Proliferation dynamics of acute myeloid leukaemia and haematopoietic progenitors competing for bone marrow space

Leukaemia progressively invades bone marrow (BM), outcompeting healthy haematopoiesis by mechanisms that are not fully understood. Combining cell number measurements with a short-timescale dual pulse labelling method, we simultaneously determine the proliferation dynamics of primitive haematopoietic compartments and acute myeloid leukaemia (AML). We observe an unchanging proportion of AML cells entering S phase per hour throughout disease progression, with substantial BM egress at high levels of infiltration. For healthy haematopoiesis, we find haematopoietic stem cells (HSCs) make a significant contribution to cell production, but we phenotypically identify a quiescent subpopulation with enhanced engraftment ability. During AML progression, we observe that multipotent progenitors maintain a constant proportion entering S phase per hour, despite a dramatic decrease in the overall population size. Primitive populations are lost from BM with kinetics that are consistent with ousting irrespective of cell cycle state, with the exception of the quiescent HSC subpopulation, which is more resistant to elimination.

H aematopoietic cells generate a turnover of billions of blood cells every day. The population-level paradigm for this system is the haematopoietic tree, a hierarchical commitment structure describing progressive amplification and differentiation with rare haematopoietic stem cells (HSCs) at its top 1 . These self-renewing, multipotent cells give rise to a cascade of increasingly lineage-restricted progenitors that are not selfrenewing. Steady-state blood cell production heavily relies on proliferation of intermediate progenitor cells, while more primitive populations such as HSCs are known to be relatively quiescent. Consequently, the mechanisms regulating HSC quiescence have been studied in great detail, and dormant HSC subpopulations have been identified [2][3][4][5] . In contrast, proliferative HSCs and downstream progenitors have been less well studied, despite their crucial role in maintaining steady-state haematopoiesis and regeneration following injury 6 . Understanding the kinetics of these cells holds vital clues about the regulation of these processes. It has, for example, been determined that HSCs become more proliferative in response to certain stresses, including infection [7][8][9][10] , but the fate of proliferative stem and progenitor cells under leukaemic stress is unknown.
Leukaemia is a form of cancer that originates from blood lineage cells. As leukaemia invades bone marrow (BM), haematopoiesis decreases to such an extent that patients typically present with symptoms such as anaemia, excessive bleeding or recurrent infections. How leukaemia outcompetes healthy haematopoiesis is only partially understood. It has been established that residual HSCs, both in murine leukaemia models and patients, are still functional in transplantation settings [11][12][13] . Furthermore, leukaemia-induced changes in BM stroma have been identified at advanced stages of disease [14][15][16][17] . Exactly how leukaemia leads to impaired haematopoiesis is yet to be determined, but potential explanations include: (I) a block in differentiation of haematopoietic progenitors 13 ; (II) a reduction in stem and progenitor cell production rates 12 ; and (III) an increase in apoptosis or emigration. We used the MLL-AF9 mouse model of acute myeloid leukaemia (AML) to better understand the contribution of each of these processes. (I) We determined absolute cell numbers of AML and haematopoietic stem and progenitor cells (HSPCs). (II) We quantified the numbers of AML cells and HSPCs entering S phase per hour. (III) We measured the proportion of apoptotic cells, and the appearance of healthy and malignant cells in blood and spleen.
Measurement of absolute numbers of healthy and apoptotic cells can be obtained by well-established flow cytometry protocols 18,19 . In contrast, quantification of cell production rates remains challenging. The proliferative behaviour of HSPCs has been studied in vivo through snapshot analysis of cell distribution across cell cycle stages 20,21 , uptake of 5-bromo-9-deoxyuridine (BrdU) or other nucleoside analogues over hours or days 22,23 , and through label retention or dilution 2,3,24 . However, cell production rates are not inferable from these methods. Dual pulse-chase nucleoside analogue labelling, pioneered by Wimber and Quastler 25 , identifies all cells that have entered into S phase in a given timeframe. We applied it to understand the cellular dynamics underlying AML growth and parallel loss of healthy haematopoiesis. We focused on HSPC populations phenotypically defined by SLAM gating 2,3,[26][27][28][29] and studied the following populations: lineage −/low c-Kit + Sca-1 + (LKS) CD150 + CD48 −/low , LKS CD150 − CD48 −/low and LKS CD48 + cells. For brevity, consistent with multiple recent publications [30][31][32][33][34] , we use the terminology HSCs, short term HSCs (ST-HSCs) and multipotent progenitors (MPPs), respectively, for each of these phenotypes. In addition, we measured absolute cell numbers, and proportions of apoptotic cells.
We found that a constant proportion of AML cells entered S phase per hour, and only AML cells in peripheral blood (PB) were less proliferative. We found that in steady-state conditions, MPPs amplify cell numbers more significantly than ST-HSCs or HSCs. The MPP compartment size is, however, comparable with the number of new cells produced every 2 weeks by HSCs and ST-HSCs. We identified that HSCs with the lowest levels of cell surface CD48 expression as the least proliferative in steady-state, and the best engrafters in transplantation. Contrary to expectation, as leukaemia invaded the BM and HSPC compartments became depleted, the fraction of MPPs entering S phase per hour remained largely unchanged. The loss of healthy cells was consistent with neutral ousting from BM by AML irrespective of cell type, with the exception of HSCs with the lowest levels of CD48, which proved more resistant to elimination.

Results
AML and HSPC cell counts during AML progression. GFP + MLL-AF9-induced blast cells from primary leukaemic mice 35 were injected into non-irradiated secondary recipient mice, which consistently developed high leukaemia burden within 3.5 weeks from injection (Fig. 1a). This approach allowed us to avoid irradiation-related injury of the haematopoietic system. To obtain detailed information on cell number dynamics, at days 12, 15, 18, 21 and 24, BM and spleen were harvested, and AML and nonmalignant mononuclear cells (MNC, excluding red blood cells) counts were determined.
AML cell numbers in BM increased exponentially from day 12 to day 18, followed by a period of slower growth in cell numbers over days 21 and 24, which constitute the final stages of the disease (Fig. 1b, left). At day 24, this resulted in 4.87 × 10 7 (±4.6 × 10 6 s.e.m) AML cells in BM contained within two tibias, two femurs and two ileac bones of each mouse analysed. BM MNC counts in the same bones started at 1.08 × 10 8 ± 3.93 × 10 6 cells and remained largely unchanged at homoeostatic levels until day 21. Over days 21-24, as the growth in AML cell numbers was tailing off, MNC counts dropped to approximately a tenth of their starting number (Fig. 1b, right).
During disease progression, the spleen weights were measured, revealing significant enlargement (Fig. 1c) as a result of colonisation by leukaemia cells (Supplementary Fig. 1) combined with extramedullary haematopoiesis. Spleen weight dynamics contrasted with the growth of AML cell numbers in BM, as there was some expansion between days 12 and 18, but a dramatic increase over days 21 and 24, coincident with slower growth of AML cell counts in BM ( Supplementary Fig. 1).
Downstream of LKS CD48 + cells are populations with a hierarchical relationship that is not yet fully characterised [37][38][39][40][41] . The dynamics of HSC, ST-HSC and MPP cell numbers mimicked those of MNCs as a whole, remaining largely unchanged until growth in AML cell numbers decreased, at which point they dropped dramatically. Interestingly, we detected significant differences between the cell populations analysed, with ST-HSCs already significantly reduced by day 21, while HSCs and MPPs were more variable at that time (Fig. 1e).
These data were consistent with multiple hypotheses. For example, the AML cell number data were compatible with a decrease in AML proliferation or increase in apoptosis from day 21. They were also compatible, however, with a model where AML proliferation continues unabated, but a lack of space in BM results in cells being displaced into PB. Similarly, HSPC numbers could decrease due to an overall decrease, or even shut down, of healthy cell proliferation and/or increased death. Alternatively, healthy cells could be eliminated by leukaemia through active mechanisms, such as "winning" competition for space. In order to distinguish between these possibilities, we reasoned that we required additional information on proliferation dynamics.
Identification of cells entering S phase. Cells undergoing S phase in the presence of BrdU incorporate it into their DNA. Continuous delivery of BrdU results in a consistent pattern of labelling 22 (solid line in Fig. 2a), where the number of BrdU + cells measured is the sum of all cells that were in S phase at the time of initial delivery plus a decreasing proportion of additional cells over time as BrdU − cells start DNA synthesis and BrdU + cells cycle further. Analysis of cells harvested shortly after a pulse of BrdU identifies all cells in S phase in that window, but in order to convert that number into a proportion of cells entering S phase per hour, one would need to divide it by the duration of S phase, which is unknown and variable. The proportion of cells entering S phase per unit time is the derivative at the origin of the curve depicting the proportion of cells labelled under continuous BrdU administration (Fig. 2a, red dashed line). In the absence of cell cycle arrest and endocycling, this serves as a proxy for the fraction of the population that is proliferating at the time of measurement. This derivative cannot be deduced from a single time-point measurement with a single label-uptake system as one must measure the offset at time zero as well as a second measurement shortly after label administration.
To overcome these limitations, we further developed a shorttimeframe, dual pulse label system that was previously established, verified and used in multiple studies, primarily focused on immune cell proliferation [42][43][44][45][46][47] . It is based on combining BrdU with another thymidine analogue, 5-ethynyl-2′-deoxyuridine (EdU). An initial intravenous EdU pulse was followed 2 h later by intravenous administration of BrdU, and 30 min later cells were harvested and analysed (Fig. 2b). All cells that were in S phase at the time of the initial EdU pulse became EdU + (blue cells, Fig. 2c, left and middle), while cells that were in S phase during both pulses became EdU + BrdU + double positive (purple cells, Fig. 2c). Cells that entered S phase in the 2 h between the initial EdU pulse and the BrdU pulse were negative for the EdU label but positive for BrdU label (EdU − BrdU + , orange cells in Fig. 2c). With this method, one obtains a measurement of the proportion of cells in a population entering S phase in a given 2-h time window. By multiplying the proportion of EdU − BrdU + cells by the overall number of cells, the number of cells in that population that entered S phase in the 2-h window can be calculated. The interval between the EdU and BrdU pulses was chosen based on two competing factors. For accurate inference, it had to be long enough to capture a sufficiently large number of cells entering S phase. However, it also had to be short enough to ensure that, in the interval between the pulses, cells could neither pass through S phase during the interval without label uptake, as that would result in an EdU − BrdU − cell in lieu of an EdU − BrdU + cell, nor take up the EdU label, transit through G2/M/G1, and re-enter S phase picking up the BrdU label, as that would result in two EdU + BrdU + cells instead of one EdU + BrdU − cell. In vitro time lapse microscopy of activated fluorescence ubiquitination cell cycle indicator (FUCCI) lymphocytes indicated a lower bound on G1 of approximately 2 h and that S/G2/M is typically greater than 4 h 48 , guiding our consideration to intervals of 2 h or less.
To confirm the method's accuracy, we analysed HSPC populations of mice that were pulsed with EdU, and received BrdU 1 or 2 h later. As expected, increasing the gap from 1 to 2 h led to a 2-fold increase in the proportions of EdU − BrdU + cells for all HSCs, ST-HSCs and MPPs populations (Fig. 2d). We chose a 2-hour labelling period as it increased the number of BrdU single positive labelled cells, and therefore provided increased accuracy for more quiescent populations.
We confirmed that we could detect differences in the rate of S phase entry by examining the mice treated with the immune- As AML cells undergo exponential growth in the BM between days 12 and 18 (Fig. 1b), and between days 18 and 24 in the spleen (Supplementary Fig. 1), this afforded us an additional opportunity to confirm the dual label method's accuracy. For a cell population undergoing negligible cell death and growing exponentially, exponential regression enables estimation of the proportion of the population proliferating per unit time. While no method exists to quantitatively determine the number of cells dying per hour in vivo, analysis of apoptotic AML cells by Annexin V and DAPI staining confirmed that these undergo little death (Fig. 3c). Thus, exponential regression from total AML cell numbers provided an estimate that~3% AML cells enter S phase per hour in both BM and spleen during the exponential growth phase. The dual label measurement provided the same estimate  . 3d), but required no assumptions on death and could be performed with a single sample harvested at a single time point. Importantly, the method could be applied to non-exponential growth phases, where population size changes could be a consequence of either slower growth or cell loss, or a combination of the two.
A constant proportion of AML cells enter S phase per hour.
The constant proportion of AML cells entering S phase per hour in BM coupled with the slow down in the growth of AML cell number indicated that from day 18 there was no longer room for all progeny to remain in that organ. We reasoned that this could explain how BM AML cell numbers grew subexponentially (Fig. 1b). Therefore, we hypothesised that a significant number of cells were displaced from the BM once most of that space had been occupied. To test that, we measured AML cell infiltration in PB. Consistent with our hypothesis, PB infiltration increased dramatically during the time when AML growth in the BM was reduced (Fig. 3e). By comparing multiple groups of mice at different stages of BM infiltration, we identified two patterns for the presence of AML cells in the PB where PB infiltration was low when BM infiltration was low (<60%), and then substantially higher when BM infiltration was high (≥60%), ( Supplementary  Fig. 3). Based on these data, the number of AML cells produced in BM that are displaced into PB could be estimated (Methods). By day 24, these would constitute a significant proportion of the increase in MNC counts in PB (Fig. 3f). We reasoned that if AML from the BM was a dominant source of AML in the PB, the division rate of PB cells could not account for expansion alone and the proportion of AML cells entering S phase per hour in PB should be lower. Consistent with this hypothesis, the proportion of AML cells entering S phase per hour in PB was approximately onethird of that in BM, at~1% per hour (Fig. 3g).
MPPs are major cell amplifiers in steady-state haematopoiesis. We turned our attention to early HSPCs, first examining steadystate haematopoiesis so that we could better understand the dynamics triggered by leukaemia growth. We acquired sufficient numbers of three phenotypically defined primitive populations (HSCs, ST-HSCs and MPPs) to allow reliable identification of their BrdU + EdU − , BrdU + EdU + , BrdU − EdU + and BrdU − EdU − fractions (Fig. 4a). We also measured the total number of cells in each population in the hind limbs of every mouse analysed (Fig. 4b). As expected, HSCs were least numerous, followed by increasingly more abundant ST-HSCs and MPPs. Knowing the proportion of cells that entered S phase per hour (Fig. 4a) and the number of cells in each population (Fig. 4b), we calculated the number of cells in each population entering S phase per hour (Fig. 4c). Of the three populations, MPPs had by far the greatest number (left panel) and fraction (right panel) of cells entering S phase per hour. The proportion of MPPs entering S phase is substantially higher than that of both HSCs and ST-HSCs, which had similar proportions of cells entering S phase. Thus, the high number of proliferative MPPs is not solely a consequence of the larger size of that population, and the MPP stage is a significant amplifier of cell numbers.
CD48 level correlates with HSC proliferation and function. The proliferation and compartment size data allowed us to directly compare HSC and ST-HSC output with MPP cell numbers. Every week LKS CD48 −/low cells (HSCs and ST-HSCs combined) produced half as many cells as there were LKS CD48 + cells (MPPs) (Fig. 4d). This finding was consistent with a recent study 6 , but did not explain the dormancy or low proliferation rate of HSCs 2-4 .
In search of these more quiescent cells, we investigated the distribution of proliferating cells (EdU − BrdU + , EdU + BrdU + and EdU + BrdU − ) within the overall LKS compartment. We noticed that they were particularly rare within the HSC compartment that is most negative for CD48 (Fig. 4e). We therefore re-gated our data based on fluorescence minus one controls to separate, on average, the 25% LKS CD150 + CD48 −/low cells that had the lowest levels of CD48. Henceforth we denote these as CD48 neg HSCs, and the remaining HSCs as CD48 low HSCs (Fig. 4f). We found that CD48 neg HSCs were turning over at a substantially lower rate than the CD48 low HSCs (Fig. 4g). This was consistent with other reports indicating that approximately 20% of LKS CD150 + CD48 −/low are label-retaining cells 2 and that not all cells within that phenotypically defined population are functional HSCs in longterm serial transplant settings 26,29 .
To investigate the functional distinction between the CD48 neg and CD48 low HSCs, we tested their engraftment ability by transplanting 100 mTomato + cells of either type together with 200 k whole BM haematopoietic cells from wild-type donors into lethally irradiated recipients. The percentage of donor derived cells was measured in the blood of CD48 neg and CD48 low HSCs recipients at weeks 8, 12 and 20 post transplant, both overall and in the B, T and Myeloid cell compartments (Fig. 4h left panel). BM constitution at week 20 was then analysed according to BM subpopulations (Fig. 4h  right panel). Recipients of CD48 neg HSCs exhibited consistently higher engraftment at all time points in both PB and BM than recipients of CD48 low HSCs, indicating that the quiescent CD48 neg population is, indeed, functionally distinct. This result was confirmed by a serial transplantation experiment, which showed that only CD48 neg HSCs have appreciable secondary engraftment (Fig. 4i).
CD48 neg HSCs are most resistant to ousting by AML. In contrast to label flux methods 2-4, 24, 30 , the dual label approach is applicable to systems with evolving dynamics such as the haematopoietic shutdown in the BM during leukaemic infiltration. It has been suggested that leukaemia induces a progressive increase of quiescence on residual haematopoietic cells, measured as increasing proportions of cells in G0 12,13 . One would therefore expect that under leukaemic stress, the proportion of primitive haematopoietic populations entering S phase per hour would progressively decrease. To investigate if this was the case, we analysed the proliferation of residual healthy early HSPCs in the MLL-AF9-driven AML model described earlier (Fig. 1).
We measured the number of CD48 neg HSCs, CD48 low HSCs, ST-HSCs and MPPs entering S phase per hour (Fig. 5a). For CD48 neg HSCs, this was below the limit of detection throughout disease progression. For CD48 low HSCs and ST-HSCs, there was no evidence of significant increase or decrease in the proportion of cells entering S phase per hour during disease progression. For MPPs, we found a strong linear correlation with the overall number of MPPs remaining in the BM. This indicated that the proportion of MPPs entering S phase per hour remained unchanged despite their decreasing size.
To better understand how the observed proliferation dynamics fitted with the observed cell loss, we focused on the last 6 days of disease, when all cell populations dramatically decrease in size. Measurement of the average cell numbers for each population (Fig. 5b, bars) established that all cell types were reduced to approximately 10% of their initial size (Fig. 5b, right y axis). In numbers, the MPP compartment decreased most dramatically, but between days 18 and 21, ST-HSCs experienced the greatest proportional decrease and CD48 − HSCs the least. Taken together, these data raised the questions: (1) why did HSCs appear increasingly quiescent in previous studies 12,13 ; (2) could there be a common mechanism leading to stem and progenitor cell loss from the BM; and (3) are ST-HSCs affected by AML differently compared to the other LKS subpopulations?
To address the first question, we focused on the relative survival of CD48 neg and CD48 low HSCs, as in steady-state, CD48 neg HSCs were the least proliferative (Fig. 4g). As these cells were being depleted at the slowest rate, we reasoned that this could explain the previously reported increasing proportion of quiescent HSCs 12,13    LKS CD48 low HSCs at the latter stages of disease (Fig. 5c). Figure  5d shows the evolution of the absolute size of the two HSC subpopulations over time (left panel), where the substantial depletion of CD48 low HSCs in advance of CD48 neg HSCs is evident, and results in the relative enrichment of the least proliferative HSCs (right panel).
To explain the declining cell numbers of primitive HSPCs, we hypothesised that all haematopoietic cells may be lost independently of cell type or proliferative state, with the only exception being CD48 neg HSCs. The sum of all MNCs (healthy and leukaemic populations) in mice at late stages of disease was 40% lower compared to age-matched controls (Fig. 5e). Assuming a   densely packed BM, this suggested that malignant cells are larger, on average, than healthy MNCs. This was confirmed qualitatively by analysing forward scatter of healthy and malignant haematopoietic cells using flow cytometry ( Supplementary Fig. 2), and quantitatively by measuring cell sizes in histological sections of BM from healthy and diseased mice. The latter data also illustrated that the BM was equally packed with haematopoietic cells regardless of steady-state or malignant haematopoiesis (Fig. 5f). The estimated cell size calculated based on MNC numbers and the measured one from histological sections were similar, with AML cells being on average between 1.5 and 2 times bigger than the average healthy cell, based on 1000 bootstrap samples from two control and three AML samples at day 24 (Fig. 5g). We therefore built a simple model in which for every new AML cell that remains in the BM, from day 18 onwards approximately 1.5-2 non-AML cells are randomly ousted without cell type preference. Under these circumstances, the rate of ousting of a cell population is proportional to its size (Methods). The resulting depletion kinetics are presented as red dashed lines in Fig. 5b and, despite the model's simplicity, provide a good explanation of the data. In this model, healthy cells can be lost by both displacement into PB or through death, and the predicted loss of ST-HSCs remains less dramatic than observed.
To understand the interplay of displacement and death as drivers of cell loss, we performed Annexin V and DAPI staining on c-Kit enriched haematopoietic cells harvested from mice at advanced stages of AML progression (Fig. 5h). We observed a modest proportion of apoptotic and dead HSCs and MPPs, equivalent to those observed in healthy control mice ( Supplementary Fig. 4), but detected a substantially higher proportion of apoptotic or dead ST-HSCs when compared with other HSPCs (Fig. 5h) and to steady-state BM (Supplementary Fig. 4). The reason why ST-HSCs were observed to be more prone to apoptosis than other HSPC populations, especially at high AML infiltration, remains to be elucidated. A combined effect of death and displacement of ST-HSCs is, however, consistent with their rate of loss being higher than that predicted by elimination alone. The data suggested that elimination from BM rather than cell death may be largely responsible for loss of other HSPCs.

Discussion
Blood production is maintained by rare HSCs through proliferative down-stream multipotent cell types that give rise to lineage committed cell populations. Here, we focused on the proliferation of early haematopoietic cells, which are responsible for blood production maintenance, analysing their rate of entry into S phase in steady-state and under leukaemic stress.
We determined the size of early haematopoietic cell populations in BM as leukaemia progressively infiltrates the BM. We further developed a dual-pulse DNA labelling method that enables in vivo measurement of the proportion of cells entering S phase per hour in multiple cell populations simultaneously. We analysed the malignant component of our system, based on the well-established MLL-AF9-driven murine model of AML. Cell number measurements revealed a decrease in AML growth in BM towards the end of the disease. However, the dual-pulse method established that a constant fraction of AML cells within the BM were entering S phase per hour. Deviation from exponential growth in cell numbers coincided with the appearance of AML cells in the PB, where AML proliferated less vigorously, and suggested a model where AML ultimately runs out of space in the BM. While our data are suggestive of the ousting hypothesis, direct confirmation would require lineage-tracing methodologies. Interestingly, the measured proportion of cells entering S phase per hour for BM AML cells is within the range observed with leukaemia cells in culture 49 . The reduced rate of entry into S phase of AML cells in PB compared with BM is supportive of current hopes of improving AML outcome through the use of mobilising agents [50][51][52] .
In steady state, we established that every week LKS CD48 −/low cells (HSCs and ST-HSCs) produced half as many cells as there were LKS CD48 + cells (MPPs), indicating that the most primitive cell population strongly contributes to steady-state haematopoiesis, in agreement with a recent study 6 . Interestingly, our measurements of the proportion of MPP and AML cells entering S phase per hour were not significantly different (p = 0.38, Welch's two-sample t-test), suggesting that mechanisms other than rapid proliferation are at the root of aggressive expansion observed in AML.
As there is ongoing speculation regarding the level of enrichment of long-term HSCs using SLAM gating 2, 3, 27-29 and a feature of our approach is that it allows retrospective re-gating, we investigated the proliferation of the subpopulation of LKS CD150 + CD48 −/low cells identified as having the lowest CD48 expression, which we called CD48 neg HSCs. These cells proved to be less proliferative than their LKS CD150 + CD48 low counterpart and thus CD48 low HSCs were the more significant contributors to steady-state blood production, consistent with a recently published study 30 . CD48 neg HSCs had greater reconstitution ability than CD48 low ones, in agreement with studies showing that quiescence is linked to better engraftment ability 3,4,8 . The LKS CD150 + CD48 −/low population has been fractionated based on CD34 expression 36 where CD34 − HSCs have been shown to be the most metabolically inactive 3,4 . Other studies reported that approximately 20% of LKS CD150 + CD48 −/low cells are label retaining 2 . This figure is consistent with our work, which provides a phenotypic identification of the least proliferative HSCs.
Overall, our data confirm that the LKS CD48 + population is a critical haematopoietic cell number amplifier, consistent with label retention studies 2, 3 . Our finding that CD48 low rather than CD48 neg HSCs are more significant contributors to homoeostatic maintenance is in line with the finding that in CD48-null mice HSCs appear more quiescent and haematopoiesis is progressively impaired 53 .
As AML cell numbers in BM grow, LKS CD150 + CD48 −/low (HSCs), LKS CD150 − CD48 −/low (ST-HSCs) and LKS CD48 + (MPPs) numbers all decreased in concert, suggesting that a process beyond increased HSC quiescence was responsible for their loss. Surprisingly, we found that the fraction of MPPs that entered S phase per hour was near constant throughout disease progression, despite the overall reduction in the population size, while there was no evidence of increased entry into S phase of other HSPCs. This suggests that shut down of healthy haematopoiesis during leukaemia may be a consequence of a distinct mechanism to the recently proposed one of leukaemia-induced quiescence 12,13 . Because an increase in cell death as disease progressed was observed only for LKS CD150 − CD48 −/low cells (ST-HSCs), our data led us to propose a model where cells of the haematopoietic system are ousted from the BM by the infiltrating leukaemia irrespectively of proliferative behaviour or phenotypic characteristics, with the exception of LKS CD150 + CD48 −/low cells with the lowest level of CD48 expression, which proved relatively resistant to removal. As these are the least proliferative HSCs, the data reveal a proportional enrichment of apparently quiescent HSCs. This is consistent with earlier studies that identified G0enriched, engraftment-able HSCs 13 that express high levels of the quiescence inducer Egr3 12 were preserved despite AML cells occupying the majority of BM. Taken together, these and our study raise the questions of how a subpopulation of HSCs survives leukaemia growth longer than any other cell type, and whether indeed quiescent cells survive or, alternatively, surviving cells are induced into quiescence.
In conclusion, we found the dual-pulse method combined with accurate total cell counts to be essential methods for gaining novel insights in the cellular dynamics underlying the competition between healthy and malignant haematopoiesis. Future studies will indicate whether similar dynamics hold true following chemotherapy treatment, haematopoietic recovery, and disease relapse. Understanding the reason for the differential dynamics of HSPC subpopulations will provide clues for better targeted interventions aimed at restoring healthy haematopoiesis in leukaemia patients.

Materials and methods
Mice. All animal experiments were approved and performed according to the standards of the animal ethics committee at Imperial College London and Sir Francis Crick Institute, and to UK Home Office regulations (ASPA 1986). C57Bl/6 WT mice were purchased from Harlan UK Ltd., and Charles River UK; CD45.1 and mTmG mice were bred and housed at Imperial College London, UK, or provided by the Sir Francis Crick Institute breeding facility (London, UK).
BrdU and EdU administration. For cell cycle analysis, EdU and BrdU were used. Both are incorporated during DNA synthesis and, combined, allow analysis of cell cycle kinetics 45 . Pulse-chase timings were optimised to achieve consistent in vivo incorporation of EdU and BrdU, and later detection in the cell populations of interest. EdU (1 mg/mouse) and BrdU (2 mg/mouse) were administered via tail vein injection 2 h apart, unless otherwise stated; 30 min later, mice were culled and the tissue was isolated. Poly-I-poly-C (Sigma) was administered intraperitoneally as a single dose of 100 µg, 24 h prior to culling.
MLL-AF9 AML model. Six-week-old CD45.1 or mTmG female mice were BM donors for the generation of AML leukaemia by retroviral transduction of purified granulocyte/monocyte progenitors (GMPs) as previously described 35 using the pMSCV-MLL-AF9-IRES-GFP construct 54 . Transduction of sorted GMPs was performed by plating sorted cells on retrovirus-coated plates in DMEM, 10% FCS, 1% non-essential amino acids, 0.1% β-mercaptoethanol, 0.5% PenStrep, 20 ng/ml SCF, 10 ng/ml IL-3, and 10 ng/ml IL-6 for 72 h. Transduction efficiency was determined by flow cytometry and each primary recipient mouse received 25,000 transduced cells intravenously. Primary recipients were sub-lethally irradiated (two doses of 3 Gy greater than 3 h apart) before transplantation. Primary blasts were harvested from the BM of primary recipients at 8-10 weeks post transplant (infiltration >90%) and banked in liquid Nitrogen. Secondary leukaemias were set up in non-irradiated recipients by injecting 100,000 AML blasts intravenously. Secondary recipients were culled for flow cytometry analysis at the indicated time points during disease development. Total MNC counts in PB were obtained by differential cell count using an Advia 2120 haematology analyser (Siemens Diagnostics).
Flow cytometry. All flow cytometry antibodies used are listed in Supplementary  Table 1, and all data were acquired on an LSRFortessa flow cytometer (BD BioSciences, CA). For GMP and HSC subpopulations sorting, red blood cells were lysed, then BM cells were stained with a biotin-conjugated lineage cocktail comprising anti-CD3, CD4, CD8, B220, Ter119, Gr1, and CD11b. Lineage positive cells were depleted by using streptavidin-conjugated magnetic beads (Miltenyi Biotech, Germany) and magnetic columns (LD columns/quadromacs holder, Miltenyi Biotech). The lineage negative-enriched cells were stained with streptavidin and antibodies against c-Kit, Sca-1, CD16/32 and CD34 before sorting for GMPs (Linc-Kit+Sca-1-CD16/32+CD34hi) using a BD ARIA III.
For the analysis of the stem and progenitor cell compartments, red cell depleted BM or PB cells were stained with the biotin-conjugated lineage antibody cocktail and c-Kit, Sca-1, CD48, CD150 antibodies, followed by streptavidin Pacific Orange (Life Technologies, CA).
BrdU and EdU were detected using a combination of the BrdU Kit (Becton Dickinson) and the Click-iT EdU Kit (Life Technologies): after performing the cell surface antigen stain, the cells were fixed and permeabilised before incubation in the Click-iT EdU reaction. DNase treatment followed and intracellular anti-BrdU stain was performed immediately before analysis by flow cytometry. For analysis of BrdU/EdU incorporation in AML blasts and residual healthy haematopoietic cells, the initial cell surface antigen staining included also anti CD45.1 antibody.
For the detection of apoptotic cells, Annexin V (APC, BB) and DAPI (Life Technologies) staining were performed according to the manufacturers' instructions on red cell lysed (to identify apoptotic AML cells), and red cell lysed, c-Kit enriched BM cells (to identify apoptotic healthy cells) from healthy control and leukaemia-burdened animals. c-Kit enrichment was performed using anti-c-Kit magnetic beads-conjugated antibody and LD columns (Miltenyi) according to the manufacturer's instructions.
Calibrite beads (BD Biosciences) were used to determine the absolute cell numbers in the populations of interest by flow cytometry. These were added to freshly isolated, untreated BM and spleen and red cell lysed blood at a concentration of 2-5% of the total cells in the tube and acquired by flow cytometry, using a white blood cell/MNC gate to exclude erythrocytes from the counts (remaining cells in BM and spleen are referred to as MNCs). Total cell numbers of different cell populations were back calculated based on their frequency in the samples analysed.
Measuring the proportion of cells entering S phase per hour. For a given population of interest, let n(t) denote the number of living cells and n s (t) denote the number of cells in S phase at time t. With e s [0,t) being all cells that enter S phase and d s [0,t) being all cells that depart S phase in the interval [0,t), then n s (t) = n s (0) + e s [0,t) − d s [0,t). With r denoting the fraction of cells that enter S phase per unit time, for t sufficiently small e s [0,t) = rn(0)t. With a single label, BrdU, continuously administered from time 0, the number of BrdU + cells at time t sufficiently small satisfies BrdU + (t) = n s (0) + e s [0,t) = n s (0) + rn(0)t. Thus to extract r, one would need measurements at two time points as one would determine the derivative of the proportion of BrdU + cells at t = 0, r ¼ lim BrdU þ t ð Þ= n 0 ð Þt ð Þ. With two labels, EdU administered at time 0 followed by a gap of length T to second label, BrdU, r can be determined directly. For any T smaller than both the length of S phase and the time to traverse G2/M/G1, one has that EdU − BrdU + (T) = e s [0,T) = rn(0)T and thus, for T sufficiently small, the fraction of cells entering S phase per unit time can be identified as the fraction of EdU + BrdU + cells, r = EdU − BrdU + (T)/(n(0)T).
CD48 neg and CD48 low HSC transplantation. CD48 neg and CD48 low HSCs were FACS sorted from the BM of mT/mG donor mice, and 100 were transplanted into lethally irradiated WT recipients (two doses of 5.5 Gy separated by more than 3 h), alongside 200,000 competitor WT BM cells. Baytril antibiotic was administered to recipient mice for 5 weeks post transplant. Multi-lineage engraftment of the recipients was monitored at 8, 12 and 20 weeks post transplant by FACS analysis of PB. Whole BM was harvested from the CD48 neg and CD48 low HSC primary recipients. Each experimental sample group was used to seed a secondary cohort of lethally irradiated WT recipients, with each mouse receiving 100 pooled tomato + HSCs (LKS CD150 + CD48 −/low ) and 200,000 competitor WT BM cells. Multi-lineage engraftment was monitored as with the primary recipients.
AML cell accumulation in PB. The numbers of MNCs in blood were estimated by measuring MNC counts per ml of blood and based on 2.5 ml total PB per mouse. The difference in numbers between days 18 and 24 was determined by subtracting the average from day 18 from each measurement at day 24.
Number of AML cells displaced from BM in late stage disease. The number of AML cells produced but unaccounted for in the BM was computed by assuming linear growth in AML BM cell numbers between days 18 and 21, where there is~4fold increase, and between days 21 and 24, where the increase is~1.9 fold. The average proportion of EdU − BrdU + AML cells proliferating per hour over days 18-24 was also determined, p AML = 0.03. The number of displaced cells was the number of newly created cells in the BM less the number required to explain the number of AML cells in the BM: 36p AML AML 18 ð Þþ2 AML 21 ð Þ þAMLð24Þ ð Þ À AML 24 ð ÞÀAML 18 ð Þ ð Þ : Bootstrap confidence intervals were determined by resampling the data 1000 times, with replacement, and repeating the above computation.
Estimation of AML and healthy haematopoietic cells size. Wild-type or Col2.3-GFP mice were lethally irradiated (two times 5.5 Gy, 3 h apart) and reconstituted with whole BM from mT/mG donors to generate chimeras in which the size of healthy mTomato+ BM cells could be measured. Alternatively, non-irradiated wild-type recipients were injected with 100,000 mTomato+ GFP+ MLL-AF9 primary blasts to measure the AML cell size in histological sections. Eight weeks after reconstitution with healthy mTomato+ BM, or at multiple stages of AML disease, tibias and ileac bones were harvested and processed as described before 16 . Briefly, the bones were fixed in periodate-lysine-paraformaldheyde (PLP), immersed in a sucrose gradient and frozen in optimal cutting temperature (OCT) compound (TissueTek). Sections were obtained with a cryostat (Leica) and the Cryojane tape transfer system (Leica Microsystems) and mounted with Prolong Diamond antifade (Invitrogen) for analysis. Miroscopic images were acquired with a confocal/ two photon hybrid microscope (Zeiss 780 NLO) equipped with a Mai Tai (Spec-traPhysics) laser and 488, 561 and 633 nm lasers. The images were analysed using ImageJ/Fiji.
Statistical analysis. Data were processed, analysed and visualised using R and MATLAB. Statistical tests were performed as described in figure legends. The outcomes were treated as significant if p < 0.05.