Deriving longitudinal tumor phylogenies from single-cell sequencing data

Tumors evolve over time and in response to treatment, leading to changes in the proportions of clones within the tumor. Single-cell sequencing of tumor samples from multiple timepoints enables the reconstruction of clonal evolution and tracking of temporal changes in tumor composition. However, the high rates of missing data in single-cell sequencing complicate evolutionary analyses, and can lead to implausible conclusions, such as the reappearance of extinct clones. We introduce PO_SCPLOWHYLLOCHRONC_SCPLOW, an algorithm that builds a longitudinal phylogeny from cancer cells sequenced from multiple timepoints and evaluates whether the constraints on clonal proportions imposed by longitudinal sampling are well supported by the data. PO_SCPLOWHYLLOCHRONC_SCPLOW relies on a novel mathematical formulation of a longitudinal perfect phylogeny, an extension of the perfect phylogeny model that is widely used in cancer evolution. We show that PO_SCPLOWHYLLOCHRONC_SCPLOW outperforms existing phylogeny inference methods on simulated single-cell sequencing data. Applied to longitudinal single-cell DNA sequencing data from an acute myeloid leukemia (AML) patient, PO_SCPLOWHYLLOCHRONC_SCPLOW constructs a longitudinal phylogeny containing rare cancer clones that persist through multiple cycles of targeted drug treatment, a crucial finding missed by existing phylogeny inference methods. The PO_SCPLOWHYLLOCHRONC_SCPLOW statistical test supports the presence of these rare clones.