Fran Supek

Whole genome analyses have shown that every cell of the body accumulates thousands of genetic changes during a lifetime, with implications for cancer initiation. We have analyzed normal kidney cells from both healthy individuals and patients with the Von Hippel Lindau (VHL) syndrome, which predisposes to kidney cancer. Using a clonal-expansion method that enables the collection of high confidence genetic and gene expression data from the same cell, we define a new mutational signature that marks a specific subset of damage-activated proximal tubule cells, as well as 96% of clear cell renal cell carcinomas. The underlying mutational process is specifically active in highly transcribed DNA and impacts the occurrence of cancer driver variants in the VHL gene. Our data from normal cells, supported by in vitro experiments, point to perturbations of oxygen availability as a causative factor. These findings link hypoxia, a frequent endogenous stress in the normal renal tissue, to excessive mutation and cancer initiation.

Programmed translational readthrough produces C-terminally extended protein isoforms via decoding of stop codons by near-cognate tRNAs. Human genes experimentally validated as readthrough targets share a CUAG motif downstream of a UGA stop codon. However, the full sequence determinants of readthrough efficiency, how they combine, and how generalisable they are across genes remain largely unexplored. Here we use deep mutational scanning to quantify ∼1,400 sequence variants for each of the three examples of human readthrough in the genes AQP4 , MAPK10 and OPRK1 . In addition to the core CUAG motif, mutations that modulate readthrough elements extend up to +27 nucleotides downstream of the stop codon and across six codons (18 nucleotides) upstream. For the downstream sequence, an additive model with a sigmoidal global epistasis function captures most of the within-gene readthrough variance for double mutants (R²=0.84-0.96), with additional contributions from a small number of strong pairwise interactions. Mutational effects nonetheless generalise poorly between genes: only the immediate -3 to +4 nucleotide window shows consistent behaviour, while mutations in more distal positions have context-dependent effects. Combinatorial assembly of sequence blocks from different genes into chimeras reveals strong interactions (epistasis) between sequences upstream and downstream of the stop codon. This study provides comprehensive quantitative maps of the sequence determinants of human programmed readthrough and suggests that three examples of programmed readthrough are located on distinct local fitness peaks, each defined by different upstream and downstream architectures built around a shared CUAG core motif.

Recurrent somatic mutations reveal cancer drivers, but in whole genomes many non-coding hotspots are passengers generated by localized mutational processes. We developed MutFormer, a transformer/convolutional neural net model that predicts base-pair-resolution somatic mutation risk from DNA sequence alone, separately for COSMIC signatures. Trained on >90 million high-confidence mutational signature-assigned SNVs from cancer genomes, MutFormer learns extended sequence determinants beyond trinucleotide context, often spanning up to ~20 nucleotides, and recovers APOBEC, UV, POLE and SBS17 sequence preferences as well as various additional mutation risk-prone motifs. We integrated MutFormer predictions with mutation burden, signature exposures and epigenetic covariates to model neutral recurrence of individual hotspots in >18,000 tumor whole genomes. Coding-region analyses calibrated the framework against known driver genes and AlphaMissense scores, supporting conservative false-discovery estimates. In non-coding regions, most recurrent hotspots were explained by passenger mutability, whereas selected outliers were enriched near cancer genes and supported by SpliceAI, PromoterAI, AlphaGenome and expression data. Prioritized candidates include splice-region or deep-intronic hotspots in BCL6 , PTEN , TCF7L2 , PBRM1 , PTPRT and VHL , and promoter hotspots in SHKBP1 , PRSS3 and BCL2 .

Mismatch repair deficiency (MMRd) is a central driver of cancer genome evolution, yet it is commonly reduced to a binary MSI-high versus microsatellite-stable call. Here we combine multiplex CRISPR/Cas12a editing, long-term mutation accumulation experiments in human cells, and whole-genome sequencing to map the mutational consequences of single and pairwise perturbations across canonical and accessory MMR genes, together with selected base excision repair and direct-repair factors. Across 72 isogenic lineages representing 58 genotypes, we recover reproducible substitution, indel and short-tandem-repeat mutational states that distinguish MSH2, MSH6, MLH1, PMS2, MSH3 and MLH3 deficiencies. A dedicated STR spectrum reveals genome-wide evidence for MSH3-driven EMAST and enables design of compact loci panels for mechanism-aware MSI classification. Experimental spectra generalize across independent cell systems and to cancer whole genome sequences, where Elastic-net classifiers accurately identify causal MMR genes in Lynch tumors and uncover common occurrences of intermediate MMRd states. Pairwise perturbations further reveal repair crosstalk reflected in mutational spectra, including enhancement of MUTYH-associated oxidative mutagenesis by MMR loss. Together, these findings recast mismatch repair deficiency as a multidimensional genomic phenotype and provide a framework for interpreting DNA repair deficiencies by bridging mutational phenotypes from experimental systems and human cancers.

Transfer RNA genes (tDNAs) are essential genomic elements that safeguard translational fidelity. Using the T2T version of the human genome we have mapped the position of human tDNAs and analyzed their individual transcriptional activities. Then we have characterized, at single base resolution, the impact of somatic mutations in human tDNAs and its relationship to the transcriptional status of each gene. We confirm that tDNAs are hotspots for somatic mutagenesis, and show that they display mutational loads that are directly proportional to their transcription rates. Highly transcribed tDNAs in tumors or healthy tissues accumulate mutations at rates up to nine-fold higher than highly transcribed protein-coding genes. Mutational loads at tDNAs are tumor-specific, and increase with patient age. Mutations at structurally conserved tRNA positions appear to be under negative selection. Anticodon nucleotides crucial for decoding frequently acquire somatic mutations, readily generating chimeric tRNA species capable of systematically introducing amino acid substitutions across the proteome. Our results reveal a previously unrecognized source of somatic heterogeneity in human cancer and aging tissues that may directly impact upon translation efficiency and fidelity, and cause cell-specific proteostasis degeneration.

The molecular consequences of protein truncating variants depend strongly on whether their transcripts are eliminated by nonsense-mediated mRNA decay (NMD), yet NMD is still predicted largely from a small set of binary positional rules. How individual premature termination codons (PTCs) engage NMD across genes and transcript contexts therefore remains incompletely resolved. Here we integrated endogenous allele-specific PTC expression from large-scale genomic data, mRNA language-model prediction and high-throughput mutational scanning to revisit the rules that govern mammalian NMD. Using allele-specific expression measurements from large human cohorts, we trained NMDetective-AI on ~14,000 somatic PTCs, and after testing it on ~1,800 germline PTCs, found that it improves on previous models, with its accuracy approaching the reproducibility of the underlying measurements. We then generated experimental maps of NMD with deep mutational scanning, including ~450 PTCs nearby the 50-nt penultimate exon boundary, ~950 engineered PTCs across 9 exon lengths to resolve long-exon escape, and ~11k PTCs across 139 genes to quantify and refine start-proximal evasion. Our results support the positional logic of mammalian NMD yet show that this logic is implemented quantitatively: the classical rules resolve into graded, gene-dependent response curves whose boundaries are shaped by transcript architecture and modulated by local sequence context. Applying this framework to population, disease and cancer datasets further identifies genes in which NMD is predicted to aggravate or ameliorate the effects of truncating variants, providing a basis for variant interpretation and for prioritizing NMD-directed therapies.

Following their success in natural language processing and protein biology, pretrained large language models have started appearing in genomics in large numbers. These genomic language models (gLMs), trained on diverse DNA and RNA sequences, promise improved performance on a variety of downstream prediction and understanding tasks. In this review, we trace the rapid evolution of gLMs, analyze current trends, and offer an overview of their application in genomic research. We investigate each gLM component in detail, from training data curation to the architecture, and highlight the present trends of increasing model complexity. We review major benchmarking efforts, suggesting that no single model dominates, and that task-specific design and pretraining data often outweigh general model scale or architecture. In addition, we discuss requirements for making gLMs practically useful for genomic research. While several applications, ranging from genome annotation to DNA sequence generation, showcase the potential of gLMs, their use highlights gaps and pitfalls that remain unresolved. This guide aims to equip researchers with a grounded understanding of gLM capabilities, limitations, and best practices for their effective use in genomics.

The accurate duplication of the genome relies on the spatiotemporal control of DNA replication initiation, yet the determinants specifying mammalian origin locations remain elusive. By developing ORIFormer, a transformer-based neural network, we decode a complex, conserved DNA sequence grammar that accurately predicts initiation sites. This approach uncovers novel determinants, which we validate by demonstrating selection and molecular effects of motif-altering genetic variants. To characterize tissue-specific usage, we developed MuSAS, a statistical genomic method leveraging widespread mutational strand asymmetries, such as from APOBEC and mismatch repair deficiency, to map initiation zones across diverse somatic tissues. Integrating these modalities reveals that while intrinsic DNA sequence features establish a high-potential landscape of constitutive origins, tissue-specific usage is governed by local chromatin accessibility acting as a permissive switch. We suggest that human replication initiation is driven by a deterministic genetic code modulated by the epigenetic landscape, providing a unified framework for understanding genome copying.

Supplementary Figure S5. Cancer types diagnosed in the carriers of germline NBN variants found to be enriched in cases compared to controls. Information about somatic loss of heterozygosity (LOH) is included. NSCLC: NonSmall Cell Lung Cancer, CUP, Cancer of Unknown Primary.

Supplementary Figure S4. Alternative somatic second hit analysis.