Welcome international healthcare professionals

This site is no longer supported and will not be updated with new content. You are welcome to browse and download all content already included in the site. Please note you will have to register your email address to access the site.

You are here

Functional Heterogeneity of Genetically Defined Subclones in Acute Myeloid Leukemia

Cancer Cell, 3, 25, pages 379 - 392


The relationships between clonal architecture and functional heterogeneity in acute myeloid leukemia (AML) samples are not yet clear. We used targeted sequencing to track AML subclones identified by whole-genome sequencing using a variety of experimental approaches. We found that virtually all AML subclones trafficked from the marrow to the peripheral blood, but some were enriched in specific cell populations. Subclones showed variable engraftment potential in immunodeficient mice. Xenografts were predominantly comprised of a single genetically defined subclone, but there was no predictable relationship between the engrafting subclone and the evolutionary hierarchy of the leukemia. These data demonstrate the importance of integrating genetic and functional data in studies of primary cancer samples, both in xenograft models and in patients.



  • AML subclones are discrete, genetically distinct entities in AML samples
  • AML subclones often have unique functional and morphological properties
  • Engraftment of AML cells in mice is not defined by evolutionary hierarchy
  • The AML founding clone is not equivalent to the AML-initiating cell in mice

Klco et al. track acute myeloid leukemia (AML) subclones identified by whole-genome sequencing and find that subclones of AML can correspond to different cellular populations within a single AML sample and can have different functional properties in vitro and in immunodeficient mice.


Although clonal heterogeneity in many tumor types has been clearly demonstrated, the functionality of tumor subclones and their relationships to “tumor initiating/stem cells” are not yet clear. Using acute myeloid leukemia (AML) as a model, we found that subclones can correspond to different cellular populations within a single AML sample and can have different functional properties in vitro and in immunodeficient mice. Specifically, xenotransplantation results in a dramatic decrease in the subclonal complexity of AML samples; engrafting subclones are not defined by recurrent mutations (e.g., FLT3) and do not reliably predict relapse. These studies suggest that engrafting potential is not uniform among AML subclones and that functional differences among subclones need to be considered in studies of primary cancer samples.


Cancer arises through an evolutionary process of somatic mutation and selection. Although it may be depicted as a linear sequence of mutational events that produces a homogeneous cell population, tumor evolution is associated with significant intratumoral heterogeneity (reviewed in Swanton, 2012 ). All cells in a tumor contain shared somatic mutations that reflect its clonal origin (the “founding clone”), but additional mutations are present in subpopulations of cells that define tumor subclones. This heterogeneity, and the presence of subclonal alterations, was recognized even in early models of tumor evolution, demonstrating that subclonal cytogenetic aberrations can define “sublines” within a tumor ( Nowell, 1976 )—which would now be defined as subclones.

New sequencing technologies have greatly improved the characterization of genetic heterogeneity in cancer. Previous work on acute myeloid leukemia (AML) and myelodysplastic syndromes (MDS) demonstrated that these myeloid disorders exhibit clonal heterogeneity that evolves upon disease progression and/or relapse (Ding et al, 2012, Ley et al, 2010, Mardis et al, 2009, Walter et al, 2012, and Welch et al, 2012). Similar observations have been made in other malignancies. Work from Gerlinger et al. (2012) found that renal cell carcinomas can show striking clonal variation within different geographic regions of a single tumor, and recent analysis of clonal architecture in breast cancer demonstrated a hierarchy that elucidated the phylogeny of mutational events within individual tumors (Ding et al, 2012, Navin et al, 2010, Nik-Zainal et al, 2012, and Shah et al, 2009).

In addition to genetic heterogeneity, functional heterogeneity also exists within a primary tumor and has largely been studied in the context of identifying cells capable of initiating tumors when transferred into immunodeficient mice. However, the relationship of these initiating cells (also referred to as cancer stem cells) to the clonal organization of a tumor is not yet clear. Previous studies of acute lymphoblastic leukemia (ALL) and colorectal cancer have begun to address this relationship: tumor subclones can be dynamic with serial passaging, and some display enhanced engraftment potential (Anderson et al, 2011, Clappier et al, 2011, Kreso et al, 2013, Notta et al, 2011, and Schmitz et al, 2011). However, studies of leukemia samples have thus far followed copy number alterations, and/or used ALL samples with single, well-defined initiating events (BCR-ABL1 or ETV6-RUNX1 gene fusions), or have used distinct clinical subsets that do not reflect the full spectrum of this disease. In addition, the regional heterogeneity of solid tumors (Ding et al, 2012, Gerlinger et al, 2012, Nik-Zainal et al, 2012, Shah et al, 2009, and Sottoriva et al, 2013) may introduce sampling bias when assessing clonal heterogeneity (especially in xenotransplantation models), making it difficult to generalize the results to other cancers. From the studies published to date, it also is not yet clear whether functional differences among tumor subclones can be observed beyond these experimental systems or whether they can be identified directly in patient samples.

In this study, we sought to explore the relationship between functional and genetic heterogeneity by following genetically defined subclones under different experimental and biological conditions in de novo AML samples with a wide range of phenotypic and genetic characteristics.


Sequencing and Somatic Mutation Identification of De Novo AMLs

We used whole-genome sequencing (WGS) to discover somatic mutations in the unfractionated bone marrow cells of 19 patients with de novo AML using previously described approaches (Ding et al, 2012, Ley et al, 2010, and Welch et al, 2012). Most of the assessed AMLs had a normal karyotype (11/18; 61%), and they encompassed a range of French-American-British (FAB) subtypes and mutational spectra ( Table 1 and Table S1 available online). All but one of the samples (AML54/UPN161510) have been analyzed previously by either exome sequencing (14 samples; Cancer Genome Atlas Research Network, 2013 ) or WGS (four samples; Ding et al, 2012, Ley et al, 2008, Ley et al, 2010, and Welch et al, 2012), although samples with existing WGS were reanalyzed to identify additional somatic variants. AML-associated single-nucleotide variants (SNV) and coding insertion-deletion (indel) mutations discovered by WGS were confirmed using targeted, deep digital sequencing with custom capture arrays ( Tables S1 , S2 , and S3 ), which demonstrated high reproducibility with repeated targeted sequencing of the same bone marrow samples ( Figure S1 A). The majority of the identified variants in each AML sample formed a variant cluster with a variant allele fraction (VAF) of 45%–50%, which corresponds to heterozygous somatic mutations present in nearly all cells in the sample; this variant cluster marks the founding clone, from which all leukemic cells descend. A smaller number of variants were present in clusters at lower VAFs and represent leukemic subclones that possess all of the founding clone variants as well as additional subclonal somatic variants that arose during the evolution of the tumor ( Figure S1 B). Although these subclones are derived from the founding clone and the subclonal variants have lower VAFs, it is important to note that, in many cases, they represent the most abundant leukemic cell population in the bone marrow.

Table 1 Features of 19 De Novo AML Samples

UPN AML No. FAB Somatic Fusions and Mutations a
933124 1 M1 DNMT3A L723fs, FLT3-ITD, NPMc SMC3 G662C
426980 28 M2 IDH2 R140L, IDH2 R140W
452198 31 M5 DNMT3A R882H, FLT3 D835H, FLT3-ITD, IDH1 R132H, NPMc,IDH2 R140Q
869586 43 M4 RUNX1 G164fs, WT1 A382fs, PHF6 G288V
161510 54 M2 IDH1 R132H, NRAS G13D, WT1 E143fs
255108 60 M0 FLT3-ITD, NPMc, WT1 e7+1, STAG2 e21+2
296361 62 M5 DNMT3A R882H, FLT3-ITD, NPMc, IDH1 R132H
303818 63 M0 SETBP1 D868N, ASXL1 Q588, KRAS I36M
348685 70 M4 FLT3-ITD, WT1 R370fs
375182 71 M5 DNMT3A R882H, FLT3-ITD, NPMc
433325 75 M2 CEPBA R343fs, DNMT3A Q515, NPMc, NRAS G13D, PTPN11 I545L, WT1 A381fs, SMC3 T1174I
492687 78 M2 FLT3-ITD, FLT3-D835Y, RUNX1 G305
498463 79 M1 DNMT3A P718L b , IDH1 R132C, NRAS G13D, EZH2 R690H
605322 83 M1 DNMT3A D781G, DNMT3A R320, FLT3 D839G, IDH1 R132C, TET2 P1115fs, TET2 E1352V, PTPN11 G503E
708512 87 M4 DNMT3A R882H, FLT3 D835E, IDH1 R132H, NPMc, KIT D816V
721214 88 M1 DNMT3A R882H, FLT3-ITD, NPMc
737451 91 M5 DNMT3A R882H, NPMc
809653 93 M1 NRAS G13D, TP53 E286G
852559 94 M1 PICALM-MLLT10 PTPN11 P491L, PHF6 F214fs

a Mutations found in significantly mutated genes ( Cancer Genome Atlas Research Network, 2013 ) are listed, in addition to ASXL1 and SETBP1 variants in AML63.

b Homozygous variant.

UPN, unique patient number.

Hierarchy of AML Mutations within Peripheral Blood Cells

AML subclones were identified using bone marrow samples from AML patients, but leukemic cells also circulate in the peripheral blood. Peripheral blood involvement can vary substantially across patients, and some studies have reported that the allelic burdens of single AML mutations can be different in the peripheral blood and bone marrow (Jilani et al, 2003, Krönke et al, 2011, and Ma et al, 2009). This raises the possibility that AML subclones may differ in their ability to leave the bone marrow and circulate in the peripheral blood. To determine whether AML subclones peripheralize in different ways, we used targeted sequencing to compare the clonal architecture of peripheral blood samples to concurrently obtained bone marrow samples. We found a similar clonal architecture in the peripheral blood in 13 of the 19 cases (68%; Figures 1 A, 1B, and S1 C); the remaining samples showed small but significant relative differences ( Figure S1 C). However, all variants detected in the bone marrow sample were present in the peripheral blood, including common recurrent coding mutations associated with AML ( Figure 1 C).


Figure 1 Mutational and Subclonal Comparison of Peripheral Blood and Bone Marrow Leukemia Samples Unfractionated bone marrow and peripheral blood leukemia samples were characterized by targeted capture followed by deep sequencing. (A) Similarity of the variant allele fractions (VAFs) of somatic variants for 19 AML cases (all variants with coverage >50× from all samples are shown). Variants present in coding regions of the genome are shown red, noncoding in gray, and sex chromosome variants in blue. (B) Four representative cases of the peripheral blood versus the bone marrow samples, demonstrating cases with strong concordance (AML31 and AML28) and more variable subclonal distributions (AML43 and AML88). (C) Comparison of the VAFs of recurrent AML mutations in paired peripheral blood and bone marrow samples, including coding mutations in DNMT3A; FLT3 (both ITD and D835); NPMc; and canonical IDH1, IDH2, and KRAS/NRAS mutations. See also Figure S1 and Tables S1 , S2 , and S3 .

The founding clone variants were commonly present at VAFs of ∼50% in the peripheral blood leukemia samples, suggesting that nearly all cells contained these leukemic variants, despite variable percentages of leukemic blasts ( Table S1 ). To determine whether hematopoietic cells other than blasts are derived from the leukemic clone, we used multicolor flow cytometry to isolate myeloid blasts (CD45dim, SSlow, and CD33+) and lymphocytes (CD45bright, SSlow, and CD33) from all 19 cases ( Figure 2 A); monocytes (CD45int/bright, SSint, and CD33bright) and mature myelomonocytic cells (CD45int, SShigh, and CD33dim/neg) were collected from a subset of cases. These different peripheral blood populations were then analyzed by targeted deep sequencing of all known somatic variants ( Table S4 ). As expected, sorted blasts contained all founding clone variants and also the vast majority of the subclonal variants identified in unfractionated marrow ( Figure 2 B). Surprisingly, somatic variants were also detected in other purified myeloid cells without blastic morphology ( Figures 2 C and S2 A), regardless of the morphologic features of the leukemia (represented by the FAB subtypes). Whereas some of the variants showed either enrichment or depletion in these cell populations (i.e., those that appear above or below the line Y = X in Figure 2 C; see also Table S5 ), the founding clone variants had VAFs of ∼50%, indicating they were present in nearly all of the cells analyzed (and thus are unlikely to represent technical contamination from flow sorting). Although AML is functionally defined by a differentiation block leading to an accumulation of blasts, some cells clonally related to the leukemia are still capable of myeloid differentiation. These mature cells are either derived from the leukemic clone or a related “preleukemic” hematopoietic stem-progenitor cell (HSPC) (Jan et al, 2012, Krönke et al, 2013, and Shlush et al, 2012).


Figure 2 Distribution of Leukemia Variants and Subclones in Peripheral Blood Cells (A) Representative sample demonstrating the isolation of different cell populations by flow cytometry using a combination of side-scatter, CD45, and CD33, as well as CD19 and CD3 when sufficient cells were available. , maturing myelomonocytic cells. Scale bar represents 10 μm. (B–D) Correlation of the bone marrow VAF with flow-sorted and enriched blasts (B), nonblast, nonlymphocytes, including maturing myelomonocytic cells and monocytes (C), and lymphocytes (D). Variants present in coding regions of the genome are shown red, noncoding in gray, and sex chromosome variants in blue. See also Figure S2 and Tables S4 , S5 , and S6 .

Other studies have shown that lymphocyte populations can harbor mutations or fusion events identified in concurrent myeloid malignancies (Miyamoto et al, 2000 and Smith et al, 2010). However, we found that the majority of the somatic mutations we identified in the unfractionated de novo leukemia samples were either not detected in peripheral blood lymphocytes or were present at very low levels (mean VAF = 1.79%; 95%ile VAF = 6.06%; Figure 2 D). This was also true for three cases with separately collected B and T lymphocytes ( Figure S2 B). In several cases, rare variants (99 of 6,805 total variants from all cases; 1.45%) demonstrated VAFs of >10%, implying they were present in a substantial fraction of the cells in the lymphocyte pool. The majority of these occurred in noncoding regions and are therefore not likely to be relevant for AML pathogenesis ( Table S6 ). In one case (AML71), a recurrent AML-associated mutation in DNMT3A at codon R882 had a VAF of 10.5% (8 out of 68 reads), but no other cases had previously identified recurrent AML mutations ( Cancer Genome Atlas Research Network, 2013 ) with VAFs >10% in the lymphocyte pools. We suspect that the mutations with higher VAFs in peripheral blood lymphocytes are present in multipotential HSPCs; these may have been acquired during embryogenesis or early hematopoietic development (i.e., not in the leukemic clone) or they arose in a preleukemic HSPC population that can contribute to lymphopoiesis. The variants with very low VAFs were probably caused by leukemic contamination of the flow-purified lymphocytes; morphologic assessment of these samples revealed that 2%–5% of the cells in the lymphocyte pools had myeloid morphology ( Figure S2 C). In addition, low levels of sequence contamination (0.5%–1%; Figure S2 D) occurred among indexed samples that were pooled together for targeted sequencing.

In sum, these data show that most AML subclones have an equal propensity to circulate into the peripheral blood. In addition, the majority of myeloid cells in the peripheral blood are derived from a common myeloid-skewed, transformed HSPC, either because this transformed cell defaults toward myeloid differentiation or because the AML-causative mutations block the potential for lymphoid differentiation. A consequence of this is that the VAFs of founding clone variants in the peripheral blood samples are influenced not only by the blast percentage, but also by the percentage of “contaminating” normal lymphocytes in the sample ( Figure S2 E).

AML Subclones Can Have Distinct Cellular Phenotypes

Although all AML subclones were represented among the different myeloid populations in the peripheral blood samples ( Table S5 ), some cases showed dramatic differences in the subclonal architecture in purified cell populations. A particularly striking example was AML31, an acute monocytic leukemia characterized by leukemic cells with predominantly monocytic features ( Figure 3 A) and only rare blasts (∼3%) in the peripheral blood. To enhance tracking of subclones in this AML across samples, we assigned leukemic variants to distinct clusters using a computational subclone identification method (C.A.M., Brian S. White, Nathan D. Dees, M.G., Obi L. Griffith, John S. Welch, Ravi Vij, Michael H. Tomasson, T.A.G., M.J.W., Matthew J. Ellis, William Schierding, J.F.D., T.J.L., E.R.M., R.K.W., and Li Ding, unpublished data) that clusters variants with similar VAFs across multiple samples. This approach identified three subclones (subclones 1–3): Subclones 1 and 2 were present in the de novo bone marrow but disappeared at relapse, and subclone 3 was rare at presentation but dominant at relapse ( Figure 3 B; Ding et al., 2012 ). Targeted sequencing of purified myeloid blasts from de novo AML31 (identified by dim CD45 expression, low side scatter [SSC], and the presence of Auer rods: Figure 3 A, inset) revealed an enrichment of variants associated with subclone 3 ( Figure 3 C), the main leukemic population present in the relapse sample. Consistent with this observation, AML31 exhibited predominantly blastic features at relapse, rather than the monocytic morphology of the de novo leukemia. In addition, cells from AML28 with flow characteristics of maturing myelomonocytic cells ( Figure 3 D) showed enrichment of subclonal variants (subclone 4 variants) previously found at disease relapse ( Ding et al., 2012 ). Characterization of the myeloid blasts and monocytes in AML87, an acute myelomonocytic leukemia (FAB M4), also revealed that different subclones had distinct morphologic properties ( Figure 3 E). Whereas such dramatic enrichment was not observed in all cases (see Figures 2 and S3 ; Table S5 ), these data demonstrate that some subclones correspond to distinct populations of cells with characteristic morphologic and/or immunophenotypic properties. This implies that some subclones are functionally distinct in their ability to differentiate into more mature cells.


Figure 3 Subclonal Enrichment in Distinct Myeloid Subpopulations (A) Cells from AML31 were flow-sorted, and distinct morphologic populations were confirmed by morphologic examination. Note Auer Rod present in the myeloblast (). Scale bar represents 10 μm. (B) Clonality plot demonstrating the relationship of the de novo AML31 tumor and the relapse leukemia ( Ding et al., 2012 ). Important coding mutations are highlighted. (C–E) Clonality plots demonstrating the relationship of the de novo leukemias to different cell populations; (C) AML31, blasts (top) and monocytes (bottom); (D) AML28, blasts (top) and maturing myelomonocytic cells (bottom); and (E) AML87, blasts (top) and monocytes (bottom). For all clonality plots, only variants in computationally identified clusters (SciClone; C.A.M., Brian S. White, Nathan D. Dees, M.G., Obi L. Griffith, John S. Welch, Ravi Vij, Michael H. Tomasson, T.A.G., M.J.W., Matthew J. Ellis, William Schierding, J.F.D., T.J.L., E.R.M., R.K.W., and Li Ding, unpublished data) that are diploid (copy number = 2) and with coverage depth >50× are shown. See also Figure S3 .

Single-Cell Genotyping Verifies Imputed AML Subclones

To confirm the identity of subclones at the single-cell level and to establish their hierarchy within the tumor, we used an amplicon-based sequencing approach to genotype the founding clone and subclonal SNVs in single cells purified by cell sorting. Leveraging the findings for AML28 (where purified blasts and maturing myelomonocytic cells were enriched for different subclonal variants; see Figure 3 D), we isolated individual cells from total myeloid cells (excluding lymphocytes) and from maturing myelomonocytic cells identified by flow cytometric and immunophenotypic features ( Figures 4 A) and verified these populations by morphologic examination ( Figure S4 A). Whole-genome amplification was then used to prepare DNA from each cell for subsequent PCR amplification and sequencing of ten known somatic mutations and nine known heterozygous SNPs (used to assess the frequency of allele-specific amplification and to establish accurate genotyping criteria; Figures S4 B–S4F). The fraction of single cells harboring each of the ten leukemic variants was in close agreement with predictions based on the variant allele fractions found in the unfractionated sample ( Figures 4 B and 4C). Founding clone variants and variants in subclones 1 and 2 were present in the majority of the purified myeloid cells, but subclone 4 variants were not detected in the total myeloid cell pool ( Figure 4 B). In contrast, subclone 4 variants were significantly enriched in the purified maturing myelomonocytic cells ( Figure 4 C). We next assembled the genotypes for each individual cell to establish the relationships of the subclonal variants within the tumor. This demonstrated that the predominant genotype among all myeloid cells included variants in subclones 1 and 2, in addition to founding clone variants; cells with subclone 1, 2, and 3 variants were the next most common ( Figure 4 D). It also established that subclone 3 arose from subclone 2 and implied the existence of an ancestral cell that contained only subclone 1 and founding clone variants. The subclone 4 genotype was most common in cells with the maturing myelomonocytic phenotype and arose directly from the founding clone, i.e., independently from subclones 1, 2, and 3 ( Figure 4 E). We also performed genotyping of single cells isolated from the unfractionated peripheral blood of AML31, which also confirmed the allele fractions and subclone genotypes imputed from the unfractionated bone marrow sample ( Figures 4 F and 4G).


Figure 4 Single-Cell Genotyping of Primary AML Samples Individual cells from AML samples 28 and 31 were isolated by flow cytometry, and single-cell genotypes were determined by whole-genome amplification and amplicon-based sequencing. (A) Flow-sorting strategy for AML28, in which individual myeloid cells (excluding lymphocytes) or maturing myelomonocytic cells were collected. (B–G) Single-cell genotyping results. (B), (C), and (F) show the proportion of cells harboring leukemia-associated variants predicted from the VAFs in unfractionated cells (in blue) and observed in individual cells (in red) for individual myeloid cells (B) and maturing myelomonocytic cells (C) from AML28 and peripheral blood from AML31 (F). For each comparison, the predictions from unfractionated cells used VAFs obtained directly from deep-sequencing read counts (and multiplied by two to correct for heterozygosity), and the observed single-cell proportions were obtained from single-cell genotyping experiments. Error bars show the 95% binomial confidence interval for each point estimate. (D), (E), and (G) show single-cell genotype frequencies from sorted myeloid cells (D) and maturing myelomonocytic cells (E) from AML28 and peripheral blood from AML31 (G). The bottom panel in each figure shows the observed single-cell genotypes, with each column representing a single observed genotype that consists of at least one of the founding clone or subclonal variants in each subclone (indicated in red). The height of the vertical bars and corresponding numbers show the frequency and absolute number of cells with the indicated genotype, respectively. Only genotypes observed in more than two cells in each single-cell experiment are shown; additional genotypes were also observed in low numbers of cells that are likely due to allele “dropout” (due to unequal amplification of the two alleles), which we estimated to be 30% using control data from heterozygous SNPs (see Supplemental Experimental Procedures ). See also Figure S4 .

Some AML Subclones Have Unique Functional Properties In Vitro

The findings noted above suggest that some subclones have distinct functional properties, including different capacities for hematopoietic differentiation in vivo. In addition, the presence of multiple subclones in some samples implies that these subclonal populations may have cell-intrinsic advantages that allow them to expand faster than the founding clone. To determine whether subclones have different growth properties in an experimental system, we used an established in vitro stromal coculture method ( Klco et al., 2013 ) to expand three different de novo AML samples with well-defined subclonal populations at presentation and relapse (AMLs 1, 31, and 43; Ding et al., 2012 ). These de novo AML samples were cultured on stromal cells in the presence of human hematopoietic cytokines for 7 days ( Figures 5 A and 5B) and then analyzed by targeted sequencing of all known variants in these genomes. Although the subclonal composition of AML1 ( Figure 5 C), AML43 ( Figure 5 D), and AML88 ( Figure S5 A) were stable after 1 week of expansion, the subclonal architecture of AML31 changed dramatically—subclone 3 variants showed substantial enrichment in culture, increasing from a mean VAF of ∼2% to almost 20% within 7 days ( Figures 5 E and S5 B). As described above, this subclone was highly enriched in myeloid blasts and was also the dominant subclone at relapse.


Figure 5 Rare Subclones Can Have Unique In Vitro Growth Properties Cells were grown in human hematopoietic cytokines in the presence or absence of MS5 stromal cells for 7 days. (A) Fold change in cell number after 7 days. (B) Percentage of 5-ethynyl-2′-deoxyuridine (EdU)-positive cells; cells were incubated with EdU for the last 18 hr of culture. Mean values (n = 3) are shown; error bars represent SD. (C–E) Subclonal architecture of AML1 (C), AML43 (D), and AML31 (E) at day 0 (label: de novo), relapse ( Ding et al., 2012 ), and day 7 of culture (label: in vitro). Each column shows the VAFs (indicated on the y axis) of founding clone and subclonal variants for each case. See also Figure S5 .

Preferential Subclone Engraftment in Immunodeficient Mice

Xenotransplantation studies are commonly used to study functional heterogeneity of primary AML samples, including the identification of phenotypes associated with leukemia-initiating cell populations. To examine the patterns of subclonal engraftment following xenotransplantation, we transferred cells from nine different AML samples into unconditioned immunodeficient mice (one million viable cells/mouse via lateral tail vein injections), including six samples that were concurrently injected into mice from the NOD-scid-IL2Rγnull (NSG) strain (Ito et al, 2002, Sanchez et al, 2009, and Sarry et al, 2011) and also the NSG-SGM3 strain ( Wunderlich et al., 2010 ), which expresses the human hematopoietic cytokines stem cell factor, granulocyte macrophage-colony stimulating factor, and interleukin 3; cells from AML1, AML62, and AML88 were injected only into NSG-SGM3 mice. A total of 73 mice (31 NSG and 42 NSG-SGM3) were injected; mice were sacrificed at 12–16 weeks or at the first sign of illness. Eight of the nine AML samples engrafted, with the NSG-SGM3 strain achieving equivalent or higher engraftment efficiencies in all cases, as expected ( Figure S6 A; Table S6 ).

To assess the subclonal composition of the xenografts, human leukemia cells were purified from 52 bone marrow xenografts by flow cytometry using antibodies to human CD45, CD33, and/or CD34 for subsequent sequencing. Flow cytometric characterization showed substantial immunophenotypic variability among the xenografts that was in part dependent on the mouse strain used. For example, AML31 xenografts (n = 11) showed higher CD34 expression in NSG-SGM3 mice compared to NSG mice ( Figure 6 A). Sequence analysis of AML31 xenografts revealed that subclone 3 was the dominant cell type in all five NSG-SGM3 mice (mean VAF 37%–46%), whereas the majority of the NSG animals (5/6) preferentially engrafted with subclone 1 ( Figure 6 B), which was the most abundant cell type in the de novo sample. This pattern supports the single-cell analysis of AML31, which showed that subclones 1 and 3 arose from the founding clone independently; accordingly, variants assigned to subclones 1 and 3 were mutually exclusive in the xenografts and were never present at similar allele fractions. These results also suggest that subclone 3 possesses a cell-intrinsic functional advantage, which appears to be enhanced by the cytokine milieu of the NSG-SGM3 mice; it was able to engraft and outcompete other subclones, despite being present in only a small fraction of the total cells in the injected sample (mean VAF: 2.5%). Notably, all the xenografts had significant subclonal restriction, and no xenografts had a subclonal architecture that was identical to that of the input leukemia.


Figure 6 Phenotype and Subclonal Composition of AML31 Xenografts Mice were injected with unfractionated peripheral blood leukemia cells and followed for 12–16 weeks, at which time cells were harvested from the bone marrow. (A) Representative examples of the immunophenotypic properties of AML31 xenografts in individual NSG and NSG-SGM3 mice. Shown are expression patterns of human CD34 and CD11b in the human myeloid leukemia cells (mCD45, hCD45+, and CD33+). (B) Human CD45+CD33+ cells were purified by cell sorting, and DNA was analyzed by targeted deep sequencing of all known somatic variants in the sample. The VAFs for all founding clone and subclonal variants are shown for each individual AML31 xenograft and the peripheral blood sample (input, far left). See also Figure S6 and Table S7 .

Immunophenotypic and sequence analysis of the xenografts from the other seven AML samples also demonstrated subclonal restriction ( Figures S6 B and S6C). In addition, immunophenotypic differences between the xenografts obtained from identical human AML samples injected into NSG and NSG-SGM3 mice were observed in other cases. For example, AML63 xenografts from NSG mice consistently displayed a CD34+CD33 immunophenotype, whereas xenografts obtained from NSG-SGM3 mice were CD34CD33+ ( Figure S6 B). In contrast to AML31, all of the AML63 xenografts were composed of the same subclone, suggesting that, in this sample, the same initiating subclone can display disparate surface antigen markers based on the cytokine milieu in which it develops.

The studies noted above tracked variants previously discovered by WGS (mean coverage ∼30×), which has a limit of detection for somatic variants of ∼10%. Thus, some rare variants that became dominant in the xenografts were probably below the limit of detection in the primary sample examined by whole-genome sequencing. In addition, previous studies have suggested that mutations can arise during passaging of human cells in immunodeficient mice ( Li et al., 2013 ). To discover additional somatic variants within the engrafted subclones, we performed targeted capture sequencing of the 264 genes that are recurrently mutated in AML and other myeloid neoplasms ( Cancer Genome Atlas Research Network, 2013 ). In most cases, there were no mutations in the xenografts that had not been identified in the initial tumor. However, we discovered a canonical IDH1 R132H mutation in all six of the AML62 xenografts (mean VAF of xenografts: 45.86%; range: 36.84%–58.44%). In retrospect, this mutation was then detected in the de novo sample (5.44%; 8 of 147 reads). Similar analysis demonstrated that this variant occurred at the level of sequencing errors (1 of 156 reads) in the relapse sample ( Figure S6 D). This implies that a rare subclone in the de novo AML sample containing this IDH1 variant preferentially engrafted in 6/6 mice, but did not contribute to the relapse in this patient; this pattern contrasts with what was observed for AML31, where a rare subclone with enhanced engraftment properties emerged at relapse. Lastly, only one xenograft (AML94, NSG3) contained a xenograft-specific mutation (TP53, e8-1; splice site mutation) that may have pathologic significance.

In sum, xenotransplantation of de novo AML samples resulted in skewing of the subclonal architecture in all samples, implying that genetically defined subclones usually represent the dominant engrafting cell population in an individual mouse. We detected subclones in 38 of the 52 xenografts from the eight AML samples. Most commonly, we observed engraftment by a single subclone (27 of 38 xenografts had only a single subclone with a mean VAF >5%). Of the 27 xenografts with monoclonal engraftment, 12 demonstrated engraftment and outgrowth of a minor subclone in the primary sample.


Studies of genetic heterogeneity in cancer thus far have focused on identifying the somatic variants that mark tumor subclones as a way to understand the origin, population dynamics, and evolutionary history of a tumor. However, it is not yet clear whether genetically defined tumor subclones possess unique phenotypic and/or functional properties that may explain some aspects of a tumor’s history and perhaps predict its future potential for relapse or resistance to therapy. Here, we used WGS followed by capture-based targeted deep sequencing to define the clonal architecture of unfractionated bone marrow cells of AML patients and then to follow these subclones after experimental manipulation. We purified individual cell populations with well-established cellular phenotypes and found that most myeloid cells in the peripheral blood at the time of AML diagnosis (even those that were morphologically nonblastic) were derived from the AML founding clone; in some cases, genetically defined subclones corresponded to distinct cell populations that could be identified by cell-surface markers. We also used xenotransplantation of unmanipulated tumor samples in immunodeficient mice (a mainstay in the experimental characterization of primary cancer samples) to understand the functional heterogeneity of tumor subclones and found that only one subclone engrafted in most mice even though multiple subclones were present in the sample that was injected. In some instances, the engrafting subclone represented only a small fraction of the injected cells (<10%), implying that some subclones have a cell-intrinsic advantage (due to increased engraftment potential, excess proliferation, and/or other factors) after transplantation. These observations, based on tracking tumor subclones with hundreds of somatic variants in primary AML samples with a variety of different initiating mutations, show that functional and phenotypic heterogeneity of subclones are manifest not only in experimental systems, but also in primary tumor samples at disease presentation.

Many studies of functional heterogeneity in leukemia have used xenotransplantation to characterize leukemia stem cells (LSCs; also known as leukemia-initiating cells), which are rare cells that are functionally defined by unique cell-surface markers as well as potential for engraftment and prolonged self-renewal in immunodeficient mice (Bonnet and Dick, 1997, Jaiswal et al, 2009, Jin et al, 2009, and Lapidot et al, 1994). More recent studies have found that engraftment is not necessarily restricted to specific cellular phenotypes, and models have been proposed where the cancer stem cell phenotype is stochastic and subject to equilibrium within the tumor cell population (Gupta et al, 2011 and Sarry et al, 2011). Our study adds another layer of complexity by showing that discreet subclones (not the founding clone) generally define the engrafting population in individual mice and that these subclones do not have equal engraftment and/or outgrowth potential. This argues against a purely stochastic model of functional diversity, because specific subclones preferentially engrafted in multiple experiments despite their low frequency in the injected sample. However, in some samples, the engrafting subclone differed across experiments and between immunodeficient mouse strains (e.g., AML31), demonstrating that different initiating populations with unique subclonal mutations can exist within a single AML sample. There was also no consistent relationship between subclone engraftment potential and the propensity of a subclone to emerge at relapse (see AMLs 62 and 71; Figure S6 C). This implies that functional and phenotypic properties of LSCs (such as engraftment, self-renewal, and chemoresistance) are experimentally variable and not directly related to the evolutionary history of the tumor, suggesting that the selective pressures imparted during xenotransplantation may not be equivalent to those imparted in vivo. In contrast to our findings, Clappier et al. (2011) established a relationship between xenotransplantation potential and T lymphoblastic leukemia/lymphoma relapse. This discrepancy may reflect inherent biological differences between these leukemias, including their disparate patterns of mutations, as well as experimental differences, such as animal conditioning, engraftment method, and cell dose. Additional studies will be needed to better define the relationship between engraftment potential and relapse for hematologic malignancies.

We anticipate that different xenotransplantation and cell manipulation approaches (i.e., extent of preconditioning, cell dose, level of immunodeficiency of the recipient mouse, transplantation route/procedure, number of passages, and timing of analysis posttransplant) could also alter the subclonal architecture of the engrafting tumor (Kelly et al, 2007, McDermott et al, 2010, Taussig et al, 2008, and Wunderlich et al, 2013). In particular, the number of injected cells may alter the competitive balance of subclones ( Notta et al., 2011 ), because the frequency of initiating cells may be variable among different subclones. It is also possible that more permissive xenotransplantation conditions (such as intrafemoral injections into irradiated recipients) may provide less selective pressure and allow for engraftment of multiple subclones. Ultimately, these different experimental approaches will need to be formally tested in light of the data presented in this study. Regardless of these uncertainties, our results do not invalidate the use of xenotransplantation models to study cancer. However, they do highlight the need for genomic characterization of tumors both before and after xenotransplantation. In fact, this study suggests that xenotransplantation may also be exploited as a means to isolate subclones for further study.

We were also able to integrate both technical and functional assays to demonstrate that subclones are unique genetic entities derived from a common ancestral cell ( Figure 7 ). Although previous studies of AML and MDS have inferred the clonal architecture in unfractionated bone marrow samples through the identification of clusters of mutations with similar VAFs (Ding et al, 2012, Walter et al, 2012, and Welch et al, 2012) and others have demonstrated the hierarchical nature of cancer at the single-cell level ( Potter et al., 2013 ), we employed multiple orthogonal approaches using the same AML samples to verify the identity and stability of imputed subclones. We observed that subclonal variants were present at the expected fractions among a population of single cells and mutations with similar VAFs in the unfractionated tumor were present in the same individual cells. Similarly, individual subclones that engrafted in immunodeficient mice contained the expected group of variants for that subclone, in addition to all founding clone mutations. Lastly, purified cell populations with well-established morphologic and immunophenotypic features corresponded to distinct subclones in some cases. Some of these resembled normal differentiating hematopoietic cells, suggesting that some AML subclones are capable of myeloid differentiation despite harboring known AML driver mutations.


Figure 7 Model of AML31 Subclonal Architecture and Predicted Phenotypes Schematic representation of the implications of AML clonal heterogeneity, based on the integrated analysis of AML31. The “% of de novo sample” values were calculated from sequencing the unfractionated AML samples and are consistent with data obtained from the interrogation of individual cells.

In this study, we were not able to define specific genetic determinants that explain the functional heterogeneity among tumor subclones. The patterns of subclone engraftment we observed in immunodeficient mice were inconsistent—some leukemias (e.g., AML31 and AML94) show variable retention (or loss) of different subclones, whereas others (e.g., AML88 and AML63) consistently engrafted a single subclone. We did not find that specific mutations consistently conferred preferential engraftment, nor did we identify new mutations in known leukemia-associated genes that were clearly responsible for enhanced engraftment, survival, or proliferation. For example, subclones with canonical mutations in the FLT3 tyrosine kinase did not preferentially engraft, despite the fact that this mutation causes potent activation of this signaling kinase and has been reported to result in higher engraftment in NSG mice ( Sanchez et al., 2009 ). The variable presence of key mutations in engrafting subclones would influence results from trials designed to test targeted chemotherapeutic agents in xenografted mice, such as FLT3 inhibitors (Smith et al, 2012 and Williams et al, 2013), or drugs targeting other mutations that often occur in subclones, such as those in RAS genes, or IDH1/IDH2 (Losman et al, 2013 and Rohle et al, 2013); clearly, the identity of the engrafting subclone (and its mutations) will be required to understand the response to a targeted drug (see Figure 7 ). However, the association of genetically defined subclones with enhanced engraftment potential suggests that some stable feature within these subclones may be responsible for their altered function. Although these genetic (or epigenetic) factors remain to be discovered, it is clear that subclones are discrete entities with important functional differences that may be genetically determined.

Although this study focused on AML, these observations likely extend to the experimental study of cancer in general, as well as its diagnosis and treatment. The presence of cancer-associated somatic mutations in cellular populations that are morphologically benign has implications for diagnostic testing of cancer samples and the use of phenotypically normal tissue for research studies. Our finding that mouse xenografts can have a skewed clonal architecture when compared to the parental tumor means that functional data obtained from these models, such as the capacity for self-renewal and chemoresistance, should not be generalized to the entire tumor or to other subclones that may contain different mutations without a rigorous genomic analysis of the xenograft. Integration of xenotransplantation and genomic data also demonstrates that the LSCs (defined functionally in immunodeficient mice) and the founding clone of a patient’s tumor (defined genetically) are not the same; in fact, there does not appear to be a consistent relationship between the cells that engraft in mice and the tumor’s evolutionary hierarchy. There is already some evidence that these conclusions generalize to other cancer models, because preferential engraftment of rare subclones has been demonstrated in xenotransplantation studies of ALL and breast cancer samples (Anderson et al, 2011, Ding et al, 2010, and Notta et al, 2011), although this observation has not been true in all studies using solid tumor models (Kreso et al, 2013 and Li et al, 2013). In this study, we observed engraftment of a rare, previously undetected subclone in one case (AML62), which was only identified through unbiased sequencing of the xenograft. It is possible that this approach, along with expression and epigenetic profiling of xenografts, may provide a mechanism for understanding the mutations and phenotypes that are under selective pressure in these model systems. The fact that primary AML samples (both from marrow and blood) contain the entire subclonal repertoire of the tumor and are therefore not subject to the same sampling biases of solid tumors suggests that this disease provides a very powerful platform for understanding the functional and genetic heterogeneity in cancer. Ultimately, functional and genetic data will have to be integrated for experimental systems to accurately model cancer and to develop therapeutic strategies to effectively treat it.

Experimental Procedures

Primary AML Samples

All cryopreserved AML samples were collected as part of a study approved by the Human Research Protection Office at Washington University School of Medicine after patients provided informed consent in accordance with the Declaration of Helsinki. Informed consent explicit for whole-genome sequencing was obtained for all patients in this study on a protocol approved by the Washington University Medical School Institutional Review Board.

Flow Cytometry

Cryopreserved AML cells were thawed as previously described ( Klco et al., 2013 ). The following human antibodies were used: anti-CD45 PerCP-Cy5.5 (eBioscience; clone 2D1), anti-CD33 phycoerythrin (PE) or antigen-presenting cell (APC) (eBioscience; clone WM-53), anti-CD19 APC (BD Biosciences; clone HIB19), anti-CD34 (PE-pool; Beckman Coulter Genomics; PN IM1459U), anti-CD15 fluorescein isothiocyanate (BD Biosciences; clone HI98), anti-CD11b BV421 (BD Biosciences; ICRF44), and CD3 V450 (eBioscience; clone OKT3). Live cell populations were discriminated initially via CD45/SSC scatterplots, and different subpopulations were defined as follows: blasts, CD45dim/SSClow, CD33+; lymphocytes, CD45bright/SSClow, CD33−; monocytes, CD45int/bright, SSCinter, CD33+; and maturing myelomonocytic cells, CD45int, SShigh, CD33dim/neg. When available in sufficient numbers, lymphocytes were further sorted into CD19+ and/or CD3+ populations. For NSG experiments, human hematopoietic cells (human CD45 positive) from the bone marrow were separated from murine cells (murine CD45) via flow cytometry; human cells were further isolated by expression of either CD33 and/or CD34. Cells were sorted on a modified Beckman Coulter MoFlo into PBS; genomic DNA (gDNA) was prepared using a QIAamp DNA mini kit (QIAGEN). For capture experiments, a minimum of 100 ng of gDNA was required for sequencing. Cytospins were performed using a Shandon Cytospin 3, and cells were stained using Wright-Giemsa (Sigma). Images were obtained with an Olympus BX51 microscope equipped with an Olympus DP26 camera. Morphologic studies were performed by a board-certified hematopathologist (J.M.K.).

Xenotransplantation Studies

Animals were used in accordance to a protocol reviewed and approved by the Institutional Animal Care and Use Committee of Washington University. Mice were produced at Washington University School of Medicine using breeders obtained from the Jackson Laboratory (NSG stock 005557; NSG-SGM3 stock 013062). Six- to ten-week-old unconditioned mice were injected with one million viable cells via lateral tail vein route. Mice were treated with antibiotics for 2 weeks after injection and then followed for 12–16 weeks. For these studies, engraftment was defined >1% human CD45 and CD33 (or CD34) positivity at the time of sacrifice. This threshold was determined rather than the standard of 0.1% ( Sarry et al., 2011 ) to ensure that sufficient material would be present for downstream analyses. For all NSG/NSG-SGM3 comparisons, cells from a single cryovial were injected at the same time into age-matched animals.

Variant Discovery and VAF Measurements

Whole-genome sequencing and capture validation was performed as described in the Supplemental Experimental Procedures . All variant count data were obtained from raw BAM files using bam-readcount ( https://github.com/genome/bam-readcount ) following filtering of reads for low mapping quality (<10), low base quality (<10), and reads with more than four mismatches. Indel variant counts were obtained separately using a custom script to align overlapping reads from the raw BAM file to short sequences with either the reference or alternate indel allele using cross_match ( http://www.phrap.org/phredphrap/phrap.html ) and tabulate variant counts based on the highest scoring alignment to these two sequences. All variant counts and fractions were manipulated and visualized in R.


This work was supported by grants to J.M.K. (K08HL116605), T.J.L. from the National Cancer Institute (P01CA101937 and RO1CA162086) and the Barnes Jewish Hospital Foundation (00335-0505-02), and R.K.W. from the National Human Genome Research Institute (U54 HG003079). Technical assistance was provided by the Alvin J. Siteman Cancer Center Tissue Procurement Core, the High Speed Cell Sorting Core, and the Molecular and Genomic Analysis Core at Washington University School of Medicine and Barnes-Jewish Hospital in St. Louis, MO, which are all supported by the National Cancer Institute Cancer Center Support Grant P30CA91842. The authors also thank Mieke Hooke for invaluable animal husbandry, Sharon Heath for clinical annotation, Julie Ritchey and Dan George for technical assistance, and David Russler-Germain for critical reading of the manuscript.

Accession Numbers

The database of Genotypes and Phenotypes accession number for the AML tumor sequence variants is phs000159.

See Supplemental Experimental Procedures for additional information.

Supplemental Information


Download file

Document S1. Supplemental Experimental Procedures, Figures S1–S6, and Table S7

Download file

Table S1. Clinical and Biological Characteristics of 19 De Novo AML Cases, WGS Coverage Parameters, and Number of Variants Identified, Related to Figure 1

Download file

Table S2. Tier1 Variants Identified by WGS, Related to Figure 1

Download file

Table S3. Tier 2 and 3 Variants Identified by WGS, Related to Figure 1

Download file

Table S4. Coverage Parameters from Capture-Based Sequencing of All Biological and Experimental Conditions, Related to Figure 2

Download file

Table S5. Relationship of Subclones in the Unfractionated Peripheral Blood to the Flow-Enriched Myeloid Populations, Related to Figure 2 Variants were assigned to subclones using sciClone (C.A.M., Brian S. White, Nathan D. Dees, M.G., Obi L. Griffith, John S. Welch, Ravi Vij, Michael H. Tomasson, T.A.G., M.J.W., Matthew J. Ellis, William Schierding, J.F.D., T.J.L., E.R.M., R.K.W., and Li Ding, unpublished data), and all values are normalized to the mean VAF of the founding clone variants. Significant differences in the distribution of each individual subclone are highlighted in red text, as determined by a cutoff of >5% mean VAF difference from the unfractionated peripheral blood and a Bonferroni-corrected p value of <0.05 (Student’s t test).

Download file

Table S6. Somatic Variants Identified by WGS of De Novo AML Present at VAF >10% in the Flow-Purified Lymphocytes, Related to Figure 2


  • Anderson et al., 2011 K. Anderson, C. Lutz, F.W. van Delft, C.M. Bateman, Y. Guo, S.M. Colman, H. Kempski, A.V. Moorman, I. Titley, J. Swansbury, et al. Genetic variegation of clonal architecture and propagating cells in leukaemia. Nature. 2011;469:356-361 Crossref.
  • Bonnet and Dick, 1997 D. Bonnet, J.E. Dick. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nat. Med.. 1997;3:730-737 Crossref.
  • Cancer Genome Atlas Research Network, 2013 Cancer Genome Atlas Research Network. Genomic and epigenomic landscapes of adult de novo acute myeloid leukemia. N. Engl. J. Med.. 2013;368:2059-2074
  • Clappier et al., 2011 E. Clappier, B. Gerby, F. Sigaux, M. Delord, F. Touzri, L. Hernandez, P. Ballerini, A. Baruchel, F. Pflumio, J. Soulier. Clonal selection in xenografted human T cell acute lymphoblastic leukemia recapitulates gain of malignancy at relapse. J. Exp. Med.. 2011;208:653-661 Crossref.
  • Ding et al., 2010 L. Ding, M.J. Ellis, S. Li, D.E. Larson, K. Chen, J.W. Wallis, C.C. Harris, M.D. McLellan, R.S. Fulton, L.L. Fulton, et al. Genome remodelling in a basal-like breast cancer metastasis and xenograft. Nature. 2010;464:999-1005 Crossref.
  • Ding et al., 2012 L. Ding, T.J. Ley, D.E. Larson, C.A. Miller, D.C. Koboldt, J.S. Welch, J.K. Ritchey, M.A. Young, T. Lamprecht, M.D. McLellan, et al. Clonal evolution in relapsed acute myeloid leukaemia revealed by whole-genome sequencing. Nature. 2012;481:506-510 Crossref.
  • Gerlinger et al., 2012 M. Gerlinger, A.J. Rowan, S. Horswell, J. Larkin, D. Endesfelder, E. Gronroos, P. Martinez, N. Matthews, A. Stewart, P. Tarpey, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N. Engl. J. Med.. 2012;366:883-892 Crossref.
  • Gupta et al., 2011 P.B. Gupta, C.M. Fillmore, G. Jiang, S.D. Shapira, K. Tao, C. Kuperwasser, E.S. Lander. Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells. Cell. 2011;146:633-644 Crossref.
  • Ito et al., 2002 M. Ito, H. Hiramatsu, K. Kobayashi, K. Suzue, M. Kawahata, K. Hioki, Y. Ueyama, Y. Koyanagi, K. Sugamura, K. Tsuji, et al. NOD/SCID/gamma(c)(null) mouse: an excellent recipient mouse model for engraftment of human cells. Blood. 2002;100:3175-3182 Crossref.
  • Jaiswal et al., 2009 S. Jaiswal, C.H. Jamieson, W.W. Pang, C.Y. Park, M.P. Chao, R. Majeti, D. Traver, N. van Rooijen, I.L. Weissman. CD47 is upregulated on circulating hematopoietic stem cells and leukemia cells to avoid phagocytosis. Cell. 2009;138:271-285 Crossref.
  • Jan et al., 2012 M. Jan, T.M. Snyder, M.R. Corces-Zimmerman, P. Vyas, I.L. Weissman, S.R. Quake, R. Majeti. Clonal evolution of preleukemic hematopoietic stem cells precedes human acute myeloid leukemia. Sci. Transl. Med.. 2012;4:149ra118 Crossref.
  • Jilani et al., 2003 I. Jilani, E. Estey, T. Manshuri, M. Caligiuri, M. Keating, F. Giles, D. Thomas, H. Kantarjian, M. Albitar. Better detection of FLT3 internal tandem duplication using peripheral blood plasma DNA. Leukemia. 2003;17:114-119 Crossref.
  • Jin et al., 2009 L. Jin, E.M. Lee, H.S. Ramshaw, S.J. Busfield, A.G. Peoppl, L. Wilkinson, M.A. Guthridge, D. Thomas, E.F. Barry, A. Boyd, et al. Monoclonal antibody-mediated targeting of CD123, IL-3 receptor alpha chain, eliminates human acute myeloid leukemic stem cells. Cell Stem Cell. 2009;5:31-42 Crossref.
  • Kelly et al., 2007 P.N. Kelly, A. Dakic, J.M. Adams, S.L. Nutt, A. Strasser. Tumor growth need not be driven by rare cancer stem cells. Science. 2007;317:337 Crossref.
  • Klco et al., 2013 J.M. Klco, D.H. Spencer, T.L. Lamprecht, S.M. Sarkaria, T. Wylie, V. Magrini, J. Hundal, J. Walker, N. Varghese, P. Erdmann-Gilmore, et al. Genomic impact of transient low-dose decitabine treatment on primary AML cells. Blood. 2013;121:1633-1643 Crossref.
  • Kreso et al., 2013 A. Kreso, C.A. O’Brien, P. van Galen, O.I. Gan, F. Notta, A.M. Brown, K. Ng, J. Ma, E. Wienholds, C. Dunant, et al. Variable clonal repopulation dynamics influence chemotherapy response in colorectal cancer. Science. 2013;339:543-548 Crossref.
  • Krönke et al., 2011 J. Krönke, R.F. Schlenk, K.O. Jensen, F. Tschürtz, A. Corbacioglu, V.I. Gaidzik, P. Paschka, S. Onken, K. Eiwen, M. Habdank, et al. Monitoring of minimal residual disease in NPM1-mutated acute myeloid leukemia: a study from the German-Austrian acute myeloid leukemia study group. J. Clin. Oncol.. 2011;29:2709-2716
  • Krönke et al., 2013 J. Krönke, L. Bullinger, V. Teleanu, F. Tschürtz, V.I. Gaidzik, M.W. Kühn, F.G. Rücker, K. Holzmann, P. Paschka, S. Kapp-Schwörer, et al. Clonal evolution in relapsed NPM1-mutated acute myeloid leukemia. Blood. 2013;122:100-108
  • Lapidot et al., 1994 T. Lapidot, C. Sirard, J. Vormoor, B. Murdoch, T. Hoang, J. Caceres-Cortes, M. Minden, B. Paterson, M.A. Caligiuri, J.E. Dick. A cell initiating human acute myeloid leukaemia after transplantation into SCID mice. Nature. 1994;367:645-648 Crossref.
  • Ley et al., 2008 T.J. Ley, E.R. Mardis, L. Ding, B. Fulton, M.D. McLellan, K. Chen, D. Dooling, B.H. Dunford-Shore, S. McGrath, M. Hickenbotham, et al. DNA sequencing of a cytogenetically normal acute myeloid leukaemia genome. Nature. 2008;456:66-72 Crossref.
  • Ley et al., 2010 T.J. Ley, L. Ding, M.J. Walter, M.D. McLellan, T. Lamprecht, D.E. Larson, C. Kandoth, J.E. Payton, J. Baty, J. Welch, et al. DNMT3A mutations in acute myeloid leukemia. N. Engl. J. Med.. 2010;363:2424-2433 Crossref.
  • Li et al., 2013 S. Li, D. Shen, J. Shao, R. Crowder, W. Liu, A. Prat, X. He, S. Liu, J. Hoog, C. Lu, et al. Endocrine-therapy-resistant ESR1 variants revealed by genomic characterization of breast-cancer-derived xenografts. Cell Rep.. 2013;4:1116-1130 Crossref.
  • Losman et al., 2013 J.A. Losman, R.E. Looper, P. Koivunen, S. Lee, R.K. Schneider, C. McMahon, G.S. Cowley, D.E. Root, B.L. Ebert, W.G. Kaelin Jr. (R)-2-hydroxyglutarate is sufficient to promote leukemogenesis and its effects are reversible. Science. 2013;339:1621-1625 Crossref.
  • Ma et al., 2009 W. Ma, H. Kantarjian, X. Zhang, I. Jilani, M.R. Sheikholeslami, A.C. Donahue, F. Ravandi, E. Estey, S. O’Brien, M. Keating, et al. Detection of nucleophosmin gene mutations in plasma from patients with acute myeloid leukemia: clinical significance and implications. Cancer Biomark.. 2009;5:51-58
  • Mardis et al., 2009 E.R. Mardis, L. Ding, D.J. Dooling, D.E. Larson, M.D. McLellan, K. Chen, D.C. Koboldt, R.S. Fulton, K.D. Delehaunty, S.D. McGrath, et al. Recurring mutations found by sequencing an acute myeloid leukemia genome. N. Engl. J. Med.. 2009;361:1058-1066 Crossref.
  • McDermott et al., 2010 S.P. McDermott, K. Eppert, E.R. Lechman, M. Doedens, J.E. Dick. Comparison of human cord blood engraftment between immunocompromised mouse strains. Blood. 2010;116:193-200 Crossref.
  • Miyamoto et al., 2000 T. Miyamoto, I.L. Weissman, K. Akashi. AML1/ETO-expressing nonleukemic stem cells in acute myelogenous leukemia with 8;21 chromosomal translocation. Proc. Natl. Acad. Sci. USA. 2000;97:7521-7526 Crossref.
  • Navin et al., 2010 N. Navin, A. Krasnitz, L. Rodgers, K. Cook, J. Meth, J. Kendall, M. Riggs, Y. Eberling, J. Troge, V. Grubor, et al. Inferring tumor progression from genomic heterogeneity. Genome Res.. 2010;20:68-80 Crossref.
  • Nik-Zainal et al., 2012 S. Nik-Zainal, P. Van Loo, D.C. Wedge, L.B. Alexandrov, C.D. Greenman, K.W. Lau, K. Raine, D. Jones, J. Marshall, M. Ramakrishna, et al., Breast Cancer Working Group of the International Cancer Genome Consortium. The life history of 21 breast cancers. Cell. 2012;149:994-1007 Crossref.
  • Notta et al., 2011 F. Notta, C.G. Mullighan, J.C. Wang, A. Poeppl, S. Doulatov, L.A. Phillips, J. Ma, M.D. Minden, J.R. Downing, J.E. Dick. Evolution of human BCR-ABL1 lymphoblastic leukaemia-initiating cells. Nature. 2011;469:362-367 Crossref.
  • Nowell, 1976 P.C. Nowell. The clonal evolution of tumor cell populations. Science. 1976;194:23-28
  • Potter et al., 2013 N.E. Potter, L. Ermini, E. Papaemmanuil, G. Cazzaniga, G. Vijayaraghavan, I. Titley, A. Ford, P. Campbell, L. Kearney, M. Greaves. Single-cell mutational profiling and clonal phylogeny in cancer. Genome Res.. 2013;23:2115-2125
  • Rohle et al., 2013 D. Rohle, J. Popovici-Muller, N. Palaskas, S. Turcan, C. Grommes, C. Campos, J. Tsoi, O. Clark, B. Oldrini, E. Komisopoulou, et al. An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells. Science. 2013;340:626-630 Crossref.
  • Sanchez et al., 2009 P.V. Sanchez, R.L. Perry, J.E. Sarry, A.E. Perl, K. Murphy, C.R. Swider, A. Bagg, J.K. Choi, J.A. Biegel, G. Danet-Desnoyers, M. Carroll. A robust xenotransplantation model for acute myeloid leukemia. Leukemia. 2009;23:2109-2117 Crossref.
  • Sarry et al., 2011 J.E. Sarry, K. Murphy, R. Perry, P.V. Sanchez, A. Secreto, C. Keefer, C.R. Swider, A.C. Strzelecki, C. Cavelier, C. Récher, et al. Human acute myelogenous leukemia stem cells are rare and heterogeneous when assayed in NOD/SCID/IL2Rγc-deficient mice. J. Clin. Invest.. 2011;121:384-395 Crossref.
  • Schmitz et al., 2011 M. Schmitz, P. Breithaupt, N. Scheidegger, G. Cario, L. Bonapace, B. Meissner, P. Mirkowska, J. Tchinda, F.K. Niggli, M. Stanulla, et al. Xenografts of highly resistant leukemia recapitulate the clonal composition of the leukemogenic compartment. Blood. 2011;118:1854-1864 Crossref.
  • Shah et al., 2009 S.P. Shah, R.D. Morin, J. Khattra, L. Prentice, T. Pugh, A. Burleigh, A. Delaney, K. Gelmon, R. Guliany, J. Senz, et al. Mutational evolution in a lobular breast tumour profiled at single nucleotide resolution. Nature. 2009;461:809-813 Crossref.
  • Shlush et al., 2012 L.I. Shlush, N. Chapal-Ilani, R. Adar, N. Pery, Y. Maruvka, A. Spiro, R. Shouval, J.M. Rowe, M. Tzukerman, D. Bercovich, et al. Cell lineage analysis of acute leukemia relapse uncovers the role of replication-rate heterogeneity and microsatellite instability. Blood. 2012;120:603-612 Crossref.
  • Smith et al., 2010 A.E. Smith, A.M. Mohamedali, A. Kulasekararaj, Z. Lim, J. Gäken, N.C. Lea, B. Przychodzen, S.A. Mian, E.E. Nasser, C. Shooter, et al. Next-generation sequencing of the TET2 gene in 355 MDS and CMML patients reveals low-abundance mutant clones with early origins, but indicates no definite prognostic value. Blood. 2010;116:3923-3932 Crossref.
  • Smith et al., 2012 C.C. Smith, Q. Wang, C.S. Chin, S. Salerno, L.E. Damon, M.J. Levis, A.E. Perl, K.J. Travers, S. Wang, J.P. Hunt, et al. Validation of ITD mutations in FLT3 as a therapeutic target in human acute myeloid leukaemia. Nature. 2012;485:260-263 Crossref.
  • Sottoriva et al., 2013 A. Sottoriva, I. Spiteri, S.G. Piccirillo, A. Touloumis, V.P. Collins, J.C. Marioni, C. Curtis, C. Watts, S. Tavaré. Intratumor heterogeneity in human glioblastoma reflects cancer evolutionary dynamics. Proc. Natl. Acad. Sci. USA. 2013;110:4009-4014 Crossref.
  • Swanton, 2012 C. Swanton. Intratumor heterogeneity: evolution through space and time. Cancer Res.. 2012;72:4875-4882 Crossref.
  • Taussig et al., 2008 D.C. Taussig, F. Miraki-Moud, F. Anjos-Afonso, D.J. Pearce, K. Allen, C. Ridler, D. Lillington, H. Oakervee, J. Cavenagh, S.G. Agrawal, et al. Anti-CD38 antibody-mediated clearance of human repopulating cells masks the heterogeneity of leukemia-initiating cells. Blood. 2008;112:568-575 Crossref.
  • Walter et al., 2012 M.J. Walter, D. Shen, L. Ding, J. Shao, D.C. Koboldt, K. Chen, D.E. Larson, M.D. McLellan, D. Dooling, R. Abbott, et al. Clonal architecture of secondary acute myeloid leukemia. N. Engl. J. Med.. 2012;366:1090-1098 Crossref.
  • Welch et al., 2012 J.S. Welch, T.J. Ley, D.C. Link, C.A. Miller, D.E. Larson, D.C. Koboldt, L.D. Wartman, T.L. Lamprecht, F. Liu, J. Xia, et al. The origin and evolution of mutations in acute myeloid leukemia. Cell. 2012;150:264-278 Crossref.
  • Williams et al., 2013 A.B. Williams, B. Nguyen, L. Li, P. Brown, M. Levis, D. Leahy, D. Small. Mutations of FLT3/ITD confer resistance to multiple tyrosine kinase inhibitors. Leukemia. 2013;27:48-55 Crossref.
  • Wunderlich et al., 2010 M. Wunderlich, F.S. Chou, K.A. Link, B. Mizukawa, R.L. Perry, M. Carroll, J.C. Mulloy. AML xenograft efficiency is significantly improved in NOD/SCID-IL2RG mice constitutively expressing human SCF, GM-CSF and IL-3. Leukemia. 2010;24:1785-1788 Crossref.
  • Wunderlich et al., 2013 M. Wunderlich, B. Mizukawa, F.S. Chou, C. Sexton, M. Shrestha, Y. Saunthararajah, J.C. Mulloy. AML cells are differentially sensitive to chemotherapy treatment in a human xenograft model. Blood. 2013;121:e90-e97 Crossref.


1 Department of Pathology & Immunology, Washington University School of Medicine, St. Louis, MO 63110, USA

2 The Genome Institute, Washington University, St. Louis, MO 63110, USA

3 Division of Oncology, Section of Stem Cell Biology, Department of Internal Medicine, Washington University School of Medicine, St. Louis, MO 63110, USA

Corresponding author

4 These authors contributed equally to this work