A comprehensive clinically informed map of dependencies in cancer cells and framework for target prioritization.

Genetic screens in cancer cell lines inform gene function and drug discovery. More comprehensive screen datasets with multi-omics data are needed to enhance opportunities to functionally map genetic vulnerabilities. Here, we construct a second-generation map of cancer dependencies by annotating 930 cancer cell lines with multi-omic data and analyze relationships between molecular markers and cancer dependencies derived from CRISPR-Cas9 screens. We identify dependency-associated gene expression markers beyond driver genes, and observe many gene addiction relationships driven by gain of function rather than synthetic lethal effects. By combining clinically informed dependency-marker associations with protein-protein interaction networks, we identify 370 anti-cancer priority targets for 27 cancer types, many of which have network-based evidence of a functional link with a marker in a cancer type. Mapping these targets to sequenced tumor cohorts identifies tractable targets in different cancer types. This target prioritization map enhances understanding of gene dependencies and identifies candidate anti-cancer targets for drug development.

Single-cell imaging of protein dynamics of paralogs reveals mechanisms of gene retention.

Gene duplication is common across the tree of life, including yeast and humans, and contributes to genomic robustness. In this study, we examined changes in the subcellular localization and abundance of proteins in response to the deletion of their paralogs originating from the whole-genome duplication event, which is a largely unexplored mechanism of functional divergence. We performed a systematic single-cell imaging analysis of protein dynamics and screened subcellular redistribution of proteins, capturing their localization and abundance changes, providing insight into forces determining paralog retention. Paralogs showed dependency, whereby proteins required their paralog to maintain their native abundance or localization, more often than compensation. Network feature analysis suggested the importance of functional redundancy and rewiring of protein and genetic interactions underlying redistribution response of paralogs. Translation of non-canonical protein isoform emerged as a novel compensatory mechanism. This study provides new insights into paralog retention and evolutionary forces that shape genomes.

Pooled Genome-Scale CRISPR Screens in Single Cells.

Assigning functions to genes and learning how to control their expression are part of the foundation of cell biology and therapeutic development. An efficient and unbiased method to accomplish this is genetic screening, which historically required laborious clone generation and phenotyping and is still limited by scale today. The rapid technological progress on modulating gene function with CRISPR-Cas and measuring it in individual cells has now relaxed the major experimental constraints and enabled pooled screening with complex readouts from single cells. Here, we review the principles and practical considerations for pooled single-cell CRISPR screening. We discuss perturbation strategies, experimental model systems, matching the perturbation to the individual cells, reading out cell phenotypes, and data analysis. Our focus is on single-cell RNA sequencing and cell sorting-based readouts, including image-enabled cell sorting. We expect this transformative approach to fuel biomedical research for the next several decades.

The interplay of DNA repair context with target sequence predictably biasses Cas9-generated mutations.

The genome engineering capability of the CRISPR/Cas system depends on the DNA repair machinery to generate the final outcome. Several genes can have an impact on mutations created, but their exact function and contribution to the result of the repair are not completely characterised. This lack of knowledge has limited the ability to comprehend and regulate the editing outcomes. Here, we measure how the absence of 21 repair genes changes the mutation outcomes of Cas9-generated cuts at 2,812 synthetic target sequences in mouse embryonic stem cells. Absence of key non-homologous end joining genes Lig4, Xrcc4, and Xlf abolished small insertions and deletions, while disabling key microhomology-mediated repair genes Nbn and Polq reduced frequency of longer deletions. Complex alleles of combined insertion and deletions were preferentially generated in the absence of Xrcc6. We further discover finer structure in the outcome frequency changes for single nucleotide insertions and deletions between large microhomologies that are differentially modulated by the knockouts. We use the knowledge of the reproducible variation across repair milieus to build predictive models of Cas9 editing results that outperform the current standards. This work improves our understanding of DNA repair gene function, and provides avenues for more precise modulation of CRISPR/Cas9-generated mutations.

Predicting Mutations Generated by Cas9, Base Editing, and Prime Editing in Mammalian Cells.

The first fruits of the CRISPR-Cas revolution are starting to enter the clinic, with gene editing therapies offering solutions to previously incurable genetic diseases. The success of such applications hinges on control over the mutations that are generated, which are known to vary depending on the targeted locus. In this review, we present the current state of understanding and predicting CRISPR-Cas cutting, base editing, and prime editing outcomes in mammalian cells. We first provide an introduction to the basics of DNA repair and machine learning that the models rely on. We then overview the datasets and methods created for characterizing edits at scale, as well as the insights that have been derived from them. The predictions generated from these models serve as a foundation for designing efficient experiments across the broad contexts where these tools are applied.

Optimized whole-genome CRISPR interference screens identify ARID1A-dependent growth regulators in human induced pluripotent stem cells.

Perturbing expression is a powerful way to understand the role of individual genes, but can be challenging in important models. CRISPR-Cas screens in human induced pluripotent stem cells (iPSCs) are of limited efficiency due to DNA break-induced stress, while the less stressful silencing with an inactive Cas9 has been considered less effective so far. Here, we developed the dCas9-KRAB-MeCP2 fusion protein for screening in iPSCs from multiple donors. We found silencing in a 200 bp window around the transcription start site in polyclonal pools to be as effective as using wild-type Cas9 for identifying essential genes, but with much reduced cell numbers. Whole-genome screens to identify ARID1A-dependent dosage sensitivity revealed the PSMB2 gene, and enrichment of proteasome genes among the hits. This selective dependency was replicated with a proteasome inhibitor, indicating a targetable drug-gene interaction. Many more plausible targets in challenging cell models can be efficiently identified with our approach.

Prediction of prime editing insertion efficiencies using sequence features and DNA repair determinants.

Most short sequences can be precisely written into a selected genomic target using prime editing; however, it remains unclear what factors govern insertion. We design a library of 3,604 sequences of various lengths and measure the frequency of their insertion into four genomic sites in three human cell lines, using different prime editor systems in varying DNA repair contexts. We find that length, nucleotide composition and secondary structure of the insertion sequence all affect insertion rates. We also discover that the 3' flap nucleases TREX1 and TREX2 suppress the insertion of longer sequences. Combining the sequence and repair features into a machine learning model, we can predict relative frequency of insertions into a site with R = 0.70. Finally, we demonstrate how our accurate prediction and user-friendly software help choose codon variants of common fusion tags that insert at high efficiency, and provide a catalog of empirically determined insertion rates for over a hundred useful sequences.

ArtSeg-Artifact segmentation and removal in brightfield cell microscopy images without manual pixel-level annotations.

Brightfield cell microscopy is a foundational tool in life sciences. The acquired images are prone to contain visual artifacts that hinder downstream analysis, and automatically removing them is therefore of great practical interest. Deep convolutional neural networks are state-of-the-art for image segmentation, but require pixel-level annotations, which are time-consuming to produce. Here, we propose ScoreCAM-U-Net, a pipeline to segment artifactual regions in brightfield images with limited user input. The model is trained using only image-level labels, so the process is faster by orders of magnitude compared to pixel-level annotation, but without substantially sacrificing the segmentation performance. We confirm that artifacts indeed exist with different shapes and sizes in three different brightfield microscopy image datasets, and distort downstream analyses such as nuclei segmentation, morphometry and fluorescence intensity quantification. We then demonstrate that our automated artifact removal ameliorates this problem. Such rapid cleaning of acquired images using the power of deep learning models is likely to become a standard step for all large scale microscopy experiments.

Live-cell microscopy or fluorescence anisotropy with budded baculoviruses-which way to go with measuring ligand binding to M4 muscarinic receptors?

M4 muscarinic acetylcholine receptor is a G protein-coupled receptor (GPCR) that has been associated with alcohol and cocaine abuse, Alzheimer's disease, and schizophrenia which makes it an interesting drug target. For many GPCRs, the high-affinity fluorescence ligands have expanded the options for high-throughput screening of drug candidates and serve as useful tools in fundamental receptor research. Here, we explored two TAMRA-labelled fluorescence ligands, UR-MK342 and UR-CG072, for development of assays for studying ligand-binding properties to M4 receptor. Using budded baculovirus particles as M4 receptor preparation and fluorescence anisotropy method, we measured the affinities and binding kinetics of both fluorescence ligands. Using the fluorescence ligands as reporter probes, the binding affinities of unlabelled ligands could be determined. Based on these results, we took a step towards a more natural system and developed a method using live CHO-K1-hM4R cells and automated fluorescence microscopy suitable for the routine determination of unlabelled ligand affinities. For quantitative image analysis, we developed random forest and deep learning-based pipelines for cell segmentation. The pipelines were integrated into the user-friendly open-source Aparecium software. Both image analysis methods were suitable for measuring fluorescence ligand saturation binding and kinetics as well as for screening binding affinities of unlabelled ligands.

Predicting base editing outcomes using position-specific sequence determinants.

CRISPR/Cas base editors promise nucleotide-level control over DNA sequences, but the determinants of their activity remain incompletely understood. We measured base editing frequencies in two human cell lines for two cytosine and two adenine base editors at ∼14 000 target sequences and find that base editing activity is sequence-biased, with largest effects from nucleotides flanking the target base. Whether a base is edited depends strongly on the combination of its position in the target and the preceding base, acting to widen or narrow the effective editing window. The impact of features on editing rate depends on the position, with sequence bias efficacy mainly influencing bases away from the center of the window. We use these observations to train a machine learning model to predict editing activity per position, with accuracy ranging from 0.49 to 0.72 between editors, and with better generalization across datasets than existing tools. We demonstrate the usefulness of our model by predicting the efficacy of disease mutation correcting guides, and find that most of them suffer from more unwanted editing than pure outcomes. This work unravels the position-specificity of base editing biases and allows more efficient planning of editing campaigns in experimental and therapeutic contexts.

Machine Learning Prediction of Resistance to Subinhibitory Antimicrobial Concentrations from Escherichia coli Genomes.

Escherichia coli is an important cause of bacterial infections worldwide, with multidrug-resistant strains incurring substantial costs on human lives. Besides therapeutic concentrations of antimicrobials in health care settings, the presence of subinhibitory antimicrobial residues in the environment and in clinics selects for antimicrobial resistance (AMR), but the underlying genetic repertoire is less well understood. Here, we used machine learning to predict the population doubling time and cell growth yield of 1,407 genetically diverse E. coli strains expanding under exposure to three subinhibitory concentrations of six classes of antimicrobials from single-nucleotide genetic variants, accessory gene variation, and the presence of known AMR genes. We predicted cell growth yields in the held-out test data with an average correlation (Spearman's ρ) of 0.63 (0.36 to 0.81 across concentrations) and cell doubling times with an average correlation of 0.59 (0.32 to 0.92 across concentrations), with moderate increases in sample size unlikely to improve predictions further. This finding points to the remaining missing heritability of growth under antimicrobial exposure being explained by effects that are too rare or weak to be captured unless sample size is dramatically increased, or by effects other than those conferred by the presence of individual single-nucleotide polymorphisms (SNPs) and genes. Predictions based on whole-genome information were generally superior to those based only on known AMR genes and were accurate for AMR resistance at therapeutic concentrations. We pinpointed genes and SNPs determining the predicted growth and thereby recapitulated many known AMR determinants. Finally, we estimated the effect sizes of resistance genes across the entire collection of strains, disclosing the growth effects for known resistance genes in each individual strain. Our results underscore the potential of predictive modeling of growth patterns from genomic data under subinhibitory concentrations of antimicrobials, although the remaining missing heritability poses a challenge for achieving the accuracy and precision required for clinical use. IMPORTANCE Predicting bacterial growth from genome sequences is important for a rapid characterization of strains in clinical diagnostics and to disclose candidate novel targets for anti-infective drugs. Previous studies have dissected the relationship between bacterial growth and genotype in mutant libraries for laboratory strains, yet no study so far has examined the predictive power of genome sequence in natural strains. In this study, we used a high-throughput phenotypic assay to measure the growth of a systematic collection of natural Escherichia coli strains and then employed machine learning models to predict bacterial growth from genomic data under nontherapeutic subinhibitory concentrations of antimicrobials that are common in nonclinical settings. We found a moderate to strong correlation between predicted and actual values for the different collected data sets. Moreover, we observed that the known resistance genes are still effective at sublethal concentrations, pointing to clinical implications of these concentrations.

Evaluating Very Deep Convolutional Neural Networks for Nucleus Segmentation from Brightfield Cell Microscopy Images.

Advances in microscopy have increased output data volumes, and powerful image analysis methods are required to match. In particular, finding and characterizing nuclei from microscopy images, a core cytometry task, remains difficult to automate. While deep learning models have given encouraging results on this problem, the most powerful approaches have not yet been tested for attacking it. Here, we review and evaluate state-of-the-art very deep convolutional neural network architectures and training strategies for segmenting nuclei from brightfield cell images. We tested U-Net as a baseline model; considered U-Net++, Tiramisu, and DeepLabv3+ as latest instances of advanced families of segmentation models; and propose PPU-Net, a novel light-weight alternative. The deeper architectures outperformed standard U-Net and results from previous studies on the challenging brightfield images, with balanced pixel-wise accuracies of up to 86%. PPU-Net achieved this performance with 20-fold fewer parameters than the comparably accurate methods. All models perform better on larger nuclei and in sparser images. We further confirmed that in the absence of plentiful training data, augmentation and pretraining on other data improve performance. In particular, using only 16 images with data augmentation is enough to achieve a pixel-wise F1 score that is within 5% of the one achieved with a full data set for all models. The remaining segmentation errors are mainly due to missed nuclei in dense regions, overlapping cells, and imaging artifacts, indicating the major outstanding challenges.

Practical segmentation of nuclei in brightfield cell images with neural networks trained on fluorescently labelled samples.

Identifying nuclei is a standard first step when analysing cells in microscopy images. The traditional approach relies on signal from a DNA stain, or fluorescent transgene expression localised to the nucleus. However, imaging techniques that do not use fluorescence can also carry useful information. Here, we used brightfield and fluorescence images of fixed cells with fluorescently labelled DNA, and confirmed that three convolutional neural network architectures can be adapted to segment nuclei from the brightfield channel, relying on fluorescence signal to extract the ground truth for training. We found that U-Net achieved the best overall performance, Mask R-CNN provided an additional benefit of instance segmentation, and that DeepCell proved too slow for practical application. We trained the U-Net architecture on over 200 dataset variations, established that accurate segmentation is possible using as few as 16 training images, and that models trained on images from similar cell lines can extrapolate well. Acquiring data from multiple focal planes further helps distinguish nuclei in the samples. Overall, our work helps to liberate a fluorescence channel reserved for nuclear staining, thus providing more information from the specimen, and reducing reagents and time required for preparing imaging experiments.

Natural variants suppress mutations in hundreds of essential genes.

The consequence of a mutation can be influenced by the context in which it operates. For example, loss of gene function may be tolerated in one genetic background, and lethal in another. The extent to which mutant phenotypes are malleable, the architecture of modifiers and the identities of causal genes remain largely unknown. Here, we measure the fitness effects of ~ 1,100 temperature-sensitive alleles of yeast essential genes in the context of variation from ten different natural genetic backgrounds and map the modifiers for 19 combinations. Altogether, fitness defects for 149 of the 580 tested genes (26%) could be suppressed by genetic variation in at least one yeast strain. Suppression was generally driven by gain-of-function of a single, strong modifier gene, and involved both genes encoding complex or pathway partners suppressing specific temperature-sensitive alleles, as well as general modifiers altering the effect of many alleles. The emerging frequency of suppression and range of possible mechanisms suggest that a substantial fraction of monogenic diseases could be managed by modulating other gene products.

Minimal genome-wide human CRISPR-Cas9 library.

CRISPR guide RNA libraries have been iteratively improved to provide increasingly efficient reagents, although their large size is a barrier for many applications. We design an optimised minimal genome-wide human CRISPR-Cas9 library (MinLibCas9) by mining existing large-scale gene loss-of-function datasets, resulting in a greater than 42% reduction in size compared to other CRISPR-Cas9 libraries while preserving assay sensitivity and specificity. MinLibCas9 provides backward compatibility with existing datasets, increases the dynamic range of CRISPR-Cas9 screens and extends their application to complex models and assays.

Genomic Epidemiology and Evolution of Escherichia coli in Wild Animals in Mexico.

Escherichia coli is a common bacterial species in the gastrointestinal tracts of warm-blooded animals and humans. Pathogenicity and antimicrobial resistance in E. coli may emerge via host switching from animal reservoirs. Despite its potential clinical importance, knowledge of the population structure of commensal E. coli within wild hosts and the epidemiological links between E. coli in nonhuman hosts and E. coli in humans is still scarce. In this study, we analyzed the whole-genome sequencing data of a collection of 119 commensal E. coli strains recovered from the guts of 55 mammal and bird species in Mexico and Venezuela in the 1990s. We observed low concordance between the population structures of E. coli isolates colonizing wild animals and the phylogeny, taxonomy, and ecological and physiological attributes of the host species, with distantly related E. coli strains often colonizing the same or similar host species and distantly related host species often hosting closely related E. coli strains. We found no evidence for recent transmission of E. coli genomes from wild animals to either domesticated animals or humans. However, multiple livestock- and human-related virulence factor genes were present in E. coli of wild animals, including virulence factors characteristic of Shiga toxin-producing E. coli (STEC) and atypical enteropathogenic E. coli (aEPEC), where several isolates from wild hosts harbored the locus of enterocyte effacement (LEE) pathogenicity island. Moreover, E. coli isolates from wild animal hosts often harbored known antibiotic resistance determinants, including those against ciprofloxacin, aminoglycosides, tetracyclines, and beta-lactams, with some determinants present in multiple, distantly related E. coli lineages colonizing very different host animals. We conclude that genome pools of E. coli colonizing the guts of wild animals and humans share virulence and antibiotic resistance genes, underscoring the idea that wild animals could serve as reservoirs for E. coli pathogenicity in human and livestock infections.IMPORTANCEEscherichia coli is a clinically important bacterial species implicated in human- and livestock-associated infections worldwide. The bacterium is known to reside in the guts of humans, livestock, and wild animals. Although wild animals are recognized as potential reservoirs for pathogenic E. coli strains, the knowledge of the population structure of E. coli in wild hosts is still scarce. In this study, we used fine resolution of whole-genome sequencing to provide novel insights into the evolution of E. coli genomes from a small yet diverse collection of strains recovered within a broad range of wild animal species (including mammals and birds), the coevolution of E. coli strains with their hosts, and the genetics of pathogenicity of E. coli strains in wild hosts in Mexico. Our results provide evidence for the clinical importance of wild animals as reservoirs for pathogenic strains and highlight the need to include nonhuman hosts in the surveillance programs for E. coli infections.

Type II and type IV toxin-antitoxin systems show different evolutionary patterns in the global Klebsiella pneumoniae population.

The Klebsiella pneumoniae species complex includes important opportunistic pathogens which have become public health priorities linked to major hospital outbreaks and the recent emergence of multidrug-resistant hypervirulent strains. Bacterial virulence and the spread of multidrug resistance have previously been linked to toxin-antitoxin (TA) systems. TA systems encode a toxin that disrupts essential cellular processes, and a cognate antitoxin which counteracts this activity. Whilst associated with the maintenance of plasmids, they also act in bacterial immunity and antibiotic tolerance. However, the evolutionary dynamics and distribution of TA systems in clinical pathogens are not well understood. Here, we present a comprehensive survey and description of the diversity of TA systems in 259 clinically relevant genomes of K. pneumoniae. We show that TA systems are highly prevalent with a median of 20 loci per strain. Importantly, these toxins differ substantially in their distribution patterns and in their range of cognate antitoxins. Classification along these properties suggests different roles of TA systems and highlights the association and co-evolution of toxins and antitoxins.

Agreement between two large pan-cancer CRISPR-Cas9 gene dependency data sets.

Genome-scale CRISPR-Cas9 viability screens performed in cancer cell lines provide a systematic approach to identify cancer dependencies and new therapeutic targets. As multiple large-scale screens become available, a formal assessment of the reproducibility of these experiments becomes necessary. We analyze data from recently published pan-cancer CRISPR-Cas9 screens performed at the Broad and Sanger Institutes. Despite significant differences in experimental protocols and reagents, we find that the screen results are highly concordant across multiple metrics with both common and specific dependencies jointly identified across the two studies. Furthermore, robust biomarkers of gene dependency found in one data set are recovered in the other. Through further analysis and replication experiments at each institute, we show that batch effects are driven principally by two key experimental parameters: the reagent library and the assay length. These results indicate that the Broad and Sanger CRISPR-Cas9 viability screens yield robust and reproducible findings.

The ribosomal P-stalk couples amino acid starvation to GCN2 activation in mammalian cells.

The eukaryotic translation initiation factor 2α (eIF2α) kinase GCN2 is activated by amino acid starvation to elicit a rectifying physiological program known as the Integrated Stress Response (ISR). A role for uncharged tRNAs as activating ligands of yeast GCN2 is supported experimentally. However, mouse GCN2 activation has recently been observed in circumstances associated with ribosome stalling with no global increase in uncharged tRNAs. We report on a mammalian CHO cell-based CRISPR-Cas9 mutagenesis screen for genes that contribute to ISR activation by amino acid starvation. Disruption of genes encoding components of the ribosome P-stalk, uL10 and P1, selectively attenuated GCN2-mediated ISR activation by amino acid starvation or interference with tRNA charging without affecting the endoplasmic reticulum unfolded protein stress-induced ISR, mediated by the related eIF2α kinase PERK. Wildtype ribosomes isolated from CHO cells, but not those with P-stalk lesions, stimulated GCN2-dependent eIF2α phosphorylation in vitro. These observations support a model whereby lack of a cognate charged tRNA exposes a latent capacity of the ribosome P-stalk to activate GCN2 in cells and help explain the emerging link between ribosome stalling and ISR activation.