Extensive tumor sequencing efforts have transformed the way in which cancer driver genes are identified. Appropriate statistical modeling is crucial for distinguishing true drivers from passenger events that accumulate during tumorigenesis but provide no fitness advantage to cancer cells. A central assumption used in discovering driver genes and specific driver mutations is that exact positional recurrence is unlikely by chance: Seeing exactly the same DNA base pair mutated recurrently across patients is taken as proof that the mutation must be under functional selection for contributing to tumor fitness. The assumption is that mutational processes, being essentially random, are unlikely to hit the exact same base pair over and over again. However, although functional selection is clearly a key cause of recurrent mutations in cancers, whether it is the only prominent cause is not known.

To distinguish driver mutations from passengers, it is critical to understand the landscape of background mutations in cancer genomes. Recent pan-cancer mutation analyses have revealed rules of mutation distribution at the smallest (one to three base pairs) and largest (megabase) scales. At the small scale, mutational processes such as those attributable to sunlight, cigarette smoke, or random DNA copying errors generate patterns known as mutational signatures at the trinucleotide level. At the opposite extreme, the cell’s nucleus is organized into two large compartments known as A and B, each consisting of multi-megabase chromatin domains. Compartment A contains gene-rich, open, active, early-replicating euchromatin. Compartment B contains gene-poor, closed, inactive, and late-replicating heterochromatin. Mutation frequency is generally higher in compartment B. Cancer genomes have been studied in detail at these two opposite scales, but less attention has been paid so far to the intervening “mesoscale.”

We investigated the influence of mesoscale genomic features on mutational recurrence. We found that mutagenesis by the cytidine deaminase APOBEC3A is uniquely sensitive to mesoscale features, specifically the ability of DNA to adopt particular “hairpin” (stem-loop) structures while transiently single-stranded. Identifying DNA loci that can form hairpins requires sequence analysis at the mesoscale (~30–base pair) level. Combining biochemistry and bioinformatics, we deduced the features of APOBEC3A’s optimal DNA substrates, revealing that cytosine bases presented in a short loop at the end of a strongly paired stem can be mutated up to 200 times as frequently as nonhairpin sites. Analyzing the most frequent APOBEC mutations in protein-coding regions of cancer genomes, we identified numerous recurrent mutations at optimal hairpins in genes unconnected to cancer. Conversely, we found that mutational hotspots at nonoptimal sites are enriched in known cancer driver genes.

Our results indicate that there are multiple possible routes to mutational hotspots in cancer. Functional mutations in oncogenes or tumor suppressors can rise to prominence through positive selection. These driver hotspots are not restricted to the “favorite” sites of any particular mutagen. In contrast, DNA sites that happen to be perfect substrates for a mutagen can give rise to “passenger hotspot mutations” that owe their prevalence to substrate optimality, not to any effects on tumor fitness. In light of these findings, we recommend caution in interpreting the long lists of putative novel cancer driver hotspots being produced by high-throughput sequencing projects.

The APOBEC cytidine�…