00 MONTH 2017 | VOL 000 | NATURE | 1 LETTER doi:10.1038/nature22993 Tracing the origins of relapse in acute myeloid leukaemia to stem cells Liran I. Shlush 1,2,3 *, Amanda Mitchell 1 *, Lawrence Heisler 4 , Sagi Abelson 1 , Stanley W. K. Ng 1 , Aaron Trotman-Grant 1 , Jessie J. F. Medeiros 1 , Abilasha Rao-Bhatia 1 , Ivana Jaciw-Zurakowsky 1 , Rene Marke 5 , Jessica L. McLeod 1 , Monica Doedens 1 , Gary Bader 6,7 , Veronique Voisin 7 , ChangJiang Xu 7 , John D. McPherson 4 , Thomas J. Hudson 4,6,8 , Jean C. Y. Wang 1,9,10 , Mark D. Minden 1,8,9,10 & John E. Dick 1,6 In acute myeloid leukaemia, long-term survival is poor as most patients relapse despite achieving remission 1 . Historically, the failure of therapy has been thought to be due to mutations that produce drug resistance, possibly arising as a consequence of the mutagenic properties of chemotherapy drugs 2 . However, other lines of evidence have pointed to the pre-existence of drug-resistant cells 3 . For example, deep sequencing of paired diagnosis and relapse acute myeloid leukaemia samples has provided direct evidence that relapse in some cases is generated from minor genetic subclones present at diagnosis that survive chemotherapy 3–5 , suggesting that resistant cells are generated by evolutionary processes before treatment 3 and are selected by therapy 6–8 . Nevertheless, the mechanisms of therapy failure and capacity for leukaemic regeneration remain obscure, as sequence analysis alone does not provide insight into the cell types that are fated to drive relapse. Although leukaemia stem cells 9,10 have been linked to relapse owing to their dormancy and self-renewal properties 11–13 , and leukaemia stem cell gene expression signatures are highly predictive of therapy failure 14,15 , experimental studies have been primarily correlative 7 and a role for leukaemia stem cells in acute myeloid leukaemia relapse has not been directly proved. Here, through combined genetic and functional analysis of purified subpopulations and xenografts from paired diagnosis/relapse samples, we identify therapy- resistant cells already present at diagnosis and two major patterns of relapse. In some cases, relapse originated from rare leukaemia stem cells with a haematopoietic stem/progenitor cell phenotype, while in other instances relapse developed from larger subclones of immunophenotypically committed leukaemia cells that retained strong stemness transcriptional signatures. The identification of distinct patterns of relapse should lead to improved methods for disease management and monitoring in acute myeloid leukaemia. Moreover, the shared functional and transcriptional stemness properties that underlie both cellular origins of relapse emphasize the importance of developing new therapeutic approaches that target stemness to prevent relapse. To track clonal dynamics during leukaemia initiation and progres- sion, we used a combined genetic and functional approach. Peripheral blood mononuclear cells collected from 11 patients with acute mye- loid leukaemia (AML) at diagnosis and relapse were obtained from our biobank. Whole-genome sequencing (WGS) was performed (coverage ~50×) on leukaemic blasts isolated from each sample (clinical and immunophenotypic data in Supplementary Tables 1 and 2, respectively). T cells purified from diagnosis samples were used as the germline reference and to identify ancestral pre-leukaemic mutations 16 . To evaluate the genetic diversity of leukaemia stem cells (LSCs), diag- nosis and relapse samples were transplanted into NOD/SCID/IL-2Rg c - null (NSG) or NSG-SGM3 (humanized cytokine) mice 17 . Human myeloid (CD45 + CD33 + ) and B-cell (CD45 + CD19 + ) populations were sorted from xenografts and genotyped by droplet digital PCR (ddPCR, sensitivity ~1 in 1,000) for a subset of the variants identified by WGS. Additionally, patient samples from diagnosis, relapse, and remission time points (where available) were sorted into four progenitor (CD33 − haematopoietic stem cells/multipotent progenitors (HSCs/MPPs), multi-lymphoid progenitors (MLPs), common myeloid progenitors/ megakaryocyte erythroid progenitors (CMPs/MEPs), and granulocyte monocyte progenitors (GMPs)) and four mature (CD45 dim CD33 + blasts, and T, B, and natural killer (NK) cells) populations using our established strategy 16 , and genotyped (5–20 variants per sample) by ddPCR (experimental design is outlined in Extended Data Fig. 1a). To investigate both the early stages of leukaemic development and the role of pre-leukaemic haematopoietic stem or progenitor cells (preL-HSPC) as a potential source of relapse 16,18 , we identified somatic variants predicted to have a damaging effect on the encoded protein (protein-damaging variants, PDVs) (Supplementary Table 3). These were defined as pre-leukaemic (preL-PDV) if they were present in T cells sorted from patient samples, or in B cells sorted from xenografts, or leukaemic (L-PDV) if absent in these populations. Patients had an average of 22.3 ± 14.6 PDVs and, remarkably, 30–50% of these PDVs were pre-leukaemic (Fig. 1a). Cell populations carrying preL-PDVs but not L-PDVs were classified as pre-leukaemic, whereas those with both preL-PDVs and L-PDVs were considered leukaemic. We found evidence of a pre-leukaemic cell population in all patients except for number 9, who bore a KMT2A (also known as MLL) translocation; the lack of cells with pre-leukaemic mutations in this patient was consistent with previous reports of few cooperating mutations in MLL acute leukaemias 19,20 . Analysis of the occurrence and variant allele frequency (VAF) of PDVs enabled estimation of the order of acquisition of muta- tions in each patient with AML (Fig. 1b and Extended Data Fig. 1b). However, because 97.5% of PDVs identified at diagnosis were also present at relapse, they were not useful for identifying the cellular origin of the relapse clone. Nevertheless, this analysis revealed considerable genetic evolution occurring during the pre-leukaemic phase. To investigate clonal dynamics between diagnosis and relapse time points, we identified somatic variants with VAF >20% in relapse blasts and <5% in blasts at diagnosis (relapse variants) (Supplementary Table 4). Relapse variants were largely intronic or intergenic, and compared with somatic mutations from the diagnosis sample were enriched for transversions (Supplementary Tables 5 and 6), consistent 1 Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario M5G 2M9, Canada. 2 Department of Immunology, Weizmann Institute of Science, Rehovot 76100, Israel. 3 Division of Hematology Rambam Healthcare Campus, Haifa 31096, Israel. 4 Ontario Institute for Cancer Research, Toronto, Ontario M5G 0A3, Canada. 5 Laboratory of Pediatric Oncology, Radboud University Medical Center, Nijmegen 6525 GA, The Netherlands. 6 Department of Molecular Genetics, University of Toronto, Toronto, Ontario M5S 1A8, Canada. 7 Donnelly Centre for Cellular and Biomolecular Research, Toronto, Ontario M5S 3E1, Canada. 8 Department of Medical Biophysics, University of Toronto, Toronto, Ontario M5G 1L7, Canada. 9 Department of Medicine, University of Toronto, Toronto, Ontario M5S 1A8, Canada. 10 Division of Medical Oncology and Hematology, University Health Network, Toronto, Ontario M5G 2M9, Canada. *These authors contributed equally to this work. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.