Chemotherapy damages the DNA of cancer cells, but seeing that damage directly is hard: mutations are spread across the genome and across many cells. Existing approaches either catch only small base substitutions or require cloning, which distorts the picture through selection. This work sits alongside studies that have already shown how single cells accumulate private mutations during replication and chromosomal instability, but until now no one had applied such an approach to the action of drugs. There are already papers on invisible connections in gene networks, on how cancer evades a drug, and on the genealogical trees of tumors, but this article takes a step toward directly measuring the mutational consequences of therapy.
The authors used single-cell sequencing with the DLP+ method, which does not require DNA amplification and therefore reflects copy number more accurately. Cells were treated with five agents: cisplatin, CX5461, pyridostatin, etoposide, and vorreloxin, each of which breaks DNA in its own way. For each cell they built an allele-specific phylogenetic tree and subtracted the state of the nearest common ancestor, yielding the "foreground": new events not inherited from the clone. Events larger than 4 megabases were classified by size, position on the chromosome (at the telomere, at the centromere, interstitial, or a whole arm), and direction (loss or gain). A hierarchical Dirichlet process then decomposed these features into eight signatures, and statistical models compared signature exposure between groups of cells.
All five DNA-damaging drugs, regardless of mechanism, increased exposure to the same signature, named TboundedArm. It consists mainly of large losses (17-38 and over 38 megabases) bounded by the telomere, plus losses of whole arms. In cells lacking the TP53 protein, exposure rose 10-11 fold compared with controls; in cells with normal TP53, it rose 9-17 fold. Drugs that do not damage DNA (palbociclib, ZM-447439) produced no such signature. In cells with a doubled genome, exposure was 1.5-1.9 times higher. In cells with inactive BRCA1, background instability already created a similar telomere-bounded signature, and the drugs did not amplify it. In breast cancer xenografts, the signature appeared with the first treatment but disappeared as resistance developed, pointing to a decline in the drug's effect on the genome rather than to habituation to mutations. The signal persisted for up to 56 days after withdrawal of cisplatin and CX5461, and in cells with a doubled genome it decayed more slowly.
This work differs from neighboring studies in that it moves mutational signature analysis from the level of the whole tumor down to the level of individual cells and focuses on large rearrangements rather than base substitutions. Neighboring papers examined gene networks, drug resistance, and tumor phylogeny, but did not offer a quantitative measure of genomic damage from a specific drug. Here it is shown that agents with different mechanisms converge on one type of chromosomal event, and that this trace can be used as a pharmacodynamic indicator. This changes the picture of how chemotherapy leaves its mark on the genome: not through unique point mutations, but through a shared large-scale rearrangement at telomeres.
The method could be extended to other classes of mutagens, including environmental substances that cause large chromosomal changes rather than substitutions of individual bases. It may become possible to link signature exposure to clinical response to therapy and to predict which tumors will respond to treatment and which will quickly become resistant. An open question remains about the mechanism behind the signal's long persistence after drug withdrawal: whether it reflects residual damage, continued mutations during replication, or the slow washout of mutated cells.
Oncology will gain a tool for measuring the genomic action of drugs in real time, which could help in dose selection and efficacy assessment. Toxicology will be able to use the same approach to test potentially mutagenic substances in cell models without cloning.
Verify the reproducibility of the signature in other cell and tumor types, as well as with drugs that have different mechanisms of DNA damage. Investigate exactly how telomere-bounded losses arise after double-strand breaks and why the signal persists for so long.
The work addresses the unresolved problem of the link between DNA damage and large-scale chromosomal instability in cancer cells. Until now it was unclear whether different classes of drugs leave unique traces or converge on common consequences; now it is shown that at the single-cell level the trace is shared, but its magnitude depends on genetic background and dose. This also ties mutational signatures to clinical resistance, which had remained a gap in understanding how tumors evolve under therapy.
Telomere-bounded events arise within 2.5 megabases of the chromosome end, but short reads cannot resolve the telomeric repeats themselves, so the authors call them "telomere-bounded" rather than "telomeric."
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.