Interrogation of cancer gene dependencies reveals paralog interactions of autosome and sex chromosome-encoded genes

Genetic dependencies associated with transcription factor activities in human cancer cell lines

Transcription factors (TFs) are important mediators of aberrant transcriptional programs in cancer cells. In this study, we focus on TF activity (TFa) as a biomarker for cell-line-selective anti-proliferative effects, in that high TFa predicts sensitivity to loss of function of a given gene (i.e., genetic dependencies [GDs]). Our linear-regression-based framework identifies 3,047 pan-cancer and 3,952 cancer-type-specific candidate TFa-GD associations from cell line data, which are then cross-examined for impact on survival in patient cohorts. One of the most prominent biomarkers is TEAD1 activity, whose associations with its predicted GDs are validated through experimental evidence as proof of concept. Overall, these TFa-GD associations represent an attractive resource for identifying innovative, biomarker-driven hypotheses for drug discovery programs in oncology.

High-throughput mRNA sequencing of human placenta shows sex differences across gestation

INTRODUCTION

Fetal sex affects fetal and maternal health outcomes in pregnancy, but this connection remains poorly understood. As the placenta is the route of fetomaternal communication and derives from the fetal genome, placental gene expression sex differences may explain these outcomes.

OBJECTIVES

We utilized next generation sequencing to study the normal human placenta in both sexes in first and third trimester to generate a normative transcriptome based on sex and gestation.

STUDY DESIGN

We analyzed 124 first trimester (T1, 59 female and 65 male) and 43 third trimester (T3, 18 female and 25 male) samples for sex differences within each trimester and sex-specific gestational differences.

RESULTS

Placenta shows more significant sexual dimorphism in T1, with 94 T1 and 26 T3 differentially expressed genes (DEGs). The sex chromosomes contributed 60.6% of DEGs in T1 and 80.8% of DEGs in T3, excluding X/Y pseudoautosomal regions. There were 6 DEGs from the pseudoautosomal regions, only significant in T1 and all upregulated in males. The distribution of DEGs on the X chromosome suggests genes on Xp (the short arm) may be particularly important in placental sex differences. Dosage compensation analysis of X/Y homolog genes shows expression is primarily contributed by the X chromosome. In sex-specific analyses of first versus third trimester, there were 2815 DEGs common to both sexes upregulated in T1, and 3263 common DEGs upregulated in T3. There were 7 female-exclusive DEGs upregulated in T1, 15 female-exclusive DEGs upregulated in T3, 10 male-exclusive DEGs upregulated in T1, and 20 male-exclusive DEGs upregulated in T3.

DISCUSSION

This is the largest cohort of placentas across gestation from healthy pregnancies defining the normative sex dimorphic gene expression and sex common, sex specific and sex exclusive gene expression across gestation. The first trimester has the most sexually dimorphic transcripts, and the majority were upregulated in females compared to males in both trimesters. The short arm of the X chromosome and the pseudoautosomal region is particularly critical in defining sex differences in the first trimester placenta. As pregnancy is a dynamic state, sex specific DEGs across gestation may contribute to sex dimorphic changes in overall outcomes.

Expanding the ligandable proteome by paralog hopping with covalent probes

More than half of the ~20,000 protein-encoding human genes have at least one paralog. Chemical proteomics has uncovered many electrophile-sensitive cysteines that are exclusive to a subset of paralogous proteins. Here, we explore whether such covalent compound-cysteine interactions can be used to discover ligandable pockets in paralogs that lack the cysteine. Leveraging the covalent ligandability of C109 in the cyclin CCNE2, we mutated the corresponding residue in paralog CCNE1 to cysteine (N112C) and found through activity-based protein profiling (ABPP) that this mutant reacts stereoselectively and site-specifically with tryptoline acrylamides. We then converted the tryptoline acrylamide-N112C-CCNE1 interaction into a NanoBRET-ABPP assay capable of identifying compounds that reversibly inhibit both N112C- and WT-CCNE1:CDK2 complexes. X-ray crystallography revealed a cryptic allosteric pocket at the CCNE1:CDK2 interface adjacent to N112 that binds the reversible inhibitors. Our findings thus provide a roadmap for leveraging electrophile-cysteine interactions to extend the ligandability of the proteome beyond covalent chemistry.

Synthetic Lethality Screening with Recursive Feature Machines

Synthetic lethality refers to a genetic interaction where the simultaneous perturbation of gene pairs leads to cell death. Synthetically lethal gene pairs (SL pairs) provide a potential avenue for selectively targeting cancer cells based on genetic vulnerabilities. The rise of large-scale gene perturbation screens such as the Cancer Dependency Map (DepMap) offers the opportunity to identify SL pairs automatically using machine learning. We build on a recently developed class of feature learning kernel machines known as Recursive Feature Machines (RFMs) to develop a pipeline for identifying SL pairs based on CRISPR viability data from DepMap. In particular, we first train RFMs to predict viability scores for a given CRISPR gene knockout from cell line embeddings consisting of gene expression and mutation features. After training, RFMs use a statistical operator known as average gradient outer product to provide weights for each feature indicating the importance of each feature in predicting cellular viability. We subsequently apply correlation-based filters to re-weight RFM feature importances and identify those features that are most indicative of low cellular viability. Our resulting pipeline is computationally efficient, taking under 3 minutes for analyzing all 17, 453 knockouts from DepMap for candidate SL pairs. We show that our pipeline more accurately recovers experimentally verified SL pairs than prior approaches. Moreover, our pipeline finds new candidate SL pairs, thereby opening novel avenues for identifying genetic vulnerabilities in cancer.

Using graph-based model to identify cell specific synthetic lethal effects

Synthetic lethal (SL) pairs are pairs of genes whose simultaneous loss-of-function results in cell death, while a damaging mutation of either gene alone does not affect the cell's survival. This makes SL pairs attractive targets for precision cancer therapies, as targeting the unimpaired gene of the SL pair can selectively kill cancer cells that already harbor the impaired gene. Limited by the difficulty of finding true SL pairs, especially on specific cell types, current computational approaches provide only limited insights because of overlooking the crucial aspects of cellular context dependency and mechanistic understanding of SL pairs. As a result, the identification of SL targets still relies on expensive, time-consuming experimental approaches. In this work, we applied cell-line specific multi-omics data to a specially designed deep learning model to predict cell-line specific SL pairs. Through incorporating multiple types of cell-specific omics data with a self-attention module, we represent gene relationships as graphs. Our approach achieves the prediction of SL pairs in a cell-specific manner and demonstrates the potential to facilitate the discovery of cell-specific SL targets for cancer therapeutics, providing a tool to unearth mechanisms underlying the origin of SL in cancer biology. The code and data of our approach can be found at https://github.com/promethiume/SLwise.

Concerning the eXclusion in human genomics: the choice of sex chromosome representation in the human genome drastically affects the number of identified variants

Over the past 30 years, a community of scientists has pieced together every base pair of the human reference genome from telomere to telomere. Interestingly, most human genomics studies omit more than 5% of the genome from their analyses. Under "normal" circumstances, omitting any chromosome(s) from an analysis of the human genome would be a cause for concern, with the exception being sex chromosomes. Sex chromosomes in eutherians share an evolutionary origin as an ancestral pair of autosomes. In humans, they share 3 regions of high-sequence identity (∼98-100%), which, along with the unique transmission patterns of the sex chromosomes, introduce technical artifacts in genomic analyses. However, the human X chromosome bears numerous important genes, including more "immune response" genes than any other chromosome, which makes its exclusion irresponsible when sex differences across human diseases are widespread. To better characterize the possible effect of the inclusion/exclusion of the X chromosome on variants called, we conducted a pilot study on the Terra cloud platform to replicate a subset of standard genomic practices using both the CHM13 reference genome and the sex chromosome complement-aware reference genome. We compared the quality of variant calling, expression quantification, and allele-specific expression using these 2 reference genome versions across 50 human samples from the Genotype-Tissue Expression consortium annotated as females. We found that after correction, the whole X chromosome (100%) can generate reliable variant calls, allowing for the inclusion of the whole genome in human genomics analyses as a departure from the status quo of omitting the sex chromosomes from empirical and clinical genomics studies.

Sex-biased genome-editing effects of CRISPR-Cas9 across cancer cells dependent on p53 status

The CRISPR-Cas9 system has emerged as the dominant technology for gene editing and clinical applications. One major concern is its off-target effect after the introduction of exogenous CRISPR-Cas9 into cells. Several previous studies have investigated either Cas9 alone or CRISPR-Cas9 interactions with p53. Here, we reanalyzed previously reported data of p53-associated Cas9 activities and observed large significant sex differences between p53-wildtype and p53-mutant cells. To expand the impact of this finding, we further examined all protein-coding genes for sex-specific dependencies in a large-scale CRISPR-Cas9 screening dataset from the DepMap project. We highlighted the p53-dependent sex bias of gene knockouts (including MYC, PIK3CA, KAT2B, KDM4E, SUV39H1, FANCB, TLR7, and APC2) across cancer types and potential mechanisms (mediated by transcriptional factors, including SOX9, FOXO4, LEF1, and RYBP) underlying this phenomenon. Our results suggest that the p53-dependent sex bias may need to be considered in future clinical applications of CRISPR-Cas9, especially in cancer.

Loss of chromosome Y in primary tumors

Certain cancer types afflict female and male patients disproportionately. The reasons include differences in male/female physiology, effect of sex hormones, risk behavior, environmental exposures, and genetics of the sex chromosomes X and Y. Loss of Y (LOY) is common in peripheral blood cells in aging men, and this phenomenon is associated with several diseases. However, the frequency and role of LOY in tumors is little understood. Here, we present a comprehensive catalog of LOY in >5,000 primary tumors from male patients in the TCGA. We show that LOY rates vary by tumor type and provide evidence for LOY being either a passenger or driver event depending on context. LOY in uveal melanoma specifically is associated with age and survival and is an independent predictor of poor outcome. LOY creates common dependencies on DDX3X and EIF1AX in male cell lines, suggesting that LOY generates unique vulnerabilities that could be therapeutically exploited.

Paralog-based synthetic lethality: rationales and applications

Tumor cells can result from gene mutations and over-expression. Synthetic lethality (SL) offers a desirable setting where cancer cells bearing one mutated gene of an SL gene pair can be specifically targeted by disrupting the function of the other genes, while leaving wide-type normal cells unharmed. Paralogs, a set of homologous genes that have diverged from each other as a consequence of gene duplication, make the concept of SL feasible as the loss of one gene does not affect the cell's survival. Furthermore, homozygous loss of paralogs in tumor cells is more frequent than singletons, making them ideal SL targets. Although high-throughput CRISPR-Cas9 screenings have uncovered numerous paralog-based SL pairs, the unclear mechanisms of targeting these gene pairs and the difficulty in finding specific inhibitors that exclusively target a single but not both paralogs hinder further clinical development. Here, we review the potential mechanisms of paralog-based SL given their function and genetic combination, and discuss the challenge and application prospects of paralog-based SL in cancer therapeutic discovery.

Analyses of Genes Critical to Tumor Survival Reveal Potential 'Supertargets': Focus on Transcription

The identification of mechanisms that underlie the biology of individual tumors is aimed at the development of personalized treatment strategies. Herein, we performed a comprehensive search of genes (termed Supertargets) vital for tumors of particular tissue origin. In so doing, we used the DepMap database portal that encompasses a broad panel of cell lines with individual genes knocked out by CRISPR/Cas9 technology. For each of the 27 tumor types, we revealed the top five genes whose deletion was lethal in the particular case, indicating both known and unknown Supertargets. Most importantly, the majority of Supertargets (41%) were represented by DNA-binding transcription factors. RNAseq data analysis demonstrated that a subset of Supertargets was deregulated in clinical tumor samples but not in the respective non-malignant tissues. These results point to transcriptional mechanisms as key regulators of cell survival in specific tumors. Targeted inactivation of these factors emerges as a straightforward approach to optimize therapeutic regimens.

Targeting synthetic lethal paralogs in cancer

Synthetic lethal interactions, where mutation of one gene renders cells sensitive to inhibition of another gene, can be exploited for the development of targeted therapeutics in cancer. Pairs of duplicate genes (paralogs) often share common functionality and hence are a potentially rich source of synthetic lethal interactions. Because the majority of human genes have paralogs, exploiting such interactions could be a widely applicable approach for targeting gene loss in cancer. Moreover, existing small-molecule drugs may exploit synthetic lethal interactions by inhibiting multiple paralogs simultaneously. Consequently, the identification of synthetic lethal interactions between paralogs could be extremely informative for drug development. Here we review approaches to identify such interactions and discuss some of the challenges of exploiting them.

Specialized Ribosomes in Health and Disease

Ribosomal heterogeneity exists within cells and between different cell types, at specific developmental stages, and occurs in response to environmental stimuli. Mounting evidence supports the existence of specialized ribosomes, or specific changes to the ribosome that regulate the translation of a specific group of transcripts. These alterations have been shown to affect the affinity of ribosomes for certain mRNAs or change the cotranslational folding of nascent polypeptides at the exit tunnel. The identification of specialized ribosomes requires evidence of the incorporation of different ribosomal proteins or of modifications to rRNA and/or protein that lead(s) to physiologically relevant changes in translation. In this review, we summarize ribosomal heterogeneity and specialization in mammals and discuss their relevance to several human diseases.

Concerning the eXclusion in human genomics: The choice of sex chromosome representation in the human genome drastically affects number of identified variants

Over the past 30 years, a community of scientists have pieced together every base pair of the human reference genome from telomere-to-telomere. Interestingly, most human genomics studies omit more than 5% of the genome from their analyses. Under 'normal' circumstances, omitting any chromosome(s) from analysis of the human genome would be reason for concern-the exception being the sex chromosomes. Sex chromosomes in eutherians share an evolutionary origin as an ancestral pair of autosomes. In humans, they share three regions of high sequence identity (~98-100%), which-along with the unique transmission patterns of the sex chromosomes-introduce technical artifacts into genomic analyses. However, the human X chromosome bears numerous important genes-including more "immune response" genes than any other chromosome-which makes its exclusion irresponsible when sex differences across human diseases are widespread. To better characterize the effect that including/excluding the X chromosome may have on variants called, we conducted a pilot study on the Terra cloud platform to replicate a subset of standard genomic practices using both the CHM13 reference genome and sex chromosome complement-aware (SCC-aware) reference genome. We compared quality of variant calling, expression quantification, and allele-specific expression using these two reference genome versions across 50 human samples from the Genotype-Tissue-Expression consortium annotated as females. We found that after correction, the whole X chromosome (100%) can generate reliable variant calls-allowing for the inclusion of the whole genome in human genomics analyses as a departure from the status quo of omitting the sex chromosomes from empirical and clinical genomics studies.

Synthetic lethal interactions of DEAD/H-box helicases as targets for cancer therapy

DEAD/H-box helicases are implicated in virtually every aspect of RNA metabolism, including transcription, pre-mRNA splicing, ribosomes biogenesis, nuclear export, translation initiation, RNA degradation, and mRNA editing. Most of these helicases are upregulated in various cancers and mutations in some of them are associated with several malignancies. Lately, synthetic lethality (SL) and synthetic dosage lethality (SDL) approaches, where genetic interactions of cancer-related genes are exploited as therapeutic targets, are emerging as a leading area of cancer research. Several DEAD/H-box helicases, including DDX3, DDX9 (Dbp9), DDX10 (Dbp4), DDX11 (ChlR1), and DDX41 (Sacy-1), have been subjected to SL analyses in humans and different model organisms. It remains to be explored whether SDL can be utilized to identity druggable targets in DEAD/H-box helicase overexpressing cancers. In this review, we analyze gene expression data of a subset of DEAD/H-box helicases in multiple cancer types and discuss how their SL/SDL interactions can be used for therapeutic purposes. We also summarize the latest developments in clinical applications, apart from discussing some of the challenges in drug discovery in the context of targeting DEAD/H-box helicases.

Functional genomic analysis of adult and pediatric brain tumor isolates

Background

Adult and pediatric tumors display stark differences in their mutation spectra and chromosome alterations. Here, we attempted to identify common and unique gene dependencies and their associated biomarkers among adult and pediatric tumor isolates using functional genetic lethal screens and computational modeling.

Methods

We performed CRISRP-Cas9 lethality screens in two adult glioblastoma (GBM) tumor isolates and five pediatric brain tumor isolates representing atypical teratoid rhabdoid tumors (ATRT), diffuse intrinsic pontine glioma, GBM, and medulloblastoma. We then integrated the screen results with machine learning-based gene-dependency models generated from data from >900 cancer cell lines.

Results

We found that >50% of candidate dependencies of 280 identified were shared between adult GBM tumors and individual pediatric tumor isolates. 68% of screen hits were found as nodes in our network models, along with shared and tumor-specific predictors of gene dependencies. We investigated network predictors associated with ADAR, EFR3A, FGFR1 (pediatric-specific), and SMARCC2 (ATRT-specific) gene dependency among our tumor isolates.

Conclusions

The results suggest that, despite harboring disparate genomic signatures, adult and pediatric tumor isolates share a preponderance of genetic dependences. Further, combining data from primary brain tumor lethality screens with large cancer cell line datasets produced valuable insights into biomarkers of gene dependency, even for rare cancers.

Importance of the Study

Our results demonstrate that large cancer cell lines data sets can be computationally mined to identify known and novel gene dependency relationships in adult and pediatric human brain tumor isolates. Gene dependency networks and lethality screen results represent a key resource for neuro-oncology and cancer research communities. We also highlight some of the challenges and limitations of this approach.

The RNA helicase DDX3 and its role in c-MYC driven germinal center-derived B-cell lymphoma

DDX3X is an RNA helicase with many functions in RNA metabolism such as mRNA translation, alternative pre-mRNA splicing and mRNA stability, but also plays a role as a regulator of transcription as well as in the Wnt/beta-catenin- and Nf-κB signaling pathways. The gene encoding DDX3X is located on the X-chromosome, but escapes X-inactivation. Hence females have two active copies and males only one. However, the Y chromosome contains the gene for the male DDX3 homologue, called DDX3Y, which has a very high sequence similarity and functional redundancy with DDX3X, but shows a more restricted protein expression pattern than DDX3X. High throughput sequencing of germinal center (GC)-derived B-cell malignancies such as Burkitt Lymphoma (BL) and Diffuse large B-cell lymphoma (DLBCL) samples showed a high frequency of loss-of-function (LOF) mutations in the DDX3X gene revealing several features that distinguish this gene from others. First, DDX3X mutations occur with high frequency particularly in those GC-derived B-cell lymphomas that also show translocations of the c-MYC proto-oncogene, which occurs in almost all BL and a subset of DLBCL. Second, DDX3X LOF mutations occur almost exclusively in males and is very rarely found in females. Third, mutations in the male homologue DDX3Y have never been found in any type of malignancy. Studies with human primary GC B cells from male donors showed that a loss of DDX3X function helps the initial process of B-cell lymphomagenesis by buffering the proteotoxic stress induced by c-MYC activation. However, full lymphomagenesis requires DDX3 activity since an upregulation of DDX3Y expression is invariably found in GC derived B-cell lymphoma with DDX3X LOF mutation. Other studies with male transgenic mice that lack Ddx3x, but constitutively express activated c-Myc transgenes in B cells and are therefore prone to develop B-cell malignancies, also showed upregulation of the DDX3Y protein expression during the process of lymphomagenesis. Since DDX3Y is not expressed in normal human cells, these data suggest that DDX3Y may represent a new cancer cell specific target to develop adjuvant therapies for male patients with BL and DLBCL and LOF mutations in the DDX3X gene.

Paralog Explorer: A resource for mining information about paralogs in common research organisms

Paralogs are genes which arose via gene duplication, and when such paralogs retain overlapping or redundant function, this poses a challenge to functional genetics research. Recent technological advancements have made it possible to systematically probe gene function for redundant genes using dual or multiplex gene perturbation, and there is a need for a simple bioinformatic tool to identify putative paralogs of a gene(s) of interest. We have developed Paralog Explorer (https://www.flyrnai.org/tools/paralogs/), an online resource that allows researchers to quickly and accurately identify candidate paralogous genes in the genomes of the model organisms D. melanogaster, C. elegans, D. rerio, M. musculus, and H. sapiens. Paralog Explorer deploys an effective between-species ortholog prediction software, DIOPT, to analyze within-species paralogs. Paralog Explorer allows users to identify candidate paralogs, and to navigate relevant databases regarding gene co-expression, protein-protein and genetic interaction, as well as gene ontology and phenotype annotations. Altogether, this tool extends the value of current ortholog prediction resources by providing sophisticated features useful for identification and study of paralogous genes.

Somatic XIST activation and features of X chromosome inactivation in male human cancers

Expression of the non-coding RNA XIST is essential for initiating X chromosome inactivation (XCI) during early development in female mammals. As the main function of XCI is to enable dosage compensation of chromosome X genes between the sexes, XCI and XIST expression are generally absent in male normal tissues, except in germ cells and in individuals with supernumerary X chromosomes. Via a systematic analysis of public sequencing data of both cancerous and normal tissues, we report that XIST is somatically activated in a subset of male human cancers across diverse lineages. Some of these cancers display hallmarks of XCI, including silencing of gene expression, reduced chromatin accessibility, and increased DNA methylation across chromosome X, suggesting that the developmentally restricted, female-specific program of XCI can be somatically accessed in male cancers.

Discovery of potent and selective HER2 inhibitors with efficacy against HER2 exon 20 insertion-driven tumors, which preserve wild-type EGFR signaling

Oncogenic alterations in human epidermal growth factor receptor 2 (HER2) occur in approximately 2% of patients with non-small cell lung cancer and predominantly affect the tyrosine kinase domain and cluster in exon 20 of the ERBB2 gene. Most clinical-grade tyrosine kinase inhibitors are limited by either insufficient selectivity against wild-type (WT) epidermal growth factor receptor (EGFR), which is a major cause of dose-limiting toxicity or by potency against HER2 exon 20 mutant variants. Here we report the discovery of covalent tyrosine kinase inhibitors that potently inhibit HER2 exon 20 mutants while sparing WT EGFR, which reduce tumor cell survival and proliferation in vitro and result in regressions in preclinical xenograft models of HER2 exon 20 mutant non-small cell lung cancer, concomitant with inhibition of downstream HER2 signaling. Our results suggest that HER2 exon 20 insertion-driven tumors can be effectively treated by a potent and highly selective HER2 inhibitor while sparing WT EGFR, paving the way for clinical translation.