Restricted cell cycle is essential for clonal evolution and therapeutic resistance of pre-leukemic stem cells

Pre-leukemic stem cells (pre-LSCs) give rise to leukemic stem cells through acquisition of additional gene mutations and are an important source of relapse following chemotherapy. We postulated that cell-cycle kinetics of pre-LSCs may be an important determinant of clonal evolution and therapeutic resistance. Using a doxycycline-inducible H2B-GFP transgene in a mouse model of T-cell acute lymphoblastic leukemia to study cell cycle in vivo, we show that self-renewal, clonal evolution and therapeutic resistance are limited to a rare population of pre-LSCs with restricted cell cycle. We show that proliferative pre-LSCs are unable to return to a cell cycle-restricted state. Cell cycle-restricted pre-LSCs have activation of p53 and its downstream cell-cycle inhibitor p21. Furthermore, absence of p21 leads to proliferation of pre-LSCs, with clonal extinction through loss of asymmetric cell division and terminal differentiation. Thus, inducing proliferation of pre-LSCs represents a promising strategy to increase cure rates for acute leukemia.

T he leukemia stem cell (LSC) concept posits the presence of a cell population with stem cell-like properties enabling their ability to generate the full heterogeneity of the tumor and fuel tumor growth during disease progression. These LSCs are intrinsically resistant to therapies via potential mechanisms that include quiescence, low reactive oxygen stress, enhanced DNA repair and expression of adenosine triphosphate-binding cassette transporters. Over recent years, genome-wide studies of matched primary and relapsed leukemic samples strongly support this model wherein the clone responsible for relapse arises from either a pre-existing LSC or an antecedent LSC clone referred to as a pre-leukemic stem cell (pre-LSC) [1][2][3] . These pre-LSCs contain the founding genetic mutation but not the full complement of mutations found at diagnosis. Although pre-LSCs retain the ability to differentiate into functional mature blood cells, they also have long-lived self-renewal capacity 4 and their presence in patient remission samples following intensive chemotherapy portends a high risk of relapse 5 . In addition to acute leukemia, cells akin to pre-LSCs underpin myelodysplastic syndromes and perhaps even clonal hematopoiesis of the elderly, which can evolve into acute leukemia over many months to years 6,7 .
Quiescence may be an important mechanism of therapeutic resistance for LSCs, particularly for therapies that rely upon cell proliferation for their activity. Clinically, this concept is exemplified in chronic myeloid leukemia where, even in the era of tyrosine kinase inhibitor therapy, the absence of cure is thought to reside with the inability to eradicate the quiescent clones of LSCs [8][9][10] . Perhaps the most convincing in vivo evidence comes from Ebinger et al. 11 , who identified a rare subpopulation of dormant and treatment-resistant cells in patient-derived xenografts. They also showed that these chemoresistant cells share the same gene expression profile with primary leukemia cells isolated from patients at minimal residual disease. Moreover, Saito et al. 12 experimentally showed that quiescent leukemic cells residing in the bone marrow niche were protected from chemotherapy. They subsequently showed that overcoming quiescence with cytokine stimulation could sensitize these leukemogenic cells to chemotherapy. However, these and other experimental in vivo studies of LSC quiescence have almost exclusively used labelretaining cell fixation assays with DNA analogs such as bromodeoxyuridine which preclude subsequent functional studies 13 . This major hurdle for the study of quiescence in hematopoietic stem and progenitor cells has recently been overcome by the generation of transgenic mice expressing a doxycycline-regulated histone H2B-GFP fusion product that is incorporated into the nucleosome during cell division 14,15 . Prospective isolation of quiescent hematopoietic stem cells (HSCs) based on cell surface markers and green fluorescent protein (GFP) retention showed that quiescent HSCs are both enriched for long-term repopulating activity and the source of proliferative HSCs during times of stress. To our knowledge, these H2B-GFP mice have been reported only once in the leukemia context. In this study, oncogenic RAS induced a bimodal effect on HSC cycling, with the quiescent but not proliferative fraction outcompeting healthy HSCs 16 . However, the relationship between quiescence and chemoresistance or clonal evolution remained to be explored.
Aberrant expression of LMO2 through chromosomal translocation or a somatically acquired neomorphic promoter occurs in 50% of T-cell acute lymphoblastic leukemia (T-ALL) 17,18 . Using a mouse model of T-ALL driven by the Lmo2 oncogene, we reported the identification of cells that fulfill the fundamental properties of pre-LSCs, namely self-renewal potential without a block in differentiation 19 . Transplant studies showed that pre-LSCs arise from immature CD4 -CD8 -CD25 + CD44 − (DN3) Tcell progenitors in Lmo2-transgenic mice (Lmo2 Tg ), as these cells were capable of long-term repopulation capacity in recipient mice. These self-renewing DN3 cells retained T-cell differentiation potential but eventually gave rise to T-ALL as they accumulated additional lesions that promote leukemia progression 20,21 . Importantly, these pre-LSCs could survive and recover after high-dose radiation 19 . Here, we have used the doxycycline-inducible H2B-GFP mouse model crossed with the Lmo2-transgenic mice to study the importance of cell cycle in pre-LSCs. We show that self-renewal, clonal evolution and therapeutic resistance are limited to a rare population of pre-LSCs with restricted cell cycle. Importantly, proliferative pre-LSCs are unable to return to a cell cycle-restricted state. Thus, inducing proliferation of pre-LSCs represents a promising strategy to increase cure rates for acute leukemia.

Results
Identification of cell cycle-restricted pre-LSCs. We crossed the TetOP-H2B-GFP KI/+ mouse line with Lmo2 Tg mice to examine the cell-cycle kinetics of pre-LSCs. Heterozygous TetOP-H2B-GFP KI/+ ;Lmo2 Tg (H2B-GFP;Lmo2 Tg ) mice were treated with doxycycline for 6 weeks to induce expression of H2B-GFP in dividing cells. We then examined GFP expression in thymocytes following withdrawal of doxycycline for 1, 2, 4 and 8 weeks (Fig. 1a), focusing on the DN3 T-cell fraction, which contains all pre-LSC activity 19,22 . At the end of the labeling period, almost all DN3 cells in both control and Lmo2 Tg mice expressed the H2B-GFP division marker, which comprised high and intermediate populations (Fig. 1b). Interestingly, a small proportion of DN3 cells in H2B-GFP;Lmo2 Tg mice remained GFP negative despite a 6-week labeling period, which suggests the presence of cells that had not divided. Consistent with this highly proliferative stage of T-cell development, withdrawal of doxycycline led to a rapid loss of GFP in DN3 thymocytes from control H2B-GFP mice such that there were no GFP hi cells beyond 2 weeks. However, in H2B-GFP;Lmo2 Tg mice, a small fraction of DN3 cells retained GFP expression for up to 8 weeks (Fig. 1b). In absolute numbers, this rare GFP hi population of cells at 8 weeks equated to 3000 cells per whole thymus ( Supplementary Fig. 1a). Consistent with the exclusive presence of pre-LSCs within the DN3 thymocyte population, no difference was observed in the proportion of GFPretaining cells within other T-cell subsets in H2B-GFP;Lmo2 Tg mice compared with control mice (Supplementary Fig. 1b). Assuming that the mean cell fluorescence halved with each cell division, we estimated a mean cycling time of 18.2 ± 2.8 h for control DN3 cells and 50.1 ± 13.8 h for Lmo2 Tg DN3. Given there was no obvious long-term plateau of GFP loss (Fig. 1c), we designated GFP hi cells as cell-cycle restricted rather than quiescent or dormant. Staining with 4′,6-diamidino-2-phenylindole (DAPI) and Ki67, an independent assay for analyzing snapshots of cell cycle, confirmed that most GFP hi cells present beyond 2week chase were in the G 0 phase (Fig. 1d), whereas less than 40% of GFP lo cells were in the G 0 phase when assayed 2 weeks after withdrawal of doxycycline (Fig. 1e). Therefore, all subsequent analyses were performed using DN3 thymocytes from mice 2 weeks after withdrawal of doxycycline and cells that did not divide during the 6-week labeling period were excluded.
Cell-cycle restriction maintains leukemogenic pre-LSCs. Serial transplantation is the gold-standard assay that defines LSCs 23 . Fluorescence-activated cell sorting (FACS)-isolated GFP hi and GFP lo DN3 cells from 2-month-old H2B-GFP;Lmo2 Tg mice were injected into sublethally irradiated CD45.1 + recipient mice to examine the importance of cell-cycle kinetics on repopulating activity (Fig. 2a). In primary recipients, GFP hi DN3 cells were able to expand 100-fold compared with 10-20-fold for GFP lo DN3 cells (Fig. 2b). This decreased capacity to generate DN3 cells may be explained in part by the enhanced differentiation of GFP lo DN3 cells into CD4 + CD8 + double-positive (DP) thymocytes (Fig. 2c), which lack self-renewal activity 19,22 . We performed serial transplant to assess long-term self-renewal capacity (Fig. 2a), the quintessential property of all stem cells. A period of 4 weeks between transplants was chosen to allow competition with normal HSCs, which take up to 3 weeks to repopulate the thymus. GFP hi DN3 cells retained an ability to expand for at least four rounds of transplantation (Fig. 2d). In contrast, GFP lo DN3 cells progressively lost the ability to regenerate DN3 cells such that by the fourth passage, there was exhaustion of their expansion potential. Thus, restricted cell cycle was a critical property of self-renewing pre-LSCs.
Given that self-renewal enables pre-LSCs to accumulate additional genetic events necessary for progression to leukemia, we postulated that only the GFP hi DN3 cells would be capable of generating T-ALL. Consistent with this idea, there was increased monoclonality in GFP hi DN3 cells as measured by Tcrβ rearrangement (Supplementary Fig. 2a) 24 . Furthermore, a proportion of secondary, tertiary and quaternary recipients of   Supplementary Fig. 2b). Given that leukemias only arise in recipients injected with GFP hi cells, our results demonstrate that restricted cell cycle is important for clonal evolution and leukemogenic potential of pre-LSCs.
HSCs can re-enter a dormant state following hematopoietic stress, including chemotherapy 25,26 . To determine if proliferative pre-LSCs were able to return to a cell cycle-restricted state, we administered doxycycline for 6 weeks to mice transplanted with GFP lo cells. Unlike normal HSCs, proliferative pre-LSCs were unable to generate GFP hi cells ( Fig. 2e and S2c). In contrast, GFP hi cells generated a cell cycle-restricted progeny, confirming that they can be maintained even in the setting of proliferative stress. To determine whether cell cycle-restricted pre-LSCs are maintained in disease progression, we determined the numbers of GFP hi DN3 cells in 6-month-old Lmo2 Tg mice. In these older mice, the GFP hi DN3 cell population had expanded 3-fold, and 10-fold in mice with overt T-ALL ( Supplementary Fig. 2d). Thus, cell cycle-restricted pre-LSCs expand with disease progression.
A recent study by Ebinger et al. 11 linked cell-cycle restriction with treatment resistance in human ALL. To determine whether cell-cycle restriction protects pre-LSCs against chemotherapeutic agents typically used for human T-ALL, we looked for enrichment of GFP hi DN3 cells following treatment of H2B-GFP;Lmo2 Tg mice with a combination of vincristine, dexamethasone and L-asparaginase (VXL) 27 . Consistent with the cell cycledependent effect of chemotherapy, the proportion of cells surviving 24 h after combination therapy was fourfold higher in the GFP hi fraction compared with the GFP lo fraction: 9% of GFP hi cells compared with 2% of GFP lo cells (Fig. 2f). A similar increased resistance of GFP hi cells was observed in response to sub-lethal irradiation ( Supplementary Fig. 2e). Given that the H2B-GFP labeling model reflects the history of the cell cycle, the enrichment for GFP hi cells observed cannot be due to therapyinduced senescence and must reflect cells that have not actively divided in the preceding 2 weeks of chase. Thus, cell-cycle restriction of pre-LSCs enhances resistance to therapeutic agents used for treatment of human T-ALL.
Expression profile of cell cycle-restricted pre-LSCs. To investigate the molecular signature of cell cycle-restricted pre-LSCs, we performed gene expression profiling of GFP hi and GFP lo DN3 cells from three independent cohorts of H2B-GFP;Lmo2 Tg mice. Overall, there were 853 genes differentially expressed more than twofold using a false discovery rate ( Cell cycle-restricted pre-LSCs acquire Notch1 mutations. One of the most striking changes identified in the GSEA was reduced expression in GFP hi DN3 cells of genes clustering on different regions of human chromosomes ( Supplementary Fig. 4a) that assembled on chromosomes 2 and 15 in mice (Fig. 4a) as compared to GFP lo thymocytes. Given that the analysis was performed using the average expression in all samples tested, this striking observation suggested that whole chromosomes were either lost in GFP hi or gained in GFP lo DN3 thymocytes. Targeted probe analysis for chromosomes 15 and 2 revealed a high frequency of trisomy 15 and 2 in GFP lo DN3 thymocytes (Fig. 4b). Genomic PCR for the IL7r and Myc loci on chromosome 15 (Supplementary Methods) confirmed increased copy number in GFP lo cells ( Supplementary Fig. 4b). Given the RNAsequencing (RNA-seq) analysis was performed on pooled samples, we also performed whole-exome sequencing (WES) on GFP hi and GFP lo DN3 thymocytes isolated from 5 individual doxycycline-pulsed 2-month-old H2B-GFP;Lmo2 Tg mice. In accordance with gene expression data, WES analysis revealed gains of chromosomes 2 and 15 as well as other numerical chromosomal alterations in the GFP lo population from all mice analyzed (Fig. 4c). Thus, cycling is associated with aneuploidy in pre-LSCs. It is postulated that the quiescent state of long-term repopulating HSCs increases the risk of acquired mutations due to the use of error-prone non-homologous end joining-mediated DNA repair 28 . In contrast, cycling HSCs or progenitors can utilize high-fidelity homologous recombination for DNA repair. To define the relationship between cell cycle and mutations in pre-LSCs, we used the RNA-seq data to identify variants differentially expressed in GFP hi and GFP lo DN3 cells. Overall, we detected 57 genes with variants predicted to be deleterious due to frameshift, splice or premature stop (Supplementary Data 2). While the majority (n = 43) were found in both GFP hi and GFP lo cell populations, frameshift mutations upstream of the PEST coding region in Notch1 were present only in the cell cyclerestricted GFP hi cells (Supplementary Data 2). Targeted sequencing of the Notch1 locus in GFP hi DN3 thymocytes isolated from 5 individual mice confirmed the presence of activating mutations of Notch1 in cell cycle-restricted pre-LSCs (Fig. 4d). In addition, a stop in Tcrg-V3 indicating Tcrg gene rearrangement was found exclusively in GFP hi cells. The expression of truncated transcripts of Tcrg-V3 represents an early marker of clonal selection in cell cycle-restricted pre-LSCs. Thus, distinct mutations occur in pre-LSCs according to their cell-cycle dynamics with recurrent activating mutations of Notch1 limited to cell cycle-restricted pre-LSCs. p21 is required for stem cell-like properties of pre-LSCs. We examined expression of the known inhibitors of cell cycle in HSCs to define the mechanism of cell-cycle restriction in pre-LSCs. Consistent with activation of p53, we found a threefold increase in the expression of Cdkn1a (p21) in the DN3 GFP hi subpopulation. In contrast, expressions of Cdkn1b (p27), Cdkn1c (p57) and the Ink4a family members were not altered in GFP hi cell ( Fig. 5a and Supplementary Fig. 5a).
To assess the importance of p21 in cell cycle-restricted pre-LSCs, we generated H2B-GFP;Lmo2 Tg mice on a p21-deficient (p21 −/− ) background. Significantly, loss of p21 led to almost complete absence of GFP hi cells by 2 weeks post doxycycline pulse ( Fig. 5b and Supplementary Fig. 5b). This loss of cell-cycle restriction was observed in total DN3, where the proportion of cells in G 0 was restored to wild-type levels ( Supplementary  Fig. 5c). Genomic quantification of the Il7r and Myc loci revealed that p21 deficiency increased copy number of chromosome 15 in p21-deficient Lmo2 Tg cells lacking p21 ( Supplementary Fig. 5d), suggesting that p21 was important for cell-cycle restriction and genomic stability in pre-LSCs. Given the absence of GFP hi cells, we performed serial transplant experiments of total DN3 thymocytes from 2-month-old mice to determine the functional consequences of loss of cell-cycle restriction (Fig. 5b). Interestingly, Lmo2 Tg DN3 cells lacking p21 were able to generate sevenfold more DN3 progeny than Lmo2 Tg DN3 cells expressing p21 (Supplementary Fig. 5e). In addition, absence of p21 promoted differentiation to DP cells (Fig. 5d)  revealed a progressive loss of DN3 repopulating activity such that there was complete loss by the fourth transplant (Fig. 5e). In contrast, the expansion of Lmo2 Tg DN3 remained relatively constant over serial transplants.
To determine if this pre-LSC exhaustion was only observable in the setting of proliferative stress related to transplantation, analogous to previous reports with p21-deficient HSCs 30 , we compared the numbers of DN3 cells in 6-and 12-month-old Lmo2 Tg and Lmo2 Tg ;p21 −/− mice. As previously reported 31 , there was a progressive increase in the numbers and proportion of DN3 cells in aged Lmo2 Tg mice ( Fig. 5f and Supplementary Fig. 5f). In contrast, absence of p21 prevented DN3 expansion with aging. Accordingly, there was marked loss of repopulating activity in 6month-old Lmo2 Tg ;p21 −/− thymus compared with age-matched Lmo2 Tg mice (Fig. 5g). Consistent with loss of pre-LSCs, there was reduced monoclonality as assessed by Tcrβ rearrangement (Supplementary Fig. 5g), and most importantly marked reduction of T-ALL penetrance in mice lacking p21 (Fig. 5h). We previously showed that Notch1 mutations are acquired during disease progression 20 . To assess the importance of cell cycle in the acquisition of Notch1 mutations, we used the RNA-seq data to identify variants differentially expressed in p21-deficient Lmo2 Tg DN3 cells, and found that Notch1 mutations were only present in Lmo2 Tg thymocytes (Supplementary Data 2). Targeted sequencing of the Notch1 locus in DN3 thymocytes isolated from 6month-old mice revealed that the presence Notch1 mutations was decreased by twofold in pre-LSCs lacking p21 ( Supplementary  Fig. 5h). In aggregate, these studies show that p21 was required for clonal evolution and leukemia progression of pre-LSCs.
To assess the role of p21 in therapeutic resistance, we measured repopulating activity in the thymus 24 h after multi-agent chemotherapy. At this time point, the proportion of surviving DN3 thymocytes was fourfold lower in Lmo2 Tg mice lacking p21 (Fig. 5i). Transplant of these chemoresistant thymocytes showed that repopulating activity was maintained in Lmo2 Tg thymocytes but markedly reduced in p21-deficient Lmo2 Tg thymocytes (Fig. 5j). The presence of p21 was also important for the resistance of Lmo2 Tg DN3 thymocytes to γ-irradiation (Supplementary Fig. 5i). Thus, p21 mediates resistance of pre-LSCs to chemotherapy and radiation.
Given the lack of cell cycle-restricted DN3 cells in Lmo2 Tg ;p21 −/− mice, we isolated total DN3 cells for gene expression profiling to gain insight into the role of p21 in pre-LSCs. We restricted studies to 6-week-old wild-type, p21 −/− , Lmo2 Tg and Lmo2 Tg ;p21 −/− mice, an age prior to loss of repopulating activity in p21deficient Lmo2 Tg mice (Fig. 5g). Principal component analysis confirmed clustering according to genotype (Fig. 6a). Comparison of wild-type DN3 cells with p21 −/− DN3 thymocytes revealed minimal changes with only 3 genes differentially expressed more than twofold (Supplementary Data 3). To determine how p21 abrogates Lmo2-induced leukemogenesis, we compared Lmo2 Tg DN3 cells with Lmo2 Tg ;p21 −/− DN3 thymocytes. Importantly, there was no difference in the expression of Lmo2 or its downstream targets responsible for self-renewal 19,32,33 such as Lyl1, Hhex and c-Kit (Fig. 6b). Overall, there were 463 differentially expressed genes: 153 increased and 310 decreased more than twofold in Lmo2 Tg ;p21 −/− DN3 cells (Fig. 6c and Supplementary Data 3). Gene ontology pathway analysis showed that the reduced genes were enriched for general metabolic pathways of transcription and translation as well as signaling (nuclear factor-κB, mitogen-activated protein kinase), G1/S transition and apoptosis (Supplementary Data 3). These changes were confirmed using GSEA, which revealed a striking reduction in genes involved in DNA replication, splicing and the proteasome (Fig. 6d). Importantly, these changes were not seen with p21-deficient DN3 cells compared with wild-type DN3 thymocytes ( Supplementary Fig. 6a). Thus, p21-mediated cellcycle restriction was required for widespread metabolic processes in the context of oncogene-transformed cells.
To understand the cellular fate of Lmo2 Tg DN3 cells in the absence of p21 (apoptosis or differentiation), we co-cultured sorted DN3 thymocytes on OP9-DL1 stroma cells, which support in vitro division and differentiation of T-cell progenitors 34 . Using this approach, we confirmed that absence of p21 promoted the differentiation of Lmo2 Tg DN3 thymocytes into DP cells (Fig. 6e). Pre-LSCs develop just prior to the β-selection checkpoint during which T-cell fate is tightly regulated by asymmetric cell division (ACD) 35 , a homeostatic cell division process also crucial for selfrenewal of HSCs 36,37 . ACD can be observed by the polarized segregation of the "differentiation fate determinant" Numb in dividing cells, which generate one identical immature/stem and one differentiated daughter cell 35,38,39 . Given that p21 has been associated with a switch from asymmetric to symmetric division in co-cultured stem cells 40 , we examined the frequency of ACD in sorted DN3 thymocytes using the Numb distribution in dividing cells (Fig. 6f). Consistent with the stem cell-like phenotype of pre-LSCs, the frequency of ACD was significantly increased in Lmo2 Tg DN3 cells compared with wild-type and p21 −/− DN3 cells ( Fig. 6f and Supplementary Fig. 6b). The increased ACD observed in Lmo2-expressing DN3 cells was significantly reduced in the absence of p21, restoring the preponderance of symmetric division observed in wild-type thymocytes. Altogether, these results show that p21-deficiency promotes differentiation of pre-LSCs at the expense of ACD, which correlates with the importance of p21 for the maintenance of self-renewing pre-LSCs during leukemia development.
We have previously shown that Lmo2 induces aberrant selfrenewal of immature T-cell progenitors without preventing T-cell differentiation 19 , and as such display features typical of pre-LSCs 43 . We now extend these findings to show that long-term self-renewal necessary for clonal evolution is limited to a rare subpopulation of cell cycle-restricted pre-LSCs. The impaired repopulating activity of GFP lo DN3 thymocytes might be explained by increased cycling leading to impaired homing. However, this is highly unlikely as DN3 cells lacking p21 had increased repopulating activity in primary transplants ( Fig. 5e) despite increased cycling. Consistent with their stem celllike properties, these cells are also more resistant to irradiation and combination chemotherapy. We did not directly examine the leukemic potential of the GFP hi cells enriched by chemotherapy; however, Ebinger et al. 11 recently showed that labelretaining cells surviving chemotherapy in human B-ALL xenografts retained leukemia-initiating potential. In sharp contrast with normal HSCs following stress-induced proliferation 25,44,45 , we show that once pre-LSCs enter a proliferative state, they are unable to return to a cell cycle-restricted state (Fig. 2e). Thus, strategies that promote cell cycle prior to chemotherapy may be able to eradicate pre-LSCs without detrimental effects on normal HSCs.
The role of p21 in leukemogenesis is controversial with both tumor suppressive and promoting activity. For example, knockdown of p21 in MLL-AF10-induced leukemia accelerated disease 46 . In contrast, PML/RAR (promyelocytic leukemia/retinoic acid receptor)-transformed HSCs required p21 for long-term Heat map of Lmo2associated upregulated gene signature in purified DN3 thymocytes from (a). c Heat map of genes differentially expressed (FDR < 0.05) in DN3 thymocytes from 6-week-old Lmo2 Tg , as compared to Lmo2 Tg ;p21 −/− as well as wild-type (WT) and p21 −/− controls. Row mean: relative expression of each gene as compared to the average expression for each genotype analyzed. d Gene set enrichment analysis (GSEA) plot of DNA replication, ribosome function (spliceosome) and proteasome genes in DN3 thymocytes from p21-deficient Lmo2 Tg mice, as compared to Lmo2 Tg mice (top panels), as well as p21 −/− DN3 cells compared to wild-type (WT) DN3 thymocytes (bottom panels). FDR false discovery rate, NES normalized enrichment score. e Immunophenotype of co-cultured DN3 thymocytes for 5 days for in vitro differentiation assays. N = 2 independent experiments, mean ± s.e.m., Student's t-test. f Representative confocal immunofluorescence of undergoing symmetric cell division (SCD, top row) and asymmetric cell division (ACD, bottom row) of DN3 thymocytes from 6-week-old mice in co-culture assays. Sorted DN3-stromal cell conjugates were fixed and co-stained with α-Tubulin (red) and the cell-fate protein Numb (green). N = 2 independent experiments, with specific numbers of cells analyzed indicated. Chi-Square test described for each genotype was performed ( Supplementary Fig. 6b); **p < 0.01, ***p < 0.001, as compared to WT; # p < 0.05, as compared to Lmo2 Tg cells, respectively self-renewal 47 . The observation that Lmo2 Tg DN3 cells lacking p21 have enhanced primary repopulating capacity but loss in subsequent transplants (Fig. 5f) provides one possible explanation for this controversy. The properties identified for cell cyclerestricted pre-LSCs have many parallels with normal HSCs under conditions of stress or aging. First, p21 is only important for HSCs in the setting of stress 30 . Second, activation of the p53-p21 axis promotes cell-cycle arrest and DNA repair in irradiated HSCs 28,48 . Third, error-prone DNA repair occurs in stress or aging HSCs due to reduced expression of genes required for highfidelity homologous recombination and components of the MCM helicase. Thus, we propose that cell cycle-restricted pre-LSCs arising from a committed progenitor behave like normal HSCs following DNA damage.
Mutation analysis of pre-LSCs identified an intriguing relationship between cell cycle and types of genomic mutations. Cell cycle of pre-LSCs was associated with a high frequency of aneuploidy. Almost half of all cells undergoing cell division had trisomy 15 and/or 2 (Fig. 4a). Although it is difficult to know which comes first (aneuploidy or cell cycle), the higher rate of aneuploidy in pre-LSCs unable to arrest (Lmo2 Tg ;p21 −/− ) suggests that cell cycle induces aneuploidy. Furthermore, studies of cell lines with trisomy generated from transgenic mice suggest that aneuploidy slows rather than promotes cell cycle 49,50 . Trisomy 15 has been reported in two other mouse models of leukemia 51,52 , suggesting the selective advantage for numerical abnormalities of chromosome 15 occurs irrespective of the oncogene or cell lineage. Acquisition of an extra copy of c-Myc is one possible explanation 51 . Decreased expression of MCM helicases, together with their reported interaction with LMO2 53 , may explain the high frequency of aneuploidy. The other striking difference was the presence of Notch1 mutations in GFP hi cells. Given these cells have a long-term selective advantage, and Notch1 enhances self-renewal 22 , this observation provides an explanation for the high frequency of Notch1 mutations in T-ALL.
Previous studies have shown that environmental cues from the niche control the balance between symmetric and asymmetric divisions of HSCs and this balance can be perturbed by the expression of oncogenes 39,54 . Using the segregation of Numb-a negative modulator of Notch1 signaling-to assess division patterning in thymocytes, we found that expression of the Lmo2 oncogene significantly increased the frequency of asymmetric division in DN3 cells. Conversely, absence of p21 reduced asymmetry and promoted differentiation in dividing Lmo2 Tg DN3 cells, restoring a division patterning similar to wild-type thymocytes. Thus, cell cycle plays a crucial role in the decision between different types of division, which ultimately effects the maintenance of the pre-LSC population. Given absence of p21 also increases aneuploidy, one unifying possibility is that the shortened duration of cell cycle leads to missegregation of differentiation fate determinants and chromosomes in dividing cells, ultimately impairing the stem cell-like properties of pre-LSCs.
In conclusion, our work provides the first in vivo evidence that cell-cycle restriction is essential for self-renewal, clonal evolution and therapeutic resistance of pre-LSCs in a murine model of T-ALL. We also demonstrate that pre-LSCs fundamentally diverge from normal HSCs with regard to their ability to return to a cell cycle-restricted state following stress-induced proliferation. We propose the H2B-GFP system will be a powerful in vivo model to determine if similar properties apply to other models of pre-LSCs and identify and test strategies to overcome quiescence of relapseinducing cells.

Methods
Mouse experiments. All experiments were approved and complied with the ethical regulations mandated by the AMREP Animal Ethics Committee. The  20 , using an antibody against Ki67 (1:10; BD Australia, Cat. no. 556027) or the isotype control, and staining DNA using DAPI (Sigma-Aldrich). Apoptosis was measured using the BD Pharmingen antibody against AnnexinV (1:40; Cat. no. 556420) and the permeable nucleic acid dye 7-aminoactinomycin D (BD Australia) following the manufacturer's protocol. FACS analysis was performed using a LSRII and a LSR Fortessa cytometers and cell sorting was performed with a FACSAria or BD Influx (BD Australia, North Ryde, NSW).
Modeling cell-cycle kinetics. The modeling cell-cycle kinetics was performed using the absolute numbers of GFP hi and total DN3 cells from 6-week-old H2B-GFP;Lmo2 Tg mice and littermate controls after 6 weeks of Doxycycline pulse, followed by 0, 1, 2, 4 or 8 weeks of chase, as indicated in Supplementary Fig. 1A.
Ratios were formed as GFP hi _DN3/DN3 and in order to stabilize the variance, the log (base 10) of the ratios was calculated as follows: logratio = log 10 ((GFP hi _DN3 + 100)/DN3). A constant (=100) was added to the numerator in the ratio to avoid taking the log of zero. The constant was chosen to be less than the smallest reported non-zero value of GFP hi _DN3 (=364). The combined data from both groups were fitted in an exponential model as Y = A + B*(R**X) in which in which Y is the log 10 of the ratio at week X. The nonlinear parameter (R) was constrained to be <1. The model has the property that at time zero (i.e., X = 0), Y = A + B and as time increases, Y asymptotes at A. The FITCURVE procedure in the GenStat statistical package 55 was used to fit increasingly complex models (to account for different values of the parameters in each group (H2B-GFP;Lmo2 and H2B-GFP). The model fitting exercise indicated that separate intercept "A" parameters were required for each group (p < 0.001) and separate slope coefficient "B" parameters were also required for each group (p < 0.001). The nonlinear parameter "R" was not significantly different between the groups (p = 0.124); nevertheless, separate nonlinear parameters were retained in the final model. The adjusted R 2 value for the final model was 83.8%. The fitted model is shown graphically in Fig. 1c. Bootstrapping the residuals from the fitted models was used to test for significant differences between the groups in their times for achieving 1-log and 2-log reductions in the ratios (n = 5000 bootstrap samples were used).
Transplantation assays. Transplantation assays were performed by intravenously injecting thymus cells into sublethally irradiated (650 Rads) isogenic Ly5.1 (Cd45.1) mice. Leukemic mice were scored positive when they presented signs of overt leukemia, which was confirmed at necropsy. Kaplan-Meier survival and statistical analysis were performed using GraphPad Prism 6.0 software (GraphPad Software Inc., San Diego, CA).
RNA analysis. Total RNAs for global gene expression were prepared from a pool of 0.7-1.0 × 10 6 sorted cells from 3 to 5 different 6-week-old mice using RNeasy extraction kit (Qiagen, Chadstone, VIC, Australia). Global RNA amplification and hybridization to the murine Agilent gene expression array was performed by the Australian Genome Research Facility Ltd (AGRF, Parkville, VIC, Australia) and analyzed using Agilent In Situ Microarray algorithm. Differentially expressed genes between DN3 thymocyte samples were determined using a linear model and the empirical Bayes method previously described 56 . Data for RNA-seq are available at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE110132. For gene expression experiments, total RNAs were prepared from 50,000 to 100,000 sorted cells from 6-week-old mice using RNeasy extraction kit (Qiagen, Chadstone, VIC, Australia). First-strand complementary DNA (cDNA) synthesis was performed by reverse transcription as previously described 57 . Primer sequences used are listed in Table 1. Real-time quantitative PCR was done with SYBR Green Master Mix (Applied Biosystems-Life Technologies Australia, Mulgrave, VIC) on Roche LightCycler480 II (Roche Diagnostics Australia, Castle Hill, NSW). Delta delta Ct values were calculated by using Ct values from Hprt and β-actin genes as reference.
Whole-exome sequencing. Genomic DNA samples were extracted from sorted GFP hi and GFP lo DN3 cells from 5 individual 2-month-old H2B-GFP;Lmo2 Tg mice using the QIAGEN DNeasy Blood & Tissue Kit (Qiagen, Chadstone, VIC, Australia). DNA quantity and quality were measured using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific Australia Pty Ltd, Scoresby, VIC, Australia). Genomic DNA amplification was performed on high-quality genomic DNA (300-1500 ng) using DETAILS by Macrogen Inc. (Macrogen Oceania, Sydney, NSW, Australia). Briefly, library construction was prepared by random fragmentation of the DNA or cDNA sample, followed by 5' and 3' adapter ligation. Library construction used DNA tagmentation, which combined fragmentation and ligation reactions into a single step, for increasing the efficiency of the library preparation process. Adapter-ligated fragments were then amplified by PCR and purified on gel. The whole exome was captured through target enrichment of DNA samples and construction of a hybridization library, using the Agilent SureSelectXT Library Prep Kit (Agilent Technologies, Santa Clara, CA, USA) according to the manufacturer's instructions. Sequencing was done with HiSeq 4000 instruments in high-output mode with TruSeq 3000 4000 SBS v3 chemistry. All runs were 101-nt paired-end reads, and data were analyzed with the HSC v3.3 software. Raw data were generated by the Illumina HiSeq 4000, which utilized HiSeq Control Software v3.3 for system control and base calling through the Real Time Analysis v2.7.3 software. The base calls binary was converted into FASTQ utilizing the Illumina bcl2fastq v2.17.1.14 protocol. Exome sequencing was performed on amplified DNA samples from sorted GFP hi and GFP lo DN3 cells from individual 2month-old H2B-GFP;Lmo2 Tg mice by Macrogen Inc. (Macrogen Oceania, Sydney, NSW, Australia).