Chromatin position effects assayed by thousands of reporters integrated in parallel.

Reporter genes integrated into the genome are a powerful tool to reveal effects of regulatory elements and local chromatin context on gene expression. However, so far such reporter assays have been of low throughput. Here, we describe a multiplexing approach for the parallel monitoring of transcriptional activity of thousands of randomly integrated reporters. More than 27,000 distinct reporter integrations in mouse embryonic stem cells, obtained with two different promoters, show ∼1,000-fold variation in expression levels. Data analysis indicates that lamina-associated domains act as attenuators of transcription, likely by reducing access of transcription factors to binding sites. Furthermore, chromatin compaction is predictive of reporter activity. We also found evidence for crosstalk between neighboring genes and estimate that enhancers can influence gene expression on average over ∼20 kb. The multiplexed reporter assay is highly flexible in design and can be modified to query a wide range of aspects of gene regulation.

Towards realizing the vision of precision medicine: AI based prediction of clinical drug response.

Accurate and individualized prediction of response to therapies is central to precision medicine. However, because of the generally complex and multifaceted nature of clinical drug response, realizing this vision is highly challenging, requiring integrating different data types from the same individual into one prediction model. We used the anti-epileptic drug brivaracetam as a case study and combine a hybrid data/knowledge-driven feature extraction with machine learning to systematically integrate clinical and genetic data from a clinical discovery dataset (n = 235 patients). We constructed a model that successfully predicts clinical drug response [area under the curve (AUC) = 0.76] and show that even with limited sample size, integrating high-dimensional genetics data with clinical data can inform drug response prediction. After further validation on data collected from an independently conducted clinical study (AUC = 0.75), we extensively explore our model to gain insights into the determinants of drug response, and identify various clinical and genetic characteristics predisposing to poor response. Finally, we assess the potential impact of our model on clinical trial design and demonstrate that, by enriching for probable responders, significant reductions in clinical study sizes may be achieved. To our knowledge, our model represents the first retrospectively validated machine learning model linking drug mechanism of action and the genetic, clinical and demographic background in epilepsy patients to clinical drug response. Hence, it provides a blueprint for how machine learning-based multimodal data integration can act as a driver in achieving the goals of precision medicine in fields such as neurology.

The small FOXP1 isoform predominantly expressed in activated B cell-like diffuse large B-cell lymphoma and full-length FOXP1 exert similar oncogenic and transcriptional activity in human B cells.

The forkhead transcription factor FOXP1 is generally regarded as an oncogene in activated B cell-like diffuse large B-cell lymphoma. Previous studies have suggested that a small isoform of FOXP1 rather than full-length FOXP1, may possess this oncogenic activity. Corroborating those studies, we herein show that activated B cell-like diffuse large B-cell lymphoma cell lines and primary activated B cell-like diffuse large B-cell lymphoma cells predominantly express a small FOXP1 isoform, and that the 5'-end of the Foxp1 gene is a common insertion site in murine lymphomas in leukemia virus- and transposon-mediated insertional mutagenesis screens. By combined mass spectrometry, (quantative) reverse transcription polymerase chain reaction/sequencing, and small interfering ribonucleic acid-mediated gene silencing, we determined that the small FOXP1 isoform predominantly expressed in activated B cell-like diffuse large B-cell lymphoma lacks the N-terminal 100 amino acids of full-length FOXP1. Aberrant overexpression of this FOXP1 isoform (ΔN100) in primary human B cells revealed its oncogenic capacity; it repressed apoptosis and plasma cell differentiation. However, no difference in potency was found between this small FOXP1 isoform and full-length FOXP1. Furthermore, overexpression of full-length FOXP1 or this small FOXP1 isoform in primary B cells and diffuse large B-cell lymphoma cell lines resulted in similar gene regulation. Taken together, our data indicate that this small FOXP1 isoform and full-length FOXP1 have comparable oncogenic and transcriptional activity in human B cells, suggesting that aberrant expression or overexpression of FOXP1, irrespective of the specific isoform, contributes to lymphomagenesis. These novel insights further enhance the value of FOXP1 for the diagnostics, prognostics, and treatment of diffuse large B-cell lymphoma patients.

Chemical profiling of the genome with anti-cancer drugs defines target specificities.

Many anticancer drugs induce DNA breaks to eliminate tumor cells. The anthracycline topoisomerase II inhibitors additionally cause histone eviction. Here, we performed genome-wide high-resolution mapping of chemotherapeutic effects of various topoisomerase I and II (TopoI and II) inhibitors and integrated this mapping with established maps of genomic or epigenomic features to show their activities in different genomic regions. The TopoI inhibitor topotecan and the TopoII inhibitor etoposide are similar in inducing DNA damage at transcriptionally active genomic regions. The anthracycline daunorubicin induces DNA breaks and evicts histones from active chromatin, thus quenching local DNA damage responses. Another anthracycline, aclarubicin, has a different genomic specificity and evicts histones from H3K27me3-marked heterochromatin, with consequences for diffuse large B-cell lymphoma cells with elevated levels of H3K27me3. Modifying anthracycline structures may yield compounds with selectivity for different genomic regions and activity for different tumor types.

Chromatin landscapes of retroviral and transposon integration profiles.

The ability of retroviruses and transposons to insert their genetic material into host DNA makes them widely used tools in molecular biology, cancer research and gene therapy. However, these systems have biases that may strongly affect research outcomes. To address this issue, we generated very large datasets consisting of ~ 120,000 to ~ 180,000 unselected integrations in the mouse genome for the Sleeping Beauty (SB) and piggyBac (PB) transposons, and the Mouse Mammary Tumor Virus (MMTV). We analyzed ~ 80 (epi)genomic features to generate bias maps at both local and genome-wide scales. MMTV showed a remarkably uniform distribution of integrations across the genome. More distinct preferences were observed for the two transposons, with PB showing remarkable resemblance to bias profiles of the Murine Leukemia Virus. Furthermore, we present a model where target site selection is directed at multiple scales. At a large scale, target site selection is similar across systems, and defined by domain-oriented features, namely expression of proximal genes, proximity to CpG islands and to genic features, chromatin compaction and replication timing. Notable differences between the systems are mainly observed at smaller scales, and are directed by a diverse range of features. To study the effect of these biases on integration sites occupied under selective pressure, we turned to insertional mutagenesis (IM) screens. In IM screens, putative cancer genes are identified by finding frequently targeted genomic regions, or Common Integration Sites (CISs). Within three recently completed IM screens, we identified 7%-33% putative false positive CISs, which are likely not the result of the oncogenic selection process. Moreover, results indicate that PB, compared to SB, is more suited to tag oncogenes.

TRIP through the chromatin: a high throughput exploration of enhancer regulatory landscapes.

Enhancers are regulatory elements that promote gene expression in a spatio-temporal way and are involved in a wide range of developmental and disease processes. Both the identification and subsequent functional dissection of enhancers are key steps in understanding these processes. Several high-throughput approaches were recently developed for these purposes; however, in almost all cases enhancers are being tested outside their native chromatin context. Until recently, the analysis of enhancer activities at their native genomic locations was low throughput, laborious and time-consuming. Here, we discuss the potential of a powerful approach, TRIP, to study the functioning of enhancers in their native chromatin environments by introducing sensor constructs directly in the genome. TRIP allows for simultaneously analyzing the quantitative readout of numerous sensor constructs integrated at random locations in the genome. The high-throughput and flexible nature of TRIP opens up potential to study different aspects of enhancer biology at an unprecedented level.

Dosage-dependent tumor suppression by histone deacetylases 1 and 2 through regulation of c-Myc collaborating genes and p53 function.

Histone deacetylases (HDACs) are epigenetic erasers of lysine-acetyl marks. Inhibition of HDACs using small molecule inhibitors (HDACi) is a potential strategy in the treatment of various diseases and is approved for treating hematological malignancies. Harnessing the therapeutic potential of HDACi requires knowledge of HDAC-function in vivo. Here, we generated a thymocyte-specific gradient of HDAC-activity using compound conditional knockout mice for Hdac1 and Hdac2. Unexpectedly, gradual loss of HDAC-activity engendered a dosage-dependent accumulation of immature thymocytes and correlated with the incidence and latency of monoclonal lymphoblastic thymic lymphomas. Strikingly, complete ablation of Hdac1 and Hdac2 abrogated lymphomagenesis due to a block in early thymic development. Genomic, biochemical and functional analyses of pre-leukemic thymocytes and tumors revealed a critical role for Hdac1/Hdac2-governed HDAC-activity in regulating a p53-dependent barrier to constrain Myc-overexpressing thymocytes from progressing into lymphomas by regulating Myc-collaborating genes. One Myc-collaborating and p53-suppressing gene, Jdp2, was derepressed in an Hdac1/2-dependent manner and critical for the survival of Jdp2-overexpressing lymphoma cells. Although reduced HDAC-activity facilitates oncogenic transformation in normal cells, resulting tumor cells remain highly dependent on HDAC-activity, indicating that a critical level of Hdac1 and Hdac2 governed HDAC-activity is required for tumor maintenance.

Computational identification of insertional mutagenesis targets for cancer gene discovery.

Insertional mutagenesis is a potent forward genetic screening technique used to identify candidate cancer genes in mouse model systems. An important, yet unresolved issue in the analysis of these screens, is the identification of the genes affected by the insertions. To address this, we developed Kernel Convolved Rule Based Mapping (KC-RBM). KC-RBM exploits distance, orientation and insertion density across tumors to automatically map integration sites to target genes. We perform the first genome-wide evaluation of the association of insertion occurrences with aberrant gene expression of the predicted targets in both retroviral and transposon data sets. We demonstrate the efficiency of KC-RBM by showing its superior performance over existing approaches in recovering true positives from a list of independently, manually curated cancer genes. The results of this work will significantly enhance the accuracy and speed of cancer gene discovery in forward genetic screens. KC-RBM is available as R-package.

Predicting seizure recurrence after an initial seizure-like episode from routine clinical notes using large language models: a retrospective cohort study.

BACKGROUND: The evaluation and management of first-time seizure-like events in children can be difficult because these episodes are not always directly observed and might be epileptic seizures or other conditions (seizure mimics). We aimed to evaluate whether machine learning models using real-world data could predict seizure recurrence after an initial seizure-like event. METHODS: This retrospective cohort study compared models trained and evaluated on two separate datasets between Jan 1, 2010, and Jan 1, 2020: electronic medical records (EMRs) at Boston Children's Hospital and de-identified, patient-level, administrative claims data from the IBM MarketScan research database. The study population comprised patients with an initial diagnosis of either epilepsy or convulsions before the age of 21 years, based on International Classification of Diseases, Clinical Modification (ICD-CM) codes. We compared machine learning-based predictive modelling using structured data (logistic regression and XGBoost) with emerging techniques in natural language processing by use of large language models. FINDINGS: The primary cohort comprised 14 021 patients at Boston Children's Hospital matching inclusion criteria with an initial seizure-like event and the comparison cohort comprised 15 062 patients within the IBM MarketScan research database. Seizure recurrence based on a composite expert-derived definition occurred in 57% of patients at Boston Children's Hospital and 63% of patients within IBM MarketScan. Large language models with additional domain-specific and location-specific pre-training on patients excluded from the study (F1-score 0·826 [95% CI 0·817-0·835], AUC 0·897 [95% CI 0·875-0·913]) performed best. All large language models, including the base model without additional pre-training (F1-score 0·739 [95% CI 0·738-0·741], AUROC 0·846 [95% CI 0·826-0·861]) outperformed models trained with structured data. With structured data only, XGBoost outperformed logistic regression and XGBoost models trained with the Boston Children's Hospital EMR (logistic regression: F1-score 0·650 [95% CI 0·643-0·657], AUC 0·694 [95% CI 0·685-0·705], XGBoost: F1-score 0·679 [0·676-0·683], AUC 0·725 [0·717-0·734]) performed similarly to models trained on the IBM MarketScan database (logistic regression: F1-score 0·596 [0·590-0·601], AUC 0·670 [0·664-0·675], XGBoost: F1-score 0·678 [0·668-0·687], AUC 0·710 [0·703-0·714]). INTERPRETATION: Physician's clinical notes about an initial seizure-like event include substantial signals for prediction of seizure recurrence, and additional domain-specific and location-specific pre-training can significantly improve the performance of clinical large language models, even for specialised cohorts. FUNDING: UCB, National Institute of Neurological Disorders and Stroke (US National Institutes of Health).

A cell-based model of Nematostella vectensis gastrulation including bottle cell formation, invagination and zippering.

The gastrulation of Nematostella vectensis, the starlet sea anemone, is morphologically simple yet involves many conserved cell behaviors such as apical constriction, invagination, bottle cell formation, cell migration and zippering found during gastrulation in a wide range of more morphologically complex animals. In this article we study Nematostella gastrulation using a combination of morphometrics and computational modeling. Through this analysis we frame gastrulation as a non-trivial problem, in which two distinct cell domains must change shape to match each other geometrically, while maintaining the integrity of the embryo. Using a detailed cell-based model capable of representing arbitrary cell-shapes such as bottle cells, as well as filopodia, localized adhesion and constriction, we are able to simulate gastrulation and associate emergent macroscopic changes in embryo shape to individual cell behaviors. We have developed a number of testable hypotheses based on the model. First, we hypothesize that the blastomeres need to be stiffer at their apical ends, relative to the rest of the cell perimeter, in order to be able to hold their wedge shape and the dimensions of the blastula, regardless of whether the blastula is sealed or leaky. We also postulate that bottle cells are a consequence of cell strain and low cell-cell adhesion, and can be produced within an epithelium even without apical constriction. Finally, we postulate that apical constriction, filopodia and de-epithelialization are necessary and sufficient for gastrulation based on parameter variation studies.

An Explainable Multimodal Neural Network Architecture for Predicting Epilepsy Comorbidities Based on Administrative Claims Data.

Epilepsy is a complex brain disorder characterized by repetitive seizure events. Epilepsy patients often suffer from various and severe physical and psychological comorbidities (e.g., anxiety, migraine, and stroke). While general comorbidity prevalences and incidences can be estimated from epidemiological data, such an approach does not take into account that actual patient-specific risks can depend on various individual factors, including medication. This motivates to develop a machine learning approach for predicting risks of future comorbidities for individual epilepsy patients. In this work, we use inpatient and outpatient administrative health claims data of around 19,500 U.S. epilepsy patients. We suggest a dedicated multimodal neural network architecture (Deep personalized LOngitudinal convolutional RIsk model-DeepLORI) to predict the time-dependent risk of six common comorbidities of epilepsy patients. We demonstrate superior performance of DeepLORI in a comparison with several existing methods. Moreover, we show that DeepLORI-based predictions can be interpreted on the level of individual patients. Using a game theoretic approach, we identify relevant features in DeepLORI models and demonstrate that model predictions are explainable in light of existing knowledge about the disease. Finally, we validate the model on independent data from around 97,000 patients, showing good generalization and stable prediction performance over time.

Multiplexed Cas9 targeting reveals genomic location effects and gRNA-based staggered breaks influencing mutation efficiency.

Understanding the impact of guide RNA (gRNA) and genomic locus on CRISPR-Cas9 activity is crucial to design effective gene editing assays. However, it is challenging to profile Cas9 activity in the endogenous cellular environment. Here we leverage our TRIP technology to integrate ~ 1k barcoded reporter genes in the genomes of mouse embryonic stem cells. We target the integrated reporters (IRs) using RNA-guided Cas9 and characterize induced mutations by sequencing. We report that gRNA-sequence and IR locus explain most variation in mutation efficiency. Predominant insertions of a gRNA-specific nucleotide are consistent with template-dependent repair of staggered DNA ends with 1-bp 5' overhangs. We confirm that such staggered ends are induced by Cas9 in mouse pre-B cells. To explain observed insertions, we propose a model generating primarily blunt and occasionally staggered DNA ends. Mutation patterns indicate that gRNA-sequence controls the fraction of staggered ends, which could be used to optimize Cas9-based insertion efficiency.

Deep learning for clustering of multivariate clinical patient trajectories with missing values.

BACKGROUND: Precision medicine requires a stratification of patients by disease presentation that is sufficiently informative to allow for selecting treatments on a per-patient basis. For many diseases, such as neurological disorders, this stratification problem translates into a complex problem of clustering multivariate and relatively short time series because (i) these diseases are multifactorial and not well described by single clinical outcome variables and (ii) disease progression needs to be monitored over time. Additionally, clinical data often additionally are hindered by the presence of many missing values, further complicating any clustering attempts. FINDINGS: The problem of clustering multivariate short time series with many missing values is generally not well addressed in the literature. In this work, we propose a deep learning-based method to address this issue, variational deep embedding with recurrence (VaDER). VaDER relies on a Gaussian mixture variational autoencoder framework, which is further extended to (i) model multivariate time series and (ii) directly deal with missing values. We validated VaDER by accurately recovering clusters from simulated and benchmark data with known ground truth clustering, while varying the degree of missingness. We then used VaDER to successfully stratify patients with Alzheimer disease and patients with Parkinson disease into subgroups characterized by clinically divergent disease progression profiles. Additional analyses demonstrated that these clinical differences reflected known underlying aspects of Alzheimer disease and Parkinson disease. CONCLUSIONS: We believe our results show that VaDER can be of great value for future efforts in patient stratification, and multivariate time-series clustering in general.

Analysis of a spatial gene expression database for sea anemone Nematostella vectensis during early development.

BACKGROUND: The spatial distribution of many genes has been visualized during the embryonic development in the starlet sea anemone Nematostella vectensis in the last decade. In situ hybridization images are available in the Kahi Kai gene expression database, and a method has been developed to quantify spatial gene expression patterns of N. vectensis. In this paper, gene expression quantification is performed on a wide range of gene expression patterns from this database and descriptions of observed expression domains are stored in a separate database for further analysis. METHODS: Spatial gene expression from suitable in situ hybridization images has been quantified with the GenExp program. A correlation analysis has been performed on the resulting numerical gene expression profiles for each stage. Based on the correlated clusters of spatial gene expression and detailed descriptions of gene expression domains, various mechanisms for developmental gene expression are proposed. RESULTS: In the blastula and gastrula stages of development in N. vectensis, its continuous sheet of cells is partitioned into correlating gene expression domains. During progressing development, these regions likely correspond to different fates. A statistical analysis shows that genes generally remain expressed during the planula stages in those major regions that they occupy at the end of gastrulation. DISCUSSION: Observed shifts in gene expression domain boundaries suggest that elongation in the planula stage mainly occurs in the vegetal ring under the influence of the gene Rx. The secondary body axis in N. vectensis is proposed to be determined at the mid blastula transition. CONCLUSIONS: Early gene expression domains in N. vectensis appear to maintain a positional order along the primary body axis. Early determination in N. vectensis occurs in two stages: expression in broad circles and rings in the blastula is consolidated during gastrulation, and more complex expression patterns appear in the planula within these broad regions. Quantification and comparison of gene expression patterns across a database can generate hypotheses about collective cell movements before these movements are measured directly.

3D hotspots of recurrent retroviral insertions reveal long-range interactions with cancer genes.

Genomically distal mutations can contribute to the deregulation of cancer genes by engaging in chromatin interactions. To study this, we overlay viral cancer-causing insertions obtained in a murine retroviral insertional mutagenesis screen with genome-wide chromatin conformation capture data. Here we find that insertions tend to cluster in 3D hotspots within the nucleus. The identified hotspots are significantly enriched for known cancer genes, and bear the expected characteristics of bona fide regulatory interactions, such as enrichment for transcription factor-binding sites. In addition, we observe a striking pattern of mutual exclusive integration. This is an indication that insertions in these loci target the same gene, either in their linear genomic vicinity or in their 3D spatial vicinity. Our findings shed new light on the repertoire of targets obtained from insertional mutagenesis screening and underline the importance of considering the genome as a 3D structure when studying effects of genomic perturbations.

Prolonged Ezh2 Depletion in Glioblastoma Causes a Robust Switch in Cell Fate Resulting in Tumor Progression.

EZH2 is frequently overexpressed in glioblastoma (GBM), suggesting an oncogenic function that could be a target for therapeutic intervention. However, reduced EZH2 activity can also promote tumorigenesis, leading to concerns about the use of EZH2 inhibitors. Here, we provide further insight about the effects of prolonged Ezh2 inhibition in glioblastoma using preclinical mouse models and primary tumor-derived human GBM cell lines. Using doxycycline-inducible shRNAs that mimic the effects of a selective EZH2 inhibitor, we demonstrate that prolonged Ezh2 depletion causes a robust switch in cell fate, including significantly enhanced proliferation, DNA damage repair, and activation of part of the pluripotency network, resulting in altered tumor cell identity and tumor progression. Short-term Ezh2 depletion significantly improved survival without the tumor progression observed upon prolonged Ezh2 depletion, suggesting that precise dosing regiments are very important. These results could be of high clinical relevance with regard to how glioblastomas should be treated with epigenetic therapies.

Applications of DNA integrating elements: Facing the bias bully.

Retroviruses and DNA transposons are an important part of molecular biologists' toolbox. The applications of these elements range from functional genomics to oncogene discovery and gene therapy. However, these elements do not integrate uniformly across the genome, which is an important limitation to their use. A number of genetic and epigenetic factors have been shown to shape the integration preference of these elements. Insight into integration bias can significantly enhance the analysis and interpretation of results obtained using these elements. For three different applications, we outline how bias can affect results, and can potentially be addressed.

A computational approach towards a gene regulatory network for the developing Nematostella vectensis gut.

BACKGROUND: The starlet sea anemone Nematostella vectensis is a diploblastic cnidarian that expresses a set of conserved genes for gut formation during its early development. During the last decade, the spatial distribution of many of these genes has been visualized with RNA hybridization or protein immunolocalization techniques. However, due to N. vectensis' curved and changing morphology, quantification of these spatial data is problematic. A method is developed for two-dimensional gene expression quantification, which enables a numerical analysis and dynamic modeling of these spatial patterns. METHODS/RESULT: In this work, first standardized gene expression profiles are generated from publicly available N. vectensis embryo images that display mRNA and/or protein distributions. Then, genes expressed during gut formation are clustered based on their expression profiles, and further grouped based on temporal appearance of their gene products in embryonic development. Representative expression profiles are manually selected from these clusters, and used as input for a simulation-based optimization scheme. This scheme iteratively fits simulated profiles to the selected profiles, leading to an optimized estimation of the model parameters. Finally, a preliminary gene regulatory network is derived from the optimized model parameters. OUTLOOK: While the focus of this study is N. vectensis, the approach outlined here is suitable for inferring gene regulatory networks in the embryonic development of any animal, thus allowing to comparatively study gene regulation of gut formation in silico across various species.

Using TRIP for genome-wide position effect analysis in cultured cells.

The influence of local chromatin context on gene expression can be explored by integrating a transcription reporter at different locations in the genome as a sensor. Here we provide a detailed protocol for analyzing thousands of reporters integrated in parallel (TRIP) at a genome-wide level. TRIP is based on tagging each reporter with a unique barcode, which is used for independent reporter expression analysis and integration site mapping. Compared with previous methods for studying position effects, TRIP offers a 100-1,000-fold higher throughput in a faster and less-labor-intensive manner. The entire experimental protocol takes ∼42 d to complete, with high-throughput sequencing and data analysis requiring an additional ∼11 d. TRIP was developed by using transcription reporters in mouse embryonic stem (mES) cells, but because of its flexibility the method can be used to probe the influence of chromatin context on a variety of molecular processes in any transfectable cell line.

Transposon mutagenesis identifies genes driving hepatocellular carcinoma in a chronic hepatitis B mouse model.

The most common risk factor for developing hepatocellular carcinoma (HCC) is chronic infection with hepatitis B virus (HBV). To better understand the evolutionary forces driving HCC, we performed a near-saturating transposon mutagenesis screen in a mouse HBV model of HCC. This screen identified 21 candidate early stage drivers and a very large number (2,860) of candidate later stage drivers that were enriched for genes that are mutated, deregulated or functioning in signaling pathways important for human HCC, with a striking 1,199 genes being linked to cellular metabolic processes. Our study provides a comprehensive overview of the genetic landscape of HCC.