How a cancer evolves and how mutations are generated are highly intertwined processes, and both are nearly impossible to observe directly. Instead, we are usually restricted to making inferences about them using data from a single snapshot in time after a cancer has formed. Writing in Nature, Aitken et al.1 show that, for a cell that has undergone DNA damage, such a snapshot provides remarkably rich information when the two DNA strands that form the double helix are considered independently.
DNA resembles a ladder, with the two ‘side rails’ often called, respectively, the Watson and Crick strands. These are fused together by ‘rungs’ of two complementary nucleotide base pairs: either cytosine (C) paired with guanine (G) or adenine (A) paired with thymine (T). When a cell divides, each daughter cell inherits either the Watson or Crick strand from the parent; this provides a template from which the other, complementary strand is replicated. Damage to a base can trigger a repair process, but if repair is not swift enough, the damaged base might be mispaired with an incorrect base during DNA replication. At the next round of cell division, when a daughter cell with such a mispaired base prepares to divide, the base complementary to the mispaired base will be added to the newly synthesized strand. This leads to a double-stranded mutation at the base pair corresponding to the original damaged base (Fig. 1).
Standard practice for genome sequencing is to consider mutations without paying attention to which of the strands received the original damage. However, when a chemical change occurs that damages a base, creating a site referred to as a lesion, this lesion is on only one of the two DNA strands of the affected base pair. Aitken and colleagues had the insight to see that, because the ‘parental’ Watson and Crick strands of an original cell that underwent DNA damage are separated into different daughter cells, when the cell divides, two cell lineages can be tracked individually by following the unique pattern of mutations that lesions on each of the parental strands generates.
To induce DNA lesions, Aitken and colleagues gave mice a large dose of the carcinogenic molecule diethylnitrosamine. This treatment predominantly caused DNA lesions at T bases in liver cells, ultimately leading to tumour growth. When the authors examined the pattern of diethylnitrosamine-induced mutations along the genome of each tumour, they found long stretches of the genome that, compared with the original, unmutated genome, were highly enriched for mutations in which T was mutated to any other base (N). These mutations were derived from T lesions on the DNA strand (let’s call it the Crick strand) that was inherited by the daughter cell, and its cellular descendants, that went on to form the tumour. Lesions on either strand can generate mutations, but it might be the case that lesions on only one of the parental strands generates tumour-promoting mutations and hence only one of the two daughter lineages forms a tumour. In the example shown in Figure 1, the lesions on T bases on the corresponding Watson parental strand received by the other daughter cell did not lead to the formation of tumour-promoting mutations, and this cellular lineage therefore did not contribute to the tumour.
The authors realized that the pattern of base-pair locations that had T-to-N mutations enabled them to pinpoint the individual Watson or Crick strand that had served as the template strand for the first cell in the tumour. This template strand carried diethylnitrosamine-induced lesions mainly at T bases, allowing the fate of each strand to be tracked individually through subsequent cell divisions (Fig. 1). These individual ‘strands of evidence’ provide remarkable information about the process of mutation and tumour evolution.
In gene expression, DNA is transcribed to produce RNA, and DNA lesions can be repaired by a process called transcription-coupled repair2. Aitken and colleagues observed that transcription-coupled repair occurred preferentially on the strand being transcribed, as opposed to the complementary strand, and that higher levels of transcription were associated with an increased frequency of repair.
The authors found that failure to repair a DNA lesion over successive cell cycles, an interesting observation in itself, provided an unexpected source of genetic diversity. Each round of DNA replication on a lesion-containing strand could lead to the incorporation of a different ‘wrong’, mispaired base opposite the lesion site in the newly synthesized strand. If this happened, it caused further, distinct mutations at the same genomic position, generating cells in the tumour each with different mutations of the same base pair.
Observing recurrent mutations at the same genomic site could be taken as evidence of convergent evolution, in which multiple individual mutational events at that base-pair site are all positively selected for during tumour growth. Instead, Aitken and colleagues’ findings indicate that recurrent mutations could result from lesion-bearing DNA strands being used as templates for DNA replication over multiple rounds of cell division.
Intriguingly, strand tracing also provides a window on the selection of mutations associated with cancer. When a cell divides, a daughter cell should inherit, at random, either DNA strand. However, when the authors tracked the prevalence of sequences corresponding to inheritance of the parental Watson or Crick strand of a particular chromosome, they noticed that the tumours contained one of these two strands more often than would be expected by chance.
The authors’ explanation for such preferential strand retention is that it occurred because the retained strand contained a diethyl nitrosamine-induced mutation in a gene that is important for tumour growth. Aitken et al. identified three potential tumour-promoting genes in this way, all of which are known to be crucial for the growth of liver tumours. Strand-by-strand analysis might be an unexpectedly useful tool for probing the tumour-promoting contribution of non-protein-coding regions of the genome, because selection can be detected without needing to know the background mutation rate — the problem of determining this rate has posed a challenge for methods previously used to study these regions3.
It might be expected that because there is strand-biased prevalence of mutations corresponding to diethylnitrosamine-induced lesions, a chromosome should be enriched for T-to-N mutations along the entire length of its DNA strand. Instead, the authors found that, for a single chromosome, the enrichment of such mutations sometimes switched over to the other strand (and could be observed as A-to-N mutations). They propose that this provides evidence of sites of a DNA-repair process called homologous recombination, in which DNA strands from a chromosome exchange with identical DNA sequences in a cell that is gearing up to divide, during an event called sister-chromatid exchange.
Sister-chromatid exchange is usually an elusive process to monitor because it involves, in theory, an exchange between two identical DNA sequences. However, because Aitken and colleagues could track individual strands through mutation patterns, they could detect evidence of such events. Interestingly, a higher frequency of these exchange events tracked with a higher diethylnitrosamine-induced mutation burden, suggesting that diethyl nitrosamine-induced damage to DNA might prompt sister-chromatid exchange4. Aitken et al. report that presumptive exchange events tended to occur in regions of the genome that were associated with lower-than-average gene expression and later replication, compared with other regions, during the cell cycle. Interestingly, these features are hallmarks of ‘fragile sites’ — genomic regions susceptible to damage during DNA replication that are known to be prone to sister-chromatid exchange5,6.
The most tantalizing question raised by this study is whether the DNA strand (the original Watson or Crick strand) in which damage first occurred can be tracked in human cancers. People are usually exposed to mutation-causing agents, such as those in cigarette smoke, for long periods of time, making it probable that DNA lesions are continually being induced. Consequently, the mutational signal arising from an individual DNA strand would probably be obscured. However, when Aitken and colleagues assessed data already available from human cancers, they found that, in rare cases of sudden, acute exposure to a mutagenic agent, most clearly observed for aristolochic acid exposure (which leads to liver, kidney and bile-duct cancer), mutational signals could be traced back to identify the strand originally damaged.
Chemotherapy also provides an example of acute exposure to mutation-causing agents, and can give rise to distinctive mutational signatures7. It will be interesting to see what can be learnt by applying Aitken and colleagues’ methods to analysing chemotherapy-treated cancers. Other examples of exposure to mutation-causing agents, such as acute radiation exposure or sunburn, might also be worth analysing using a strand-by-strand approach. Furthermore, such analysis could clarify the timeline of exposure to a mutation-causing agent: a single, large exposure might generate a mutational signal that could be assigned to individual DNA strands, whereas repeated exposures might cause a progressively less distinct signal. Similarly, analysing individual strands might provide insight into the rate of lesion-repair processes, and offer a new means of studying defects in DNA-repair processes in cancer, such as defective-mismatch repair.
Tracking individual DNA strands is a reductive approach offering a powerful way to study DNA replication and repair processes that have been challenging to observe. Aitken and colleagues’ study shows the potential of this method for tackling the complexities of cancer — it seems that the maxim that the whole is greater than the sum of the parts does not apply to individual DNA strands.
Nature 583, 207-209 (2020)