Genome-scale quantification and prediction of pathogenic stop codon readthrough by small molecules.

Premature termination codons (PTCs) cause ~10-20% of inherited diseases and are a major mechanism of tumor suppressor gene inactivation in cancer. A general strategy to alleviate the effects of PTCs would be to promote translational readthrough. Nonsense suppression by small molecules has proven effective in diverse disease models, but translation into the clinic is hampered by ineffective readthrough of many PTCs. Here we directly tackle the challenge of defining drug efficacy by quantifying the readthrough of ~5,800 human pathogenic stop codons by eight drugs. We find that different drugs promote the readthrough of complementary subsets of PTCs defined by local sequence context. This allows us to build interpretable models that accurately predict drug-induced readthrough genome-wide, and we validate these models by quantifying endogenous stop codon readthrough. Accurate readthrough quantification and prediction will empower clinical trial design and the development of personalized nonsense suppression therapies.

Energetic portrait of the amyloid beta nucleation transition state.

Amyloid protein aggregates are pathological hallmarks of more than fifty human diseases including the most common neurodegenerative disorders. The atomic structures of amyloid fibrils have now been determined, but the process by which soluble proteins nucleate to form amyloids remains poorly characterised and difficult to study, even though this is the key step to understand to prevent the formation and spread of aggregates. Here we use massively parallel combinatorial mutagenesis, a kinetic selection assay, and machine learning to reveal the transition state of the nucleation reaction of amyloid beta, the protein that aggregates in Alzheimer's disease. By quantifying the nucleation of >140,000 proteins we infer the changes in activation energy for all 798 amino acid substitutions in amyloid beta and the energetic couplings between >600 pairs of mutations. This unprecedented dataset provides the first comprehensive view of the energy landscape and the first large-scale measurement of energetic couplings for a protein transition state. The energy landscape reveals that the amyloid beta nucleation transition state contains a short structured C-terminal hydrophobic core with a subset of interactions similar to mature fibrils. This study demonstrates the feasibility of using mutation-selection-sequencing experiments to study transition states and identifies the key molecular species that initiates amyloid beta aggregation and, potentially, Alzheimer's disease.

Interpretably deep learning amyloid nucleation by massive experimental quantification of random sequences.

Insoluble amyloid aggregates are the hallmarks of more than fifty human diseases, including the most common neurodegenerative disorders. The process by which soluble proteins nucleate to form amyloid fibrils is, however, quite poorly characterized. Relatively few sequences are known that form amyloids with high propensity and this data shortage likely limits our capacity to understand, predict, engineer, and prevent the formation of amyloid fibrils. Here we quantify the nucleation of amyloids at an unprecedented scale and use the data to train a deep learning model of amyloid nucleation. In total, we quantify the nucleation rates of >100,000 20-amino-acid-long peptides. This large and diverse dataset allows us to train CANYA, a convolution-attention hybrid neural network. CANYA is fast and outperforms existing methods with stable performance across diverse prediction tasks. Interpretability analyses reveal CANYA's decision-making process and learned grammar, providing mechanistic insights into amyloid nucleation. Our results illustrate the power of massive experimental analysis of random sequence-spaces and provide an interpretable and robust neural network model to predict amyloid nucleation.

A complete map of specificity encoding for a partially fuzzy protein interaction.

Thousands of human proteins function by binding short linear motifs embedded in intrinsically disordered regions. How affinity and specificity are encoded in these binding domains and the motifs themselves is not well understood. The evolvability of binding specificity - how rapidly and extensively it can change upon mutation - is also largely unexplored, as is the contribution of 'fuzzy' dynamic residues to affinity and specificity in protein-protein interactions. Here we report the first complete map of specificity encoding for a globular protein domain. Quantifying >200,000 energetic interactions between a PDZ domain and its ligand identifies 20 major energetically coupled pairs of sites that control specificity. These are organized into six modules, with most mutations in each module reprogramming specificity for a single position in the ligand. Nine of the major energetic couplings controlling specificity are between structural contacts and 11 have an allosteric mechanism of action. The dynamic tail of the ligand is more robust to mutation than the structured residues but contributes additively to binding affinity and communicates with structured residues to enable changes in specificity. Our results quantify the binding specificities of >1,800 globular proteins to reveal how specificity is encoded and provide a direct comparison of the encoding of affinity and specificity in structured and dynamic molecular recognition.

Guidelines for releasing a variant effect predictor.

Computational methods for assessing the likely impacts of mutations, known as variant effect predictors (VEPs), are widely used in the assessment and interpretation of human genetic variation, as well as in other applications like protein engineering. Many different VEPs have been released to date, and there is tremendous variability in their underlying algorithms and outputs, and in the ways in which the methodologies and predictions are shared. This leads to considerable challenges for end users in knowing which VEPs to use and how to use them. Here, to address these issues, we provide guidelines and recommendations for the release of novel VEPs. Emphasising open-source availability, transparent methodologies, clear variant effect score interpretations, standardised scales, accessible predictions, and rigorous training data disclosure, we aim to improve the usability and interpretability of VEPs, and promote their integration into analysis and evaluation pipelines. We also provide a large, categorised list of currently available VEPs, aiming to facilitate the discovery and encourage the usage of novel methods within the scientific community.

An extension of the Walsh-Hadamard transform to calculate and model epistasis in genetic landscapes of arbitrary shape and complexity.

Accurate models describing the relationship between genotype and phenotype are necessary in order to understand and predict how mutations to biological sequences affect the fitness and evolution of living organisms. The apparent abundance of epistasis (genetic interactions), both between and within genes, complicates this task and how to build mechanistic models that incorporate epistatic coefficients (genetic interaction terms) is an open question. The Walsh-Hadamard transform represents a rigorous computational framework for calculating and modeling epistatic interactions at the level of individual genotypic values (known as genetical, biological or physiological epistasis), and can therefore be used to address fundamental questions related to sequence-to-function encodings. However, one of its main limitations is that it can only accommodate two alleles (amino acid or nucleotide states) per sequence position. In this paper we provide an extension of the Walsh-Hadamard transform that allows the calculation and modeling of background-averaged epistasis (also known as ensemble epistasis) in genetic landscapes with an arbitrary number of states per position (20 for amino acids, 4 for nucleotides, etc.). We also provide a recursive formula for the inverse matrix and then derive formulae to directly extract any element of either matrix without having to rely on the computationally intensive task of constructing or inverting large matrices. Finally, we demonstrate the utility of our theory by using it to model epistasis within both simulated and empirical multiallelic fitness landscapes, revealing that both pairwise and higher-order genetic interactions are enriched between physically interacting positions.

The energetic and allosteric landscape for KRAS inhibition.

Thousands of proteins have been validated genetically as therapeutic targets for human diseases1. However, very few have been successfully targeted, and many are considered 'undruggable'. This is particularly true for proteins that function via protein-protein interactions-direct inhibition of binding interfaces is difficult and requires the identification of allosteric sites. However, most proteins have no known allosteric sites, and a comprehensive allosteric map does not exist for any protein. Here we address this shortcoming by charting multiple global atlases of inhibitory allosteric communication in KRAS. We quantified the effects of more than 26,000 mutations on the folding of KRAS and its binding to six interaction partners. Genetic interactions in double mutants enabled us to perform biophysical measurements at scale, inferring more than 22,000 causal free energy changes. These energy landscapes quantify how mutations tune the binding specificity of a signalling protein and map the inhibitory allosteric sites for an important therapeutic target. Allosteric propagation is particularly effective across the central ß-sheet of KRAS, and multiple surface pockets are genetically validated as allosterically active, including a distal pocket in the C-terminal lobe of the protein. Allosteric mutations typically inhibit binding to all tested effectors, but they can also change the binding specificity, revealing the regulatory, evolutionary and therapeutic potential to tune pathway activation. Using the approach described here, it should be possible to rapidly and comprehensively identify allosteric target sites in many proteins.

Dominance vs epistasis: the biophysical origins and plasticity of genetic interactions within and between alleles.

An important challenge in genetics, evolution and biotechnology is to understand and predict how mutations combine to alter phenotypes, including molecular activities, fitness and disease. In diploids, mutations in a gene can combine on the same chromosome or on different chromosomes as a "heteroallelic combination". However, a direct comparison of the extent, sign, and stability of the genetic interactions between variants within and between alleles is lacking. Here we use thermodynamic models of protein folding and ligand-binding to show that interactions between mutations within and between alleles are expected in even very simple biophysical systems. Protein folding alone generates within-allele interactions and a single molecular interaction is sufficient to cause between-allele interactions and dominance. These interactions change differently, quantitatively and qualitatively as a system becomes more complex. Altering the concentration of a ligand can, for example, switch alleles from dominant to recessive. Our results show that intra-molecular epistasis and dominance should be widely expected in even the simplest biological systems but also reinforce the view that they are plastic system properties and so a formidable challenge to predict. Accurate prediction of both intra-molecular epistasis and dominance will require either detailed mechanistic understanding and experimental parameterization or brute-force measurement and learning.

SUNi mutagenesis: Scalable and uniform nicking for efficient generation of variant libraries.

Multiplexed assays of variant effects (MAVEs) have made possible the functional assessment of all possible mutations to genes and regulatory sequences. A core pillar of the approach is generation of variant libraries, but current methods are either difficult to scale or not uniform enough to enable MAVEs at the scale of gene families or beyond. We present an improved method called Scalable and Uniform Nicking (SUNi) mutagenesis that combines massive scalability with high uniformity to enable cost-effective MAVEs of gene families and eventually genomes.

An atlas of amyloid aggregation: the impact of substitutions, insertions, deletions and truncations on amyloid beta fibril nucleation.

Multiplexed assays of variant effects (MAVEs) guide clinical variant interpretation and reveal disease mechanisms. To date, MAVEs have focussed on a single mutation type-amino acid (AA) substitutions-despite the diversity of coding variants that cause disease. Here we use Deep Indel Mutagenesis (DIM) to generate a comprehensive atlas of diverse variant effects for a disease protein, the amyloid beta (Aß) peptide that aggregates in Alzheimer's disease (AD) and is mutated in familial AD (fAD). The atlas identifies known fAD mutations and reveals that many variants beyond substitutions accelerate Aß aggregation and are likely to be pathogenic. Truncations, substitutions, insertions, single- and internal multi-AA deletions differ in their propensity to enhance or impair aggregation, but likely pathogenic variants from all classes are highly enriched in the polar N-terminal region of Aß. This comparative atlas highlights the importance of including diverse mutation types in MAVEs and provides important mechanistic insights into amyloid nucleation.

The impact of rare germline variants on human somatic mutation processes.

Somatic mutations are an inevitable component of ageing and the most important cause of cancer. The rates and types of somatic mutation vary across individuals, but relatively few inherited influences on mutation processes are known. We perform a gene-based rare variant association study with diverse mutational processes, using human cancer genomes from over 11,000 individuals of European ancestry. By combining burden and variance tests, we identify 207 associations involving 15 somatic mutational phenotypes and 42 genes that replicated in an independent data set at a false discovery rate of 1%. We associate rare inherited deleterious variants in genes such as MSH3, EXO1, SETD2, and MTOR with two phenotypically different forms of DNA mismatch repair deficiency, and variants in genes such as EXO1, PAXIP1, RIF1, and WRN with deficiency in homologous recombination repair. In addition, we identify associations with other mutational processes, such as APEX1 with APOBEC-signature mutagenesis. Many of the genes interact with each other and with known mutator genes within cellular sub-networks. Considered collectively, damaging variants in the identified genes are prevalent in the population. We suggest that rare germline variation in diverse genes commonly impacts mutational processes in somatic cells.

Mapping the energetic and allosteric landscapes of protein binding domains.

Allosteric communication between distant sites in proteins is central to biological regulation but still poorly characterized, limiting understanding, engineering and drug development1-6. An important reason for this is the lack of methods to comprehensively quantify allostery in diverse proteins. Here we address this shortcoming and present a method that uses deep mutational scanning to globally map allostery. The approach uses an efficient experimental design to infer en masse the causal biophysical effects of mutations by quantifying multiple molecular phenotypes-here we examine binding and protein abundance-in multiple genetic backgrounds and fitting thermodynamic models using neural networks. We apply the approach to two of the most common protein interaction domains found in humans, an SH3 domain and a PDZ domain, to produce comprehensive atlases of allosteric communication. Allosteric mutations are abundant, with a large mutational target space of network-altering 'edgetic' variants. Mutations are more likely to be allosteric closer to binding interfaces, at glycine residues and at specific residues connecting to an opposite surface within the PDZ domain. This general approach of quantifying mutational effects for multiple molecular phenotypes and in multiple genetic backgrounds should enable the energetic and allosteric landscapes of many proteins to be rapidly and comprehensively mapped.

Higher order genetic interactions switch cancer genes from two-hit to one-hit drivers.

The classic two-hit model posits that both alleles of a tumor suppressor gene (TSG) must be inactivated to cause cancer. In contrast, for some oncogenes and haploinsufficient TSGs, a single genetic alteration can suffice to increase tumor fitness. Here, by quantifying the interactions between mutations and copy number alterations (CNAs) across 10,000 tumors, we show that many cancer genes actually switch between acting as one-hit or two-hit drivers. Third order genetic interactions identify the causes of some of these switches in dominance and dosage sensitivity as mutations in other genes in the same biological pathway. The correct genetic model for a gene thus depends on the other mutations in a genome, with a second hit in the same gene or an alteration in a different gene in the same pathway sometimes representing alternative evolutionary paths to cancer.

Neuronal perception of the social environment generates an inherited memory that controls the development and generation time of C. elegans.

An old and controversial question in biology is whether information perceived by the nervous system of an animal can "cross the Weismann barrier" to alter the phenotypes and fitness of their progeny. Here, we show that such intergenerational transmission of sensory information occurs in the model organism, C. elegans, with a major effect on fitness. Specifically, that perception of social pheromones by chemosensory neurons controls the post-embryonic timing of the development of one tissue, the germline, relative to others in the progeny of an animal. Neuronal perception of the social environment thus intergenerationally controls the generation time of this animal.

The genetic landscape for amyloid beta fibril nucleation accurately discriminates familial Alzheimer's disease mutations.

Plaques of the amyloid beta (Aß) peptide are a pathological hallmark of Alzheimer's disease (AD), the most common form of dementia. Mutations in Aß also cause familial forms of AD (fAD). Here, we use deep mutational scanning to quantify the effects of >14,000 mutations on the aggregation of Aß. The resulting genetic landscape reveals mechanistic insights into fibril nucleation, including the importance of charge and gatekeeper residues in the disordered region outside of the amyloid core in preventing nucleation. Strikingly, unlike computational predictors and previous measurements, the empirical nucleation scores accurately identify all known dominant fAD mutations in Aß, genetically validating that the mechanism of nucleation in a cell-based assay is likely to be very similar to the mechanism that causes the human disease. These results provide the first comprehensive atlas of how mutations alter the formation of any amyloid fibril and a resource for the interpretation of genetic variation in Aß.

Alzheimer's disease is the most common form of dementia, affecting more than 50 million people worldwide. Despite more than 400 clinical trials, there are still no effective drugs that can prevent or treat the disease. A common target in Alzheimer's disease trials is a small protein called amyloid beta. Amyloid beta proteins are 'sticky' molecules. In the brains of people with Alzheimer's disease, they join to form first small aggregates and then long chains called fibrils, a process which is toxic to neurons. Specific mutations in the gene for amyloid beta are known to cause rare, aggressive forms of Alzheimer's disease that typically affect people in their fifties or sixties. But these are not the only mutations that can occur in amyloid beta. In principle, any part of the protein could undergo mutation. And given the size of the human population, it is likely that each of these mutations exists in someone alive today. Seuma et al. reasoned that studying these mutations could help us understand the process by which amyloid beta forms new aggregates. Using an approach called deep mutational scanning, Seuma et al. mutated each point in the protein, one at a time. This produced more than 14,000 different versions of amyloid beta. Seuma et al. then measured how quickly these mutants were able to form aggregates by introducing them into yeast cells. All the mutations known to cause early-onset Alzheimer's disease accelerated amyloid beta aggregation in the yeast. But the results also revealed previously unknown properties that control how fast aggregation occurs. In addition, they highlighted a number of positions in the amyloid beta sequence that act as 'gatekeepers'. In healthy brains, these gatekeepers prevent amyloid beta proteins from sticking together. When mutated, they drive the protein to form aggregates. This comprehensive dataset will help researchers understand how proteins form toxic aggregates, which could in turn help them find ways to prevent this from happening. By providing an 'atlas' of all possible amyloid beta mutations, the dataset will also help clinicians interpret any new mutations they encounter in patients. By showing whether or not a mutation speeds up aggregation, the atlas will help clinicians predict whether that mutation increases the risk of Alzheimer's disease.

To NMD or Not To NMD: Nonsense-Mediated mRNA Decay in Cancer and Other Genetic Diseases.

The nonsense-mediated mRNA decay (NMD) pathway degrades some but not all mRNAs bearing premature termination codons (PTCs). Decades of work have elucidated the molecular mechanisms of NMD. More recently, statistical analyses of large genomic datasets have allowed the importance of known and novel 'rules of NMD' to be tested and combined into methods that accurately predict whether PTC-containing mRNAs are degraded or not. We discuss these genomic approaches and how they can be applied to identify diseases and individuals that may benefit from inhibition or activation of NMD. We also discuss the importance of NMD for gene editing and tumor evolution, and how inhibiting NMD may be an effective strategy to increase the efficacy of cancer immunotherapy.

Mutations primarily alter the inclusion of alternatively spliced exons.

Genetic analyses and systematic mutagenesis have revealed that synonymous, non-synonymous and intronic mutations frequently alter the inclusion levels of alternatively spliced exons, consistent with the concept that altered splicing might be a common mechanism by which mutations cause disease. However, most exons expressed in any cell are highly-included in mature mRNAs. Here, by performing deep mutagenesis of highly-included exons and by analysing the association between genome sequence variation and exon inclusion across the transcriptome, we report that mutations only very rarely alter the inclusion of highly-included exons. This is true for both exonic and intronic mutations as well as for perturbations in trans. Therefore, mutations that affect splicing are not evenly distributed across primary transcripts but are focussed in and around alternatively spliced exons with intermediate inclusion levels. These results provide a resource for prioritising synonymous and other variants as disease-causing mutations.

Biophysical ambiguities prevent accurate genetic prediction.

A goal of biology is to predict how mutations combine to alter phenotypes, fitness and disease. It is often assumed that mutations combine additively or with interactions that can be predicted. Here, we show using simulations that, even for the simple example of the lambda phage transcription factor CI repressing a gene, this assumption is incorrect and that perfect measurements of the effects of mutations on a trait and mechanistic understanding can be insufficient to predict what happens when two mutations are combined. This apparent paradox arises because mutations can have different biophysical effects to cause the same change in a phenotype and the outcome in a double mutant depends upon what these hidden biophysical changes actually are. Pleiotropy and non-monotonic functions further confound prediction of how mutations interact. Accurate prediction of phenotypes and disease will sometimes not be possible unless these biophysical ambiguities can be resolved using additional measurements.

DiMSum: an error model and pipeline for analyzing deep mutational scanning data and diagnosing common experimental pathologies.

Deep mutational scanning (DMS) enables multiplexed measurement of the effects of thousands of variants of proteins, RNAs, and regulatory elements. Here, we present a customizable pipeline, DiMSum, that represents an end-to-end solution for obtaining variant fitness and error estimates from raw sequencing data. A key innovation of DiMSum is the use of an interpretable error model that captures the main sources of variability arising in DMS workflows, outperforming previous methods. DiMSum is available as an R/Bioconda package and provides summary reports to help researchers diagnose common DMS pathologies and take remedial steps in their analyses.