Single-cell sequencing reveals the origin and the order of mutation acquisition in T-cell acute lymphoblastic leukemia

Abstract

Next-generation sequencing has provided a detailed overview of the various genomic lesions implicated in the pathogenesis of T-cell acute lymphoblastic leukemia (T-ALL). Typically, 10-20 protein-altering lesions are found in T-ALL cells at diagnosis. However, it is currently unclear in which order these mutations are acquired and in which progenitor cells this is initiated. To address these questions, we used targeted single-cell sequencing of total bone marrow cells and CD34⁺CD38^- multipotent progenitor cells for four T-ALL cases. Hierarchical clustering detected a dominant leukemia cluster at diagnosis, accompanied by a few smaller clusters harboring only a fraction of the mutations. We developed a graph-based algorithm to determine the order of mutation acquisition. Two of the four patients had an early event in a known oncogene (MED12, STAT5B) among various pre-leukemic events. Intermediate events included loss of 9p21 (CDKN2A/B) and acquisition of fusion genes, while NOTCH1 mutations were typically late events. Analysis of CD34⁺CD38^- cells and myeloid progenitors revealed that in half of the cases somatic mutations were detectable in multipotent progenitor cells. We demonstrate that targeted single-cell sequencing can elucidate the order of mutation acquisition in T-ALL and that T-ALL development can start in a multipotent progenitor cell.

Fig. 1

Somatic variant identification by bulk sequencing of four primary T-ALL samples. (A) Overview of the number and different types of somatic variants identified in each T-ALL patient by bulk whole-genome and RNA sequencing, subsequently used for targeted single-cell sequencing. (B) Violin plots illustrating the variant allele frequency distributions of the bulk mutations identified per patient

Fig. 2

Targeted single-cell sequencing of four primary T-ALL samples. (A) Schematic workflow of the protocol describing the targeted single-cell sequencing. (B) Histograms of the locus drop-out rates (i), allelic drop-out rates (ii) or locus and allelic drop-out rates combined (iii). Quality control consisted of removing all cells with more than one-third of SNPs affected by locus and allelic drop-out combined, indicated by the gray area in panel (iii). (C) Bar chart of the absolute numbers of single leukemic cells isolated per patient together with the percentage of cells retained for analysis after quality control

Fig. 3

T-ALL patient samples have limited heterogeneity at presentation. Heatmaps of the somatic variations detected per patient: X09 (n = 187 cells), XB37 (n = 115 cells), XB41 (n = 251 cells) and XB47 (n = 96 cells). Columns represent single cells, rows represent the somatic variations. The order of both the cells and the variations is based on hierarchical clustering with the Jaccard distance as metric. Presence of a variation is indicated in red, absence in black, whereas gray represents variations with <10 reads (i.e., no data available). Gene names from known oncogenic drivers are colored red. Percentages indicate the relative number of cells attributed to each clone

Fig. 4

Single-cell RNA sequencing reveals transcriptional uniformity of T-ALL cells. a tSNE analysis and cluster allocation for the single cells per patient. Cluster allocation is described in more detail in Supplemental Methods and Supplemental Fig. 6. b Violin plots showing the normalized expression of several cluster of differentiation markers for the leukemic T cells in each patient. Expression levels correspond with the immunophenotype established with flow cytometry at the time of diagnosis (data not shown). CD19 and CD33 expression represent negative controls

Fig. 5

Multiple mutations can be present in multipotent progenitor cells. a Bar chart of the absolute numbers of single CD34⁺CD38^- multipotent progenitor cells isolated per patient and the percentage of cells accepted for analysis after quality control. b Heatmaps of the variations in single CD34⁺CD38^- multipotent progenitor cells isolated from patient X09, XB37, XB41 or XB47 taken at diagnosis (i) and at remission (iii). Sanger sequencing was performed on bulk DNA extracted from 2000 to 5000 myeloid progenitor cells sorted from the diagnostic samples (ii) to confirm the presence of the mutations found in the multipotent progenitor cells at diagnosis. Deletions and fusion genes were not evaluated in the bulk myeloid progenitor DNA to prevent false-positive results caused by few contaminating leukemic cells, and are therefore colored white in the graph. The order of both the cells and the variations is based on hierarchical clustering with the Jaccard distance as metric. *These variations were initially considered somatic mutations, based on the WGS results of the remission sample. However, we could confirm the presence of these SNPs with PCR on the bulk remission samples

Fig. 6

Single-cell data illuminate the mutational hierarchy in T-ALL patient samples. The order of mutation acquisition based on the newly developed graph-based algorithm for patient X09, XB37, XB41 and XB47. The algorithm evaluated all single-cell information available from both diagnostic leukemic and CD34⁺CD38^- multipotent progenitor cells and stipulated the most probable order of events. Its output (including the 100 most probable order of events per patient) is provided in Supplemental table S9. Percentages on the right represent clones detected at diagnosis per patient, whereas the stars represent different steps in mutation accumulation. Events that happened together or are closely related in time are represented by their respective gene names and written above each star