Lineage Tracing in Chronic Lymphocytic Leukemia Reveals Clones with Stable Gene Expression States That Differentially Respond to Therapy

Cancer's ability to evolve and adapt is a major challenge to therapeutic success. Fueling this evolution is vast tumor heterogeneity, with constituent clones varying in their genetics, epigenetics and response to therapy. As a field, however, we have yet to couple the genetic alterations present within individual clones to their transcriptional or functional outputs. Here, we applied a novel adaptation of clone tracing that integrates DNA barcoding with single-cell RNA sequencing (scRNA-seq) to HG3, a CLL cell line harboring del(13q) and no other known cancer drivers, to model in vitro responses to front-line chemotherapy with fludarabine and cyclophosphamide at clone-level resolution.

To generate a high-diversity barcode library compatible with scRNA-seq, a random pool of 20 base pair DNA barcodes was introduced into the 3'UTR of a reporter gene in a lentiviral expression vector. This viral barcode library was transduced into HG3 cells (at MOI 0.1 to minimize multiple barcode-tagging of cells) and 1.2x10⁶ barcoded cells were sorted and expanded to establish the parental barcoded HG3 population. We subsequently treated this barcoded population with an LD95 combined dose of fludarabine and mafosfamide (the in vitro analog of cyclophosphamide) in 8 parallel experiments. Cell barcodes were sequenced prior to treatment (TP1) and following outgrowth from treatment (TP2) for analysis of clonal composition. 10,000 cells each from TP1 and from 2 of 8 parallel replicates at TP2 were processed for scRNA-seq.

We observed a massive decrease in viability across all 8 replicates, with regrowth occurring at 20 days post-treatment. Barcode analysis revealed a marked decrease in clonal diversity from TP1 to TP2 (11,827 to 2,622 ± 380, n=8; or ~78%), and clones that survived treatment did so consistently such that 94% of surviving cells in each replicate had a clonal identity that was present in all 8 replicates. Analysis of clonally-resolved transcriptional profiles revealed that clones consistently fell into one of two stable gene expression states (clusters) prior to treatment, with nominal intermixing between populations. Treatment predominantly selected for clones comprising the smaller of these two clusters (TP1-'high tolerance'), with only a minimal number of resistant clones originating from the larger cluster at TP1 (TP1-'low tolerance'). Pathway and gene set enrichment analysis of these two TP1 clusters demonstrated that TP1-high tolerance had a stark upregulation of common CLL signaling pathways (i.e. WNT and CXCR4, an inflammatory/migratory phenotype) and a reliance on chromatin modification pathways. TP1-low tolerance, on the other hand, exhibited upregulated type 2 antigen presentation and prostaglandin biosynthesis/metabolism which has a known role in driving inflammation and migration in adjacent cells (Wang et al, BMJ 2006). These gene expression states remained stable after treatment, but with the added upregulation of well-described mechanisms of resistance to cyclophosphamide, with TP1-high tolerance exhibiting upregulation of GSTP1, a glutathione S-transferase that is a main mediator of cyclophosphamide metabolism, and TP1-low tolerance exhibiting upregulation of members of the ALDH family thought to ameliorate toxicity from chemotherapy-induced ROS (Andersson et al, Acta Oncologica 1995).

Through this work, we resolved the underlying clonal composition of a CLL cell line and observed that the constituent clones exhibit stable and discrete gene expression states that differentially respond to chemotherapy. We noted two different axes of resistance - a more successful avenue that relies on WNT and CXCR4 signaling as well as cyclophosphamide clearance for resistance, and a less-successful avenue that involves clearance of reactive oxygen species for survival. The intersection of these two critical CLL pathways in in vitro resistance to first-line CLL therapy is of particular interest given the FAT1 (WNT regulator) mutations and CXCR4 upregulation frequently seen in chemo-refractory CLL (Messina et al, Blood 2014; Burger et al, Blood 2006), and future efforts will assess the stability and interplay of these two pathways in patient samples collected upon relapse to fludarabine and cyclophosphamide. Further, our approach provides a template for the high-resolution study of tens of thousands of clones and their respective phenotypes in a mixed leukemic population.