Abstract
How tumors evolve affects cancer progression, therapy response, and relapse. However, whether tumor evolution is driven primarily by selectively advantageous or neutral mutations remains under debate. Resolving this controversy has so far been limited by the use of bulk sequencing data. Here, we leverage the high resolution of single-cell DNA sequencing (scDNA-seq) to test for clock-like, neutral evolution. Under neutrality, different cell lineages evolve at a similar rate, accumulating mutations according to a molecular clock. We developed and benchmarked a test of the somatic clock based on single-cell phylogenies and applied it to 22 scDNA-seq datasets. We rejected the clock in 10/13 cancer and 5/9 healthy datasets. The clock rejection in seven cancer datasets could be related to known driver mutations. Our findings demonstrate the power of scDNA-seq for studying somatic evolution and suggest that some cancer and healthy cell populations are driven by selection while others seem to evolve under neutrality.
Introduction
The nature of somatic evolution—whether it is predominantly neutral or adaptive—is a central question in cancer biology. Neutral evolution implies that mutations accumulate at a constant rate without conferring fitness advantages, whereas adaptive evolution involves selective sweeps where specific "driver" mutations allow certain clones to dominate. Historically, bulk sequencing has struggled to distinguish these modes because it averages genetic signals across millions of cells, often missing the subtle signatures of early clonal expansion.
The Poisson Tree (PT) Test
To overcome the limitations of bulk data, we developed the Poisson Tree (PT) test, a statistical framework designed for single-cell DNA sequencing (scDNA-seq) data. This approach constructs explicit phylogenies of individual cells and models mutation counts on each branch using a Poisson distribution.
Under the null hypothesis of a "molecular clock" (neutral evolution), all lineages should accumulate mutations at approximately the same rate, resulting in an "ultrametric" tree where the distance from the root to any living cell is roughly equal. Deviations from this clock suggest selection, where lineages with growth advantages divide more frequently and accumulate more mutations. Crucially, the PT test incorporates a weighting system to account for technical noise typical of scDNA-seq, such as Allele Dropout (ADO), ensuring that technical errors are not mistaken for biological selection.
Results and Insights
We applied the PT test to 22 scDNA-seq datasets across various tissues, including blood, bladder, lung, prostate, breast, colon, and kidney.
Selection is Widespread in Cancer
The molecular clock was rejected in 10 out of 13 cancer datasets, indicating that adaptive evolution is the dominant mode in most tumors. In seven of these cases, known driver mutations (e.g., PIK3CA, TP53, NRAS) were identified on internal branches, likely providing the fitness boost that caused the lineage to deviate from the clock.
Non-Neutral Evolution in Healthy Tissues
Surprisingly, the clock was also rejected in 5 out of 9 healthy datasets, including neurons and colon tissue. This suggests that even in normal aging, certain cell lineages gain competitive advantages, potentially serving as precursors to malignancy.
Comparison with Bulk Methods
Benchmarking against simulated data showed that the PT test achieves over 90% power to detect non-clock evolution with as few as 10 cells. It significantly outperformed existing bulk-based methods, such as the $1/f$ test, highlighting the superior resolution of single-cell phylogenies for detecting selection.
Discussion
Our findings demonstrate that scDNA-seq provides the necessary resolution to resolve long-standing debates about somatic evolution. While selection appears prevalent, the detection of neutrality in some samples suggests that both modes can exist depending on the tissue context and stage of progression.
Limitations remain, particularly regarding technical noise and sampling bias. Extremely high false-negative rates can mask biological signals, and localized mutation rate changes might mimic selection. Future research integrating phylogenetic tests with spatial data will further clarify how the physical environment of a tissue influences these selective sweeps.
Summary updated