Cancer drug sensitivity estimation using modular deep Graph Neural Networks.

Computational drug sensitivity models have the potential to improve therapeutic outcomes by identifying targeted drugs components that are tailored to the transcriptomic profile of a given primary tumor. The SMILES representation of molecules that is used by state-of-the-art drug-sensitivity models is not conducive for neural networks to generalize to new drugs, in part because the distance between atoms does not generally correspond to the distance between their representation in the SMILES strings. Graph-attention networks, on the other hand, are high-capacity models that require large training-data volumes which are not available for drug-sensitivity estimation. We develop a modular drug-sensitivity graph-attentional neural network. The modular architecture allows us to separately pre-train the graph encoder and graph-attentional pooling layer on related tasks for which more data are available. We observe that this model outperforms reference models for the use cases of precision oncology and drug discovery; in particular, it is better able to predict the specific interaction between drug and cell line that is not explained by the general cytotoxicity of the drug and the overall survivability of the cell line. The complete source code is available at https://zenodo.org/doi/10.5281/zenodo.8020945. All experiments are based on the publicly available GDSC data.

Neurofibromin 1 controls metabolic balance and Notch-dependent quiescence of murine juvenile myogenic progenitors.

Patients affected by neurofibromatosis type 1 (NF1) frequently show muscle weakness with unknown etiology. Here we show that, in mice, Neurofibromin 1 (Nf1) is not required in muscle fibers, but specifically in early postnatal myogenic progenitors (MPs), where Nf1 loss led to cell cycle exit and differentiation blockade, depleting the MP pool resulting in reduced myonuclear accretion as well as reduced muscle stem cell numbers. This was caused by precocious induction of stem cell quiescence coupled to metabolic reprogramming of MPs impinging on glycolytic shutdown, which was conserved in muscle fibers. We show that a Mek/Erk/NOS pathway hypersensitizes Nf1-deficient MPs to Notch signaling, consequently, early postnatal Notch pathway inhibition ameliorated premature quiescence, metabolic reprogramming and muscle growth. This reveals an unexpected role of Ras/Mek/Erk signaling supporting postnatal MP quiescence in concert with Notch signaling, which is controlled by Nf1 safeguarding coordinated muscle growth and muscle stem cell pool establishment. Furthermore, our data suggest transmission of metabolic reprogramming across cellular differentiation, affecting fiber metabolism and function in NF1.

Pre-Training on In Vitro and Fine-Tuning on Patient-Derived Data Improves Deep Neural Networks for Anti-Cancer Drug-Sensitivity Prediction.

Large-scale databases that report the inhibitory capacities of many combinations of candidate drug compounds and cultivated cancer cell lines have driven the development of preclinical drug-sensitivity models based on machine learning. However, cultivated cell lines have devolved from human cancer cells over years or even decades under selective pressure in culture conditions. Moreover, models that have been trained on in vitro data cannot account for interactions with other types of cells. Drug-response data that are based on patient-derived cell cultures, xenografts, and organoids, on the other hand, are not available in the quantities that are needed to train high-capacity machine-learning models. We found that pre-training deep neural network models of drug sensitivity on in vitro drug-sensitivity databases before fine-tuning the model parameters on patient-derived data improves the models' accuracy and improves the biological plausibility of the features, compared to training only on patient-derived data. From our experiments, we can conclude that pre-trained models outperform models that have been trained on the target domains in the vast majority of cases.

Epirubicin Alters DNA Methylation Profiles Related to Cardiotoxicity.

BACKGROUND: Epirubicin (EPI) is an important anticancer drug that is well-known for its cardiotoxic side effect. Studying epigenetic modification such as DNA methylation can help to understand the EPI-related toxic mechanisms in cardiac tissue. In this study, we analyzed the DNA methylation profile in a relevant human cell model and inspected the expression of differentially methylated genes at the transcriptome level to understand how changes in DNA methylation could affect gene expression in relation to EPI-induced cardiotoxicity. METHODS: Human cardiac microtissues were exposed to either therapeutic or toxic (IC20) EPI doses during 2 weeks. The DNA and RNA were collected from microtissues in triplicates at 2, 8, 24, 72, 168, 240, and 336 hours of exposure. Methylated DNA immunoprecipitation-sequencing (MeDIP-seq) analysis was used to detect DNA methylation levels in EPI-treated and control samples. The MeDIP-seq data were analyzed and processed using the QSEA package with a recently published workflow. RNA sequencing (RNA-seq) was used to measure global gene expression in the same samples. RESULTS: After processing the MeDIP-seq data, we detected 35, 37, 15 candidate genes which show strong methylated alterations between all EPI-treated, EPI therapeutic and EPI toxic dose-treated samples compared to control, respectively. For several genes, gene expressions changed compatibly reflecting the DNA methylation regulation. CONCLUSIONS: The observed DNA methylation modifications provide further insights into the EPI-induced cardiotoxicity. Multiple differentially methylated genes under EPI treatment, such as SMARCA4, PKN1, RGS12, DPP9, NCOR2, SDHA, POLR2A, and AGPAT3, have been implicated in different cardiac dysfunction mechanisms. Together with other differentially methylated genes, these genes can be candidates for further investigations of EPI-related toxic mechanisms. Data Repository: The data has been generated by the HeCaToS project (http://www.ebi.ac.uk/biostudies) under accession numbers S-HECA433 and S-HECA434 for the MeDIP-seq data and S-HECA11 for the RNA-seq data. The R code is available on Github (https://github.com/NhanNguyen000/MeDIP).

Matching anticancer compounds and tumor cell lines by neural networks with ranking loss.

Computational drug sensitivity models have the potential to improve therapeutic outcomes by identifying targeted drug components that are likely to achieve the highest efficacy for a cancer cell line at hand at a therapeutic dose. State of the art drug sensitivity models use regression techniques to predict the inhibitory concentration of a drug for a tumor cell line. This regression objective is not directly aligned with either of these principal goals of drug sensitivity models: We argue that drug sensitivity modeling should be seen as a ranking problem with an optimization criterion that quantifies a drug's inhibitory capacity for the cancer cell line at hand relative to its toxicity for healthy cells. We derive an extension to the well-established drug sensitivity regression model PaccMann that employs a ranking loss and focuses on the ratio of inhibitory concentration and therapeutic dosage range. We find that the ranking extension significantly enhances the model's capability to identify the most effective anticancer drugs for unseen tumor cell profiles based in on in-vitro data.

Altered DNA Methylation Profiles in SF3B1 Mutated CLL Patients.

Mutations in splicing factor genes have a severe impact on the survival of cancer patients. Splicing factor 3b subunit 1 (SF3B1) is one of the most frequently mutated genes in chronic lymphocytic leukemia (CLL); patients carrying these mutations have a poor prognosis. Since the splicing machinery and the epigenome are closely interconnected, we investigated whether these alterations may affect the epigenomes of CLL patients. While an overall hypomethylation during CLL carcinogenesis has been observed, the interplay between the epigenetic stage of the originating B cells and SF3B1 mutations, and the subsequent effect of the mutations on methylation alterations in CLL, have not been investigated. We profiled the genome-wide DNA methylation patterns of 27 CLL patients with and without SF3B1 mutations and identified local decreases in methylation levels in SF3B1mut CLL patients at 67 genomic regions, mostly in proximity to telomeric regions. These differentially methylated regions (DMRs) were enriched in gene bodies of cancer-related signaling genes, e.g., NOTCH1, HTRA3, and BCL9L. In our study, SF3B1 mutations exclusively emerged in two out of three epigenetic stages of the originating B cells. However, not all the DMRs could be associated with the methylation programming of B cells during development, suggesting that mutations in SF3B1 cause additional epigenetic aberrations during carcinogenesis.

Large-scale literature mining to assess the relation between anti-cancer drugs and cancer types.

BACKGROUND: There is a huge body of scientific literature describing the relation between tumor types and anti-cancer drugs. The vast amount of scientific literature makes it impossible for researchers and physicians to extract all relevant information manually. METHODS: In order to cope with the large amount of literature we applied an automated text mining approach to assess the relations between 30 most frequent cancer types and 270 anti-cancer drugs. We applied two different approaches, a classical text mining based on named entity recognition and an AI-based approach employing word embeddings. The consistency of literature mining results was validated with 3 independent methods: first, using data from FDA approvals, second, using experimentally measured IC-50 cell line data and third, using clinical patient survival data. RESULTS: We demonstrated that the automated text mining was able to successfully assess the relation between cancer types and anti-cancer drugs. All validation methods showed a good correspondence between the results from literature mining and independent confirmatory approaches. The relation between most frequent cancer types and drugs employed for their treatment were visualized in a large heatmap. All results are accessible in an interactive web-based knowledge base using the following link: https://knowledgebase.microdiscovery.de/heatmap . CONCLUSIONS: Our approach is able to assess the relations between compounds and cancer types in an automated manner. Both, cancer types and compounds could be grouped into different clusters. Researchers can use the interactive knowledge base to inspect the presented results and follow their own research questions, for example the identification of novel indication areas for known drugs.

PWD/Ph-Encoded Genetic Variants Modulate the Cellular Wnt/ß-Catenin Response to Suppress Apc Min-Triggered Intestinal Tumor Formation.

Genetic predisposition affects the penetrance of tumor-initiating mutations, such as APC mutations that stabilize ß-catenin and cause intestinal tumors in mice and humans. However, the mechanisms involved in genetically predisposed penetrance are not well understood. Here, we analyzed tumor multiplicity and gene expression in tumor-prone Apc Min/+ mice on highly variant C57BL/6J (B6) and PWD/Ph (PWD) genetic backgrounds. (B6 × PWD) F1 APC Min offspring mice were largely free of intestinal adenoma, and several chromosome substitution (consomic) strains carrying single PWD chromosomes on the B6 genetic background displayed reduced adenoma numbers. Multiple dosage-dependent modifier loci on PWD chromosome 5 each contributed to tumor suppression. Activation of ß-catenin-driven and stem cell-specific gene expression in the presence of Apc Min or following APC loss remained moderate in intestines carrying PWD chromosome 5, suggesting that PWD variants restrict adenoma initiation by controlling stem cell homeostasis. Gene expression of modifier candidates and DNA methylation on chromosome 5 were predominantly cis controlled and largely reflected parental patterns, providing a genetic basis for inheritance of tumor susceptibility. Human SNP variants of several modifier candidates were depleted in colorectal cancer genomes, suggesting that similar mechanisms may also affect the penetrance of cancer driver mutations in humans. Overall, our analysis highlights the strong impact that multiple genetic variants acting in networks can exert on tumor development. SIGNIFICANCE: These findings in mice show that, in addition to accidental mutations, cancer risk is determined by networks of individual gene variants.

Combined Targeted Resequencing of Cytosine DNA Methylation and Mutations of DNA Repair Genes with Potential Use for Poly(ADP-Ribose) Polymerase 1 Inhibitor Sensitivity Testing.

Current molecular tumor diagnostics encompass panel sequencing to detect mutations, copy number alterations, and rearrangements. However, tumor suppressor genes can also be inactivated by methylation within their promoter region. These epigenetic alterations are so far rarely assessed in the clinical setting. Therefore, we established the AllCap protocol facilitating the combined detection of mutations and DNA methylation at the coding and promoter regions of 342 DNA repair genes in one experiment. We demonstrate the use of the protocol by applying it to ovarian cancer cell lines with different responsiveness to poly(ADP-ribose) polymerase inhibition. BRCA1, ATM, ATR, and EP300 mutations and methylation of the BRCA1 promoter were detected as potential predictors for therapy response. The required amount of input DNA was optimized, and the application to formalin-fixed, paraffin-embedded tissue samples was verified to improve the clinical applicability. Thus, by adding DNA methylation values to panel resequencings, the AllCap assay will add another important level of information to clinical tests and will improve stratification of patients for systemic therapies.

Epigenomic profiling of non-small cell lung cancer xenografts uncover LRP12 DNA methylation as predictive biomarker for carboplatin resistance.

BACKGROUND: Non-small cell lung cancer (NSCLC) is the most common cause of cancer-related deaths worldwide and is primarily treated with radiation, surgery, and platinum-based drugs like cisplatin and carboplatin. The major challenge in the treatment of NSCLC patients is intrinsic or acquired resistance to chemotherapy. Molecular markers predicting the outcome of the patients are urgently needed. METHODS: Here, we employed patient-derived xenografts (PDXs) to detect predictive methylation biomarkers for platin-based therapies. We used MeDIP-Seq to generate genome-wide DNA methylation profiles of 22 PDXs, their parental primary NSCLC, and their corresponding normal tissues and complemented the data with gene expression analyses of the same tissues. Candidate biomarkers were validated with quantitative methylation-specific PCRs (qMSP) in an independent cohort. RESULTS: Comprehensive analyses revealed that differential methylation patterns are highly similar, enriched in PDXs and lung tumor-specific when comparing differences in methylation between PDXs versus primary NSCLC. We identified a set of 40 candidate regions with methylation correlated to carboplatin response and corresponding inverse gene expression pattern even before therapy. This analysis led to the identification of a promoter CpG island methylation of LDL receptor-related protein 12 (LRP12) associated with increased resistance to carboplatin. Validation in an independent patient cohort (n = 35) confirmed that LRP12 methylation status is predictive for therapeutic response of NSCLC patients to platin therapy with a sensitivity of 80% and a specificity of 84% (p < 0.01). Similarly, we find a shorter survival time for patients with LRP12 hypermethylation in the TCGA data set for NSCLC (lung adenocarcinoma). CONCLUSIONS: Using an epigenome-wide sequencing approach, we find differential methylation patterns from primary lung cancer and PDX-derived cancers to be very similar, albeit with a lower degree of differential methylation in primary tumors. We identify LRP12 DNA methylation as a powerful predictive marker for carboplatin resistance. These findings outline a platform for the identification of epigenetic therapy resistance biomarkers based on PDX NSCLC models.

Therapeutic targeting of ependymoma as informed by oncogenic enhancer profiling.

Genomic sequencing has driven precision-based oncology therapy; however, the genetic drivers of many malignancies remain unknown or non-targetable, so alternative approaches to the identification of therapeutic leads are necessary. Ependymomas are chemotherapy-resistant brain tumours, which, despite genomic sequencing, lack effective molecular targets. Intracranial ependymomas are segregated on the basis of anatomical location (supratentorial region or posterior fossa) and further divided into distinct molecular subgroups that reflect differences in the age of onset, gender predominance and response to therapy. The most common and aggressive subgroup, posterior fossa ependymoma group A (PF-EPN-A), occurs in young children and appears to lack recurrent somatic mutations. Conversely, posterior fossa ependymoma group B (PF-EPN-B) tumours display frequent large-scale copy number gains and losses but have favourable clinical outcomes. More than 70% of supratentorial ependymomas are defined by highly recurrent gene fusions in the NF-κB subunit gene RELA (ST-EPN-RELA), and a smaller number involve fusion of the gene encoding the transcriptional activator YAP1 (ST-EPN-YAP1). Subependymomas, a distinct histologic variant, can also be found within the supratetorial and posterior fossa compartments, and account for the majority of tumours in the molecular subgroups ST-EPN-SE and PF-EPN-SE. Here we describe mapping of active chromatin landscapes in 42 primary ependymomas in two non-overlapping primary ependymoma cohorts, with the goal of identifying essential super-enhancer-associated genes on which tumour cells depend. Enhancer regions revealed putative oncogenes, molecular targets and pathways; inhibition of these targets with small molecule inhibitors or short hairpin RNA diminished the proliferation of patient-derived neurospheres and increased survival in mouse models of ependymomas. Through profiling of transcriptional enhancers, our study provides a framework for target and drug discovery in other cancers that lack known genetic drivers and are therefore difficult to treat.

QSEA-modelling of genome-wide DNA methylation from sequencing enrichment experiments.

Genome-wide enrichment of methylated DNA followed by sequencing (MeDIP-seq) offers a reasonable compromise between experimental costs and genomic coverage. However, the computational analysis of these experiments is complex, and quantification of the enrichment signals in terms of absolute levels of methylation requires specific transformation. In this work, we present QSEA, Quantitative Sequence Enrichment Analysis, a comprehensive workflow for the modelling and subsequent quantification of MeDIP-seq data. As the central part of the workflow we have developed a Bayesian statistical model that transforms the enrichment read counts to absolute levels of methylation and, thus, enhances interpretability and facilitates comparison with other methylation assays. We suggest several calibration strategies for the critical parameters of the model, either using additional data or fairly general assumptions. By comparing the results with bisulfite sequencing (BS) validation data, we show the improvement of QSEA over existing methods. Additionally, we generated a clinically relevant benchmark data set consisting of methylation enrichment experiments (MeDIP-seq), BS-based validation experiments (Methyl-seq) as well as gene expression experiments (RNA-seq) derived from non-small cell lung cancer patients, and show that the workflow retrieves well-known lung tumour methylation markers that are causative for gene expression changes, demonstrating the applicability of QSEA for clinical studies. QSEA is implemented in R and available from the Bioconductor repository 3.4 (www.bioconductor.org/packages/qsea).

Simultaneous deletion of the methylcytosine oxidases Tet1 and Tet3 increases transcriptome variability in early embryogenesis.

Dioxygenases of the TET (Ten-Eleven Translocation) family produce oxidized methylcytosines, intermediates in DNA demethylation, as well as new epigenetic marks. Here we show data suggesting that TET proteins maintain the consistency of gene transcription. Embryos lacking Tet1 and Tet3 (Tet1/3 DKO) displayed a strong loss of 5-hydroxymethylcytosine (5hmC) and a concurrent increase in 5-methylcytosine (5mC) at the eight-cell stage. Single cells from eight-cell embryos and individual embryonic day 3.5 blastocysts showed unexpectedly variable gene expression compared with controls, and this variability correlated in blastocysts with variably increased 5mC/5hmC in gene bodies and repetitive elements. Despite the variability, genes encoding regulators of cholesterol biosynthesis were reproducibly down-regulated in Tet1/3 DKO blastocysts, resulting in a characteristic phenotype of holoprosencephaly in the few embryos that survived to later stages. Thus, TET enzymes and DNA cytosine modifications could directly or indirectly modulate transcriptional noise, resulting in the selective susceptibility of certain intracellular pathways to regulation by TET proteins.

The histone deacetylase SIRT6 controls embryonic stem cell fate via TET-mediated production of 5-hydroxymethylcytosine.

How embryonic stem cells (ESCs) commit to specific cell lineages and yield all cell types of a fully formed organism remains a major question. ESC differentiation is accompanied by large-scale histone and DNA modifications, but the relations between these epigenetic categories are not understood. Here we demonstrate the interplay between the histone deacetylase sirtuin 6 (SIRT6) and the ten-eleven translocation enzymes (TETs). SIRT6 targets acetylated histone H3 at Lys 9 and 56 (H3K9ac and H3K56ac), while TETs convert 5-methylcytosine into 5-hydroxymethylcytosine (5hmC). ESCs derived from Sirt6 knockout (S6KO) mice are skewed towards neuroectoderm development. This phenotype involves derepression of OCT4, SOX2 and NANOG, which causes an upregulation of TET-dependent production of 5hmC. Genome-wide analysis revealed neural genes marked with 5hmC in S6KO ESCs, thereby implicating TET enzymes in the neuroectoderm-skewed differentiation phenotype. We demonstrate that SIRT6 functions as a chromatin regulator safeguarding the balance between pluripotency and differentiation through Tet-mediated production of 5hmC.

Epigenomic analysis of primary human T cells reveals enhancers associated with TH2 memory cell differentiation and asthma susceptibility.

A characteristic feature of asthma is the aberrant accumulation, differentiation or function of memory CD4(+) T cells that produce type 2 cytokines (TH2 cells). By mapping genome-wide histone modification profiles for subsets of T cells isolated from peripheral blood of healthy and asthmatic individuals, we identified enhancers with known and potential roles in the normal differentiation of human TH1 cells and TH2 cells. We discovered disease-specific enhancers in T cells that differ between healthy and asthmatic individuals. Enhancers that gained the histone H3 Lys4 dimethyl (H3K4me2) mark during TH2 cell development showed the highest enrichment for asthma-associated single nucleotide polymorphisms (SNPs), which supported a pathogenic role for TH2 cells in asthma. In silico analysis of cell-specific enhancers revealed transcription factors, microRNAs and genes potentially linked to human TH2 cell differentiation. Our results establish the feasibility and utility of enhancer profiling in well-defined populations of specialized cell types involved in disease pathogenesis.

MEDIPS: genome-wide differential coverage analysis of sequencing data derived from DNA enrichment experiments.

MOTIVATION: DNA enrichment followed by sequencing is a versatile tool in molecular biology, with a wide variety of applications including genome-wide analysis of epigenetic marks and mechanisms. A common requirement of these diverse applications is a comparison of read coverage between experimental conditions. The amount of samples generated for such comparisons ranges from few replicates to hundreds of samples per condition for epigenome-wide association studies. Consequently, there is an urgent need for software that allows for fast and simple processing and comparison of sequencing data derived from enriched DNA. RESULTS: Here, we present a major update of the R/Bioconductor package MEDIPS, which allows for an arbitrary number of replicates per group and integrates sophisticated statistical methods for the detection of differential coverage between experimental conditions. Our approach can be applied to a diversity of quantitative sequencing data. In addition, our update adds novel functionality to MEDIPS, including correlation analysis between samples, and takes advantage of Bioconductor's annotation databases to facilitate annotation of specific genomic regions. AVAILABILITY AND IMPLEMENTATION: The latest version of MEDIPS is available as version 1.12.0 and part of Bioconductor 2.13. The package comes with a manual containing detailed description of its functionality and is available at http://www.bioconductor.org.

DNA-methylome analysis of mouse intestinal adenoma identifies a tumour-specific signature that is partly conserved in human colon cancer.

Aberrant CpG methylation is a universal epigenetic trait of cancer cell genomes. However, human cancer samples or cell lines preclude the investigation of epigenetic changes occurring early during tumour development. Here, we have used MeDIP-seq to analyse the DNA methylome of APC(Min) adenoma as a model for intestinal cancer initiation, and we present a list of more than 13,000 recurring differentially methylated regions (DMRs) characterizing intestinal adenoma of the mouse. We show that Polycomb Repressive Complex (PRC) targets are strongly enriched among hypermethylated DMRs, and several PRC2 components and DNA methyltransferases were up-regulated in adenoma. We further demonstrate by bisulfite pyrosequencing of purified cell populations that the DMR signature arises de novo in adenoma cells rather than by expansion of a pre-existing pattern in intestinal stem cells or undifferentiated crypt cells. We found that epigenetic silencing of tumour suppressors, which occurs frequently in colon cancer, was rare in adenoma. Quite strikingly, we identified a core set of DMRs, which is conserved between mouse adenoma and human colon cancer, thus possibly revealing a global panel of epigenetically modified genes for intestinal tumours. Our data allow a distinction between early conserved epigenetic alterations occurring in intestinal adenoma and late stochastic events promoting colon cancer progression, and may facilitate the selection of more specific clinical epigenetic biomarkers.

RNA-Seq provides new insights in the transcriptome responses induced by the carcinogen benzo[a]pyrene.

Whole-genome transcriptome measurements are pivotal for characterizing molecular mechanisms of chemicals and predicting toxic classes, such as genotoxicity and carcinogenicity, from in vitro and in vivo assays. In recent years, deep sequencing technologies have been developed that hold the promise of measuring the transcriptome in a more complete and unbiased manner than DNA microarrays. Here, we applied this RNA-seq technology for the characterization of the transcriptomic responses in HepG2 cells upon exposure to benzo[a]pyrene (BaP), a well-known DNA damaging human carcinogen. Based on EnsEMBL genes, we demonstrate that RNA-seq detects ca 20% more genes than microarray-based technology but almost threefold more significantly differentially expressed genes. Functional enrichment analyses show that RNA-seq yields more insight into the biology and mechanisms related to the toxic effects caused by BaP, i.e., two- to fivefold more affected pathways and biological processes. Additionally, we demonstrate that RNA-seq allows detecting alternative isoform expression in many genes, including regulators of cell death and DNA repair such as TP53, BCL2 and XPA, which are relevant for genotoxic responses. Moreover, potentially novel isoforms were found, such as fragments of known transcripts, transcripts with additional exons, intron retention or exon-skipping events. The biological function(s) of these isoforms remain for the time being unknown. Finally, we demonstrate that RNA-seq enables the investigation of allele-specific gene expression, although no changes could be observed. Our results provide evidence that RNA-seq is a powerful tool for toxicology, which, compared with microarrays, is capable of generating novel and valuable information at the transcriptome level for characterizing deleterious effects caused by chemicals.