Uncertainty Quantification and Interpretability for Clinical Trial Approval Prediction.

Background: Clinical trial is a crucial step in the development of a new therapy (e.g., medication) and is remarkably expensive and time-consuming. Forecasting the approval of clinical trials accurately would enable us to circumvent trials destined to fail, thereby allowing us to allocate more resources to therapies with better chances. However, existing approval prediction algorithms did not quantify the uncertainty and provide interpretability, limiting their usage in real-world clinical trial management. Methods: This paper quantifies uncertainty and improves interpretability in clinical trial approval predictions. We devised a selective classification approach and integrated it with the Hierarchical Interaction Network, the state-of-the-art clinical trial prediction model. Selective classification, encompassing a spectrum of methods for uncertainty quantification, empowers the model to withhold decision-making in the face of samples marked by ambiguity or low confidence. This approach not only amplifies the accuracy of predictions for the instances it chooses to classify but also notably enhances the model's interpretability. Results: Comprehensive experiments demonstrate that incorporating uncertainty markedly enhances the model's performance. Specifically, the proposed method achieved 32.37%, 21.43%, and 13.27% relative improvement on area under the precision-recall curve over the base model (Hierarchical Interaction Network) in phase I, II, and III trial approval predictions, respectively. For phase III trials, our method reaches 0.9022 area under the precision-recall curve scores. In addition, we show a case study of interpretability that helps domain experts to understand model's outcome. The code is publicly available at https://github.com/Vincent-1125/Uncertainty-Quantification-on-Clinical-Trial-Outcome-Prediction. Conclusion: Our approach not only measures model uncertainty but also greatly improves interpretability and performance for clinical trial approval prediction.

p53 governs an AT1 differentiation programme in lung cancer suppression.

Lung cancer is the leading cause of cancer deaths worldwide1. Mutations in the tumour suppressor gene TP53 occur in 50% of lung adenocarcinomas (LUADs) and are linked to poor prognosis1-4, but how p53 suppresses LUAD development remains enigmatic. We show here that p53 suppresses LUAD by governing cell state, specifically by promoting alveolar type 1 (AT1) differentiation. Using mice that express oncogenic Kras and null, wild-type or hypermorphic Trp53 alleles in alveolar type 2 (AT2) cells, we observed graded effects of p53 on LUAD initiation and progression. RNA sequencing and ATAC sequencing of LUAD cells uncovered a p53-induced AT1 differentiation programme during tumour suppression in vivo through direct DNA binding, chromatin remodelling and induction of genes characteristic of AT1 cells. Single-cell transcriptomics analyses revealed that during LUAD evolution, p53 promotes AT1 differentiation through action in a transitional cell state analogous to a transient intermediary seen during AT2-to-AT1 cell differentiation in alveolar injury repair. Notably, p53 inactivation results in the inappropriate persistence of these transitional cancer cells accompanied by upregulated growth signalling and divergence from lung lineage identity, characteristics associated with LUAD progression. Analysis of Trp53 wild-type and Trp53-null mice showed that p53 also directs alveolar regeneration after injury by regulating AT2 cell self-renewal and promoting transitional cell differentiation into AT1 cells. Collectively, these findings illuminate mechanisms of p53-mediated LUAD suppression, in which p53 governs alveolar differentiation, and suggest that tumour suppression reflects a fundamental role of p53 in orchestrating tissue repair after injury.

Computational workflow for integrative analyses of DNA replication timing, epigenomic, and transcriptomic data.

Temporal profiling of DNA replication timing (RT) in combination with chromatin modifications, chromatin accessibility, and gene expression provides new insights into the causal relationships between chromatin and RT during cell cycle. Here, we describe a protocol for in-depth integrative computational analyses of Repli-seq, ATAC-seq, RNA-seq, and ChIP-seq or CUT&RUN data for multiple marks at various time points across cell cycle and changes in their interrelationships upon an experimental perturbation (e.g., knockdown or overexpression of a regulatory protein). For complete details on the use and execution of this protocol, please refer to Van Rechem et al. (2021).

Phosphoproteomic mapping reveals distinct signaling actions and activation of muscle protein synthesis by Isthmin-1.

The secreted protein isthmin-1 (Ism1) mitigates diabetes by increasing adipocyte and skeletal muscle glucose uptake by activating the PI3K-Akt pathway. However, while both Ism1 and insulin converge on these common targets, Ism1 has distinct cellular actions suggesting divergence in downstream intracellular signaling pathways. To understand the biological complexity of Ism1 signaling, we performed phosphoproteomic analysis after acute exposure, revealing overlapping and distinct pathways of Ism1 and insulin. We identify a 53% overlap between Ism1 and insulin signaling and Ism1-mediated phosphoproteome-wide alterations in ~450 proteins that are not shared with insulin. Interestingly, we find several unknown phosphorylation sites on proteins related to protein translation, mTOR pathway, and, unexpectedly, muscle function in the Ism1 signaling network. Physiologically, Ism1 ablation in mice results in altered proteostasis, including lower muscle protein levels under fed and fasted conditions, reduced amino acid incorporation into proteins, and reduced phosphorylation of the key protein synthesis effectors Akt and downstream mTORC1 targets. As metabolic disorders such as diabetes are associated with accelerated loss of skeletal muscle protein content, these studies define a non-canonical mechanism by which this antidiabetic circulating protein controls muscle biology.

Cells need energy to survive and carry out their role in the body. They do this by breaking down molecules, like sugar, into substances that can fuel the creation of new compounds, like proteins or lipids. This process, known as metabolism, involves a series of interconnecting chemical reactions which are organized into pathways. Metabolic pathways contain proteins that catalyze each sequential reaction. Hormones can change the activity of these proteins by adding a chemical group called a phosphate. This reversible modification can majorly impact the metabolism of cells, resulting in changes to the body's tissues. The hormone insulin, for instance, alters a well-known metabolic pathway that triggers skeletal muscle cells to produce more proteins, leading to stronger and larger muscles. In 2021, a group of scientists discovered a molecule made by fat cells, called Isthmin-1, also activates components in this pathway. Similar to insulin, Isthmin-1 encourages muscle and fat cells to take up sugar. However, it also prevents the liver from accumulating excess fat, suggesting Isthmin-1 may trigger a different cascade of molecules to insulin. To investigate this possibility, Zhao et al. ­ including some of the researchers involved in the 2021 study ­ exposed cells grown in the laboratory to Isthmin-1 or insulin and looked for phosphates on all their proteins. This revealed that only 53% of the proteins Isthmin-1 modifies are also altered by insulin. Of the proteins unique to Isthmin-1, several had known roles in making and maintaining proteins in muscle cells. To understand more about the role of this newly discovered pathway, Zhao et al. genetically engineered mice to lack the gene that codes for Isthmin-1. This decreased the size and strength of the mice's muscle fibers and reduced the signals that normally lead to skeletal muscle growth. These findings suggest that Isthmin-1 regulates skeletal muscle size via a metabolic pathway that is slightly different to the one activated by insulin. Many metabolic disorders are associated with muscle loss, like diabetes, and this newly discovered network of proteins could further our understanding of how to prevent and treat these diseases.

The Interaction of SWI/SNF with the Ribosome Regulates Translation and Confers Sensitivity to Translation Pathway Inhibitors in Cancers with Complex Perturbations.

Subunits from the chromatin remodelers mammalian SWItch/Sucrose Non-Fermentable (mSWI/SNF) are mutated, deleted, or amplified in more than 40% of cancers. Understanding their functions in normal cells and the consequences of cancerous alterations will provide insight into developing new targeted therapies. Here we examined whether mSWI/SNF mutations increase cellular sensitivity to specific drugs. Taking advantage of the DepMap studies, we demonstrate that cancer cells harboring mutations of specific mSWI/SNF subunits exhibit a genetic dependency on translation factors and are sensitive to translation pathway inhibitors. Furthermore, mSWI/SNF subunits were present in the cytoplasm and interacted with the translation initiation machinery, and short-term inhibition and depletion of specific subunits decreased global translation, implicating a direct role for these factors in translation. Depletion of specific mSWI/SNF subunits also increased sensitivity to mTOR-PI3K inhibitors. In patient-derived breast cancer samples, mSWI/SNF subunits expression in both the nucleus and the cytoplasm was substantially altered. In conclusion, an unexpected cytoplasmic role for mSWI/SNF complexes in translation suggests potential new therapeutic opportunities for patients afflicted by cancers demonstrating alterations in their subunits. SIGNIFICANCE: This work establishes direct functions for mSWI/SNF in translation and demonstrates that alterations in mSWI/SNF confer a therapeutic vulnerability to translation pathway inhibitors in cancer cells.

DNA replication fork speed underlies cell fate changes and promotes reprogramming.

Totipotency emerges in early embryogenesis, but its molecular underpinnings remain poorly characterized. In the present study, we employed DNA fiber analysis to investigate how pluripotent stem cells are reprogrammed into totipotent-like 2-cell-like cells (2CLCs). We show that totipotent cells of the early mouse embryo have slow DNA replication fork speed and that 2CLCs recapitulate this feature, suggesting that fork speed underlies the transition to a totipotent-like state. 2CLCs emerge concomitant with DNA replication and display changes in replication timing (RT), particularly during the early S-phase. RT changes occur prior to 2CLC emergence, suggesting that RT may predispose to gene expression changes and consequent reprogramming of cell fate. Slowing down replication fork speed experimentally induces 2CLCs. In vivo, slowing fork speed improves the reprogramming efficiency of somatic cell nuclear transfer. Our data suggest that fork speed regulates cellular plasticity and that remodeling of replication features leads to changes in cell fate and reprogramming.

Protocol to isolate cells in four stages of S phase for high-resolution replication-timing sequencing.

Traditional replication timing (RT) experiments divide S phase into two phases: early and late. However, there is an increasing awareness that variation in RT can occur during the course of S phase and impact our understanding of RT patterns and regulation. Here, we describe a RT protocol in RPE-1 cells for collecting four phases within S and the library preparation that takes advantage of a commercial kit for methyl-DNA. This step allows BrdU-labeled DNA sequencing and assessment of RT genome wide. For complete details on the use and execution of this protocol, please refer to Van Rechem et al. (2021).

A cell-sorting-based protocol for cell cycle small-scale ChIP sequencing.

Classic approaches to characterizing cell cycle leverage chemicals or altered nucleotide pools, which could impact chromatin states at specific phases of the cell cycle. Such approaches could induce metabolic alterations and/or DNA damage, which could reshape protein recruitment and histone modifications. In this protocol, we describe ways to fix and sort cells across the cell cycle based on their DNA content. We further detail immunoprecipitation and library preparation, allowing analysis of the epigenome by chromatin immunoprecipitation sequencing (ChIP-seq) for small numbers of cells. For complete details on the use and execution of this protocol, please refer to Van Rechem et al. (2021).

Collective regulation of chromatin modifications predicts replication timing during cell cycle.

Replication timing (RT) associates with genome architecture, while having a mixed relationship to histone marks. By profiling replication at high resolution and assessing broad histone marks across the cell cycle at the resolution of RT with and without genetic perturbation, we address the causal relationship between histone marks and RT. Four primary chromatin states, including an uncharacterized H3K36me2 state, emerge and define 97% of the mappable genome. RT and local replication patterns (e.g., initiation zones) quantitatively associate with chromatin states, histone mark dynamics, and spatial chromatin structure. Manipulation of broad histone marks and enhancer elements by overexpressing the histone H3 lysine 9/36 tri-demethylase KDM4A impacts RT across 11% of the genome. Broad histone modification changes were strong predictors of the observed RT alterations. Lastly, replication within H3K36me2-enriched neighborhoods is sensitive to KDM4A overexpression and is controlled at a megabase scale. These studies establish a role for collective chromatin mark regulation in modulating RT.

Integrated multi-omics analysis of RB-loss identifies widespread cellular programming and synthetic weaknesses.

Inactivation of RB is one of the hallmarks of cancer, however gaps remain in our understanding of how RB-loss changes human cells. Here we show that pRB-depletion results in cellular reprogramming, we quantitatively measured how RB-depletion altered the transcriptional, proteomic and metabolic output of non-tumorigenic RPE1 human cells. These profiles identified widespread changes in metabolic and cell stress response factors previously linked to E2F function. In addition, we find a number of additional pathways that are sensitive to RB-depletion that are not E2F-regulated that may represent compensatory mechanisms to support the growth of RB-depleted cells. To determine whether these molecular changes are also present in RB1-/- tumors, we compared these results to Retinoblastoma and Small Cell Lung Cancer data, and identified widespread conservation of alterations found in RPE1 cells. To define which of these changes contribute to the growth of cells with de-regulated E2F activity, we assayed how inhibiting or depleting these proteins affected the growth of RB1-/- cells and of Drosophila E2f1-RNAi models in vivo. From this analysis, we identify key metabolic pathways that are essential for the growth of pRB-deleted human cells.

The lysine demethylase KDM4A controls the cell-cycle expression of replicative canonical histone genes.

Chromatin modulation provides a key checkpoint for controlling cell cycle regulated gene networks. The replicative canonical histone genes are one such gene family under tight regulation during cell division. These genes are most highly expressed during S phase when histones are needed to chromatinize the new DNA template. While this fact has been known for a while, limited knowledge exists about the specific chromatin regulators controlling their temporal expression during cell cycle. Since histones and their associated mutations are emerging as major players in diseases such as cancer, identifying the chromatin factors modulating their expression is critical. The histone lysine tri-demethylase KDM4A is regulated over cell cycle and plays a direct role in DNA replication timing, site-specific rereplication, and DNA amplifications during S phase. Here, we establish an unappreciated role for the catalytically active KDM4A in directly regulating canonical replicative histone gene networks during cell cycle. Of interest, we further demonstrate that KDM4A interacts with proteins controlling histone expression and RNA processing (i.e., hnRNPUL1 and FUS/TLS). Together, this study provides a new function for KDM4A in modulating canonical histone gene expression.

Histone Lysine Methylation Dynamics Control EGFR DNA Copy-Number Amplification.

Acquired chromosomal DNA copy gains are a feature of many tumors; however, the mechanisms that underpin oncogene amplification are poorly understood. Recent studies have begun to uncover the importance of epigenetic states and histone lysine methyltransferases (KMT) and demethylases (KDM) in regulating transient site-specific DNA copy-number gains (TSSG). In this study, we reveal a critical interplay between a myriad of lysine methyltransferases and demethylases in modulating H3K4/9/27 methylation balance to control extrachromosomal amplification of the EGFR oncogene. This study further establishes that cellular signals (hypoxia and EGF) are able to directly promote EGFR amplification through modulation of the enzymes controlling EGFR copy gains. Moreover, we demonstrate that chemical inhibitors targeting specific KMTs and KDMs are able to promote or block extrachromosomal EGFR amplification, which identifies potential therapeutic strategies for controlling EGFR copy-number heterogeneity in cancer, and, in turn, drug response. SIGNIFICANCE: This study identifies a network of epigenetic factors and cellular signals that directly control EGFR DNA amplification. We demonstrate that chemical inhibitors targeting enzymes controlling this amplification can be used to rheostat EGFR copy number, which uncovers therapeutic opportunities for controlling EGFR DNA amplification heterogeneity and the associated drug response.This article is highlighted in the In This Issue feature, p. 161.

The Histone Deacetylase SIRT6 Restrains Transcription Elongation via Promoter-Proximal Pausing.

Transcriptional regulation in eukaryotes occurs at promoter-proximal regions wherein transcriptionally engaged RNA polymerase II (Pol II) pauses before proceeding toward productive elongation. The role of chromatin in pausing remains poorly understood. Here, we demonstrate that the histone deacetylase SIRT6 binds to Pol II and prevents the release of the negative elongation factor (NELF), thus stabilizing Pol II promoter-proximal pausing. Genetic depletion of SIRT6 or its chromatin deficiency upon glucose deprivation causes intragenic enrichment of acetylated histone H3 at lysines 9 (H3K9ac) and 56 (H3K56ac), activation of cyclin-dependent kinase 9 (CDK9)-that phosphorylates NELF and the carboxyl terminal domain of Pol II-and enrichment of the positive transcription elongation factors MYC, BRD4, PAF1, and the super elongation factors AFF4 and ELL2. These events lead to increased expression of genes involved in metabolism, protein synthesis, and embryonic development. Our results identified SIRT6 as a Pol II promoter-proximal pausing-dedicated histone deacetylase.

METTL13 Methylation of eEF1A Increases Translational Output to Promote Tumorigenesis.

Increased protein synthesis plays an etiologic role in diverse cancers. Here, we demonstrate that METTL13 (methyltransferase-like 13) dimethylation of eEF1A (eukaryotic elongation factor 1A) lysine 55 (eEF1AK55me2) is utilized by Ras-driven cancers to increase translational output and promote tumorigenesis in vivo. METTL13-catalyzed eEF1A methylation increases eEF1A's intrinsic GTPase activity in vitro and protein production in cells. METTL13 and eEF1AK55me2 levels are upregulated in cancer and negatively correlate with pancreatic and lung cancer patient survival. METTL13 deletion and eEF1AK55me2 loss dramatically reduce Ras-driven neoplastic growth in mouse models and in patient-derived xenografts (PDXs) from primary pancreatic and lung tumors. Finally, METTL13 depletion renders PDX tumors hypersensitive to drugs that target growth-signaling pathways. Together, our work uncovers a mechanism by which lethal cancers become dependent on the METTL13-eEF1AK55me2 axis to meet their elevated protein synthesis requirement and suggests that METTL13 inhibition may constitute a targetable vulnerability of tumors driven by aberrant Ras signaling.

Cross-talk between Lysine-Modifying Enzymes Controls Site-Specific DNA Amplifications.

Acquired chromosomal DNA amplifications are features of many tumors. Although overexpression and stabilization of the histone H3 lysine 9/36 (H3K9/36) tri-demethylase KDM4A generates transient site-specific copy number gains (TSSGs), additional mechanisms directly controlling site-specific DNA copy gains are not well defined. In this study, we uncover a collection of H3K4-modifying chromatin regulators that function with H3K9 and H3K36 regulators to orchestrate TSSGs. Specifically, the H3K4 tri-demethylase KDM5A and specific COMPASS/KMT2 H3K4 methyltransferases modulate different TSSG loci through H3K4 methylation states and KDM4A recruitment. Furthermore, a distinct chromatin modifier network, MLL1-KDM4B-KDM5B, controls copy number regulation at a specific genomic locus in a KDM4A-independent manner. These pathways comprise an epigenetic addressing system for defining site-specific DNA rereplication and amplifications.

E2F/DP Prevents Cell-Cycle Progression in Endocycling Fat Body Cells by Suppressing dATM Expression.

To understand the consequences of the complete elimination of E2F regulation, we profiled the proteome of Drosophila dDP mutants that lack functional E2F/DP complexes. The results uncovered changes in the larval fat body, a differentiated tissue that grows via endocycles. We report an unexpected mechanism of E2F/DP action that promotes quiescence in this tissue. In the fat body, dE2F/dDP limits cell-cycle progression by suppressing DNA damage responses. Loss of dDP upregulates dATM, allowing cells to sense and repair DNA damage and increasing replication of loci that are normally under-replicated in wild-type tissues. Genetic experiments show that ectopic dATM is sufficient to promote DNA synthesis in wild-type fat body cells. Strikingly, reducing dATM levels in dDP-deficient fat bodies restores cell-cycle control, improves tissue morphology, and extends animal development. These results show that, in some cellular contexts, dE2F/dDP-dependent suppression of DNA damage signaling is key for cell-cycle control and needed for normal development.

Hypoxia drives transient site-specific copy gain and drug-resistant gene expression.

Copy number heterogeneity is a prominent feature within tumors. The molecular basis for this heterogeneity remains poorly characterized. Here, we demonstrate that hypoxia induces transient site-specific copy gains (TSSGs) in primary, nontransformed, and transformed human cells. Hypoxia-driven copy gains are not dependent on HIF1α or HIF2α; however, they are dependent on the KDM4A histone demethylase and are blocked by inhibition of KDM4A with a small molecule or the natural metabolite succinate. Furthermore, this response is conserved at a syntenic region in zebrafish cells. Regions with site-specific copy gain are also enriched for amplifications in hypoxic primary tumors. These tumors exhibited amplification and overexpression of the drug resistance gene CKS1B, which we recapitulated in hypoxic breast cancer cells. Our results demonstrate that hypoxia provides a biological stimulus to create transient site-specific copy alterations that could result in heterogeneity within tumors and cell populations. These findings have major implications in our understanding of copy number heterogeneity and the emergence of drug resistance genes in cancer.

Lysine demethylase KDM4A associates with translation machinery and regulates protein synthesis.

UNLABELLED: Chromatin-modifying enzymes are predominantly nuclear; however, these factors are also localized to the cytoplasm, and very little is known about their role in this compartment. In this report, we reveal a non-chromatin-linked role for the lysine-specific demethylase KDM4A. We demonstrate that KDM4A interacts with the translation initiation complex and affects the distribution of translation initiation factors within polysome fractions. Furthermore, KDM4A depletion reduced protein synthesis and enhanced the protein synthesis suppression observed with mTOR inhibitors, which paralleled an increased sensitivity to these drugs. Finally, we demonstrate that JIB-04, a JmjC demethylase inhibitor, suppresses translation initiation and enhances mTOR inhibitor sensitivity. These data highlight an unexpected cytoplasmic role for KDM4A in regulating protein synthesis and suggest novel potential therapeutic applications for this class of enzyme. SIGNIFICANCE: This report documents an unexpected cytoplasmic role for the lysine demethylase KDM4A. We demonstrate that KDM4A interacts with the translation initiation machinery, regulates protein synthesis and, upon coinhibition with mTOR inhibitors, enhances the translation suppression and cell sensitivity to these therapeutics.