Characterization of Rare, Dormant, and Therapy-Resistant Cells in Acute Lymphoblastic Leukemia

1After initially successful chemotherapy, relapse frequently jeopardizes the outcome of cancer patients. To improve the prognosis of ALL patients, treatment strategies that eliminate tumor cells at minimal residual disease (MRD) and prevent relapse are required. Toward a better understanding of the underlying biology, we established preclinical mouse models mimicking MRD and relapse in patients. Primary and surrogate MRD cells shared major similarities in expression profiles, demonstrating the suitability of our model. MRD cells revealed major functional plasticity in vivo and treatment resistance was reversible; MRD cells became sensitive toward treatment once released from their in vivo environment. Effective therapeutic strategies might aim at dissociating persistent cells from their protective niche to prevent relapse in ALL patients.

2Relapse represents a major threat for patients with cancer. After initially successful treatment, rare tumor cells might survive and re-initiate the malignant disease with dismal outcome. Acute lymphoblastic leukemia (ALL) is associated with poor prognosis in infants and adult patients and is the most frequent malignancy in children. In many patients, the majority of ALL cells respond to chemotherapy but a minority display resistance, survive therapy, and cause relapse with poor outcome.

3Despite its clinical importance, basic biologic conditions underlying relapse remain partially elusive. For example, it is unclear whether relapse-inducing cells exist before onset of treatment or develop as result of therapy, and whether permanent or reversible characteristics determine relapse-inducing cells. Of translational importance, understanding basic mechanisms opens perspectives for effective therapies to eradicate relapse-inducing cells.

4Relapse-inducing cells, by their clinical definition, self-renew and give rise to entire tumors indicating tumor-initiating potential, a typical characteristic of cancer stem cells. In numerous tumor entities including acute myeloid leukemia, cancer stem cells were identified as a biologically distinct subpopulation that displays specific surface markers, has leukemia-inducing potential in mice, and gives rise to a hierarchy of descendant cells that lack such properties. In ALL, however, many different subpopulations display stem cell properties; neither a stem cell hierarchy nor phenotypic markers defining stem cells could be identified. Thus, up to now, stemness represents an insufficient criterion to define the subpopulation of relapse-inducing cells in ALL.

5An additional feature of relapse-inducing cells is their treatment resistance, as, again by definition, they survive chemotherapy and eventually give rise to relapse with decreased chemosensitivity. Resistance against chemotherapy is closely related to dormancy as chemotherapy mainly targets proliferation-associated processes that are inactive in dormant cells. Dormant cells, by definition, do not divide or divide very slowly over prolonged periods of time, might survive chemotherapy, persist in minimal residual disease (MRD), and give rise to relapse. Indeed, an increased frequency of non-dividing tumor cells has been described in patients after chemotherapy for defined subtypes of ALL.

6So far, technical obstacles have hampered characterizing phenotypic and functional features of relapse-inducing cells in ALL in detail. Established ALL cell lines represent inappropriate models as they display continuous proliferation. In patients, relapse-inducing cells are very rare and defining cell surface markers that reliably identify these rare ALL cells from the multiplicity of normal bone marrow cells remains intricate, at least in certain ALL subtypes. Moreover, primary ALL cells do not grow ex vivo, disabling their amplification in culture.

7An attractive possibility to experimentally study patients' tumor cells in vivo is the patient-derived xenograft (PDX) model, which uses immuno-compromised mice to expand tumor cells from patients. As shown previously, PDX ALL cells retain important characteristics of primary ALL cells. While PDX models are mostly used for preclinical treatment trials, we used them here to study relapse-inducing cells in ALL.

8To characterize the challenging subpopulation of relapse-inducing cells in ALL, we used the individualized xenograft mouse model as a preclinical model, molecular cell marking as an unbiased approach, and in vivo dormancy as a functional benchmark. To mimic the heterogeneity of ALL, samples from nine different ALL patients were studied including children and adults, B cell precursor-ALL and T-ALL, first diagnosis, and relapse (Table S1).

9To study ALL growth starting very early after disease onset in the PDX mouse model, the technical challenge consisted in reliably enriching very low numbers of human ALL cells from mouse bone marrow. As expression levels of endogenous surface antigens across potentially relevant, but yet undefined, subpopulations are unknown, we used lentiviral transduction for unbiased molecular marking and in vivo imaging (Figure 1A).

10PDX ALL cells were lentivirally transduced to express a luciferase for in vivo imaging, an artificial antigen (truncated nerve growth factor receptor [NGFR]) for magneto-activated cell sorting and a red fluorochrome for cell sorting by flow cytometry (Figures S1A and S1B). Transgenes allowed effective and reliable enrichment of minute numbers of PDX cells from mouse bone marrow in this two-step procedure. Quantification of PDX cells isolated with the magnetic-activated cell sorting (MACS)/fluorescence-activated cell sorting approach closely correlated with other methods monitoring leukemic proliferation, such as in vivo imaging and flow cytometry-based quantification of leukemia cells (Figure S1C). Quality controls showed that the procedure was highly efficient and reliable with minor cell loss (Table S2).

11The procedure enabled addressing basic questions with translational potential in ALL biology. Homing capacity of PDX cells to mouse bone marrow differed by more than two orders of magnitude between the nine samples studied (Figure 1B). Homing efficiency decreased significantly when smaller cell numbers were injected (Figure S1D). These data argue in favor of sample-specific characteristics determining homing, and against the presence of a preformed, fixed number of leukemia homing sites within the niche. Spontaneous growth of PDX ALL cells in mouse bone marrow was logarithmic over the first 2 weeks of in vivo growth (Figures 1C and S1C). Growth slowed down thereafter and as early as at 10% blasts in bone marrow, when space restriction appears unlikely to be causative. Model selection indicated overall logistic growth which is typical for insufficient nutrient supply (Figure S1E). Thus, PDX ALL cells show sample-specific homing followed by logistic growth in mouse bone marrow.

12Proliferation-dependent dyes such as bromodeoxyuridine (BrdU) and carboxyfluorescein diacetate succinimidyl ester (CFSE) remain stable in mice over several months, enabling the characterization of a heterogeneous growth pattern in normal hematopoiesis. We adapted the use of these dyes in PDX tumor models. As BrdU staining requires the permeabilization and destroying of cells, fluorescent CFSE was mainly used as it allows flow cytometric enrichment of living cells for functional experiments including re-transplantation. Loss of CFSE was used to distinguish subpopulations of slowly and rapidly growing cells (Figures 1D and S1F) that were called label-retaining cells (LRC) and non-label-retaining cells (non-LRC), respectively. LRC were defined as those cells that had undergone at most three CFSE bisections resembling cell divisions (see the Supplemental Experimental Procedures for details). Loss of CFSE tightly correlated with increase in PDX cell numbers and loss of BrdU (Figures 1E and S1G) and confirmed that PDX ALL cells grow in vivo, but not ex vivo (Figure S1H). Thus, CFSE staining represents a reliable approach to monitor proliferation of PDX ALL cells in mice.

13Importantly, CFSE staining disclosed the existence of a rare fraction of PDX ALL cells that hardly divided over prolonged periods of time (Figure 2A). LRC, by definition, had undergone no more than three cell divisions within 21 days, during which the leukemia burden had risen by several orders of magnitude so that mice would succumb to leukemia within a few days. In all nine PDX ALL samples studied, LRC were identified after prolonged periods of leukemic growth; (Figures 2B and S2A).

14Thus, similarly to normal hematopoiesis, PDX ALL contains a rare subpopulation of LRC. LRC might resemble the dormant tumor cells described in ALL patients (Figure S2B). As an advantage over work with primary cells, our preclinical approach allows repetitive work on pure, vivid LRC, which gave us the chance to functionally and phenotypically characterize this interesting population.

15Both normal hematopoietic stem cells and leukemia stem cells were reported to preferentially localize close to the endosteum, where a supportive niche might exist. We also found that LRCs preferentially localized close to the endosteum (Figures 3A–3C and S3), suggesting that they might use the same niche as normal hematopoietic stem cells and cancer stem cells.

16We therefore asked whether LRC might resemble cancer stem cells. To compare leukemia-initiating potential between LRC and non-LRC, we performed limiting dilution transplantation assays and monitored engraftment by bioluminescence in a total of 83 mice (Table S3). To our surprise, we found highly similar stem cell frequencies in LRC and non-LRC and similar engraftment rates after transplantation of, e.g., ten cells per mouse (Figure 3D). The 95% confidence interval of the estimated frequency of leukemia-inducing cells ranged between 1/19 and 1/84 cells for LRC and between 1/40 and 1/179 cells in non-LRC of ALL-265 (Table S3). Similar findings were obtained for ALL-199 (Table S3). Thus, although only LRC display typical characteristics of stem cells such as reduced proliferation rate and localization close to the endosteum, LRC and non-LRC exhibited similar leukemia-initiating potential.

17Dormant cells are known for their resistance against drug treatment, complicating elimination by anti-cancer therapy. We compared in vivo drug response of LRC and non-LRC by transplanting CFSE-labeled PDX ALL cells, treating mice with systemic chemotherapy on day 7 and analyzing surviving LRC and non-LRC on day 10 (Figure 4A). Chemotherapy reduced the overall leukemic burden by over 90% (Figures 4B and S4A) and eradicated most non-LRC. As a prominent difference, most LRC survived chemotherapy so that LRC increased in relative proportions (Figures 4C–4E and S4B–S4D). A 10- to 100-fold less efficient elimination of LRC compared with non-LRC became obvious across all PDX ALL samples tested that were derived from either primary disease or relapse, suggesting that this phenomenon is not restricted to a certain disease stage. Treatment-surviving LRC harbored leukemia-initiating potential as they gave rise to leukemias upon re-transplantation at a kinetic similar to that of untreated LRC (Figures 4F and S4E).

18Taken together, LRC share the most important functional features that impede the cure of cancer: (1) dormancy, (2) in vivo drug resistance, and (3) leukemia-initiating potential. LRC might thus serve as preclinical surrogate for relapse-inducing cells in ALL.

19We then evaluated whether LRC adequately resemble challenging cells in patients. For a broad, unbiased comparison between LRC and non-LRC, RNA sequencing (RNA-seq) was performed on single cells and bulk populations (Figure 5A). Data from single cells correlated with data from bulk populations and different ALL PDX samples showed similar expression profiles (Figures S5A and S5B). Preliminary expression arrays on pools of 40 LRC and non-LRC showed mainly similar results (data not shown).

20Single LRC differed consistently from single non-LRC as revealed by clustering differently expressed genes (Figures 5B and Table S4) and by a principle component analysis of the most variable genes (Figure 5C). Single LRC also had an overall reduced RNA content (Figure S5C), indicating a less active metabolism that is a prerequisite of dormant cells. We combined single-cell and bulk data of all six sample pairs to identify differently expressed genes (Table S5). Enrichment analysis revealed that genes expressed less in LRC were most strongly enriched in cell cycle and DNA replication and that genes more expressed in LRC were most strongly enriched in cell adhesion (Figures 5D, S5D, and Table S6). Hence, expression profiling of single cells and in bulk confirmed the quiescent state of LRC and an LRC signature of at least 2-fold differently expressed genes ranked by their significance (Figures 5E and Table S5) was used for further comparisons.

21Relapse often results from treatment-resistant tumor cells that survive chemotherapy and persist at MRD. MRD cells contain a major fraction of dormant tumor cells. Here, we hypothesized that LRC might represent surrogates for MRD cells.

22To experimentally test this hypothesis, we established a preclinical model of MRD for ALL-265 and ALL-199. When untreated control samples were harvested at advanced leukemia, they contained a leukemic burden of ∼30% human blasts in mouse bone marrow, mimicking the situation at diagnosis. Remaining mice received a systemic treatment with conventional chemotherapeutic drugs over 2–3 weeks (Figure 6A), which needs careful dosing as supportive therapy is mainly unfeasible in mice. A combination treatment of vincristine and cyclophosphamide reduced tumor burden substantially according to in vivo imaging (Figures 6B, 6C, and S6A). Postmortem analysis revealed that chemotherapy had reduced leukemic burden by more than two orders of magnitude to ∼0.1% leukemia cells in bone marrow. This resembled not only complete morphologic, but also complete molecular remission criteria (Figures 6D and S6B). MRD cells revealed relapse-inducing potential as they re-grew in mice when treatment was stopped (Figure S6C).

23MRD cells were isolated from mouse bone marrow using expressed transgenes as above, and RNA sequencing of single cells and bulk samples was performed. Resulting transcriptomes showed marked differences between MRD and untreated control cells (Figure S6D). Enrichment analysis revealed significantly reduced expression of MYC and E2F target genes in MRD compared with untreated cells. Genes expressed less in MRD cells were most strongly enriched in cell cycle and DNA replication, while genes expressed more in MRD cells were most strongly enriched in cell adhesion (Figures 6E and S6E). This suggests a dormant phenotype of MRD cells similar to the dormant phenotype seen in LRC (Figure 5D). KEGG pathway analysis highlighted that MRD cells were of dormant nature and expressed increased adhesion molecules (Figure S6E). Indeed, single MRD cells clustered together with single LRC in a principal component analysis separated from non-LRC and cells from untreated mice (Figure S6F). Accordingly, the LRC signature (Figure 5E and Table S5) was strongly enriched in MRD cells and genes in MRD and LRC cells were similarly regulated compared with their respective controls (Figure 6F). This suggests that LRC mimic MRD cells in our preclinical mouse model.

24To relate these findings to the clinical situation, expression profiles from primary tumor cells from five children and two adults with B cell precursor (BCP) ALL were profiled at diagnosis and at MRD (Figure 7A and Table S7). Children and adults were treated according to the BFM-2009 and GMALL-0703 protocols, respectively, and MRD cells were enriched by flow cytometry at days 33 and 71 of treatment, respectively. In adults, we chose BCR-ABL-positive ALL and enriched the subpopulation of StemB cells at MRD, as had shown that these cells exhibit a dormant phenotype. As dormancy in StemB cells might have persisted for a long period during treatment in patients, LRC might especially resemble StemB cells at MRD. We could obtain single-cell transcriptomes from one patient and one bulk transcriptome from another patient. K-means clustering and principal component analysis revealed that single StemB cells clustered together with single LRC and MRD cells, while single non-LRC clustered together with single untreated control cells (Figures 7B and 7C). The bulk StemB sample was distinct from diagnostic tumor cells of untreated adult patients with BCR-ABL-positive ALL (Figure S7A). Although limited by small cell and sample numbers, the data indicate that LRC resemble the dormant subpopulation of StemB cells in adult ALL patients at MRD.

25This is also supported when comparing the LRC profiles with further published transcriptomes. Genes differently expressed in CD34-positive chronic myeloid leukemia cells, in leukemia stem cells, in hematopoietic stem cells, as well as in pediatric ALL cells with high risk of relapse were all significantly enriched in LRC versus non-LRC cells (Figures 7D, S7B, and S7C).

26To further analyze the similarity of LRC to MRD cells from patients, we generated bulk transcriptomes of primary samples from five children with BCP-ALL before the onset of treatment and three matched MRD samples collected 33 days after the onset of treatment. Expression profiles differed significantly between diagnosis and MRD (Figure 7E and Table S8) and MRD cells regulated genes in the same direction as LRC compared with their respective controls, as revealed by a significant overlap of up- and downregulated genes (hypergeometric test, p = 1.9 × 10−23) and by a significant enrichment of the LRC signature (p < 0.001; Figure 7F). Finally, we combined these transcriptomes with all bulk samples isolated from the LRC and MRD mouse models and analyzed them unsupervised in a principal component analysis (Figure 7G). The first principal component separated all dormant and drug-resistant cells (PDX-LRC, PDX-MRD, and primary MRD) from all control cells (PDX-non-LRC, PDX untreated, and primary diagnosis).

27In summary, we show that a distinct subpopulation of LRC exists in our ALL PDX model that combines the unfavorable characteristics of stemness, drug resistance, and dormancy. These LRC show high similarities to MRD cells in our mouse model and to MRD cells in ALL patients. Hence, LRC might represent preclinical surrogates for relapse-inducing cells in patients and could be used to develop therapeutic strategies to prevent relapse.

28As the first step toward therapies, we studied whether unfavorable drug resistance and dormancy represented permanent or reversible features in LRC. Dormancy and drug resistance might exist as genuine, constant biological characteristics of a special ALL subpopulation or as reversible functional phenotypes of putatively every ALL cell depending on the context.

29To address this question, LRC and non-LRC were dissociated from their environment, isolated, and re-transplanted into recipient mice (Figures 8A and S8A). When non-LRC were re-stained with CFSE and re-transplanted at high numbers, they gave rise to an identical LRC population as re-transplanted bulk cells (Figures 8B and S8A); transplantation of high cell numbers of LRC was impossible, as only low numbers of LRC can be recovered from mice. When low cell numbers were re-transplanted, LRC, non-LRC, and bulk cells initiated identical leukemic growth in mice as monitored by bioluminescence in vivo imaging (Figures 8C and S8A). These data indicate that dormancy represents a reversible feature of LRC, as LRC lose their dormant nature once they are retrieved from their specific environment and transferred into a different surrounding.

30As dormancy emerged as a reversible phenotype, we asked whether drug resistance might be equally reversible. Isolated LRC and non-LRC or MRD and previously untreated cells from the PDX mouse model were treated ex vivo with common ALL chemotherapy drugs or drug controls. Here, the technical challenge lay in the very minor cell numbers of LRC and MRD that can be isolated from mice and used for ex vivo experiments (Figure S8B). Co-culture with feeder cells resembling bone marrow stroma reduced drug response in all samples, suggesting the influence of the bone marrow environment on drug resistance (Figures S8C–S8F). Ex vivo, neither LRC nor MRD cells displayed increased drug resistance compared with their respective controls (Figures 8D and S8G).

31Taken together, LRC and MRD cells showed a marked gain in drug sensitivity ex vivo compared with in vivo after isolation from the bone marrow environment. Both LRC and MRD cells lost their enhanced drug resistance, distinguishing them from non-LRC or untreated cells, once they were retrieved from their in vivo environment and cultured ex vivo (Figure 8E). Dormancy was reversible in LRC and drug resistance was reversible in both LRC and MRD cells. As LRC might represent surrogates for relapse-inducing cells in patients, our data suggest that the interaction between LRC and their environment represents an attractive therapeutic target for preventing relapse. Relapse-inducing cells might gain sensitivity toward treatment once mobilized from their in vivo environment.

32The present work aimed at a better understanding of the cells that induce relapse in ALL and thereby limit prognosis of patients. We identified a rare, long-term dormant subpopulation termed LRC exhibiting the adverse characteristics of dormancy, in vivo drug resistance, and leukemia-initiating properties. LRC highly resemble primary MRD cells from adult and pediatric patients with ALL. MRD cells require preferential eradication by anti-leukemia treatment. LRC in preclinical models can now be used as surrogates for relapse-inducing cells in patients for developing therapies to prevent relapse. Upon removal from their in vivo environment, LRC lost dormancy and drug resistance, suggesting a reversible nature of adverse characteristics and an important role for the interaction between ALL and the environment. The data suggest that drug resistance and dormancy are linked and represent an acquired stem-like phenotype. Our data imply developing treatment approaches that dissociate ALL cells from their protective niche to sensitize them toward anti-leukemia treatment.

33Here, we provide a preclinical tool to study dormant human ALL cells in vivo and show that long-term resting cells exist in ALL. This fact was previously unknown, as primary patients' samples allow quantifying non-cycling cells in a snapshot at a given moment, but fail to distinguish between short- and long-term resting cells. As monitoring functionally defined cellular subpopulations such as LRC in longitudinal studies is still impossible in patients, our preclinical model enables the gaining of insights into ALL biology that cannot be obtained in patients: here the presence of long-term resting cells in ALL. Beyond its use in preclinical treatment trials, PDX models harbor major potential in basic research and enable unique insights into disease biology.

34The emergence of relapse is a complex process involving genetic and non-genetic factors. Early relapse might be caused by a putatively pre-existing clone with additional mutations responsible for drug resistance, especially in adult patients. The genetic stability of most cases of ALL suggests that many relapses may not be mediated by mutational mechanisms. Late relapse might be caused by persisting, dormant tumor cells in the absence of additional mutations, and relapse cells often respond to the identical drugs used to treat the primary disease. LRC represent surrogates for late relapse and relapse in the absence of additional mutations, as often seen in children.

35The fact that LRC exist might explain why ALL patients benefit from maintenance therapy, even in prognostically favorable, chemo-sensitive ALL subtypes. ALL patients are routinely treated with oral low-dose chemotherapy from end of intensive chemotherapy until, e.g., 2 years after diagnosis, and maintenance therapy improves patients' prognosis. Low-dose maintenance therapy might act by removing LRC-type ALL cells with relapse-inducing potential that remained quiescent over prolonged periods of time and turned on their cell cycle at late time points in the months following intensive chemotherapy.

36Tumor cells often display both dormancy and drug resistance. It is unclear whether either dormancy or drug resistance is pivotal in respect to the other, so that dormancy is a consequence of resistance or vice versa. Our two complementary mouse models show that LRC were defined by their dormant nature and displayed drug resistance, while MRD cells were defined by their ability to survive drug treatment and displayed a dormant phenotype. Thus, both characteristics might be equally sufficient to determine each other and coincide interdependently.

37Our study shows that ALL consists of functionally heterogeneous cells regarding proliferation rate and drug resistance, similar to the functional heterogeneity shown in other tumor entities. As LRC did not substantially participate in proliferation during growth of leukemia over weeks, in our model LRC existed before onset of therapy and were not developed as a consequence of treatment. As both LRC and non-LRC contain similar amounts of stem cells, but show different sensitivity toward drug treatment in vivo, our data imply that stemness and drug resistance are not directly connected in ALL.

38So how does a rare subpopulation acquire the three clinically challenging features dormancy, resistance, and stemness? LRC might represent a cell subpopulation with genuinely different biology harboring distinct intrinsic, constant characteristics, or being an LRC might represent a reversible, temporary, functional phenotype depending on circumstances. In the first case, LRC and non-LRC might be organized in a hierarchical way similar to the known stem cell hierarchy existing in many tumors including AML. In the second case, ALL cells might mimic the phenotypic reversibility of normal hematopoiesis, where long-term dormant hematopoietic stem cells start cycling in response to stress for a defined period of time and turn back into dormancy later.

39Our data favor the second scenario as LRC exhibit their specific characteristics as reversible, temporary, transient functional phenotypes. Re-transplantation experiments showed that formerly dormant LRC started proliferating as soon as they were dissociated from their in vivo environment and transferred into next recipient mice. Upon re-transplantation, LRC converted into non-LRC, while certain non-LRC converted into LRC. Both LRC and non-LRC thus harbored plasticity to switch between slow and rapid proliferation depending on the current context. This fact might explain the area of overlap between LRC and non-LRC detected in single-cell RNA sequencing.

40Besides proliferation, drug resistance also proved to be a transient characteristic. Drug-treatment experiments showed that LRC lost their in vivo drug resistance upon ex vivo culture. The discrepancy between drug sensitivity ex vivo and in vivo might at least partly explain the limited predictability of ex vivo drug-screening tests for the outcome of cancer patients. Thus, localization of LRC to the bone marrow niche influences both dormancy and drug resistance.

41These insights have translational implications. For diagnostics, as LRC lose their clinically relevant characteristics upon release from their niche, rapid sample processing might be critical for reliable profiling, which represents a challenge in clinical routine. Our data at least in part explain the limited power of in vitro assays using, e.g., proliferating cell lines, for studies on MRD cells or primary leukemia cells for drug testing in the absence of feeders. Most importantly for putative treatment strategies, the transient nature of the adverse characteristics of LRC suggests aiming at removing MRD cells from their protective environment to sensitize them toward treatment. The interaction between MRD cells and their bone marrow niche represents a promising target for therapeutic approaches to prevent relapse. Beyond the tumor cell itself, its interaction with the environment represents a suitable therapeutic target. As a caveat, a persistent therapeutic inhibition of the bone marrow niche might be required over prolonged periods of time, as in principle each and every remaining non-LRC ALL cell could convert into a drug-resistant LRC, as soon as it gets access to the protective niche.

42At this point, we can only speculate which signals might determine whether an ALL cell behaves like an LRC or a non-LRC. In theory, external as well as internal factors or conditions might be influential; stimuli might be sent or received either stochastically or within a well-regulated process. As our studies were restricted to bone marrow, the bone marrow niche is a likely candidate for a regulatory function and requires investigatory work. Further research is required to address these important questions. Obvious candidates for therapeutic intervention are cell surface molecules expressed on LRC, the inhibition of which might release cells from their environment. Similarly, niche cells could be targeted to aim at reducing environmental support.

43Our study shows that ALL growing in vivo contains a rare subpopulation of LRC that exhibits typical challenging adverse characteristics of relapse induction, which proved to be of a reversible nature. Our model might help to develop future anti-leukemia treatment strategies allowing the eradication of the precarious subpopulation of drug-resistant stem cells to prevent relapse and improve the prognosis of patients with ALL.

44Written informed consent was obtained from all patients and from parents/carers in the cases where the patients were minors. The study was performed in accordance with the ethical standards of the responsible committee on human experimentation (written approval by Ethikkommission des Klinikums der Ludwig-Maximilians-Universität München, Ethikkommission@med.unimuenchen.de, April 15, 2008, number 068-08) and with the Helsinki Declaration of 1975, as revised in 2000.

45All animal trials were performed in accordance with the current ethical standards of the official committee on animal experimentation under the written approvals by Regierung von Oberbayern, poststelle@reg-ob.bayern.de, May 10, 2007 number 55.2-1-54-2531-2-07 and August 8, 2010 number 55.2-1-54-2531-95-10.

46PDX ALL cells were genetically engineered as described using lentiviruses to express the transgenes’ truncated NGFR, a red fluorochrome, and luciferase; cells were stained with BrdU and/or CFSE before re-transplantation of fresh cells into mice.

47For determining the fraction of dormant PDX ALL cells, mouse bone marrow was harvested from numerous bones and enriched for human PDX ALL cells using NGFR for MACS and the red fluorochrome for flow cytometry cell sorting. LRC were discriminated from non-LRC using CFSE staining as shown in Figure 1D. CFSE mean fluorescence intensity (MFI) was measured at day 3 after injection, when bleaching had ceased, and defined cells before the onset of proliferation (“0 divisions”). Day 3 CFSE MFI was divided by factor 2 to calculate CFSE bisections mimicking cell divisions. Seven CFSE MFI bisections or more were defined as entire loss of the CFSE signal characterizing non-LRC. The LRC gate was set to include all cells harboring high CFSE signal of below three bisections of the maximum CFSE MFI (Figure 1D).

48Single cells were isolated at 4°C and processed on the Fluidigm C1 platform. In brief, 500 cells were loaded on the 10–17 μm mRNA-seq IFC (Fluidigm) with External RNA Controls Consortium spike-in controls. Cell lysis, reverse transcription, and pre-amplification of cDNA was done on-chip using the SMARTer Ultra Low RNA Kit for C1 (Clontech). Harvested cDNA libraries of the samples (2.5 μL) were used as input for tagmentation with the Nextera XT Sample Preparation Kit (Illumina) at half the volume of Illumina's protocol. Barcoding PCR was performed for 12 cycles. Equal amounts of libraries were pooled.

49Single-cell Smart-seq and bulk Smart-seq2 libraries were sequenced at 1 × 50 bases on an Illumina HiSeq1500. SCRB-seq and UMI-seq libraries were sequenced paired-end with 16 cycles on the first read to decode sample barcodes and unique molecular identifiers and 50 cycles on the second read into the cDNA fragment.

50S.E. and E.Z.Ö. planned, performed, and analyzed the experiments and designed the data presentation; C.Z., E.Z.Ö., S.P., and W.E. generated and analyzed RNA-seq data, W.E. participated in writing the manuscript; S.T. and C.C.A. established the mouse model, first detected LRC, and started their characterization; M.G. started establishing the MRD PDX model; A.S. guided the work of E.Z.Ö.; M.D. performed enrichment of pediatric MRD cells provided by R.P.G.; C.L., V.A.T., and T.E. performed enrichment of adult StemB cells provided by H.P.; H.P.H. and K.So. performed immunohistochemistry of primary bone marrow biopsies; K.Sp. and W.H. provided primary adult samples of Figure S2; B.P., S.K., M.H., and B.K. performed and analyzed gene expression array data; J.H. performed the mathematical analysis of Figure S1E; O.G. participated in designing the experiments, guiding the study, and writing the manuscript; I.J. initiated and guided the study and wrote the manuscript.