From genomic variation to accelerated genome evolution: Integrating genomics, imaging and AI
Group Leader and Senior Scientist
jan.korbel [at] embl.de
ORCID: 0000-0002-2798-3794
EditStructural variants (SVs) are large-scale changes in the sequence architecture of the genome. They include deletions, duplications, insertions, inversions, translocations and more complex rearrangements of DNA such as chromothripsis events. Although less numerous than single-nucleotide variants, SVs affect substantially larger stretches of the genome and collectively account for most of the bases that differ between human genomes. By altering gene dosage, disrupting genes or regulatory regions, causing gene fusions and reorganising chromosomes, they can profoundly influence genome function, human diversity, and disease.
Structural variants can be inherited, or they can arise in individual cells during a person’s lifetime, resulting in genetic mosaicism and leading to disease such as cancer. Our research examines how these somatic changes form, how cell populations respond to them and how they contribute to chromosome instability, cellular diversity, ageing and cancer development.
We study these questions from complementary angles, combining genomics and multi-omics, advanced imaging and computational as well as artificial intelligence approaches. Long-read sequencing enables us to reconstruct complex genomic rearrangements that are difficult to resolve with conventional technologies. Multi-omics approaches help us connect changes in genome structure with their effects on gene regulation and cellular state. Closed-loop adaptive-feedback microscopy approaches such as MAGIC (see further below) allow us to identify and investigate dynamic, rare or transient cellular events, while machine-learning helps us recognise patterns across large and complex datasets. By integrating these methodologies, we aim to understand structural variation across scales, from DNA sequence and chromosome architecture to individual cells and tissues, and to uncover the mechanisms through which genome instability shapes human biology and disease.
Video 1: Jan Korbel introduces the lab’s research.
MAGIC (machine-learning-assisted genomics and imaging convergence) combines autonomous live-cell imaging, on-the-fly machine learning, targeted photolabelling and single-cell genomics to study chromosomal alterations as they form.
Chromosome instability (CIN) is the ongoing generation of structural and numerical chromosome alterations, often caused by errors during cell division. It is a major force in cancer genome evolution. When a chromosome or chromosome fragment fails to segregate correctly, it can be entrapped outside the main nucleus in a small compartment known as a micronucleus. DNA contained in micronuclei is particularly vulnerable: it can undergo further segregation errors or extensive fragmentation and rearrangement, including chromothripsis. However, the earliest steps connecting mitotic errors with newly formed chromosome alterations, and the frequency at which these alterations arise spontaneously have been difficult to observe systematically.
MAGIC (machine-learning-assisted genomics and imaging convergence) combines autonomous live-cell imaging, real-time machine-learning classification, targeted photolabelling and single-cell genomics. The platform identifies and follows living cells with abnormal nuclear features, including micronuclei, and then links these features to newly formed chromosome alterations. Applying MAGIC to near-diploid, non-transformed human cell lines allowed the researchers to track chromosome alterations across successive cell cycles.
In these experimental models, dicentric chromosomes (chromosomes with two centromeres) frequently marked the beginning of chromosome-alteration events. During cell division, dicentric chromosomes can form bridges between daughter cells, break and generate additional alterations over subsequent cell cycles. This establishes a mechanistic connection between an initial segregation error, the formation of abnormal nuclear structures and the continuing evolution of the affected chromosome.
We also measured a baseline rate of spontaneous chromosome-alteration formation and found that it approximately doubled in TP53-deficient cells. Chromosome losses occurred more frequently than gains. Experiments introducing DNA breaks at different positions along a chromosome produced distinct outcomes, including stable isochromosomes, the coordinated segregation and amplification of chromosome segments lacking a centromere, and complex rearrangements.
Observing chromosome alterations as they form is important because their initial spectrum differed from the patterns detected after cells had undergone selection. Established cell populations therefore preserve only part of the mutational process. By linking live-cell behaviour, including micronucleus formation, with the genomic alterations found in the same cells, MAGIC enables researchers to investigate how individual errors during cell division develop into simple or complex chromosome changes relevant to cancer evolution.
Video 2: A short introduction to MAGIC (Machine Learning-Assisted Genomics and Imaging Convergence).
One of our most recent studies describes chromophagy, an autophagy pathway that removes whole micronuclei and limits the transmission of unstable chromosomal material to daughter cells, thereby causing chromosome losses and suppressing chromosomal instability in individual cells.
Chromosomal instability (CIN) is defined by the generation of structural and numerical chromosome abnormalities (CAs) often through mitotic errors. This phenomenon is a central driver of cancer genome evolution. When a chromosome is missegregated during cell division, it can become trapped in a small, separate nucleus called a micronucleus. Micronuclei can promote repeated segregation errors and catastrophic rearrangements such as chromothripsis.
Our goal was to study whether cells have a mechanism that limits this source of chromosome instability.
Building on this framework, our recent work has uncovered a cellular mechanism that actively shapes these outcomes: selective autophagy of whole micronuclei, termed “chromophagy”. Using a live-cell chromatin acidification sensor, this study shows that micronuclei with nuclear envelope defects arising at mitotic exit are recognized and degraded via the autophagy pathway, resulting in the direct elimination of the entrapped chromosome from the cell and its loss from subsequent daughter-cell lineages. This establishes chromophagy as a direct determinant of genome composition, distinct from micronucleus rupture and chromothripsis.
Cosenza, M.R., Gaiatto, A., Erarslan Uysal, B. et al. Origins of chromosome instability unveiled by coupled imaging and genomics. Nature 648, 383–393 (2025). https://doi.org/10.1038/s41586-025-09632-5
Watson, N.A., Melli, M., Cosenza, M.R. et al. Selective autophagy of whole micronuclei suppresses chromosomal instability. bioRxiv (2026). https://doi.org/10.64898/2026.04.03.716211
Long-read sequencing of 1,019 individuals from 26 populations reveals previously under-characterised forms of structural variation and provides an open reference for studying human genetic diversity and prioritising variants in patient genomes.
Structural variants are differences between human genomes in which sections of DNA are deleted, duplicated, inserted, inverted or repeated. These variants account for a substantial proportion of the DNA differences between individuals and contribute to both common and rare diseases. However, many structural variants have remained difficult to detect, particularly in repetitive and complex regions of the genome.
This study characterises structural variation in 1,019 individuals representing 26 populations from five continental regions. By analysing longer stretches of DNA, we identified more than 100,000 sequence-resolved structural variants and genotyped approximately 300,000 variable-number tandem repeats. The resulting dataset captures common, rare and population-associated variation more comprehensively than previous population-scale surveys based on short-read sequencing, with particularly improved representation of insertions.
The findings provide insights into how different types of structural variants arise. Analyses of variant breakpoints indicate that several DNA-repair processes contribute to their formation and to recurrent deletions. The study also shows how mobile genetic elements can move additional stretches of DNA to new genomic locations.
By incorporating this variation into a pangenome reference, we have expanded the representation of structural diversity across human populations. The openly available resource we created supports research into human genetic variation and provides a reference for filtering and prioritising candidate structural variants in patient genomes. It therefore contributes to a more accurate interpretation of genomic variation in both population research and genomic medicine.
Blood-forming stem and progenitor cells acquire mosaic structural variants throughout life. In this study, we observed expanded cell populations carrying these variants only in donors over 60 and identified cell-type-specific molecular effects of mosaic structural variants in these cells.
Changes to the genome occur in cells throughout life, including in people without diagnosed disease. Some of these changes are structural variants, in which sections of DNA are deleted, duplicated, inverted or otherwise rearranged. When a structural variant is present in only a subset of cells, it is described as mosaic.
In this study we examined mosaic structural variants in blood-forming stem and progenitor cells from healthy donors of different ages. Our findings show that these variants arise continuously throughout life. However, we observed expanded groups of cells carrying the same variant only in individuals over 60 years of age within the study cohort. Cells already carrying mosaic structural variants were more likely to contain additional genetic alterations.
The effects depended on cell type and genomic location. The variants skew cell differentiation towards some cell types but not others. Expanded cell populations carrying structural variants were frequently enriched in progenitor cells associated with the myeloid blood-cell lineage. Although the variants affected different regions of the genome, they altered several biological pathways connected with aging of the blood system, clonal hematopoiesis and leukemia, including RAS and JAK–STAT signalling and lipid metabolism.
These findings demonstrate how mosaic structural variants can influence the cellular and molecular characteristics of the aging blood system. They provide a basis for further investigation of the relationships between structural genome variation, aging and susceptibility to disease in otherwise healthy tissues.
Our laboratory develops and applies integrated experimental and computational methods for studying genome variation in individual cells.
Our laboratory develops and applies integrated experimental and computational methods for studying genome variation in individual cells. OP-style Strand-seq adapts one-pot library preparation to microlitre-volume workflows, while EVAS is being developed to automate the quality assessment of Strand-seq libraries. For downstream analysis, MosaiCatcher v2 provides a reproducible framework for detecting and analysing structural variants; strandtools supports copy-number analysis, annotation and visualisation; and scNOVA uses nucleosome-occupancy information captured by Strand-seq to connect structural variants with cell-specific molecular changes. MAGIC extends this approach by combining live-cell imaging, machine learning and targeted cell selection with single-cell genomics, enabling genomic alterations to be linked to cellular features observed in real time. Together, these methods support an integrated workflow from library preparation and quality control to structural-variant detection, functional interpretation and the connection of genomic changes with cellular behaviour. We provide long term support and constantly improve our widely used DELLY tool for discovering structural variation in short and long-read datasets.
