Self-Supervised Learning Reveals Clinically Relevant Histomorphological Patterns for Therapeutic Strategies in Colon Cancer.

Self-supervised learning (SSL) automates the extraction and interpretation of histopathology features on unannotated hematoxylin-and-eosin-stained whole-slide images (WSIs). We trained an SSL Barlow Twins-encoder on 435 TCGA colon adenocarcinoma WSIs to extract features from small image patches. Leiden community detection then grouped tiles into histomorphological phenotype clusters (HPCs). HPC reproducibility and predictive ability for overall survival was confirmed in an independent clinical trial cohort (N=1213 WSIs). This unbiased atlas resulted in 47 HPCs displaying unique and sharing clinically significant histomorphological traits, highlighting tissue type, quantity, and architecture, especially in the context of tumor stroma. Through in-depth analysis of these HPCs, including immune landscape and gene set enrichment analysis, and association to clinical outcomes, we shed light on the factors influencing survival and responses to treatments like standard adjuvant chemotherapy and experimental therapies. Further exploration of HPCs may unveil new insights and aid decision-making and personalized treatments for colon cancer patients.

Brain and cancer associated binding domain mutations provide insight into CTCF's relationship with chromatin and its ability to act as a chromatin organizer.

Although only a fraction of CTCF motifs are bound in any cell type, and few occupied sites overlap cohesin, the mechanisms underlying cell-type specific attachment and ability to function as a chromatin organizer remain unknown. To investigate the relationship between CTCF and chromatin we applied a combination of imaging, structural and molecular approaches, using a series of brain and cancer associated CTCF mutations that act as CTCF perturbations. We demonstrate that binding and the functional impact of WT and mutant CTCF depend not only on the unique binding properties of each protein, but also on the genomic context of bound sites and enrichment of motifs for expressed TFs abutting these sites. Our studies also highlight the reciprocal relationship between CTCF and chromatin, demonstrating that the unique binding properties of WT and mutant proteins have a distinct impact on accessibility, TF binding, cohesin overlap, chromatin interactivity and gene expression programs, providing insight into their cancer and brain related effects.

Inflammation in the tumor-adjacent lung as a predictor of clinical outcome in lung adenocarcinoma.

Approximately 30% of early-stage lung adenocarcinoma patients present with disease progression after successful surgical resection. Despite efforts of mapping the genetic landscape, there has been limited success in discovering predictive biomarkers of disease outcomes. Here we performed a systematic multi-omic assessment of 143 tumors and matched tumor-adjacent, histologically-normal lung tissue with long-term patient follow-up. Through histologic, mutational, and transcriptomic profiling of tumor and adjacent-normal tissue, we identified an inflammatory gene signature in tumor-adjacent tissue as the strongest clinical predictor of disease progression. Single-cell transcriptomic analysis demonstrated the progression-associated inflammatory signature was expressed in both immune and non-immune cells, and cell type-specific profiling in monocytes further improved outcome predictions. Additional analyses of tumor-adjacent transcriptomic data from The Cancer Genome Atlas validated the association of the inflammatory signature with worse outcomes across cancers. Collectively, our study suggests that molecular profiling of tumor-adjacent tissue can identify patients at high risk for disease progression.

MethNet: a robust approach to identify regulatory hubs and their distal targets in cancer.

Aberrations in the capacity of DNA/chromatin modifiers and transcription factors to bind non-coding regions can lead to changes in gene regulation and impact disease phenotypes. However, identifying distal regulatory elements and connecting them with their target genes remains challenging. Here, we present MethNet, a pipeline that integrates large-scale DNA methylation and gene expression data across multiple cancers, to uncover novel cis regulatory elements (CREs) in a 1Mb region around every promoter in the genome. MethNet identifies clusters of highly ranked CREs, referred to as 'hubs', which contribute to the regulation of multiple genes and significantly affect patient survival. Promoter-capture Hi-C confirmed that highly ranked associations involve physical interactions between CREs and their gene targets, and CRISPRi based scRNA Perturb-seq validated the functional impact of CREs. Thus, MethNet-identified CREs represent a valuable resource for unraveling complex mechanisms underlying gene expression, and for prioritizing the verification of predicted non-coding disease hotspots.

Cell-type-specific prediction of 3D chromatin organization enables high-throughput in silico genetic screening.

Investigating how chromatin organization determines cell-type-specific gene expression remains challenging. Experimental methods for measuring three-dimensional chromatin organization, such as Hi-C, are costly and have technical limitations, restricting their broad application particularly in high-throughput genetic perturbations. We present C.Origami, a multimodal deep neural network that performs de novo prediction of cell-type-specific chromatin organization using DNA sequence and two cell-type-specific genomic features-CTCF binding and chromatin accessibility. C.Origami enables in silico experiments to examine the impact of genetic changes on chromatin interactions. We further developed an in silico genetic screening approach to assess how individual DNA elements may contribute to chromatin organization and to identify putative cell-type-specific trans-acting regulators that collectively determine chromatin architecture. Applying this approach to leukemia cells and normal T cells, we demonstrate that cell-type-specific in silico genetic screening, enabled by C.Origami, can be used to systematically discover novel chromatin regulation circuits in both normal and disease-related biological systems.

Lysophosphatidic Acid Is a Proinflammatory Stimulus of Renal Tubular Epithelial Cells.

Chronic kidney disease (CKD) refers to a spectrum of diseases defined by renal fibrosis, permanent alterations in kidney structure, and low glomerular-filtration rate. Prolonged epithelial-tubular damage involves a series of changes that eventually lead to CKD, highlighting the importance of tubular epithelial cells in this process. Lysophosphatidic acid (LPA) is a bioactive lipid that signals mainly through its six cognate LPA receptors and is implicated in several chronic inflammatory pathological conditions. In this report, we have stimulated human proximal tubular epithelial cells (HKC-8) with LPA and 175 other possibly pathological stimuli, and simultaneously detected the levels of 27 intracellular phosphoproteins and 32 extracellular secreted molecules with multiplex ELISA. This quantification revealed a large amount of information concerning the signaling and the physiology of HKC-8 cells that can be extrapolated to other proximal tubular epithelial cells. LPA responses clustered with pro-inflammatory stimuli such as TNF and IL-1, promoting the phosphorylation of important inflammatory signaling hubs, including CREB1, ERK1, JUN, IκΒα, and MEK1, as well as the secretion of inflammatory factors of clinical relevance, including CCL2, CCL3, CXCL10, ICAM1, IL-6, and IL-8, most of them shown for the first time in proximal tubular epithelial cells. The identified LPA-induced signal-transduction pathways, which were pharmacologically validated, and the secretion of the inflammatory factors offer novel insights into the possible role of LPA in CKD pathogenesis.

Deep Learning and Pathomics Analyses Reveal Cell Nuclei as Important Features for Mutation Prediction of BRAF-Mutated Melanomas.

Image-based analysis as a method for mutation detection can be advantageous in settings when tumor tissue is limited or unavailable for direct testing. In this study, we utilize two distinct and complementary machine-learning methods of analyzing whole-slide images for predicting mutated BRAF. In the first method, whole-slide images of melanomas from 256 patients were used to train a deep convolutional neural network to develop a fully automated model that first selects for tumor-rich areas (area under the curve = 0.96) and then predicts for mutated BRAF (area under the curve = 0.71). Saliency mapping was performed and revealed that pixels corresponding to nuclei were the most relevant to network learning. In the second method, whole-slide images were analyzed using a pathomics pipeline that first annotates nuclei and then quantifies nuclear features, showing that mutated BRAF nuclei were significantly larger and rounder than BRAF‒wild-type nuclei. Finally, we developed a model that combines clinical information, deep learning, and pathomics that improves the predictive performance for mutated BRAF to an area under the curve of 0.89. Not only does this provide additional insights on how BRAF mutations affect tumor structural characteristics, but machine learning‒based analysis of whole-slide images also has the potential to be integrated into higher-order models for understanding tumor biology.

Prediction of combination therapies based on topological modeling of the immune signaling network in multiple sclerosis.

BACKGROUND: Multiple sclerosis (MS) is a major health problem, leading to a significant disability and patient suffering. Although chronic activation of the immune system is a hallmark of the disease, its pathogenesis is poorly understood, while current treatments only ameliorate the disease and may produce severe side effects. METHODS: Here, we applied a network-based modeling approach based on phosphoproteomic data to uncover the differential activation in signaling wiring between healthy donors, untreated patients, and those under different treatments. Based in the patient-specific networks, we aimed to create a new approach to identify drug combinations that revert signaling to a healthy-like state. We performed ex vivo multiplexed phosphoproteomic assays upon perturbations with multiple drugs and ligands in primary immune cells from 169 subjects (MS patients, n=129 and matched healthy controls, n=40). Patients were either untreated or treated with fingolimod, natalizumab, interferon-ß, glatiramer acetate, or the experimental therapy epigallocatechin gallate (EGCG). We generated for each donor a dynamic logic model by fitting a bespoke literature-derived network of MS-related pathways to the perturbation data. Last, we developed an approach based on network topology to identify deregulated interactions whose activity could be reverted to a "healthy-like" status by combination therapy. The experimental autoimmune encephalomyelitis (EAE) mouse model of MS was used to validate the prediction of combination therapies. RESULTS: Analysis of the models uncovered features of healthy-, disease-, and drug-specific signaling networks. We predicted several combinations with approved MS drugs that could revert signaling to a healthy-like state. Specifically, TGF-ß activated kinase 1 (TAK1) kinase, involved in Transforming growth factor ß-1 proprotein (TGF-ß), Toll-like receptor, B cell receptor, and response to inflammation pathways, was found to be highly deregulated and co-druggable with all MS drugs studied. One of these predicted combinations, fingolimod with a TAK1 inhibitor, was validated in an animal model of MS. CONCLUSIONS: Our approach based on donor-specific signaling networks enables prediction of targets for combination therapy for MS and other complex diseases.

H3K27ac bookmarking promotes rapid post-mitotic activation of the pluripotent stem cell program without impacting 3D chromatin reorganization.

During self-renewal, cell-type-defining features are drastically perturbed in mitosis and must be faithfully reestablished upon G1 entry, a process that remains largely elusive. Here, we characterized at a genome-wide scale the dynamic transcriptional and architectural resetting of mouse pluripotent stem cells (PSCs) upon mitotic exit. We captured distinct waves of transcriptional reactivation with rapid induction of stem cell genes and transient activation of lineage-specific genes. Topological reorganization at different hierarchical levels also occurred in an asynchronous manner and showed partial coordination with transcriptional resetting. Globally, rapid transcriptional and architectural resetting associated with mitotic retention of H3K27 acetylation, supporting a bookmarking function. Indeed, mitotic depletion of H3K27ac impaired the early reactivation of bookmarked, stem-cell-associated genes. However, 3D chromatin reorganization remained largely unaffected, suggesting that these processes are driven by distinct forces upon mitotic exit. This study uncovers principles and mediators of PSC molecular resetting during self-renewal.

A Deep Learning Framework for Predicting Response to Therapy in Cancer.

A major challenge in cancer treatment is predicting clinical response to anti-cancer drugs on a personalized basis. Using a pharmacogenomics database of 1,001 cancer cell lines, we trained deep neural networks for prediction of drug response and assessed their performance on multiple clinical cohorts. We demonstrate that deep neural networks outperform the current state in machine learning frameworks. We provide a proof of concept for the use of deep neural network-based frameworks to aid precision oncology strategies.

EpiMethylTag: simultaneous detection of ATAC-seq or ChIP-seq signals with DNA methylation.

Activation of regulatory elements is thought to be inversely correlated with DNA methylation levels. However, it is difficult to determine whether DNA methylation is compatible with chromatin accessibility or transcription factor (TF) binding if assays are performed separately. We developed a fast, low-input, low sequencing depth method, EpiMethylTag, that combines ATAC-seq or ChIP-seq (M-ATAC or M-ChIP) with bisulfite conversion, to simultaneously examine accessibility/TF binding and methylation on the same DNA. Here we demonstrate that EpiMethylTag can be used to study the functional interplay between chromatin accessibility and TF binding (CTCF and KLF4) at methylated sites.

NSD2 overexpression drives clustered chromatin and transcriptional changes in a subset of insulated domains.

CTCF and cohesin play a key role in organizing chromatin into topologically associating domain (TAD) structures. Disruption of a single CTCF binding site is sufficient to change chromosomal interactions leading to alterations in chromatin modifications and gene regulation. However, the extent to which alterations in chromatin modifications can disrupt 3D chromosome organization leading to transcriptional changes is unknown. In multiple myeloma, a 4;14 translocation induces overexpression of the histone methyltransferase, NSD2, resulting in expansion of H3K36me2 and shrinkage of antagonistic H3K27me3 domains. Using isogenic cell lines producing high and low levels of NSD2, here we find oncogene activation is linked to alterations in H3K27ac and CTCF within H3K36me2 enriched chromatin. A logistic regression model reveals that differentially expressed genes are significantly enriched within the same insulated domain as altered H3K27ac and CTCF peaks. These results identify a bidirectional relationship between 2D chromatin and 3D genome organization in gene regulation.

Non-lethal proteasome inhibition activates pro-tumorigenic pathways in multiple myeloma cells.

Multiple myeloma (MM) is a haematological malignancy being characterized by clonal plasma cell proliferation in the bone marrow. Targeting the proteasome with specific inhibitors (PIs) has been proven a promising therapeutic strategy and PIs have been approved for the treatment of MM and mantle-cell lymphoma; yet, while outcome has improved, most patients inevitably relapse. As relapse refers to MM cells that survive therapy, we sought to identify the molecular responses induced in MM cells after non-lethal proteasome inhibition. By using bortezomib (BTZ), epoxomicin (EPOX; a carfilzomib-like PI) and three PIs, namely Rub999, PR671A and Rub1024 that target each of the three proteasome peptidases, we found that only BTZ and EPOX are toxic in MM cells at low concentrations. Phosphoproteomic profiling after treatment of MM cells with non-lethal (IC10 ) doses of the PIs revealed inhibitor- and cell type-specific readouts, being marked by the activation of tumorigenic STAT3 and STAT6. Consistently, cytokine/chemokine profiling revealed the increased secretion of immunosuppressive pro-tumorigenic cytokines (IL6 and IL8), along with the inhibition of potent T cell chemoattractant chemokines (CXCL10). These findings indicate that MM cells that survive treatment with therapeutic PIs shape a pro-tumorigenic immunosuppressive cellular and secretory bone marrow microenvironment that enables malignancy to relapse.

Machine learning and data mining frameworks for predicting drug response in cancer: An overview and a novel in silico screening process based on association rule mining.

A major challenge in cancer treatment is predicting the clinical response to anti-cancer drugs on a personalized basis. The success of such a task largely depends on the ability to develop computational resources that integrate big "omic" data into effective drug-response models. Machine learning is both an expanding and an evolving computational field that holds promise to cover such needs. Here we provide a focused overview of: 1) the various supervised and unsupervised algorithms used specifically in drug response prediction applications, 2) the strategies employed to develop these algorithms into applicable models, 3) data resources that are fed into these frameworks and 4) pitfalls and challenges to maximize model performance. In this context we also describe a novel in silico screening process, based on Association Rule Mining, for identifying genes as candidate drivers of drug response and compare it with relevant data mining frameworks, for which we generated a web application freely available at: https://compbio.nyumc.org/drugs/. This pipeline explores with high efficiency large sample-spaces, while is able to detect low frequency events and evaluate statistical significance even in the multidimensional space, presenting the results in the form of easily interpretable rules. We conclude with future prospects and challenges of applying machine learning based drug response prediction in precision medicine.

Targeting Mitochondrial Structure Sensitizes Acute Myeloid Leukemia to Venetoclax Treatment.

The BCL2 family plays important roles in acute myeloid leukemia (AML). Venetoclax, a selective BCL2 inhibitor, has received FDA approval for the treatment of AML. However, drug resistance ensues after prolonged treatment, highlighting the need for a greater understanding of the underlying mechanisms. Using a genome-wide CRISPR/Cas9 screen in human AML, we identified genes whose inactivation sensitizes AML blasts to venetoclax. Genes involved in mitochondrial organization and function were significantly depleted throughout our screen, including the mitochondrial chaperonin CLPB. We demonstrated that CLPB is upregulated in human AML, it is further induced upon acquisition of venetoclax resistance, and its ablation sensitizes AML to venetoclax. Mechanistically, CLPB maintains the mitochondrial cristae structure via its interaction with the cristae-shaping protein OPA1, whereas its loss promotes apoptosis by inducing cristae remodeling and mitochondrial stress responses. Overall, our data suggest that targeting mitochondrial architecture may provide a promising approach to circumvent venetoclax resistance. SIGNIFICANCE: A genome-wide CRISPR/Cas9 screen reveals genes involved in mitochondrial biological processes participate in the acquisition of venetoclax resistance. Loss of the mitochondrial protein CLPB leads to structural and functional defects of mitochondria, hence sensitizing AML cells to apoptosis. Targeting CLPB synergizes with venetoclax and the venetoclax/azacitidine combination in AML in a p53-independent manner.See related commentary by Savona and Rathmell, p. 831.This article is highlighted in the In This Issue feature, p. 813.

MAPK pathway and B cells overactivation in multiple sclerosis revealed by phosphoproteomics and genomic analysis.

Dysregulation of signaling pathways in multiple sclerosis (MS) can be analyzed by phosphoproteomics in peripheral blood mononuclear cells (PBMCs). We performed in vitro kinetic assays on PBMCs in 195 MS patients and 60 matched controls and quantified the phosphorylation of 17 kinases using xMAP assays. Phosphoprotein levels were tested for association with genetic susceptibility by typing 112 single-nucleotide polymorphisms (SNPs) associated with MS susceptibility. We found increased phosphorylation of MP2K1 in MS patients relative to the controls. Moreover, we identified one SNP located in the PHDGH gene and another on IRF8 gene that were associated with MP2K1 phosphorylation levels, providing a first clue on how this MS risk gene may act. The analyses in patients treated with disease-modifying drugs identified the phosphorylation of each receptor's downstream kinases. Finally, using flow cytometry, we detected in MS patients increased STAT1, STAT3, TF65, and HSPB1 phosphorylation in CD19+ cells. These findings indicate the activation of cell survival and proliferation (MAPK), and proinflammatory (STAT) pathways in the immune cells of MS patients, primarily in B cells. The changes in the activation of these kinases suggest that these pathways may represent therapeutic targets for modulation by kinase inhibitors.

A Computational Platform and Guide for Acceleration of Novel Medicines and Personalized Medicine.

In the era of big data and informatics, computational integration of data across the hierarchical structures of human biology enables discovery of new druggable targets of disease and new mode of action of a drug. We present herein a computational framework and guide of integrating drug targets, gene expression data, transcription factors, and prior knowledge of protein interactions to computationally construct the signaling network (mode of action) of a drug. In a similar manner, a disease network is constructed using its disease targets. And then, drug candidates are computationally prioritized by computationally ranking the closeness between a disease network and a drug's signaling network. Furthermore, we describe the use of the most perturbed HLA genes to assess the safety risk for immune-mediated adverse reactions such as Stevens-Johnson syndrome/toxic epidermal necrolysis.

Phosphoprotein patterns predict trametinib responsiveness and optimal trametinib sensitisation strategies in melanoma.

Malignant melanoma is a highly aggressive form of skin cancer responsible for the majority of skin cancer-related deaths. Recent insight into the heterogeneous nature of melanoma suggests more personalised treatments may be necessary to overcome drug resistance and improve patient care. To this end, reliable molecular signatures that can accurately predict treatment responsiveness need to be identified. In this study, we applied multiplex phosphoproteomic profiling across a panel of 24 melanoma cell lines with different disease-relevant mutations, to predict responsiveness to MEK inhibitor trametinib. Supported by multivariate statistical analysis and multidimensional pattern recognition algorithms, the responsiveness of individual cell lines to trametinib could be predicted with high accuracy (83% correct predictions), independent of mutation status. We also successfully employed this approach to case specifically predict whether individual melanoma cell lines could be sensitised to trametinib. Our predictions identified that combining MEK inhibition with selective targeting of c-JUN and/or FAK, using siRNA-based depletion or pharmacological inhibitors, sensitised resistant cell lines and significantly enhanced treatment efficacy. Our study indicates that multiplex proteomic analyses coupled with pattern recognition approaches could assist in personalising trametinib-based treatment decisions in the future.

Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning.

Visual inspection of histopathology slides is one of the main methods used by pathologists to assess the stage, type and subtype of lung tumors. Adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) are the most prevalent subtypes of lung cancer, and their distinction requires visual inspection by an experienced pathologist. In this study, we trained a deep convolutional neural network (inception v3) on whole-slide images obtained from The Cancer Genome Atlas to accurately and automatically classify them into LUAD, LUSC or normal lung tissue. The performance of our method is comparable to that of pathologists, with an average area under the curve (AUC) of 0.97. Our model was validated on independent datasets of frozen tissues, formalin-fixed paraffin-embedded tissues and biopsies. Furthermore, we trained the network to predict the ten most commonly mutated genes in LUAD. We found that six of them-STK11, EGFR, FAT1, SETBP1, KRAS and TP53-can be predicted from pathology images, with AUCs from 0.733 to 0.856 as measured on a held-out population. These findings suggest that deep-learning models can assist pathologists in the detection of cancer subtype or gene mutations. Our approach can be applied to any cancer type, and the code is available at https://github.com/ncoudray/DeepPATH .

Selective cytotoxicity of the herbal substance acteoside against tumor cells and its mechanistic insights.

Natural products are characterized by extreme structural diversity and thus they offer a unique source for the identification of novel anti-tumor agents. Herein, we report that the herbal substance acteoside being isolated by advanced phytochemical methods from Lippia citriodora leaves showed enhanced cytotoxicity against metastatic tumor cells; acted in synergy with various cytotoxic agents and it sensitized chemoresistant cancer cells. Acteoside was not toxic in physiological cellular contexts, while it increased oxidative load, affected the activity of proteostatic modules and suppressed matrix metalloproteinases in tumor cell lines. Intraperitoneal or oral (via drinking water) administration of acteoside in a melanoma mouse model upregulated antioxidant responses in the tumors; yet, only intraperitoneal delivery suppressed tumor growth and induced anti-tumor-reactive immune responses. Mass-spectrometry identification/quantitation analyses revealed that intraperitoneal delivery of acteoside resulted in significantly higher, vs. oral administration, concentration of the compound in the plasma and tumors of treated mice, suggesting that its in vivo anti-tumor effect depends on the route of administration and the achieved concentration in the tumor. Finally, molecular modeling studies and enzymatic activity assays showed that acteoside inhibits protein kinase C. Conclusively, acteoside holds promise as a chemical scaffold for the development of novel anti-tumor agents.