ab initio from synergistic gene pairs derived from survival-labelled gene expression data. To obtain SyNet, we evaluate synergy for all 69 million pairwise combinations of genes resulting in a network that is specific to the dataset and phenotype under study and can be used to in a NOP model. We evaluated SyNet and 11 other networks on a compendium dataset of >4000 survival-labelled breast cancer samples. For this purpose, we used cross-study validation which more closely emulates real world application of these outcome predictors. We find that SyNet is the only network that truly improves performance, stability and interpretability in several existing NOPs. We show that SyNet overlaps significantly with existing gene networks, and can be confidently predicted (~85% AUC) from graph-topological descriptions of these networks, in particular the breast tissue-specific network. Due to its data-driven nature, SyNet is not biased to well-studied genes and thus facilitates post-hoc interpretation. We find that SyNet is highly enriched for known breast cancer genes and genes related to e.g. histological grade and tamoxifen resistance, suggestive of a role in determining breast cancer outcome.