Single-Cell Analysis of Clonal Dynamics in Childhood ALL Reveals a Key Role for Transcriptional Intratumor Heterogeneity in Driving Resistance to Chemotherapy

Tumors exhibit intratumor heterogeneity in both genetic make-up and phenotypic traits such as morphology, cell cycle status and gene expression. Although heterogeneity has been long recognized, its biological and clinical significance is not fully understood. To determine the relative contribution of genetic versus epigenetic heterogeneity to therapeutic resistance and relapse in childhood acute lymphoblastic leukemia (cALL), we have developed a mouse model that affords "real-time" longitudinal analysis of subclonal dynamics.

The mouse model allows the engraftment of primary leukemic cells from an individual patient into multiple xenograft recipients, half of which are then treated with a combination of vincristine and dexamethasone. By assessing the reproducibility of independent outcomes, one can distinguish between deterministic and stochastic mechanisms of selection during therapy. We have used automated multicolor fluorescence in situ hybridization (mFISH) for the most frequent drivers of the leukemia (ETV6, RUNX1, PAX5, P16) to track the fate of individual genetic subclones. By comparing clonal composition in control and treated mice pre- and post-chemotherapy we demonstrate that, while treatment of sensitive ALLs results in a striking reduction in leukemic burden, the overall extent of genetic diversity (measured by mean of the Shannon index of diversity) is unaffected. This suggests that resistance in ALL may be largely independent of genetic variegation. In light of recent data demonstrating aberrant RAG activation in ETV6-RUNX1 leukemia, we reasoned that our results could reflect convergent evolution. Because RAG mediated deletions recurrently target the same key B-cell pathways, it is possible that genetically distinct clones could display convergent phenotypes. To test this hypothesis, and extend our mutational analysis to whole genome resolution, we performed single-cell whole genome sequencing. We sequenced matching diagnostic and relapse samples, and further validated our mFISH data. This reveals unprecedented levels of intratumor heterogeneity, with individual cells carrying on average 37 copy number variations (CNVs), a large number of which are private. By resolving the genomic breakpoints of lesions to known drivers of disease we formally proved that convergent evolution can happen. Thus, we identified clones carrying the same lesion but distinct breakpoints, highlighting their independent origins.

Functional analysis based on limiting dilution secondary transplantation assays showed that chemotherapy enriches for cells with tumor propagating potential. Transcriptome analysis (RNAseq) of treated and untreated cells revealed that the former have distinct signatures shared amongst metastatic sites. Chemo-resistant cells displayed differences in cell cycle status and transcription rate, as well as differential activation of key signaling pathways (MAPK, Rap1 and Jak-Stat), and expression of cell adhesion molecules. Through gene set enrichment analysis we observed that resistant cells have features characteristic of more primitive hematopoietic cells, including concomitant up-regulation of hematopoietic stem cells and lympho-myeloid progenitors markers. Interestingly, following secondary transplantation global gene expression of treated cells largely reverted to a treatment-naïve signature.

Overall this analysis provides novel insight into the contribution of genetic and epigenetic heterogeneity to resistance to cytotoxic chemotherapy. In the case of cALL phenotypic heterogeneity appears to play a larger role than genetic diversity, particularly with regards to cell cycle state and developmental stage. Our data suggest that resistance is driven by a pre-existing population of immature cells with a distinct global gene expression signature and higher tumor propagating potential, and that genetic heterogeneity in cALL arises through independent acquisition of CNVs affecting the same gene or genes within the same pathway, thereby driving convergent phenotypic evolution. Such a view is consistent with the observation that distinct subclones appear to respond to chemotherapy-derived environmental pressure in a similar manner.