Evi1 defines leukemia-initiating capacity and tyrosine kinase inhibitor resistance in chronic myeloid leukemia

Relapse of chronic myeloid leukemia (CML) is triggered by stem cells with a reconstituting capacity similar to that of hematopoietic stem cells (HSCs) and CML stem cells are a source of resistance in drug therapy with tyrosine kinase inhibitors (TKIs). Ecotropic viral integration site 1 (EVI1), a key transcription factor in HSC regulation, is known to predict poor outcomes in myeloid malignancies, however, incapability of prospective isolation of EVI1-high leukemic cells precludes the functional evaluation of intraindividual EVI1-high cells. Introduction of CML into Evi1-internal ribosomal entry site (IRES)-green fluorescent protein (GFP) knock-in mice, a versatile HSC-reporter strain, enables us to separate Evi1-high CML cells from the individual. Evi1-IRES-GFP allele models of CML in chronic phase (CML-CP), by retroviral overexpression of BCR–ABL and by crossing BCR–ABL transgenic mice, revealed that Evi1 is predominantly enriched in the stem cell fraction and associated with an enhanced proliferative as well as a leukemia-initiating capacity and that Evi1-high CML-CP cells exhibit resistance to TKIs. Overexpressing BCR–ABL and NUP98–HOXA9 in Evi1-IRES-GFP knock-in mice to model CML in blast crisis (CML-BC), in which Evi1-high cells turned to be a major population as opposed to a minor population in CML-CP models, showed that Evi1-high CML-BC cells have a greater potential to recapitulate the disease and appear resistant to TKIs. Furthermore, given that Evi1 heterozygosity ameliorates CML-CP and CML-BC development and that the combination of Evi1 and BCR–ABL causes acute myeloid leukemia resembling CML-BC, Evi1 could regulate CML development as a potent driver. In addition, in human CML-CP cases, we show that EVI1 is highly expressed in stem cell-enriched CD34+CD38–CD90+ fraction at single-cell level. This is the first report to clarify directly that Evi1-high leukemic cells themselves possess the superior potential to Evi1-low cells in oncogenic self-renewal, which highlights the role of Evi1 as a valuable and a functional marker of CML stem cells.


INTRODUCTION
As multipotent hematopoietic stem cells (HSCs) reside at the apex of hematopoietic hierarchy, leukemic progeny from leukemic stem cells (LSCs) shape the bulk of the tumor with intact capacity of LSCs to self-renew. Physiological and biological similarities of LSCs to those of HSCs are supposed to be the main causes of the difficulties in establishment of LSC-targeted therapy. 1,2 Chronic myeloid leukemia (CML) is a myeloproliferative disorder by BCR-ABL, which can transform HSCs into LSCs with a limitless capacity for self-renewal, whereas LSCs of acute myeloid leukemia (AML) are mainly composed of more differentiated progenitor cells. [3][4][5] The relentless march of CML from chronic phase (CP) to blast crisis (BC) phase can result in fatal survival outcomes. 6 Despite substantial prognostic improvement of CML by a specific debulking of tumor burden with a tyrosine kinase inhibitor (TKI) targeting ABL kinase, imatinib-treated CP patients can relapse and progress to BC because of the remnant CML stem cells. [7][8][9][10] Although recent findings have started to unveil the biological nature of CML stem cells, [11][12][13][14][15][16] further elucidation of the mechanisms controlling the self-renewal of these cells is still needed.
Defects in the molecular components that control hematopoiesis severely perturb normal development, one of which is ecotropic viral integration site 1 (Evi1), a predictor of poor outcomes in myeloid malignancies such as AML, myelodysplastic syndrome and CML-BC. [17][18][19][20][21] In normal hematopoiesis, Evi1 is restricted to embryonic and adult HSCs 22 and cumulative data have placed Evi1 as one of 'stemness' genes. [23][24][25] The recent gene expression profiling of bulk samples or selective populations have shown that EVI1 is one of LSC signature genes in AML and that stem cell-enriched CML CD34+ cells have high EVI1, underlining the relevance of EVI1 and LSCs. 26,27 As EVI1 is an oncogenic transcription factor, [28][29][30][31][32][33][34] prospective isolation of EVI1-high leukemic cells from clinical patients is unfeasible, so is the functional assessment of EVI1-high cells compared with EVI1-low cells intraindividually.
In this study, we aimed to cover in depth the regulation of CML stem cells by Evi1. Our single-cell analysis of primitive or differentiated subsets from primary CML-CP samples show that EVI1 is highly expressed in stem cell-enriched CD34+CD38-CD90+ cells. Furthermore, we have established multiple CML mice model with Evi1-Internal Ribosomal Entry Site (IRES)-green fluorescent protein (GFP) knock-in allele, in which Evi1-high CML cells can be separated directly and prospectively using a single GFP 35 to evaluate their capacity for leukemia development.
Evi1-IRES-GFP allele leukemia animals provide us for the first time with a more definite hierarchical map in CML hematopoiesis and, through loss-and gain-of-function studies we could evaluate the functional role of Evi1 in CML. This study could also determine whether Evi1-high CML cells could have resistance to TKI therapy.
Evi1-high CML-CP LSK cells have a higher proliferative potential As LSK cells showed a heterogeneous pattern of Evi1 in BM of Evi1-reporter CML-CP mice, we sorted the Evi1-high or Evi1-low fractions from KuOr+ LSK cells (Figure 2a). Evi1-high LSK cells showed a higher colony-forming potential in vitro, the sizes of which were larger than those from Evi1-low cells (Figures 2b and c).
Co-culture of LSK cells with OP-9 cells revealed a higher clonogenic potential of Evi1-high LSK cells (Supplementary Figure S2e). Transcripts of BCR-ABL were comparable between Evi1-high and Evi1-low LSK populations (Figure 2d), suggesting little relevance of BCR-ABL to the enhanced proliferation of Evi1-high CML cells. These results suggest that Evi1-high CML-CP LSK cells have the proliferative advantage.
Evi1-high CML LSK cells have SLAM LSK marker profile with CML-initiating potential and TKI resistance Generating primary CML-CP mice by BCR-ABL retrovirus with irradiated recipients inevitably leads to damaged BM microenvironment and likely overestimation of extramedullary hematopoiesis. To clarify the precise behavior of Evi1-high CML LSK cells in vivo, we crossed Evi1 +/GFP mice with p210 BCR-ABL transgenic mice (BCR-ABL tg/− ), 36 Figure S3d). Strikingly, analysis of CD150/signaling lymphocyte activation molecule (SLAM) markers in LSK cells of Evi1 +/GFP BCR-ABL tg/− mice clearly showed that Evi1high LSK cells had a higher percentage of CD150+CD48-cells, a well-established marker phenotype of long-term stem cells (Figures 2f and g). Evi1-high CML LSK cells were the most in G0/ G1 phase (quiescent) in CML setting, similar to the trend of subpopulations from normal Evi1 +/GFP mice (Figure 2h). Colonyforming assay showed the advantageous phenotype of Evi1-high CML LSK cells to proliferate in vitro (Supplementary Figure S3e). From these data, Evi1-high Evi1 +/GFP BCR-ABL tg/− LSK cells had the most immature immunophenotype.
To evaluate the stem cell activity of Evi1-high CML LSK cells in vivo, we next performed BM transplantation (BMT) experiments, in which 5 × 10 3 Evi1-high or Evi1-low LSK cells were injected intravenously into sublethally irradiated recipients (Figure 3a). FCM analyses of peripheral blood (PB) in recipients revealed the efficient engraftment of donor cells from Evi1-high LSK-transplanted cohort as contrasted with recipients of Evi1-low LSK cells (Figure 3b). The successive engraftment and multilineage reconstitution up to 16 weeks after BMT clearly indicated the higher repopulating capacity of Evi1-high LSK cells (Figure 3c, Supplementary Figure S3f). With such a prolonged engraftment, recipients with Evi1-high LSK cells died of CML (4 deaths per 6 mice), while those with Evi1-low LSK cells had no incidence of CML (0 death per 6 mice), revealing high CML-initiating potential of Evi1-high CML LSK cells (Figure 3d).
Drug resistance is one of critical characteristics of LSCs, which motivated us to conduct in vivo therapeutical interventions to Evi1 +/GFP BCR-ABL tg/− mice. Oral administration of nilotinib, a potent BCR-ABL inhibitor, to Evi1 +/GFP BCR-ABL tg/− mice for 7 days resulted in amelioration of BM cellularity and splenomegaly, which underlay the efficacy of the drug (Supplementary Figure S3c). In LSK fraction, although nilotinib-treated Evi1 +/GFP BCR-ABL tg/− mice showed the reduced number of Evi1-low LSK cells in BM compared with vehicle-treated Evi1 +/GFP BCR-ABL tg/− mice, that of Evi1-high LSK cells had no change irrespective of nilotinib treatment, which implied nilotinib resistance of Evi1-high CML cells (Figure 3e). The relatively nilotinib-resistant aspect of these cells was confirmed by in vitro colony-forming assay (Supplementary Figure S3g). FCM analysis of the residual cells in BM and spleen of nilotinib-treated Evi1 +/GFP BCR-ABL tg/− mice revealed a marked increase in the percentage of CD150+ CD48-fraction of Evi1-high LSK cells (Supplementary Figure S3h).
The treatment had no impact on normal Evi1 +/GFP mice as to Evi1-positive rate (data not shown). These data of Evi1 +/GFP BCR-ABL tg/− mice revealed CML stem cell activity and nilotinib resistance of Evi1-high cells.
Evi1 heterozygosity impairs CML development Based on these findings, we next crossed Evi1 heterozygous knock-out (Evi1 +/− ) mice with BCR-ABL tg/− mice to clarify whether  loss of Evi1 would affect CML development ( Figure 3f). Surprisingly, 1-year follow-up revealed a significantly prolonged survival of Evi1 +/− BCR-ABL tg/− mice (4 deaths per 7 cases), with an obvious contrast to the high lethality of Evi1 +/+ BCR-ABL tg/− mice (12 deaths per 14 cases; Figure 3g, P = 0.0106). All dead mice in both cohorts had granulocytosis and mild splenoegaly (data not shown). Given that Evi1 +/− mice are fertile with no sign of BM failure over a year, 25 these data suggest that Evi1 has a distinctive role in CML development.
BCR-ABL and NUP98-HOXA9 induce myeloid BC of CML with Evi1 upregulation The BC phase of CML is characterized by high mortality and resistance to both TKI therapy and conventional chemotherapy, with HSC transplantation being the only therapy providing appreciable efficacy. Among CML-BC cases with high EVI1, there exist not only 3q-rearranged cases but also cases without 3q rearrangement, possibly suggesting unknown EVI1 regulation. To evaluate a leukemogenic function of Evi1-high cells in CML-BC, BCR-ABL and NUP98-HOXA9 were co-transduced 37 into 5FU-primed BM of Evi1 +/GFP mice to establish Evi1-reporter CML-BC mice (Figure 4a). Evi1-reporter CML-BC mice had an obviously high level of Evi1-positive cells in the BM compared with Evi1-reporter CML-CP mice (Figures 4b and e). Evi1-reporter CML-BC mice died at 14 days after BMT and showed emergence of blast cells in PB as well as in BM and spleen, a AML-like phenotype, which was markedly different from that of CML-CP model.

Evi1-high CML-BC cells are enriched in the progenitor fraction
The morphological analyses of BM in Evi1-reporter CML-BC mice showed that Evi1-high cells mostly comprises leukemic blasts, while more than half of Evi1-low cells were differentiated cells (Figures 4c and d). From FCM analyses, most of Evi1-high CML-BC cells were negative for Gr-1, as opposed to Gr-1+ immunophenotype of Evi1-low cells. In Lin-cells of these mice, Evi1-high cells had a higher percentage of c-kit (Figure 4b). Totally, Evi1-high cells were rich in Lin-c-kit+ (LK) fraction of myeloid progenitor immunophenotype, the level of which was apparently higher than that of CML-CP (Figure 4e). Evi1-high LK cells had higher colony-forming capacity, indicating their proliferative advantage or      Table S4). Under in vitro nilotinib treatment, Evi1-high LK cells showed a relatively higher proliferative capacity than Evi1-low cells (Supplementary Figure S4c). Seven-day in vivo nilotinib treatment resulted in the significant improvement of hepatosplenomegaly and only Evi1low LK cells showed a significant decrease in number, indicating that Evi1-high LK cells are more resistant to TKI than the Evi1-low counterpart (Figure 5b,Supplementary Figure S4d). Even residual Evi1-high LK cells in BM after nilotinib still possessed leukemogenic potential (Supplementary Figure S4e). All of CML-BC mice treated with nilotinib relapsed with the expansion of Evi1-high cells after the drug cessation, reflecting a limited role of TKI for CML-BC (data not shown).  We next tested whether loss of Evi1 would affect the development of CML-BC. 5FU-primed Evi1 +/+ or Evi1 +/− BM cells transduced with BCR-ABL and NUP98-HOXA9 were injected into lethally irradiated recipients (Figure 5c). All of the primary recipients developed CML-BC regardless of Evi1 dosage, with a slight delayed onset in Evi1 +/− cohort (Figure 5d). In the secondary recipients receiving 5 × 10 4 CML-BC BM cells, however, Evi1 +/− CML-BC cells showed a significantly attenuated leukemogenic potential, whereas Evi1 +/+ CML-BC retained the full capacity (Figure 5e). These results indicate that the development of CML-BC is dependent on Evi1 dosage.
Combination of BCR-ABL and Evi1 induces AML Having established that Evi1 was upregulated in vivo by BCR-ABL and NUP98-HOXA9, we analyzed a role for Evi1 as a direct driver in CML-BC. Co-transduction of BCR-ABL and Evi1 caused lethality in transplanted mice (median survival: 24 days after BMT, Figures 6a and f). These recipients showed marked predominance of leukemic blasts in BM, spleen and PB, with a neutrophilic component, hypercellular BM and hepatosplenomegaly (Figures 6b-d). Splenomegaly was more obvious in the recipients of BCR-ABL plus Evi1 than those of either BCR-ABL alone or BCR-ABL plus NUP98-HOXA9, whereas liver sizes were comparable (Figure 6e). The disease was transplantable with the shorter latency and BM cells of the recipients with BCR-ABL and Evi1 were skewed to the myeloid lineage (Gr-1+) and Lin-c-kit+ cells were dominant in the primitive fractions (Figures 6f and g). Based on the criteria of Bethesda proposals, 38 the disease was finally diagnosed as 'Myeloid leukemia with maturation'. Thus, even without NUP98--HOXA9, activation of Evi1 can induce AML, resembling myeloid BC in CML, in collaboration with BCR-ABL in mice. Given that the duration of CML-CP development by BCR-ABL retrovirus is about 3-4 weeks, and Evi1 alone could induce myelodysplastic syndrome/AML in mice after 6 months or more, 39,40 these results indicate that Evi1 can have a causative role in blastic transformation of CML.

DISCUSSION
Discovery strategies aimed to identify novel driver oncogenic lesions have succeeded in enrichment of the catalog of therapeutic targets, the most striking of which is BCR-ABL where TKI therapy has conferred tremendous benefits to CML patients with a sustained debulking of tumor burden. Even in the era of TKI treatment in CML, blastic transformation can occur with the translocation involving EVI1 locus, 41,42 which is crucial because the clinical outcomes of CML plummet from CP (80%) to BC (20%). 43,44 Allogeneic HSC transplantation is so far the only potential remedy for high EVI1 cases of CML-BC as well as AML, emphasizing the need for new therapy targeting EVI1. 18,21,45,46 Our hypothesis stems from the fact that EVI1 is a valuable prognostic factor of myeloid malignancies as well as a critical regulator of HSCs. To overcome the incapability of prospective separation of intraindividual EVI1-high cells in human leukemias, we used our original Evi1 +/GFP mice, developing two types of CML-CP model to show that Evi1 is a valuable marker for CML stem cells. Both in retrovirally developed CML-CP mice and BCR-ABL tg/− mice, Evi1 is restricted to a small population with primitive immunophenotypic markers, especially in the latter, the most primitive profile like SLAM-LSK represents Evi1-high cells. When seen against previous studies, 47-49 the novelty of this study lies in that high Evi1 could distinguish CML-CP stem cells even from SLAM-LSK cells or non-SLAM-LSK cells. This point can be translated to human CML cases like that high EVI1 means the increasing number of CML stem cells. Evi1-high CML LSK cells have a superior proliferative potential in vitro, a superior leukemia-initiating capacity in vivo and nilotinib resistance. The resistant aspect of Evi1-high cells to nilotinib would fit into clinical data that high EVI1 is related to TKI resistance. 17 The in vivo quiescent status of Evi1-high CML-CP cells, which could proliferate aggressively in vitro, may be controlled by hypoxic BM niche microenvironment. 50,51 In line with less dependence of CML stem cells on BCR-ABL, 2,52,53 Evi1-high CML-CP LSK cells showed a comparable BCR-ABL to their Evi1-low counterpart, reflecting little addiction to BCR-ABL (Figure 2d).
In accordance with Evi1-reporter CML-CP model, our single-cell analysis of clinical CML-CP cases revealed the highest EVI1 in stem cell-enriched CD34+CD38-CD90+ cells, reflecting CML-CP as a stem cell disease (Figure 1a). Using BCR-ABL-specific primers, we could effectively distinguish CML-CP cells (BCR-ABL+) from remnant normal cells (BCR-ABL-). These human data could reinforce the value of murine CML data.
We also examined retroviral Evi1-reporter CML-BC mice and showed that a sizable fraction of LK cells have distinct Evi1 expression in sharp contrast to the CML-CP model. CML-BC stem cells are exclusively enriched in Evi1-high LK cells with resistance to nilotinib. The impact of combining the CML-BC with Evi1 upregulation is in strong contrast to that of the simple BCR-ABL BMT model, which only leads to CML-CP in mice 3-4 weeks after inoculation (Figure 4h). Taken together, this is the first report that visualizes Evi1 upregulation in in vivo leukemia models. The finding is consistent with the previous report that showed the transcriptional regulation of Evi1 by NUP98-HOXA9 in vitro. 54 In our CML-BC model, HOXA9 showed no difference between Evi1high and Evi1-low LK cells, eliminating the possible dependence of Evi1-high cells on NUP98-HOXA9 (Supplementary Figure S4b).
In our Evi1 knock-out studies of CML-CP and BC mice, Evi1 heterozygosity alleviated the disease (Figures 3g and 5e), highlighting a functional role of Evi1 in CML pathogenesis. Although complete loss of Evi1 causes embryonic lethality in mice, 25,55 Evi1 +/− mice have no sign of BM failure over a year and are fertile with a decreased size and function of HSCs. 25,35 EVI1 has been reported to be relevant to BCR-ABL tyrosine kinase activity. 27 Collectively, it is supposed that Evi1 reduction may permit the reversal of CML at the partial expense of HSCs.
We extended this study to establish a new model of AML by BCR-ABL and Evi1. It is meaningful that overexpression of Evi1 itself is the driver in blastic transformation of CML not only by Evi1-related fusions. Cuenco et al. 56 have previously reported that EVI1 with BCR-ABL induces not leukemias but a fatal myeloproliferative disorder in mice. In contrast to the usage of human EVI1 complementary DNA (in MSCV retroviral vector) in murine BMT model of Cuenco's study, we utilized murine Evi1 for Evi1 overexpression. Murine Evi1 especially in pMYs vector is suitable for establishing myeloid leukemia in mouse BMT model (Jones et al. 42 and unpublished), which resulted in a new AML model by Evi1 plus BCR-ABL. These findings can remind us of the importance of controlling Evi1 to impair CML-BC development by BCR-ABL and NUP98-HOXA9. As authentic Evi1 targets, 22,33,34 such as Gata2, Pten and Pbx1, showed no difference in expression levels between Evi1-high and Evi1-low CML cells (data not shown), the further exploration of CML-specific Evi1 targets would be warranted.
As opposed to Evi1-reporter CML models, in AML model by MLL-ENL retrovirus, Evi1-high MLL-ENL leukemic cells showed no advantage in leukemia initiation compared with Evi1-low cells (Supplementary Figure S5). Other Evi1-reporter AML models by MOZ-TIF2 and TP+AE never generated Evi1-high fraction, suggesting the high affinity of Evi1 for stem cell disease such as CML. Although Evi1 could not enrich MLL-ENL AML LSCs in our model, it is possible that Evi1 reduction in the bulk leukemia cells would be a key in amelioration of MLL-related leukemia as previously investigated. 25 Exact introduction of these oncogenes to HSCs by different approaches such as via a transgene or a knocking-in technology could unravel the relation between AML stem cells and Evi1.
In conclusion, high Evi1 can define the population of CML stem cells that are resistant to nilotinib. This is the first report that uncovers the importance of leukemia cells with high Evi1 intraindividually. Combinatorial analyses of Evi1-IRES-GFP allele CML animals and single cells from primary CML-CP patients covered in depth the critical regulation of CML stem cells by Evi1. To elucidate the complex cellular circuits involving Evi1, a further genetic and epigenetic investigation of Evi1-high cells would unveil precise mechanisms of the phenotype in leukemias. Single-cell gene expression analysis   Primary BM samples of CML-CP patients were stained with anti-CD34, anti-CD38, anti-CD33 and anti-CD90 (Supplementary Table S2), and single-cell sorted by FACSAriaII (BD Biosciences, San Jose, CA, USA) gating on each population excluding cell duplets directly into individual wells of 96-well plates filled with 5 μl RT/Specific Target Amplification master mix solution (Fluidigm, San Francisco, CA, USA) immediately followed by reverse transcription and gene-specific pre-amplification with CellsDirect One-Step qRT-PCR kit (Life Technologies, Carlsbad, CA, USA) in the same plates. Each complementary DNA was subjected to quantitative real-time PCR by LightCycler 480 (Roche, Basel, Switzerland) using SYBR Green (TAKARA, Otsu, Japan) with the primers inside of the pre-amplified complementary DNA. Gene-specific primers were listed in Supplementary Table S3. Only samples with a specific product peak in the melting curve analysis of housekeeping genes were taken for further analysis. This study was approved by ethical committee of the University of Tokyo.

MATERIALS AND METHODS
Population quantitative real-time-PCR Target gene expression for bulk samples or sorted populations was evaluated by LightCycler 480 as described previously. 35 All assays were performed in triplicate and expression levels relative to 18s ribosomal RNA were determined. Primer sequences are listed in Supplementary Table S3. Mice Evi1-IRES-GFP knock-in (Evi1 +/GFP ) mice, Evi1 knock-out (Evi1 +/− ) mice and p210 BCR-ABL transgenic (BCR-ABL tg/− ) mice on a C57BL/6 (Ly5.2) background were used and genotyped as previously described. 25,35,57 CML development of BCR-ABL tg/− mice was confirmed by leukocyte elevation (>10 000 cells/μl) and >80% increase of Gr-1+ cells in PB. Ly5.1 mice were purchased from SRL Inc. (Tokyo, Japan). All mice were kept at the Animal Center for Biomedical Research of the University of Tokyo.

Flow cytometry
FACSAriaII was used for cell sorting of stained cells and LSRII (BD Biosciences) was used for other analyses. Data were analyzed with FlowJo (TreeStar, Ashland, OR, USA). In experiments with the Evi1-IRES-GFP knockin mouse, a 'fluorescence minus one' littermate control was analyzed in parallel to set GFP gates. 35 Antibodies are listed in Supplementary Table S2.

Nilotinib treatment
For in vivo treatment, nilotinib (AMN107-AA) was diluted at 10 mg/ml in 10% 1-methyl-2-pyrrolidone (Sigma-Aldrich, St Louis, MO, USA), 90% polyethylenglycol 300 (Sigma-Aldrich) and administered by oral gavage at 75 mg/kg once a day for a week. A mixture without nilotinib was used as vehicle. For in vitro treatment, nilotinib diluted in dimethylsulphoxide (Sigma-Aldrich) was used at the final concentration of 1 or 5 μmol/l.

Colony-forming assay
Cells (1 × 10 3 ) were plated into MethoCult GF M3434 (StemCell Technologies, Vancouver, BC, Canada) as described previously. 59 The number of colonies was counted at day 7. Images were taken with a Nikon Eclipse TE2000-U (Nikon, Tokyo, Japan).

Statistical analysis
Statistical significance of differences between parameters was assessed by a two-tailed unpaired t-test. The overall survival of mice was analyzed with a Mantel-Cox test according to the Kaplan-Meier method.

CONFLICT OF INTEREST
MK is a member of the advisory committees of Bristol-Myers and Novartis, and has also received research funding and honoraria from the two companies. The remaining authors declare no conflict of interest.