A Suite of Foundation Models Captures the Contextual Interplay Between Codons.

In the canonical genetic code, many amino acids are assigned more than one codon. Work by us and others has shown that the choice of these synonymous codon is not random, and carries regulatory and functional consequences. Existing protein foundation models ignore this context-dependent role of coding sequence in shaping the protein landscape of the cell. To address this gap, we introduce cdsFM, a suite of codon-resolution large language models, including both EnCodon and DeCodon models, with up to 1B parameters. Pre-trained on 60 million protein-coding sequences from more than 5,000 species, our models effectively learn the relationship between codons and amino acids, recapitualing the overall structure of the genetic code. In addition to outperforming state-of-the-art genomic foundation models in a variety of zero-shot and few-shot learning tasks, the larger pre-trained models were superior in predicting the choice of synonymous codons. To systematically assess the impact of synonymous codon choices on protein expression and our models' ability to capture these effects, we generated a large dataset measuring overall and surface expression levels of three proteins as a function of changes in their synonymous codons. We showed that our EnCodon models could be readily fine-tuned to predict the contextual consequences of synonymous codon choices. Armed with this knowledge, we applied EnCodon to existing clinical datasets of synonymous variants, and we identified a large number of synonymous codons that are likely pathogenic, several of which we experimentally confirmed in a cell-based model. Together, our findings establish the cdsFM suite as a powerful tool for decoding the complex functional grammar underlying the choice of synonymous codons.

Tumor Cell Spatial Organization Directs EGFR/RAS/RAF Pathway Primary Therapy Resistance through YAP Signaling.

Non-small cell lung cancers (NSCLC) harboring common mutations in EGFR and KRAS characteristically respond transiently to targeted therapies against those mutations, but invariably, tumors recur and progress. Resistance often emerges through mutations in the therapeutic target or activation of alternative signaling pathways. Mechanisms of acute tumor cell resistance to initial EGFR (EGFRi) or KRASG12C (G12Ci) pathway inhibition remain poorly understood. Our study reveals that acute response to EGFR/RAS/RAF-pathway inhibition is spatial and culture context specific. In vivo, EGFR mutant tumor xenografts shrink by > 90% following acute EGFRi therapy, and residual tumor cells are associated with dense stroma and have increased nuclear YAP. Interestingly, in vitro EGFRi induced cell cycle arrest in NSCLC cells grown in monolayer, while 3D spheroids preferentially die upon inhibitor treatment. We find differential YAP nuclear localization and activity, driven by the distinct culture conditions, as a common resistance mechanism for selective EGFR/KRAS/BRAF pathway therapies. Forced expression of the YAPS127A mutant partially protects cells from EGFR-mediated cell death in spheroid culture. These studies identify YAP activation in monolayer culture as a non-genetic mechanism of acute EGFR/KRAS/BRAF therapy resistance, highlighting that monolayer vs spheroid cell culture systems can model distinct stages of patient cancer progression.

Functional screen for mediators of onco-mRNA translation specificity.

Oncogenic protein dosage is tightly regulated to enable cancer cells to adapt and survive. Whether this is regulated at the level of translational control and the key factors in cis and trans remain unknown. The Myc oncogene is a central paradigm of an exquisitely regulated oncogene and a major driver of pancreatic ductal adenocarcinoma (PDAC). Using a functional genome-wide CRISPRi screen in PDAC cells, we identified activators of selective MYC translation through its 5' untranslated region (5'UTR) and validated four RNA binding proteins (RBPs), including epitranscriptome modifiers. Among these RBPs, our top hit was RBM42, which is highly expressed in PDAC and predicts poor survival. Combining polysome sequencing and CLIP-seq analyses, we find that RBM42 binds and selectively regulates the translation of MYC and a precise, yet vital suite of pro-oncogenic transcripts, including JUN and EGFR . Mechanistically, employing IP-mass spectrometry analysis, we find that RMB42 is a novel ribosome-associated protein (RAP). Using DMS-Seq and mutagenesis analysis, we show that RBM42 directly binds and remodels the MYC 5'UTR RNA structure, facilitating the formation of the translation pre-initiation complex. Importantly, RBM42 is necessary for human PDAC cell growth and fitness and PDAC tumorigenesis in xenograft mouse models in a Myc-dependent manner in vivo . In PDAC patient samples, RBM42 expression is correlated with Myc protein levels and transcriptional activity. This work transforms our understanding of the translational code in cancer and offers a new therapeutic opening to target the expression of oncogenes.

pyPAGE: A framework for Addressing biases in gene-set enrichment analysis-A case study on Alzheimer's disease.

Inferring the driving regulatory programs from comparative analysis of gene expression data is a cornerstone of systems biology. Many computational frameworks were developed to address this problem, including our iPAGE (information-theoretic Pathway Analysis of Gene Expression) toolset that uses information theory to detect non-random patterns of expression associated with given pathways or regulons. Our recent observations, however, indicate that existing approaches are susceptible to the technical biases that are inherent to most real world annotations. To address this, we have extended our information-theoretic framework to account for specific biases and artifacts in biological networks using the concept of conditional information. To showcase pyPAGE, we performed a comprehensive analysis of regulatory perturbations that underlie the molecular etiology of Alzheimer's disease (AD). pyPAGE successfully recapitulated several known AD-associated gene expression programs. We also discovered several additional regulons whose differential activity is significantly associated with AD. We further explored how these regulators relate to pathological processes in AD through cell-type specific analysis of single cell and spatial gene expression datasets. Our findings showcase the utility of pyPAGE as a precise and reliable biomarker discovery in complex diseases such as Alzheimer's disease.

Learning chemical sensitivity reveals mechanisms of cellular response.

Chemical probes interrogate disease mechanisms at the molecular level by linking genetic changes to observable traits. However, comprehensive chemical screens in diverse biological models are impractical. To address this challenge, we develop ChemProbe, a model that predicts cellular sensitivity to hundreds of molecular probes and drugs by learning to combine transcriptomes and chemical structures. Using ChemProbe, we infer the chemical sensitivity of cancer cell lines and tumor samples and analyze how the model makes predictions. We retrospectively evaluate drug response predictions for precision breast cancer treatment and prospectively validate chemical sensitivity predictions in new cellular models, including a genetically modified cell line. Our model interpretation analysis identifies transcriptome features reflecting compound targets and protein network modules, identifying genes that drive ferroptosis. ChemProbe is an interpretable in silico screening tool that allows researchers to measure cellular response to diverse compounds, facilitating research into molecular mechanisms of chemical sensitivity.

Integrative identification of non-coding regulatory regions driving metastatic prostate cancer.

Large-scale sequencing efforts have been undertaken to understand the mutational landscape of the coding genome. However, the vast majority of variants occur within non-coding genomic regions. We designed an integrative computational and experimental framework to identify recurrently mutated non-coding regulatory regions that drive tumor progression. Applying this framework to sequencing data from a large prostate cancer patient cohort revealed a large set of candidate drivers. We used (1) in silico analyses, (2) massively parallel reporter assays, and (3) in vivo CRISPR interference screens to systematically validate metastatic castration-resistant prostate cancer (mCRPC) drivers. One identified enhancer region, GH22I030351, acts on a bidirectional promoter to simultaneously modulate expression of the U2-associated splicing factor SF3A1 and chromosomal protein CCDC157. SF3A1 and CCDC157 promote tumor growth in vivo. We nominated a number of transcription factors, notably SOX6, to regulate expression of SF3A1 and CCDC157. Our integrative approach enables the systematic detection of non-coding regulatory regions that drive human cancers.

Systematic identification of post-transcriptional regulatory modules.

In our cells, a limited number of RNA binding proteins (RBPs) are responsible for all aspects of RNA metabolism across the entire transcriptome. To accomplish this, RBPs form regulatory units that act on specific target regulons. However, the landscape of RBP combinatorial interactions remains poorly explored. Here, we perform a systematic annotation of RBP combinatorial interactions via multimodal data integration. We build a large-scale map of RBP protein neighborhoods by generating in vivo proximity-dependent biotinylation datasets of 50 human RBPs. In parallel, we use CRISPR interference with single-cell readout to capture transcriptomic changes upon RBP knockdowns. By combining these physical and functional interaction readouts, along with the atlas of RBP mRNA targets from eCLIP assays, we generate an integrated map of functional RBP interactions. We then use this map to match RBPs to their context-specific functions and validate the predicted functions biochemically for four RBPs. This study provides a detailed map of RBP interactions and deconvolves them into distinct regulatory modules with annotated functions and target regulons. This multimodal and integrative framework provides a principled approach for studying post-transcriptional regulatory processes and enriches our understanding of their underlying mechanisms.

RPL22 is a tumor suppressor in MSI-high cancers and a splicing regulator of MDM4.

Microsatellite instability-high (MSI-H) tumors are malignant tumors that, despite harboring a high mutational burden, often have intact TP53. One of the most frequent mutations in MSI-H tumors is a frameshift mutation in RPL22, a ribosomal protein. Here, we identified RPL22 as a modulator of MDM4 splicing through an alternative splicing switch in exon 6. RPL22 loss increases MDM4 exon 6 inclusion and cell proliferation and augments resistance to the MDM inhibitor Nutlin-3a. RPL22 represses the expression of its paralog, RPL22L1, by mediating the splicing of a cryptic exon corresponding to a truncated transcript. Therefore, damaging mutations in RPL22 drive oncogenic MDM4 induction and reveal a common splicing circuit in MSI-H tumors that may inform therapeutic targeting of the MDM4-p53 axis and oncogenic RPL22L1 induction.

Glycosphingolipid synthesis mediates immune evasion in KRAS-driven cancer.

Cancer cells frequently alter their lipids to grow and adapt to their environment1-3. Despite the critical functions of lipid metabolism in membrane physiology, signalling and energy production, how specific lipids contribute to tumorigenesis remains incompletely understood. Here, using functional genomics and lipidomic approaches, we identified de novo sphingolipid synthesis as an essential pathway for cancer immune evasion. Synthesis of sphingolipids is surprisingly dispensable for cancer cell proliferation in culture or in immunodeficient mice but required for tumour growth in multiple syngeneic models. Blocking sphingolipid production in cancer cells enhances the anti-proliferative effects of natural killer and CD8+ T cells partly via interferon-γ (IFNγ) signalling. Mechanistically, depletion of glycosphingolipids increases surface levels of IFNγ receptor subunit 1 (IFNGR1), which mediates IFNγ-induced growth arrest and pro-inflammatory signalling. Finally, pharmacological inhibition of glycosphingolipid synthesis synergizes with checkpoint blockade therapy to enhance anti-tumour immune response. Altogether, our work identifies glycosphingolipids as necessary and limiting metabolites for cancer immune evasion.

Identifying a gene signature of metastatic potential by linking pre-metastatic state to ultimate metastatic fate.

Identifying the key molecular pathways that enable metastasis by analyzing the eventual metastatic tumor is challenging because the state of the founder subclone likely changes following metastatic colonization. To address this challenge, we labeled primary mouse pancreatic ductal adenocarcinoma (PDAC) subclones with DNA barcodes to characterize their pre-metastatic state using ATAC-seq and RNA-seq and determine their relative in vivo metastatic potential prospectively. We identified a gene signature separating metastasis-high and metastasis-low subclones orthogonal to the normal-to-PDAC and classical-to-basal axes. The metastasis-high subclones feature activation of IL-1 pathway genes and high NF-κB and Zeb/Snail family activity and the metastasis-low subclones feature activation of neuroendocrine, motility, and Wnt pathway genes and high CDX2 and HOXA13 activity. In a functional screen, we validated novel mediators of PDAC metastasis in the IL-1 pathway, including the NF-κB targets Fos and Il23a, and beyond the IL-1 pathway including Myo1b and Tmem40. We scored human PDAC tumors for our signature of metastatic potential from mouse and found that metastases have higher scores than primary tumors. Moreover, primary tumors with higher scores are associated with worse prognosis. We also found that our metastatic potential signature is enriched in other human carcinomas, suggesting that it is conserved across epithelial malignancies. This work establishes a strategy for linking cancer cell state to future behavior, reveals novel functional regulators of PDAC metastasis, and establishes a method for scoring human carcinomas based on metastatic potential.

A systematic search for RNA structural switches across the human transcriptome.

RNA structural switches are key regulators of gene expression in bacteria, but their characterization in Metazoa remains limited. Here, we present SwitchSeeker, a comprehensive computational and experimental approach for systematic identification of functional RNA structural switches. We applied SwitchSeeker to the human transcriptome and identified 245 putative RNA switches. To validate our approach, we characterized a previously unknown RNA switch in the 3' untranslated region of the RORC (RAR-related orphan receptor C) transcript. In vivo dimethyl sulfate (DMS) mutational profiling with sequencing (DMS-MaPseq), coupled with cryogenic electron microscopy, confirmed its existence as two alternative structural conformations. Furthermore, we used genome-scale CRISPR screens to identify trans factors that regulate gene expression through this RNA structural switch. We found that nonsense-mediated messenger RNA decay acts on this element in a conformation-specific manner. SwitchSeeker provides an unbiased, experimentally driven method for discovering RNA structural switches that shape the eukaryotic gene expression landscape.

High Throughput Variant Libraries and Machine Learning Yield Design Rules for Retron Gene Editors.

The bacterial retron reverse transcriptase system has served as an intracellular factory for single-stranded DNA in many biotechnological applications. In these technologies, a natural retron non-coding RNA (ncRNA) is modified to encode a template for the production of custom DNA sequences by reverse transcription. The efficiency of reverse transcription is a major limiting step for retron technologies, but we lack systematic knowledge of how to improve or maintain reverse transcription efficiency while changing the retron sequence for custom DNA production. Here, we test thousands of different modifications to the retron-Eco1 ncRNA and measure DNA production in pooled variant library experiments, identifying regions of the ncRNA that are tolerant and intolerant to modification. We apply this new information to a specific application: the use of the retron to produce a precise genome editing donor in combination with a CRISPR-Cas9 RNA-guided nuclease (an editron). We use high-throughput libraries in S. cerevisiae to additionally define design rules for editrons. We extend our new knowledge of retron DNA production and editron design rules to human genome editing to achieve the highest efficiency retron-Eco1 editrons to date.

Accurate isoform quantification by joint short- and long-read RNA-sequencing.

Accurate quantification of transcript isoforms is crucial for understanding gene regulation, functional diversity, and cellular behavior. Existing RNA sequencing methods have significant limitations: short-read (SR) sequencing provides high depth but struggles with isoform deconvolution, whereas long-read (LR) sequencing offers isoform resolution at the cost of lower depth, higher noise, and technical biases. Addressing this gap, we introduce Multi-Platform Aggregation and Quantification of Transcripts (MPAQT), a generative model that combines the complementary strengths of different sequencing platforms to achieve state-of-the-art isoform-resolved transcript quantification, as demonstrated by extensive simulations and experimental benchmarks. By applying MPAQT to an in vitro model of human embryonic stem cell differentiation into cortical neurons, followed by machine learning-based modeling of transcript abundances, we show that untranslated regions (UTRs) are major determinants of isoform proportion and exon usage; this effect is mediated through isoform-specific sequence features embedded in UTRs, which likely interact with RNA-binding proteins that modulate mRNA stability. These findings highlight MPAQT's potential to enhance our understanding of transcriptomic complexity and underline the role of splicing-independent post-transcriptional mechanisms in shaping the isoform and exon usage landscape of the cell.

Focal adhesion kinase-YAP signaling axis drives drug-tolerant persister cells and residual disease in lung cancer.

Targeted therapy is effective in many tumor types including lung cancer, the leading cause of cancer mortality. Paradigm defining examples are targeted therapies directed against non-small cell lung cancer (NSCLC) subtypes with oncogenic alterations in EGFR, ALK and KRAS. The success of targeted therapy is limited by drug-tolerant persister cells (DTPs) which withstand and adapt to treatment and comprise the residual disease state that is typical during treatment with clinical targeted therapies. Here, we integrate studies in patient-derived and immunocompetent lung cancer models and clinical specimens obtained from patients on targeted therapy to uncover a focal adhesion kinase (FAK)-YAP signaling axis that promotes residual disease during oncogenic EGFR-, ALK-, and KRAS-targeted therapies. FAK-YAP signaling inhibition combined with the primary targeted therapy suppressed residual drug-tolerant cells and enhanced tumor responses. This study unveils a FAK-YAP signaling module that promotes residual disease in lung cancer and mechanism-based therapeutic strategies to improve tumor response.

Direct transposition of native DNA for sensitive multimodal single-molecule sequencing.

Concurrent readout of sequence and base modifications from long unamplified DNA templates by Pacific Biosciences of California (PacBio) single-molecule sequencing requires large amounts of input material. Here we adapt Tn5 transposition to introduce hairpin oligonucleotides and fragment (tagment) limiting quantities of DNA for generating PacBio-compatible circular molecules. We developed two methods that implement tagmentation and use 90-99% less input than current protocols: (1) single-molecule real-time sequencing by tagmentation (SMRT-Tag), which allows detection of genetic variation and CpG methylation; and (2) single-molecule adenine-methylated oligonucleosome sequencing assay by tagmentation (SAMOSA-Tag), which uses exogenous adenine methylation to add a third channel for probing chromatin accessibility. SMRT-Tag of 40 ng or more human DNA (approximately 7,000 cell equivalents) yielded data comparable to gold standard whole-genome and bisulfite sequencing. SAMOSA-Tag of 30,000-50,000 nuclei resolved single-fiber chromatin structure, CTCF binding and DNA methylation in patient-derived prostate cancer xenografts and uncovered metastasis-associated global epigenome disorganization. Tagmentation thus promises to enable sensitive, scalable and multimodal single-molecule genomics for diverse basic and clinical applications.

Stem-cell states converge in multistage cutaneous squamous cell carcinoma development.

Stem cells play a critical role in cancer development by contributing to cell heterogeneity, lineage plasticity, and drug resistance. We created gene expression networks from hundreds of mouse tissue samples (both normal and tumor) and integrated these with lineage tracing and single-cell RNA-seq, to identify convergence of cell states in premalignant tumor cells expressing markers of lineage plasticity and drug resistance. Two of these cell states representing multilineage plasticity or proliferation were inversely correlated, suggesting a mutually exclusive relationship. Treatment of carcinomas in vivo with chemotherapy repressed the proliferative state and activated multilineage plasticity whereas inhibition of differentiation repressed plasticity and potentiated responses to cell cycle inhibitors. Manipulation of this cell state transition point may provide a source of potential combinatorial targets for cancer therapy.

Identification of the extracellular membrane protein ENPP3 as a major cGAMP hydrolase and innate immune checkpoint.

2'3'-Cyclic guanosine monophosphate (GMP)-AMP (cGAMP) is a second messenger synthesized upon detection of cytosolic double-stranded DNA (dsDNA) and passed between cells to facilitate downstream immune signaling. Ectonucleotide pyrophosphatase phosphodiesterase I (ENPP1), an extracellular enzyme, was the only metazoan hydrolase known to regulate cGAMP levels to dampen anti-cancer immunity. Here, we uncover ENPP3 as the second and likely the only other metazoan cGAMP hydrolase under homeostatic conditions. ENPP3 has a tissue expression pattern distinct from ENPP1's and accounts for all cGAMP hydrolysis activity in ENPP1-deficient mice. Importantly, we also show that, as with ENPP1, selectively abolishing ENPP3's cGAMP hydrolysis activity results in diminished cancer growth and metastasis of certain tumor types in a stimulator of interferon genes (STING)-dependent manner. Both ENPP1 and ENPP3 are extracellular enzymes, suggesting the dominant role that extracellular cGAMP must play as a mediator of cell-cell innate immune communication. Our work demonstrates that ENPP1 and ENPP3 non-redundantly dampen extracellular cGAMP-STING signaling, pointing to ENPP3 as a target for cancer immunotherapy.

Systematic annotation of orphan RNAs reveals blood-accessible molecular barcodes of cancer identity and cancer-emergent oncogenic drivers.

From extrachromosomal DNA to neo-peptides, the broad reprogramming of the cancer genome leads to the emergence of molecules that are specific to the cancer state. We recently described orphan non-coding RNAs (oncRNAs) as a class of cancer-specific small RNAs with the potential to play functional roles in breast cancer progression1. Here, we report a systematic and comprehensive search to identify, annotate, and characterize cancer-emergent oncRNAs across 32 tumor types. We also leverage large-scale in vivo genetic screens in xenografted mice to functionally identify driver oncRNAs in multiple tumor types. We have not only discovered a large repertoire of oncRNAs, but also found that their presence and absence represent a digital molecular barcode that faithfully captures the types and subtypes of cancer. Importantly, we discovered that this molecular barcode is partially accessible from the cell-free space as some oncRNAs are secreted by cancer cells. In a large retrospective study across 192 breast cancer patients, we showed that oncRNAs can be reliably detected in the blood and that changes in the cell-free oncRNA burden captures both short-term and long-term clinical outcomes upon completion of a neoadjuvant chemotherapy regimen. Together, our findings establish oncRNAs as an emergent class of cancer-specific non-coding RNAs with potential roles in tumor progression and clinical utility in liquid biopsies and disease monitoring.

Revealing the grammar of small RNA secretion using interpretable machine learning.

Small non-coding RNAs can be secreted through a variety of mechanisms, including exosomal sorting, in small extracellular vesicles, and within lipoprotein complexes. However, the mechanisms that govern their sorting and secretion are not well understood. Here, we present ExoGRU, a machine learning model that predicts small RNA secretion probabilities from primary RNA sequences. We experimentally validated the performance of this model through ExoGRU-guided mutagenesis and synthetic RNA sequence analysis. Additionally, we used ExoGRU to reveal cis and trans factors that underlie small RNA secretion, including known and novel RNA-binding proteins (RBPs), e.g., YBX1, HNRNPA2B1, and RBM24. We also developed a novel technique called exoCLIP, which reveals the RNA interactome of RBPs within the cell-free space. Together, our results demonstrate the power of machine learning in revealing novel biological mechanisms. In addition to providing deeper insight into small RNA secretion, this knowledge can be leveraged in therapeutic and synthetic biology applications.

Translational adaptation in breast cancer metastasis and emerging therapeutic opportunities.

Breast cancer's tendency to metastasize poses a critical barrier to effective treatment, making it a leading cause of mortality among women worldwide. A growing body of evidence is showing that translational adaptation is emerging as a key mechanism enabling cancer cells to thrive in the dynamic tumor microenvironment (TME). Here, we systematically summarize how breast cancer cells utilize translational adaptation to drive metastasis, highlighting the intricate regulation by specific translation machinery and mRNA attributes such as sequences and structures, along with the involvement of tRNAs and other trans-acting RNAs. We provide an overview of the latest findings and emerging concepts in this area, discussing their potential implications for therapeutic strategies in breast cancer.