Genetic hierarchy and temporal variegation in the clonal history of acute myeloid leukaemia

Abstract

In acute myeloid leukaemia (AML) initiating pre-leukaemic lesions can be identified through three major hallmarks: their early occurrence in the clone, their persistence at relapse and their ability to initiate multilineage haematopoietic repopulation and leukaemia in vivo. Here we analyse the clonal composition of a series of AML through these characteristics. We find that not only DNMT3A mutations, but also TET2, ASXL1 mutations, core-binding factor and MLL translocations, as well as del(20q) mostly fulfil these criteria. When not eradicated by AML treatments, pre-leukaemic cells with these lesions can re-initiate the leukaemic process at various stages until relapse, with a time-dependent increase in clonal variegation. Based on the nature, order and association of lesions, we delineate recurrent genetic hierarchies of AML. Our data indicate that first lesions, variegation and treatment selection pressure govern the expansion and adaptive behaviour of the malignant clone, shaping AML in a time-dependent manner.

ARTICLE

Received 16 Feb 2016 | Accepted 5 Jul 2016 | Published 18 Aug 2016

Pierre Hirsch1,2,3,4,*, Yanyan Zhang5,6,*, Ruoping Tang4, Virginie Joulin5,6, Hlne Boutroux1,2,3,7, Elodie Pronier6,

Hannah Moatti1,2,3, Pascale Flandrin1,2,3, Christophe Marzac8, Dominique Bories9, Fanny Fava1, Hayat Mokrani6, Aline Betems6, Florence Lorre10, Rmi Favier8, Frdric Fger8, Mohamad Mohty1,3, Luc Douay1,2,8,Ollivier Legrand1,2,3,4, Chrystle Bilhou-Nabera1,2,3,8, Fawzia Louache5,6,** & Franois Delhommeau1,2,3,8,**

In acute myeloid leukaemia (AML) initiating pre-leukaemic lesions can be identied through three major hallmarks: their early occurrence in the clone, their persistence at relapse and their ability to initiate multilineage haematopoietic repopulation and leukaemia in vivo. Here we analyse the clonal composition of a series of AML through these characteristics. We nd that not only DNMT3A mutations, but also TET2, ASXL1 mutations, core-binding factor and MLL translocations, as well as del(20q) mostly full these criteria. When not eradicated by AML treatments, pre-leukaemic cells with these lesions can re-initiate the leukaemic process at various stages until relapse, with a time-dependent increase in clonal variegation. Based on the nature, order and association of lesions, we delineate recurrent genetic hierarchies of AML. Our data indicate that rst lesions, variegation and treatment selection pressure govern the expansion and adaptive behaviour of the malignant clone, shaping AML in a time-dependent manner.

1 Sorbonne Universits, UPMC Univ Paris 06, UMR_S 938, CDR Saint-Antoine, F-75012 Paris, France. 2 INSERM, UMR_S 938, CDR Saint-Antoine, F-75012 Paris, France. 3 Sorbonne Universits, UPMC Univ Paris 06, GRC n7, Groupe de Recherche Clinique sur les Myloprolifrations Aigus et Chroniques

MYPAC, F-75012 Paris, France. 4 AP-HP, Hpital St Antoine, Service dHmatologie clinique et de thrapie cellulaire, F-75012 Paris, France. 5 Institut National de la Sant et de la Recherche Mdicale (INSERM), UMRS 1170, CNRS GDR 3697 Micronit, 94805 Villejuif, France. 6 Institut Gustave Roussy, Univ Paris-Sud, Universit Paris Saclay, 94805 Villejuif, France. 7 Department of Pediatric Hematology and Oncology, AP-HP, Hpital Armand-Trousseau, F-75012 Paris, France. 8 AP-HP, Hpital Saint-Antoine & Hpital Armand-Trousseau, Service dhmatologie biologique, F-75012 Paris, France. 9 AP-HP, Hpital Henri Mondor, Unit dHmatologie Molculaire, F-94010 Crteil, France. 10 AP-HP, Hpital Saint-Antoine, Laboratoire commun de biologie et gntique molculaires, F-75012 Paris, France. * These authors contributed equally to this work. ** These authors jointly supervised this work. Correspondence and requests for materials should be addressed to F.D. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

DOI: 10.1038/ncomms12475 OPEN

Genetic hierarchy and temporal variegationin the clonal history of acute myeloid leukaemia

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

Acute myeloid leukaemia (AML) emerges from haematopoietic stem/progenitor cells (HSPCs) that acquire multiple genomic or chromosomal aberrations.

It has been proposed to afliate these heterogeneous aberrations to eight distinct functional groups according to their known or putative consequences on signalling, chromatin modication, DNA methylation, the cohesin complex, transcription factors, NPM1, the splicing machinery and tumor suppressors1,2. Some of these genetic lesions, including DNMT3A, TET2 and ASXL1 mutations, are considered as initiating events and are believed to result in clonal expansion of mutant HSPCs, leading to clonal haematopoiesis of indeterminate potential (CHIP)36, a condition that predisposes to the subsequent acquisition of leukaemic mutations. This model of leukaemogenesis ts with the results from previous studies that investigated the clonal evolution of AML with normal karyotype, showing for instance that DNMT3A mutations precede NPM1 or FLT3 mutations79.

DNMT3A mutations genetically dene pre-leukaemic stem cells in mouse xenotransplantation models of AML because they are sufcient to provide to mutant HSPCs a multilineage repopulation potential8. However, as DNMT3A mutations account for less than 30% of individuals with CHIP3 or adult AMLs2, the genetic variety of these diseases raises the question whether other chromosomal or genetic lesions behave as initiating pre-leukaemic events. In a particular subtype of AML, acute promyelocytic leukaemia, the PML/RARA fusion has been shown to be the initiating lesion10. In other types of AMLs, some lesions like TET2 or IDH2 mutations, as well as core-binding factor (CBF) or MLL translocations, are also thought to be initiating events, as they occur early in the clonal history, or can lead to pre-leukaemic haematopoiesis in xenotransplantation models7,1114.

In the present work, we examine how the models of initiation and clonal evolution already described in some subtypes of AML could be generalized to most cases. We analyse 74 non-promyelocytic AMLs, including 53 consecutive unselected cases reecting the genetic variety of AMLs, in order to identify the events that t with three major hallmarks of pre-leukaemic initiating events: (1) their early occurrence in the clone7,8,10,11, (2) their persistence at relapse8,11 and (3) their ability to provide a multilineage selective advantage to mutant over normal HSPCs in vivo7,8. By an integrated clonal reconstruction of AML using cytogenetic, molecular, targeted sequencing, whole exome sequencing and single-cell-derived colony genotyping data, we show that DNMT3A mutations, but also TET2, ASXL1, as well as CBF, MLL and chromosome 20q rearrangements are the rst driver events in most AMLs, can persist in remission, and are retained when patients experience relapse. Analysis of early and late relapse samples reveals that clones with these stable alterations variegate increasingly with time from diagnosis to recurrence. Moreover, cells from patients with these early lesions can repopulate bone marrow of xenotransplanted NOD/SCID/IL-2Rgc-null (NSG) mice with leukaemic or non-leukaemic engraftment, a functional signature of expanding pre-leukaemic stem cells8. Finally, our determination of clonal composition allows us to delineate genetic hierarchies that suggest distinct mechanisms of disease initiation and clonal expansion.

ResultsClonal composition of AML. To identify early putative initiating events, we analysed the clonal composition of 53 consecutive AMLs (Supplementary Fig. 1) by routine cytogenetic and PCR techniques, and targeted resequencing of 122 genes recurrently mutated in AML (Supplementary Tables 1 and 2). Additional whole-exome sequencing was performed in four MLL-rearranged

cases where only a few lesions were identied by targeted resequencing, one normal karyotype AML (UPN 2014-018), and one AML with monosomy 7 (UPN 2014-009) (Supplementary Tables 3,4, Supplementary Data 1). As multiple lesions can lead to various functional consequences1,2, we arbitrarily re-classied them from eight to four categories according to their main putative functional consequences (Supplementary Table 5). The rst category comprises lesions disturbing epigenetic regulation: mutations in DNA methylation and chromatin modiers, MLL and CBF translocations1517, and del(20q) (ref. 18). The second category includes mutations in splicing factors, transcription factors, and in NPM1, all impairing haematopoietic differentiation1922. The third category refers to mutations in genes that regulate proliferation and the fourth category to all other events.

Twenty-six out of the 53 samples were seeded in methylcellu-lose cultures to obtain single-cell-derived colonies. Colonies from 15 out of these 26 AMLs were next analysed by molecular and cytogenetic methods to infer the architecture of the principal clones from the existence of colonies with none, all or combinations of lesions detected in the bulk DNA. The 11 remaining cases were not further analysed due to either a lack of material or an insufcient number of detectable lesions.

We rst focused on eight patients with normal karyotype and mutations in TET2, DNMT3A or ASXL1, the most frequently mutated genes in CHIP35. In ve patients, TET2 mutations were found in rst position, preceding DNMT3A mutations in three double-mutant cases. In two other patients without TET2 mutations, DNMT3A mutations were found rst (Fig. 1a,b, Supplementary Fig. 2). The last patient had an ASXL1 mutation followed by lesions in EZH2, RUNX1 and FLT3 (Fig. 1b, Supplementary Fig. 2). In three cases (UPN2014-001, UPN2014-020, UPN2014-022), we found evidence for branching subclones involving variants in TET2, FLT3 and NRAS, all detected with low allele frequencies in the bulk AML DNA (Supplementary Table 3).

We then analysed the seven AMLs with abnormal karyotypes. Informative polymorphisms and uorescent in situ hybridization (FISH) were used to detect losses of heterozygosity (LOH) and translocations, respectively. In one case (UPN 2014-015), del(20q) was the founder event, followed by a splicing machinery mutation in U2AF1, and a mutation in the tyrosine kinase RET (Fig. 1c). In a second patient (UPN 2014-008), del(16q) was found after an U2AF1 mutation, and preceding an FLT3-internal tandem duplication (ITD) with LOH (Fig. 1e, Supplementary Fig. 3). In one patient with monosomy 7 (UPN 2014-009), the 122 gene-panel revealed mutations in SF3B1, PTPN11 and FLT3 (Supplementary Table 3). Exome sequencing did not identify any additional candidate initiating variant (Supplementary Table 4, Supplementary Data 1). Single-cell-derived colony analysis showed that the rst event was the mutation in SF3B1, followed by the PTPN11 one, the loss of chromosome 7 and the FLT3 mutation. In four other patients (UPN 2014-003, UPN 2013-004, UPN 2014-019 and UPN 2013-001), MLL translocations were identied as rst events by FISH, and RAS or FLT3 pathway mutations as last lesions by single-cell-derived colonies (Fig. 1d,e, Supplementary Fig. 3).

Taken together, single-cell-derived colony analyses revealed that mutations in epigenetic regulators, MLL rearrangements, and del(20q) are early events. Conversely, in 12/15 cases, proliferation-associated events, including FLT3 and RAS pathway mutations, were found as last events.

Early event retention and temporal changes in relapse AML. A second characteristic of pre-leukaemic lesions is that they persist at relapse8,11. To investigate this, we listed the changes in

2 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

a

Colonies

* *

* * * * *

* *

* *

*

*

* * *

Mutations

TET2 p.Y1294C

DNMT3A p.R882H

NPM1 type A

KRAS p.Q61H

NRAS p.Q61K

TET2 p.L627fs

*

Fluorescence intensity (delta Rn)

Cycle number

b

UPN 2014-001

UPN2014-022 CTCF?

SRSF2

UPN 2014-016

UPN 2014-024

UPN 2015-022

UPN 2014-041

PTPN11

NRAS?

UPN 2013-002

UPN 2014-020

FLT3-LOH

c

Del(20q)

U2AF1 p.S34F

RETp.G733D

* * *

* *

*

* * * * * *

*

UPN 2014-015

Del(20q) U2AF1 RET

d

MLL t(11;19)

2014-003

JAG1 p.T713M

CDC27 p.G88E

FLT3 p.A680V

RALBP1 p.C113fs

MLL t(11;19)

JAG1

RALBP1

CDC27

FLT3

e

UPN 2013-004

MLLt(10;11) XRCC5 ASXL1 KRAS SOS2

UPN 2014-008

U2AF1 Del(16q) FLT3-ITD FLT3-LOH

UPN 2014-019

MLL t(9;11)

MLL t(11;19)

HNRNPU

UPN 2014-009

SF3B1

PTPN11 FLT3

UPN 2013-001

Figure 1 | Single-cell-derived colony analysis identies a recurrent order of acquisition of AML mutations and chromosomal rearrangements. (a) Left panel: allele specic (AS) PCR amplication plots of mutant (red) and wild type (blue) alleles in seven colonies from patient UPN2014-001. Asterisks indicate mutant allele detection. Right panel: results of AS assays (rows) in 87 colonies (columns). Mutant colonies are indicated by coloured boxes. Blue: lesions in epigenetic regulators; cyan: lesions involving NPM1, transcription or splicing factors; green: proliferative lesions; pink: other lesions. (b) Mutation order in eight patients with normal karyotype AML. Dashed arrows and hatched circles indicate positions that could not be determined. (c) Analysis of 85 colonies from patient UPN2014-015 as in a. To detect del(20q), an informative single-nucleotide polymorphism (SNP)(rs11556379) was used. The inferred order of lesions is shown. (d) Genotyping analyses, as in a, and FISH analyses of colonies from patient UPN2014-003. Dissociations of the red and green signals indicate MLL rearrangements. Nuclei were stained with 4,6-diamidino-2-phenylindole. Bars indicate 10 mm. Right panel: results of FISH and AS assays, as in a. The inferred order of lesions is shown. (e) Order of lesions in ve additional abnormal karyotype AMLs.

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

clonal composition between diagnosis and relapse in 22 patients, including ve additional patients who had experienced relapse over 5 years after diagnosis (Supplementary Fig. 1). In all patients but two, from one to 20 genomic or chromosomal changes were detected (Fig. 2a, Supplementary Tables 1,6,7, Supplementary Fig. 4, Supplementary Data 2). In contrast to epigenetic modifying events, which were mostly unchanged in the relapse clones (16 changes out of 55 lesions), late proliferative events changed frequently (28 changes out of 36 events, Po0.0001, Fishers exact test) (Fig. 2b). Moreover, all epigenetic lesions identied as rst events in our previous colony analyses DNMT3A, TET2, ASXL1 mutations, MLL rearrangement and chromosome 20q deletionswere retained at relapse (Fig. 2a, Supplementary Table 7). We also observed that, while the number of retained lesions decreased with time to relapse (means.e.m.: 4.10.6, n 17 before 5 years versus

1.60.7, n 5 after 5 years, P 0.016, MannWhitney test), the

number of variegating lesions increased in the meantime (means.e.m.: 2.50.4, n 17 before 5 years versus 8.03.3,

n 5 after 5 years P 0.023, MannWhitney test) (Fig. 2c,

Supplementary Table 7 and Supplementary Fig. 4). Of note, four out the ve patients with late relapses had multiple changes in clonal composition. In one patient (UPN2015-003), we concluded that a second disease, with fully distinct genotype and karyotype from those of the initial leukaemia, had occurred. In the three other patients, persistent DNMT3A mutant clones had lost most mutations, including an NPM1 mutation (UPN2015-014), an IDH2 mutation (UPN2015-004), and a RUNX1 mutation (UPN2015-005). At relapse, these founding clones had re-evolved with several changes including new mutations in NPM1, IDH1 and RUNX1, respectively (Fig. 2a, Supplementary Table 7 and Supplementary Fig. 4). None of the relapse specic mutations was detected by deep sequencing with a 0.001 sensitivity in the diagnosis sample (Supplementary Table 8). In patient UPN2015-014, the analysis of sequential samples showed that the relapse arose after the emergence of an IDH2 mutant sub-clone, which had subsequently acquired a second NPM1 type A mutation identical to that of the diagnosis clone (Fig. 2a, Supplementary Table 8). These results suggest that leukaemia treatments had reset the clones back to their

a

b

20

18

Diagnosis Relapse Diagnosis Relapse

Variegating lesions (n =82)

Stable lesions (n =77)

5 m

14 m

16

28

2013006

CCND3

2014033 2014038

5 m 39

8

2015047

2014009

14 m

18

12

Lesions in epigenetic regulatorsMutations in NPM1, in transcription or splicing factors Mutations in proliferation/signalling pathwaysOther lesions

6 m

2014001

16 m

2014043

2014017

6 m

8 m

17 m

c

ASXL1 EZH2 EZH2

13

2013002

12

8 m

(11;21) JAK2 SETBP1 FLT3-ITD

2014016

11

35 m

10

Variegating lesions Stable lesions

2014021

t

9

2014039

8

(4;8;X)

9 m

5.4 y

7

2014042

2015014

6

5

6.3 y

2014016

9 m

NOTCH1

4

3

6.6 y

2

2014027

2015005

10 m

1

Time to relapse

RUNX1#1 DSCAM LOH Chr.21

7 ZRSR2RUNX1#2 PTPN11

1 Years

TET2

EZH2BCORASXL1 (6;19;16)

2014037

11 m

11 m

NRAS

2015003

2

BCORL1#2

3

4

5

2014015

11 m

6

7

2014029

2015004

8

9

(1;6)

10

, in transcription or splicing factors

Figure 2 | Temporal variegation govern AML relapse. (a) Clonal composition of 22 AMLs at diagnosis and relapse. Internally tangent circles represent successive lesions. Arrows indicate subclonal evolution inferred as indicated in the material and methods section. Delay from diagnosis to relapse is indicated in the upper right corner of each panel (m, months; y, years). (b) Numbers of stable and variegating lesions between diagnosis and relapse samples in the 22 AMLs. Colours are as in (a) with TP53 mutations included in the group of other lesions. (c) Changes in clonal composition as a function of time to relapse. Droplets above and below the x axis represent the number of variegating and stable lesions, respectively. Paired droplets of single patients are connected by vertical lines. Droplets outlined by dashed lines indicate lesions from one patient whose relapse lesions (top droplet) were all different from diagnostic ones (bottom droplet). Grey circles indicate cases with no changes in clonal composition.

4 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

pre-leukaemic states, with evolutionary potentials similar to those of the ancestral clones.

According to these observations, one may expect early relapses developed from unchanged clones to be drug resistant, and late relapses emerging from the pre-leukaemic pool to be drug sensitive. We investigated this in the 19 out of 22 relapse AMLs who received intensive chemotherapy or 5-azacytidine as salvage therapy (Supplementary Table 1). We did not nd any signicant association between the response to the second treatment and either the clonal variation or the delay to relapse. Larger studies are mandatory to fully address this question.

To assess the persistence of relapse reservoirs after treatment, we performed Sanger sequencing (Fig. 3a), FISH (Fig. 3b), quantitative reverse transcriptionPCR or deep sequencing analyses (Fig. 3c,d) in remission samples from 12 patients, including two (UPN2014-022 and UPN2014-041) who were still in remission 21 and 27 months after diagnosis, respectively, and 10 who further experienced relapse. Early DNMT3A, but also TET2 and ASXL1 mutations, as well as MLL and del(20q) rearrangements were all detected in these samples, whereas other lesions were inconstantly present (Fig. 3, Supplementary Table 8).

Taken together, these results show that cells carrying these early initiating lesions are frequently not eradicated by the treatment, and are able to re-initiate the disease.

NSG repopulation by AML cells with early initiating lesions. The pre-leukaemic potential of AML lesions can be assessed in xenotransplantation models7,8. We thus injected 5 106

T-depleted mononuclear cells from 38 AML blood samples, including 19 samples from our prospective cohort and 19 from other patients (Supplementary Fig. 1), to sublethally irradiated NSG mice to analyse their repopulation capacities with respect to cytogenetic and mutational patterns. Among these samples, 13 led to overt leukaemic engraftment with a clonal composition of human cell population remarkably mirroring the one of injected samples (Fig. 4a, Supplementary Figs 5 and 6, Supplementary Table 9). Fifteen samples repopulated NSG bone marrow with more than 0.5% of human CD45 (hCD45 ) cells but no overt leukaemia (Fig. 4a,b). Lymphoid or lympho-myeloid repopulation was detected in 13 of these 28 engrafting samples, including samples with DNMT3A, TET2, ASXL1 mutations, CBF or MLL rearrangements at injection. In total, of the 38 AML samples, most samples with TET2 mutations (7/8), DNMT3A mutations (9/12), ASXL1 mutations (6 /7) and MLL fusions (3/4) led to either leukaemic or non-leukaemic repopulation (Fig. 4a,b, Supplementary Figs 5 and 6).

As TET2 mutations were associated with DNMT3A or ASXL1 mutations in 6/7 engrafting samples, and preceded DNMT3A mutations in some patients, we next asked whether isolated TET2 loss could induce pre-leukaemic expansion. CD34 cells from cord blood were transduced with lentiviruses expressing small hairpin RNA (shRNA) designed to knockdown TET2 or scramble control23 and injected into NSG mice. Four months after injection, TET2 knocked-down cells showed a 2.6-fold greater repopulation (means.e.m.: 506% of hCD45 cells in mouse bone marrow) than control cells (195%), with both lymphoid and myeloid potentials (Fig. 4c,d, Supplementary Fig. 7a). Moreover, TET2 knocked-down human cells sorted from primary recipients bone marrow led to lympho-myeloid repopulation 12 weeks after injection into secondary recipients (Supplementary Fig. 7b). This result shows that TET2 depletion is sufcient to improve the multi-lineage repopulation of NSG bone marrow with intact self-renewing capacity, a signature of preleukaemic stem cell function, as previously dened for DNMT3A mutations8. It has been proposed that the expansion of HSPCs with DNMT3A mutations favours subsequent accumulation of

additional mutations, driving progression to AML8. Altogether, our xenotransplantation results support this model for other putative pre-leukaemic lesions, including TET2 mutations, ASXL1 mutations, and MLL rearrangements.

Clonal composition denes distinct AML genetic hierarchies. By combining our data, we then attempted to reconstruct the clonal phylogeny of all our AML samples (Supplementary Fig. 8). When available, single-cell-derived colony analyses conrmed in most cases the order of lesions inferred from cytogenetic, molecular and sequencing data obtained on the bulk material (Supplementary Tables 10 and 11) indicating a strong correlation between the order of events inferred from variant allele frequencies (VAFs) and those inferred from colony analysis. In 49/72 cases at diagnosis, the rst events observed were mutations in epigenetic regulators, CBF translocations, MLL rearrangements or chromosome 20q deletions, with a frequent (n 27) early

accumulation of epigenetic events in a dominant clone. This rst hit was recurrently followed by lesion affecting NPM1, transcription factors, or the splicing machinery, and then mutations in signalling pathways (Fig. 5a, Supplementary Table 10).

Depending on initial lesions, distinct genetic hierarchies were delineated (Fig. 5b). First, 27/72 patients had a genetic hierarchy reminiscent of CHIP, with DNMT3A, TET2 or ASXL1 mutations co-occurring with either mutations in NPM1 or in major hematopoietic transcription factors: RUNX1, CEBPA, and GATA2. In a second group of 10 patients, we found mutations in NPM1 or in these transcription factors, but no mutations in DNMT3A, TET2 or ASXL1. A third group of 14 patients had CBF, MLL or chromosome 20q rearrangements in rst position, but no mutations in NPM1 or in the aforementioned transcription factors. Finally, in 21/72 patients, neither CBF, MLL, 20q rearrangements nor mutations in NPM1, RUNX1, GATA2 or CEBPA were retrieved. Beside cases with germline variants predisposing to AMLthree DDX41 mutations24, one CSF3R mutation and one 14q32.2 duplication25, this group comprised all seven patients with TP53 mutations (Fig. 5b, Supplementary Fig. 9). Within the latter cases, we found co-occurring DNMT3A mutations in three out of four de novo AMLs but not in two therapy-related AMLs and one secondary AML. As VAFs indicated that TP53 and DNMT3A mutations occurred in a unique dominant clone (Supplementary Tables 3 and 11), we analysed 155 and 51 single-cell-derived colonies from patients UPN2014-027 and UPN2015-027 (Fig. 6a,b). We failed in conclusively determining which of the TP53 or DNMT3A mutations occurred rst because all 138 and 51 mutant colonies from the two patients, respectively, had both mutations. These results contrasted with those previously obtained from the seven DNMT3A or TET2 mutant AMLs with normal karyotype, where subsets of colonies with single founding mutations were detected, as a reection of an expansion step preceding the acquisition of the second event (Fig. 6c).

DiscussionIn the present study, we reconstructed the clonal composition of AML with a focus on known driver lesions and their putative pre-leukaemic initiating potential. We identify distinct routes for AML initiation and evolution in adults. The main one involves mutations in epigenetic regulators, such as DNMT3A, TET2 and ASXL1. Lesions in these genes have been detected in 68% of individuals with CHIP and haematologic driver mutations35 but in smaller proportions of de novo non-promyelocytic AMLs in our series (31/59, 53%) and in The Cancer Genome Atlas study2 (66/180, 37%). This suggests that several other pre-leukaemic expanding lesions initiate other AMLs. First, del(20q) may be a

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

a

UPN2014039

UPN2014043

TET2p.C1358Y

UPN2014022

TET2

DNMT3Ap.R882H

TET2p.R1261C

DNMT3Ap.W409*

TET2p.P818Lfs*2

p.W1003*fs*1

Diagnosis

AAGYGG

AATTGC

GGWAAA

WWYYCC

Remission

ATGRGC GTRCCG

ATGGGC GTGCCG

AAGCGGAATTGCAAGCGG

ATGRGC GTGCCG

GGWAAA TTCCCC

AATCGC

GGAAAA TTCCCC

Wild type sequences

b

Diagnosis

Remission

c

UPN2014-015

1

0.1

0.01

0.0001

1

0.1

0.01

Fraction of interphasic nuclei with del(20q)

U2AF1

RET

2014015

VAF

0.001

Diagnosis

Remission

Relapse (322 days)

d

TET2

BCORL1#2

UPN2014039

UPN2015-014

1

0.1

0.01

0.001

0.0001

UPN2014037

EZH2 BCOR

BCORL1#1

CEPBA

1

0.1

0.01

1

0.1

0.01

ASXL1

TET2

ZRSR2

VAF

DNMT3A

NPM1

DNMT3A

NPM1

IDH2

SMC1A

PTPN11

RUNX1

WT1

FLT3

0.001

0.001

Diagnosis

Remission

Relapse (322 days)

Diagnosis

Remission

Relapse (253 days)

Diagnosis

Remission Relapse

(1966 days)

UPN2014-038

UPN2014-041

UPN2014-001

1

1 TET2#1

VAF

FLT3-ITD

NRAS

PTPN11

1

0.1

0.01

0.001

0.0001

0.00001

Diagnosis

MLL-AF6 / GUS

Transcript ratio

0.1

IDH1

DNMT3A

0.1

0.01

0.001

0.0001

PTPN11

NRAS

TET2#2

DNMT3A

0.01

0.001

<0.0001

NPM1

NRAS

NPM1

KRAS

0

100

200

300

Days

Diagnosis

Remission (1, 2, 6 courses)

248 days

Diagnosis

Remission (1, 2 courses)

Relapse (173 days)

Haematopoietic stem cell transplantation

Maximum background error rate

Figure 3 | Retention of lesions in remission samples denes reservoirs for persistent clonal haematopoiesis and relapse (a) Sequence traces of paired diagnostic and remission samples from three AML patients with DNMT3A and TET2 mutations. Arrows indicate the detection of mutant alleles.(b) FISH analysis of diagnosis and remission samples from patient UPN2014-015 with persistence of del(20q). Presence of one instead of two red green

signals indicates del(20q)(outlined nuclei). Bars indicate 10 mm. (c) Kinetics of fraction of interphasic nuclei with del(20q) and VAFs of U2AF1 and RET mutations determined by targeted or deep sequencing in samples from patient UPN2014-015. Black dashed lines over shaded areas delimit the maximum background error rate for single nucleotide variant detection of patient specic mutations by deep sequencing, and the 0.01 limit of detection for the gene panel targeted sequencing. Different time points of follow-up are marked by arrows. (d) Kinetics of lesions in six additional patients as in (c). MLL-AF6 transcript quantication in patient UPN2014-038 was monitored by quantitative reverse transcriptionPCR with a limit of detection of 0.01.

bona de CHIP lesion, as it was found in non-tumoral cells from patients with multiple myeloma devoid of myeloid malignancy26. Second, RUNX1/RUNX1T1 translocations were detected in

Guthrie cards from new-borns who developed AML several years later27. Third, both CBF and MLL translocations promote non-leukaemic repopulation of immunocompromised mice by

6 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

a

Mice with human lymphoid cells < 0.5%

Mice with human lymphoid cells 0.5%

100

80

60

% of hCD45+ cells

40

20

5

43 21

0.5%

0

Del(20q)

MLL ASXL1 DNMT3A

TET2

UPN

2014002

2015007

2015051

2015048

2014003

2015020

2015050

2015019

2015004

2015054

2015058

2015053

2014040

2015061

2015008

2014013

2014019

2013004

2015060

2015018

2014020

2015059

2015062

2015021

2015055

2015057

2015056

2014007

2014042

2014038

2015049

2015052

2014041

2014039

2013003

2014043

2014037

2015001

No engraftment

No overt leukaemic engraftment

Overt leukaemic engraftment

b c d

SSC

CD19

Experiment #1 Experiment #2 Experiment #3

ShRNA scramble

ShRNA

TET2

1,000

800 600 400 200

0

hCD45

1,000

800 600 400 200

0

10

CD33

10

10

10

10

1,000

800 600 400 200

0

UPN2014043UPN2015021

SSC-A CD19 CD33

100

90

80

10

10

10

10

10

1,000

800 600 400 200

1,000

800 600 400 200

0

70

10

% of hCD45+cells

10

UPN2014019

10

10

10

10

10

10

10

10

60

10

10

50

10

10

40

10

10

30

0 10 10 10 10

10

10

10

20

10

10

10

0

10

ShRNA

scramble

P <.001

ShRNA

TET2

10 10 10 10 10 10 10 10

hCD45

0 10 10 10 10 10 10 10

Figure 4 | AMLs with early epigenetic lesions have leukaemic and non-leukaemic repopulation capacities in NSG mice. (a) Percentages of human CD45 (hCD45 ) cells in the bone marrow of NSG mice 843 weeks after injection of 5 106 mononuclear cells from 38 AML blood samples. The

presence of candidate pre-leukaemic lesions in injected cells is indicated in the co-mutation table. (b) Flow cytometric analysis of NSG bone marrow repopulated with non-leukaemic (UPN2014-019, UPN2015-021) and leukaemic (UPN2014-043) cells. (c) Repopulation of NSG bone marrow by shRNA TET2 or scramble transduced cord blood CD34 cells. Bars indicate the median (MannWhitney test). (d) Flow cytometric analysis of the bone marrow from two representative mice transplanted with control (shRNA scramble) and TET2 knocked-down (shRNA TET2) cells.

human HSPCs1214. In line with these observations, we found that del(20q), CBF and MLL translocations are early AML events retained in the clone when patients experience relapse. Finally, in some AMLs without these lesions, we found early mutations in IDH2, which have also been proposed to be pre-leukaemic lesions8,11, or early mutations in SF3B1, U2AF1, TP53 and JAK2, all detected in a few individuals with CHIP3,4. In our work, we did not perform whole-genome sequencing but focused on 122

genes recurrently mutated in AML with additional whole exome sequencing in seven patients. Thus, we cannot exclude that some lesions missed by our strategy can be rst drivers. However, our results recapitulate those from larger studies with whole-genome and whole-exome analyses2. Moreover, when analyzing non-promyelocytic AMLs from the TCGA cohort, we were able to retrieve the four genetic hierarchies found in our study with roughly similar proportions of cases.

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

a

100%

*

7 10

33 Lesions in epigenetic regulators

Mutations in NPM1, intranscription or splicing factors Other lesions

Mutations in proliferation/ signalling pathways

Percentage of events

*

* *

80%

13 8

8

60%

13

49

40%

15

20%

34 15

7

23

0%

First

Early

events

Late

event

Intermediate

Last event

b

CHIP Initiating chromosomal aberrations

NPM1/RUNX1/GATA2/CEBPA mutations without CHIP lesions

DNMT3A NPM1 FLT3-ITD FLT3-LOH

ASXL1 SRSF2 BCOR ETV6 RUNX1

UPN 2014042

Del(20q)

TET2#1 TET2#2 NPM1

UPN 2014036

UPN 2015049

UPN 2014014

GATA2 CEBPA

UPN2015051 UPN

2014014

CBFB/MYH11

SMC1A RUNX1 SF3B1 UPN

2015058

MLL/ELL HNRNPU

UPN 2013001

Mutations in master CHIP genes (DNMT3A, TET2, ASXL1), mutations in NPM1, RUNX1, GATA2, CEBPA

Pre-leukaemic chromosomal aberrationsNo mutation in NPM1, RUNX1, GATA2, CEBPA

No mutation in master CHIP genes, mutations in NPM1, RUNX1, GATA2, CEBPA

Karyotype

Normal Not done Gain/loss of chromosome X/Y

Inv(16) t(8;21)

MLL Del(20q) Complex Otheraberrations

Epigenetic lesions

DNMT3A TET2 ASXL1

IDH2 IDH1 Others

NPM1,

transcription factors (TFs), splice

NPM1 RUNX1

GATA2 CEBPA Other TFs Splicing machinery

Proliferation events

FLT3 RAS pathway

KIT Others

TP53

Other events

ATM OtherDNA repair mutations

Cohesin NOTCH pathway

PHF6 Others

AML predisposition

AML sub-type

Germline variants

De novo AML tAML sAML

No putative

pre-leukaemic chromosomal aberration

No mutation in NPM1, RUNX1,

GATA2, CEBPA

TP53 mutations

Germline predisposition to AML

TP53 mutated

TP53 p.R150W /p.C143Y t(8;5) add(3p) del(7q) del(12p)

?add(14q) der(15) 15

JAK2 JAK2LOHTP53

del(5q) del(1p) der(17) add(22p)x2 7

t-AML De novo AML

De novo

UPN 2014014

UPN 2015003

UPN 2015057

NRAS del(6q) t(?x;8) SH2B2 TP53 SF3B1 add(6)

TP53 p.Y181C 16 del(5q)DNMT3A t(3;7)

IDH1 TP53 p.A120V

UPN 2014027

Germline predisposition to AML

Germline DDX41 p.E268 fs BRCC3 CDK19 SRSF2 DDX41

OGTCUX1

UPN 2014018

Germline CSF3R p.W545* IDH2 SRSF2

UPN 2014025

Order of lesions

First Early Late Last

2 events

First + early First + late First + last Early + late

Late + last

Early + last

5 | Founding lesions dene distinct genetic hierarchies and clonal histories in AML. (a) In 49 patients with lesions in epigenetic regulators rst, subsequent events were classied as early intermediate (second to median), late intermediate (median 1 to penultimate) and last event. Histograms

show the distribution of events at each chronological position. * indicate Po0.05, Fishers exact test. (b) Co-mutation table of 72 AMLs at diagnosis. The bottom left colour code indicates the position of each lesion as dened in a; hatched boxes mean Z2 lesions at distinct positions. Groups of patients with distinct genetic hierarchies were dened according to the co-occurrence or exclusion of mutations in the three master genes involved in CHIPDNMT3A,

TET2, ASXL1, mutations in NPM1 and in haematopoietic transcription factorsRUNX1, GATA2, CEBPA and pre-leukaemic chromosomal aberrationsMLL and CBF rearrangements, del(20q). Clonal composition of representative AMLs with distinct genetic hierarchies are shown in boxes surrounding the mutation table. s-AML, secondary AML; t-AML, therapy-related AML. Internally tangent circles represent successive events. Colours are as in Fig. 2a.

Figure

8 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

a

Colonies (n =84)

*

*

TP53 p.Y181C

*

*

* * *

Lesions

DNMT3A p.L508P

Ch.7q LOH

*

IDH1 p.R132H

*

Fluorescence intensity (delta Rn)

TP53 p.L330R

DNMT3A p.N838D

TP53 p.L111R

Chromosome 18

Chromosome 5

Chromosome 12

Chromosome 16

Chromosome 11

Cycle number

b

Colonies (n =51)

Mutations

LOH

c

UPN

2014024 2014022 20140162014001

2015022

2014020

Wild-type

#1 #2

3

TET2

BCLAF1

7

2014041 2014027

11

29

15

2

3

17

2015027

0

Figure 6 | Capture of early clonal expansion steps by single-cell-derived colony analysis. (a,b) Genotyping of single cell derived colonies from two de novo AMLsUPN2014-027, (a) and UPN2015-027, (b)with concomitant DNMT3A and TP53 mutations, as in Fig. 1. Losses of heterozygosity (LOH) were detected using informative SNPs. Asterisks indicate mutant allele detection or LOH. (c) Mutational patterns of individual colonies from patients with early initiating lesions. Cell sorting strategies before culture are indicated above each table. Dotted lines separate wild-type colonies, colonies with one mutation, and colonies with Z2 mutations/lesions. The number of colonies of each category is indicated below the tables.

Most of the aforementioned early lesions may lead to HSPC expansion and pre-leukaemic clonal haematopoiesis that requires additional hits for the onset of AML. In abnormally expanding HSPCs, the excess in mitoses may increase the absolute number of unrepaired replication errors, thereby favouring the accumulation of subsequent mutations. Depending on the genetic hierarchy of AMLthat is, the nature of the initial lesions

and the number of additional mutations necessary to trigger a full blown disease, distinct latencies from the onset of clonal haematopoiesis to AML may be expected. For instance, the typical CHIP-derived hierarchy of AML comprises early lesions in DNMT3A, TET2 and ASXL1 that aggregate together or with other mutations in epigenetic regulators, followed by mutations in NPM1 or in haematopoietic transcription factors, and then by

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

lesions in signalling pathways such as FLT3 and RAS. In other AMLs, CBF and MLL rearrangements induce both epigenetic and haematopoietic transcription factor deregulation1517. This may explain the requirement for less additional lesions, and may result in a shorter latency than in CHIP-derived diseases for the progression to acute leukaemia (Fig. 7a). Nevertheless, functional studies remain necessary to understand the mechanisms of lesion cooperation and their precise consequences in the dynamics of AML clones.

In one third of our patients, neither recurrent CBF or MLL translocations, nor typical CHIP hierarchy or mutations in NPM1, RUNX1, CEBPA and GATA2 were found. This group of patients was enriched in cases with germline predisposition to AML and TP53 mutated cases, including three de novo AMLs with concomitant DNMT3A mutations. TP53 mutations have a major role in therapy-related AML, in which the previous exposure to chemotherapy for a rst cancer leads to the preferential expansion of pre-existing heterozygous mutant

HSPCs28. Such an exposure does not exist in de novo AMLs, and our analysis of DNMT3A and TP53 mutations in single-cell-derived colonies suggests that co-occurring lesions may be essential to provide an expansion capacity to TP53 mutant HSPCs. In line with this, among the 39 TP53 mutations detected in 1,125 individuals with CHIP35, nine had co-occurred with other mutations, including two in DNMT3A and four in TET2.

Finally, we established a link between the number of clonal changes and the delay from diagnosis to relapse. In our AML series, as in other series of AMLs and acute lymphoblastic leukaemias29,30, relapses arose from the persistence of a clone in which a few lesions changed, or in which only the rst founding one was retained. In this last situation, we observed that some ancestral clones were able to re-evolve in a way similar to the one that built the initial genetic hierarchy, triggering relapse up to 10 years after the rst AML diagnosis. In patients who achieve complete remission (CR), this return to a pre-leukaemic condition raises the question whether the clone will progressively

a

Initiation of AML

Post-treatment evolution of AML

AML AML

Complex karyotype

Complex karyotype

AML

Time Time

b

Treatment

Early relapse

Few changes in the clone

Late relapse

Multiple changes in the clone

Long-term remission

Retention of the founding lesion

Cure

Eradication of the founding clone

Time

Figure 7 | The action of time and treatment selection pressure in the initiation and evolution of AML. (a) Schematic representation of the initiation and evolution of the malignant clone (coloured areas) at the expense of normal HSPCs (grey area) from the acquisition of the rst event to the onset of AML. Left diagrams show AMLs emerging after the occurrence of the rst lesions which endow HSPCs with enhanced expansion capacity. Right diagrams depict AMLs with TP53 mutations. For therapy-related AMLs (t-AML), the occurrence and treatment of a rst cancer, followed by HSPC depletion and recovery, are indicated. The nature of each lesion is indicated as follows: Epigenetic 1: mutations in DNMT3A, TET2, ASXL1; Epigenetic 2: Epigenetic 1 or other mutations in epigenetic modiers; Gene fusions: CBF or MLL translocations; NPM1/TF: mutations in NPM1 or haematopoietic transcription factors; Prolif.: mutation in proliferation/signalling pathways; TP53: TP53 mutations. (b) Post-treatment evolution of AML, according to various outcomes. CHIP, clonal haematopoiesis of indeterminate potential.

10 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

re-accumulate new driver lesions or will stay dormant as seen in the vast majority of individuals with CHIP who never develop any haematopoietic malignancy (Fig. 7b). Understanding the mechanisms by which the clone will leave dormancy to re-initiate leukaemia may provide powerful biomarkers and therapeutic targets to detect and prevent late relapses.

Methods

Samples. Bone marrow (BM) and blood samples were collected at the time of diagnosis, CR and relapse from AML patients after informed consent in accordance with the declaration of Helsinki and the local ethics committee of Saint-Antoine hospital. Mononuclear cells were obtained after coll separation and cryopreserved at the tumour bank of Saint-Antoine Hospital, Paris. Fresh or frozen samples were used for further cellular and molecular experiments. BM cells were sorted to enrich the samples in leukaemic or non-leukaemic progenitor cells. CD34 , CD38 ,

CD34 or CD34 CD38- cell sorting strategies were established according to initial leukaemia immuno-phenotype. Blood cells from CR were sorted to isolate the CD3 fraction. Immuno-magnetic cell sorting was performed using CD3,

CD34 or CD38 MicroBeads and columns (MACS, Miltenyi Biotec) according to the manufacturers protocols.

Umbilical cord blood samples were collected from healthy new-borns with mothers consent in accordance with the institutional review boards of the Etablissement Franais du Sang, Crteil France, and the Institut National de la Sant et de la Recherche Mdicale, Paris, France. CD34 cells were isolated using a dextran/coll based procedure followed by immuno-magnetic separation (MACS,

Miltenyi Biotec).

Targeted sequencing. A panel of 122 genes mutated in AML and myeloid malignancies was designed (Supplementary Table 2). Amplicon libraries were obtained from 225 ng of BM DNA, using HaloPlex Target Enrichment System (Agilent technologies), according to the manufacturers protocol. Sequencing was performed using a MiSeq sequencer (Illumina) using the manufacturer recommendations. Raw data from targeted sequencing have been deposited at EGA and are available on request (EGA study number: EGAS00001001779).

Results were analysed after alignment of the reads using the SureCall Software version 3.0.1.4 from Agilent Technologies. BWA MEM algorithm was used for alignment and Illumina SNPPET SNP Caller algorithm was used to identify single nucleotide polymorphism (SNP) and indel variants. Minimum allele frequency for variant calling was set at 5% with a minimum local depth at 40. All variants were manually checked using IGV 2.3 software. Variants identied in diagnosis or relapse samples were manually checked in paired relapse or diagnosis samples using IGV software to study clonal evolution between samples and to rule out low frequency variants. For all samples, average depth in target regions was 619 (range 2701650) and 98.26% (range 93.599.5) of target regions were covered by at least 100 reads.

Detection of described polymorphism was performed by an in-house software using Ensembl database. Non-described variants of potential interest with VAF 410% were sequenced using the Sanger method in the diagnosis BM sample and in the CD3 fraction from CR or non-sorted CR samples, in order to rule out artifacts and non-somatic variants. All primers are described in Supplementary

Information (Supplementary Table 12).

Exome sequencing. Exome sequencing was done after sequence capture, enrichment and elution according to the manufacturers instructions (SureSelect, Agilent) without modication except for library preparation (NEBNext Ultra kit, New England Biolabs): 600 ng of each genomic DNA were fragmented by sonication and puried to yield fragments of 150200 bp. Paired-end adaptor oligonucleotides from the NEBNext Ultra kit were ligated on repaired A tailed fragments, then puried and enriched by eight PCR cycles. In all, 1,200 ng of these puried libraries were then hybridized to the SureSelect oligo probe capture library for 72 h. After hybridization, washing and elution, the eluted fraction was PCR-amplied with nine cycles, puried and quantied by QPCR to obtain sufcient DNA template for downstream applications. Each eluted-enriched DNA sample was then sequenced on an Illumina HISEQ 2000 as paired-end 75b reads.

Image analysis and base calling was performed using Illumina Real Time Analysis (RTA 1.17.21.3) with default parameters.

The bioinfomatic analysis of sequencing data was based on the Illumina pipeline (CASAVA 1.8.2). CASAVA performs alignment of a sequencing run to a reference genome (hg19), calls the SNPs based on the allele calls and read depth, and detects variants (SNPs and Indels). The alignment algorithm used is ELANDv2 (performs multiseed and gapped alignments). Only the positions included in the bait coordinates were conserved. Genetic variation annotation was realized from IntegraGen in-house pipeline, consisting in annotation of genes (RefSeq) and known polymorphisms (dbSNP 132, 1000Genome, EVS) followed by a mutation characterization (exonic, intronic, silent, nonsensey.). For each position, the exomic frequencies (homozygous and heterozygous) were determined from all the exomes already sequenced at Integragen, and the exome results provided by 1000G, EVS and HapMap. Minimum average depth on the exome is around 70. Samples

from diagnosis and CR were paired-analysed and manually checked with IGV v2.3 to identify somatic variants. Conrmed somatic variants are shown in supplemental data. Raw data from exome sequencing have been deposited at EGA and are available on request (EGA study number: EGAS00001001779).

Methyl cellulose assays. Sorted cells from diagnostic samples were seeded in triplicate at 100 to 300,000 cells per 1 ml culture dish in 2% standard methylcellulose medium (Stem Cells Technologies) supplemented with 37% fetal calf serum, 12% bovine serum albumin, 1% L-glutamine, 10 6 M b-mercaptoethanol, 1 IU ml 1 of erythropoietin, 50 ng ml 1 of stem cell factor, 25 ng ml 1 of FLT3 ligand,10 ng ml 1 of interleukin 3, 10 ng ml 1 of interleukin 6, 10 ng ml 1 of granulocyte-colony stimulating factor, 5 ng ml 1 of granulocyte-macrophage colony-stimulating factor, 10 ng ml 1 of thrombopoietin, 100 IU ml 1 penicillin and 100 mg ml 1 streptomycin. Colonies were counted after 14 days and picked. When appropriate, after a wash in PBS, individual colonies were splitted into two cell suspensions to perform both genotyping and FISH analyses. The rst cell suspension was gently deposited onto 18-well immunouorescence slides. Slides were dried and xed for further FISH analyses. The second cell suspension, or whole colonies when no translocation was detected in bulk AML, was stored at 80 C for further geno-

typing assays.

Single-cell-derived colony genotyping assays. DNA from individual colonies was prepared from a 50 ml lysis buffer containing 0.5 ml tween 20, 18.5 mg recombinant proteinase K (Thermo scientic) and H2O, after incubation at 56 C for 1 h and 95 C for 15 min. Genotyping was performed using custom Taqman SNP genotyping assays (Life technologies) with 5 ml colony DNA, using a 7,500 fast real-time PCR system (Applied Biosystem), according to the manufacturers protocol. For chromosomal deletion investigation, SNP genotyping assays of described SNP of the deleted area were performed, using MYBL1 p.I624M for del(20q), FANCA p.G809D for del(16q) and CUX1 p.A418T for del(7). Sequences of probes and primers are described in Supplementary Information (Supplementary Table 13).

FLT3-ITD and HNRNPU large indel detection. The mutational status for FLT3-ITD and HNRNPU at diagnosis and in colonies was determined using high-resolution sizing of uorescent dye-labelled PCR amplication. Beyond its contribution to mutation detection, high-resolution sizing also allowed to estimate the allelic burden by measuring peak height ratios (mutant/wild-type mutant) on

a uorescence scale. PCR probes and primers are described in Supplementary Information (Supplementary Table 12).

Cytogenetic and FISH analyses. Conventional cytogenetic analysis was performed in diagnosis and relapse samples on R-banding metaphases obtained from 24 h unstimulated culture using standard procedures. Karyotypes were interpreted according to the recommendations of the International System for Human Cytogenetic Nomenclature. Chromosomal rearrangements were conrmed by FISH, using the following probes: MLL breakapart probe (LPH 013), AML1 (RUNX1) breakapart probe (LPH 027), del (5q) probe (LPH 024), del(7q) deletion probe (LPH 025), del(20q) deletion probe (LPH 020), inv(16) probe (LPH 022), fast FISH X, Y an 18 (LPF 002) (all from Cytocell Ltd, Cambridge UK). NUP98 (11p15) break probe (Kreatech), Vysis LSI D7S522 (7q31)/CEP7 dual colour probe (Abbott) and XL 7q22 / 7q36 deletion (MetaSystems) were also used. Hybridization signals were scored on a BX61 uorescence microscope with an UPLAN FLN 100 /1.30 oil immersion lens (Olympus, Rungis, France) in 176500 nuclei and

analysed using CytoLabView BandView/FISHView Analysis (VDS)6.0 software (Applied Spectral Imaging Ltd, Yokneam, Israel). The same FISH analyses were performed on colonies, except that fewer nuclei were analysed.

MLL/AF6 transcript detection and quantication. Total RNA was isolated from Ficoll isolated mononuclear cells by use of isothiocyanate guanidium method (Extract All, Eurobio, France). RNA (1 ug) was reverse transcribed using Moloney murine leukaemia virus reverse transcriptase. complementary DNA synthesis was performed with the following primers (forward: 50-GAGGATCCTGCCCCAAAG

AAAAG-30; reverse: 50-GGGAGAGGACAGCATTCGC-30). Sanger sequencing of the PCR product was performed in order to conrm the MLL and AF6 exons implicated into the fusion transcript. The patient follow-up was monitored using real time quantitative PCR of the MLL-AF6 fusion according to the European Leukemianet recommendation31, using the following primers and probe (forward: 50-GTCCAGAGCAGAGCAAACAGAAA-30; reverse: 50-GAAAATAAAATCTCA

TCACTCCATGG-30; probe: 50-CTCCCCGCCCAAGTATCCCTGTAAA-30) adapted from published data32 and normalized to GUS reference gene expression. Sensitivity reached 10 5.

CBFB/MYH11 transcript detection and quantication. CBFB/MYH11 transcript detection and quantication were performed according to the European Leukemianet recommendation31.

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 11

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475

Deep targeted sequencing. A targeted-resequencing panel including all variants detected at diagnosis and relapse in 11 patients was designed. Amplicon libraries were obtained from 57.6 ng of DNA, using HaloPlex HS Target Enrichment System (Agilent technologies), according to the manufacturer protocol. With this method, after an initial enzymatic digestion, DNA fragments are captured with target probes, and hybridized with two indexes. The rst index is sample specic. The second index is a unique random sequence of 10 nucleotides, with a specic sequence for each DNA fragment of the sample. After PCR amplication at the end of library preparation, amplicon families bearing the same indexes can be identied, allowing a better detection of PCR and sequencing errors, and a better sensitivity than standard HaloPlex target enrichment system. Sequencing was performed using a MiSeq sequencer (Illumina) using manufacturer recommendations. Results were analysed after alignment of the reads using the SureCall Software version 3.0.2.1 from Agilent Technologies. Variants detected at diagnosis and relapse were manually checked in the CR sample, using IGV software version2.3 in order to identify low frequency variants. The background error rate at a single nucleotide variant position was calculated as the ratio of the sum of amplicon families with non-reference or non-mutant bases to the total number of amplicon families at this position. For all patient-specic samples, the maximum background error rate was dened as the maximum value of all background error rates obtained at target positions. As indels are not subject to sequencing error-dependent miscalling, the sensitivity of indel detection was set as 1/(number of amplicon families).

Xenograft experiments with AML cells. NSG mice were bred and maintained under specic pathogen free conditions with acidied water (pH 5.3) at the animal facility of Gustave Roussy Institute. Animal experiments were performed in accordance with guidelines established by the Institutional Animal Committee. Peripheral blood mononuclear cells from AML patients were depleted in CD3 cells by RosetteSep human CD3 depletion cocktail (StemCell Technologies) and 5 106 cells were intravenously injected to female mice (68 weeks old) 24 h after

irradiation at 2.5 Gy from a 137Cs source33. Mice were analysed at 843 weeks post-injection. Cells from mouse BM were stained with rat anti-mouse CD45 (Biolegend) and mouse anti-human CD45, anti-human CD19, anti-human CD33 and anti-human CD3 antibodies (all from BD Pharmingen; clones and uorochromes are indicated in Supplementary Table 12). Stained cells were analysed on FACSort or FACSCanto II cytometers (BD Biosciences). The presence of o0.5% of human CD45 (hCD45 ) population was considered as non-engraftment. The presence of 45% of hCD45 cells with major CD33 population (475% of the hCD45 cells) was considered as overt AML engraftment. The presence of 0.5 to 5% of hCD45 or the presence of 45% of hCD45 with o75% of CD33 in the hCD45 population was considered as non-overt or non-leukaemic engraftment.

Xenograft experiments with TET2-depleted CD34 cells. CD34 cells from three to ten distinct cord blood samples were pooled and transduced with lentiviruses (pRRLsin-PGK-eGFP-WPRE, Genethon, Evry, France) expressing the green uorescent protein and either a short hairpin RNA targeting TET2 (shRNA-TET2, 50-GGGTAAGCCAAGAAAGAAA-30) or a scramble sequence (shRNA-scramble, 50-GCCGGCAGCTAGCGACGCCAT-30) as control23. Twenty-four hours after transduction, 2 105 cells were intravenously injected to

sublethally irradiated NSG female mice (68 weeks old). Mice were killed 1517 weeks after injection, and repopulation of mouse bone marrow (femurs and tibias) by human cells was assessed by ow cytometry, using APC-conjugated mouse anti-human CD45, PE-conjugated anti-human CD19, PE-conjugated anti-CD33 (all from BD Pharmingen). Antibodies are listed in the Supplementary Information (Supplementary Table 14).

For NGS experiments and for secondary transplantation, bone marrow of repopulated mice was enriched in human cells using a mouse/human chimera isolation kit (Stem Cell Technologies).

Determination of clonal composition. To determine the order of chromosomal and genomic lesions in each AML diagnosis or relapse sample, we combined the following quantitative results: (1) frequencies of karyotype metaphases harbouring each chromosomal aberration, (2) frequencies of interphasic nuclei with specic translocations, rearrangements, deletions or gains, (3) VAFs from exome or targeted sequencing runs for somatic gene mutations (single nucleotide variants and indels), (4) VAFs from exome or targeted sequencing for SNPs in sequenced regions with LOHs or copy number variations, (5) Peak height ratios of high-resolution sizing of uorescent dye-labelled PCR amplication for FLT3-ITDs. Except for FLT3-ITD peak height ratios, 95% condence intervals were calculated by using the numbers of analysed nuclei and read depths as sample sizes for cytogenetic and NGS quantications, respectively. These quantitative values were converted to fractions of cells harbouring the lesions (Variant Cell Fraction, VCF), taking into account LOHs, with or without copy number variations, as well as the gender of the patient for lesions on chromosome X. Then, lesions were ordered according to VCFs with correction in case of variant alleles involving genes mapped on chromosomes with imbalances or LOH. When available, we integrated the results from colony analyses performed at diagnosis and the order of events

from quantitative data from distinct time point samples (diagnosis/CR/relapse) to correct or rene the clonal phylogeny.

Data availability. Raw exome and targeted sequencing data have been deposited at the European Genome-phenome Archive (EGA, https://ega-archive.org

Web End =https://ega-archive.org )under accession number EGAS00001001779. All other relevant data is contained within the article or supplementary les, or available from the author upon request.

References

1. Dhner, H., Weisdorf, D. J. & Bloomeld, C. D. acute myeloid leukemia.N. Engl. J. Med. 373, 11361152 (2015).2. Cancer Genome Atlas Research Network. Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N. Engl. J. Med. 368, 20592074 (2013).

3. Genovese, G. et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N. Engl. J. Med. 371, 24772487 (2014).

4. Jaiswal, S. et al. Age-related clonal hematopoiesis associated with adverse outcomes. N. Engl. J. Med. 371, 24882498 (2014).

5. Xie, M. et al. Age-related mutations associated with clonal hematopoietic expansion and malignancies. Nat. Med. 20, 14721478 (2014).

6. Busque, L. et al. Recurrent somatic TET2 mutations in normal elderly individuals with clonal hematopoiesis. Nat. Genet. 44, 11791181 (2012).

7. Jan, M. et al. Clonal evolution of preleukemic hematopoietic stem cells precedes human acute myeloid leukemia. Sci. Transl. Med. 4, 149ra118 (2012).

8. Shlush, L. I. et al. Identication of pre-leukaemic haematopoietic stem cells in acute leukaemia. Nature 506, 328333 (2014).

9. Klco, J. M. et al. Functional heterogeneity of genetically dened subclones in acute myeloid leukemia. Cancer Cell 25, 379392 (2014).

10. Welch, J. S. et al. The origin and evolution of mutations in acute myeloid leukemia. Cell 150, 264278 (2012).

11. Corces-Zimmerman, M. R., Hong, W.-J., Weissman, I. L., Medeiros, B. C. & Majeti, R. Preleukemic mutations in human acute myeloid leukemia affect epigenetic regulators and persist in remission. Proc. Natl Acad. Sci. USA 111, 25482553 (2014).

12. Moriya, K. et al. Development of a multi-step leukemogenesis model of MLL-rearranged leukemia using humanized mice. PLoS ONE 7, e37892 (2012).

13. Basecke, J. et al. AML1/ETO promotes the maintenance of early hematopoietic progenitors in NOD/SCID mice but does not abrogate their lineage specic differentiation. Leuk. Lymphoma 46, 265272 (2005).

14. Wunderlich, M., Krejci, O., Wei, J. & Mulloy, J. C. Human CD34 cells expressing the inv(16) fusion protein exhibit a myelomonocytic phenotype with greatly enhanced proliferative ability. Blood 108, 16901697 (2006).

15. Figueroa, M. E. et al. DNA methylation signatures identify biologically distinct subtypes in acute myeloid leukemia. Cancer Cell 17, 1327 (2010).

16. Li, Y. et al. Epigenetic silencing of microRNA-193a contributes to leukemogenesis in t(8;21) acute myeloid leukemia by activating the PTEN/PI3K signal pathway. Blood 121, 499509 (2013).

17. Alvarez, S. et al. DNA methylation proles and their relationship with cytogenetic status in adult acute myeloid leukemia. PLoS ONE 5, e12197 (2010).

18. Aziz, A. et al. Cooperativity of imprinted genes inactivated by acquired chromosome 20q deletions. J. Clin. Invest. 123, 21692182 (2013).

19. El Hajj, H. et al. Retinoic acid and arsenic trioxide trigger degradation of mutated NPM1, resulting in apoptosis of AML cells. Blood 125, 34473454 (2015).

20. Dolatshad, H. et al. Disruption of SF3B1 results in deregulated expression and splicing of key genes and pathways in myelodysplastic syndrome hematopoietic stem and progenitor cells. Leukemia 29, 10921103 (2015).

21. Bolli, N. et al. Expression of the cytoplasmic NPM1 mutant (NPMc ) causes

the expansion of hematopoietic cells in zebrash. Blood 115, 33293340 (2010).22. Sportoletti, P. et al. The human NPM1 mutation A perturbs megakaryopoiesis in a conditional mouse model. Blood 121, 34473458 (2013).

23. Pronier, E. et al. Inhibition of TET2-mediated conversion of 5-methylcytosine to 5-hydroxymethylcytosine disturbs erythroid and granulomonocytic differentiation of human hematopoietic progenitors. Blood 118, 25512555 (2011).

24. Polprasert, C. et al. Inherited and somatic defects in DDX41 in myeloid neoplasms. Cancer Cell 27, 658670 (2015).

25. Saliba, J. et al. Germline duplication of ATG2B and GSKIP predisposes to familial myeloid malignancies. Nat. Genet. 47, 11311140 (2015).

26. White, J. S., Zordan, A., Batzios, C. & Campbell, L. J. Deletion(20q) as the sole abnormality in plasma cell myeloma is not associated with plasma cells as identied by cIg FISH. Cancer Genet. 205, 644652 (2012).

27. Wiemels, J. L. et al. In utero origin of t(8;21) AML1-ETO translocations in childhood acute myeloid leukemia. Blood 99, 38013805 (2002).

28. Wong, T. N. et al. Role of TP53 mutations in the origin and evolution of therapy-related acute myeloid leukaemia. Nature 518, 552555 (2015).

12 NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12475 ARTICLE

29. Ding, L. et al. Clonal evolution in relapsed acute myeloid leukaemia revealed by whole-genome sequencing. Nature 481, 506510 (2012).

30. Ma, X. et al. Rise and fall of subclones from diagnosis to relapse in pediatric B-acute lymphoblastic leukaemia. Nat. Commun. 6, 6604 (2015).

31. Gabert, J. et al. Standardization and quality control studies of real-time quantitative reverse transcriptase polymerase chain reaction of fusion gene transcripts for residual disease detection in leukemia-a Europe Against Cancer program. Leukemia 17, 23182357 (2003).

32. Scholl, C. et al. Development of a real-time RT-PCR assay for the quantication of the most frequent MLL/AF9 fusion types resulting from translocation t(9;11)(p22;q23) in acute myeloid leukemia. Genes Chromosomes Cancer 38, 274280 (2003).

33. Zhang, Y. et al. CXCR4 inhibitors selectively eliminate CXCR4-expressing human acute myeloid leukemia cells in NOG mouse model. Cell Death Dis. 3, e396 (2012).

Acknowledgements

This project was funded by the ARC foundation (NEML20110602421), the Rgion Ile-de-France (N2012-2-eml-06-UPMC_12016710), the Association Laurette Fugain (NJ15I409 to FD)(NALF/06-05 to F.L.), the Institut National du Cancer to F.L. (N2010-1-RT-04) and to P.H., and the Institut Universitaire dIngnierie en Sant. We would like to thank The Cancer Genome Atlas (TCGA) for providing free access to their somatic mutational data. We would like to thank Simona Lapusan, Franoise Isnard, Anne-Claire Mamez, Paul Coppo for their help in collecting samples, Christelle Mazurier and Christine Nguyen for their technical advices and assistance, Mlanie Letexier, Jean-Paul Saraiva and Emmanuel Martin for exome sequencing and analysis, Nicole Casadevall and William Vainchenker for helpful discussions and critical reading of the manuscript.

Author contributions

P.H. performed cell culture, genotyping and NGS experiments, interpreted the results and wrote the manuscript. Y.Z., F.L. performed xenograft experiments, interpreted the

results and wrote the manuscript. V.J. performed xenograft experiments. R.T. designed NGS assays and participated in material collection. H.B., H. Moatti and P.F. contributed in cell culture experiments and genotyping of colonies. E.P., H.M. and A.B. performed shTET2 experiments. C.M. and D.B. performed standard molecular analysis. F.F. and F.L. contributed in material collection, sample preparation and in standard molecular analyses. R.F. performed RUNX1 Sanger sequencing. F.F. performed ow cytometry analysis of patient samples. M.M. and O.L. contributed in follow-up of patients. C.B.N. performed cytogenetic and FISH analyses. F.D. designed the research, interpreted the results and wrote the manuscript. All authors contributed in manuscript review

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no conict of interest.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions

Web End =reprintsandpermissions /

How to cite this article: Hirsch, P. et al. Genetic hierarchy and temporal variegation in the clonal history of acute myeloid leukaemia. Nat. Commun. 7:12475 doi: 10.1038/ncomms12475 (2016).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

r The Author(s) 2016

NATURE COMMUNICATIONS | 7:12475 | DOI: 10.1038/ncomms12475 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 13