Treatment-induced arteriolar revascularization and miR-126 enhancement in bone marrow niche protect leukemic stem cells in AML

Background During acute myeloid leukemia (AML) growth, the bone marrow (BM) niche acquires significant vascular changes that can be offset by therapeutic blast cytoreduction. The molecular mechanisms of this vascular plasticity remain to be fully elucidated. Herein, we report on the changes that occur in the vascular compartment of the FLT3-ITD+ AML BM niche pre and post treatment and their impact on leukemic stem cells (LSCs). Methods BM vasculature was evaluated in FLT3-ITD+ AML models (MllPTD/WT/Flt3ITD/ITD mouse and patient-derived xenograft) by 3D confocal imaging of long bones, calvarium vascular permeability assays, and flow cytometry analysis. Cytokine levels were measured by Luminex assay and miR-126 levels evaluated by Q-RT-PCR and miRNA staining. Wild-type (wt) and MllPTD/WT/Flt3ITD/ITD mice with endothelial cell (EC) miR-126 knockout or overexpression served as controls. The impact of treatment-induced BM vascular changes on LSC activity was evaluated by secondary transplantation of BM cells after administration of tyrosine kinase inhibitors (TKIs) to MllPTD/WT/Flt3ITD/ITD mice with/without either EC miR-126 KO or co-treatment with tumor necrosis factor alpha (TNFα) or anti-miR-126 miRisten. Results In the normal BM niche, CD31+Sca-1high ECs lining arterioles have miR-126 levels higher than CD31+Sca-1low ECs lining sinusoids. We noted that during FLT3-ITD+ AML growth, the BM niche lost arterioles and gained sinusoids. These changes were mediated by TNFα, a cytokine produced by AML blasts, which induced EC miR-126 downregulation and caused depletion of CD31+Sca-1high ECs and gain in CD31+Sca-1low ECs. Loss of miR-126high ECs led to a decreased EC miR-126 supply to LSCs, which then entered the cell cycle and promoted leukemia growth. Accordingly, antileukemic treatment with TKI decreased the BM blast-produced TNFα and increased miR-126high ECs and the EC miR-126 supply to LSCs. High miR-126 levels safeguarded LSCs, as shown by more severe disease in secondary transplanted mice. Conversely, EC miR-126 deprivation via genetic or pharmacological EC miR-126 knock-down prevented treatment-induced BM miR-126high EC expansion and in turn LSC protection. Conclusions Treatment-induced CD31+Sca-1high EC re-vascularization of the leukemic BM niche may represent a LSC extrinsic mechanism of treatment resistance that can be overcome with therapeutic EC miR-126 deprivation. Graphic abstract Supplementary Information The online version contains supplementary material available at 10.1186/s13045-021-01133-y.


Background
Acute myeloid leukemia (AML) is a hematopoietic malignancy characterized by somatic mutations occurring in the hematopoietic stem cells (HSCs) and progenitor cells that block hematopoietic differentiation and promote accumulation of leukemic "blasts" in the bone marrow (BM) and/or other extramedullary organs [1]. To date, despite a deep molecular understanding of the pathogenesis, the development of molecular targeting therapeutics and the broadened use of allogeneic HSC transplantation, the overall outcome of AML patients remains poor. Disease refractoriness to initial therapy or post-remission disease relapse [2] are likely due to persistence of treatment-resistant leukemic stem cells (LSCs) [3]. These are primitive leukemic cells capable of unlimited self-renewal and disease initiation [4,5] and reside in a leukemic BM niche that also comprises several types of non-hematopoietic cells and that preferentially supports homeostasis and competitive growth of LSCs over those of HSCs [6,7].
Mechanisms of treatment resistance in cancer are multifaceted and often result from the acquisition of genetic mutations that enable malignant cells to escape the therapeutic pressure. Recently, other non-genetic mechanisms of treatment resistance have been also described [8]. While these reported mechanisms have been mainly reported as intrinsic to malignant cells, it is possible that they also include mechanisms that are extrinsic to malignant cells, such as those involving the microenvironment and that protect malignant cells during treatment exposure [7].
Utilizing the FMS-like tyrosine kinase 3 (FLT3) gene internal tandem duplication (ITD) (FLT3-ITD) knockin mouse and FLT3-ITD+ AML patient-derived xenograft (PDX) models that recapitulate features of human FLT3-ITD+ AML, we report here on previously unrecognized non-genetic, extrinsic mechanisms of treatment resistance in LSCs that involve the vascular compartment of the leukemic BM niche and that are mediated by a TNFα-miR-126 axis in the BM endothelial cells (ECs). FLT3-ITD occurs in approximately 25% of AML patients and the mutated gene encodes a mutant receptor with aberrant, ligand-independent tyrosine kinase (TK) activity that confers growth and survival advantages to leukemic blasts [9]. FLT3-ITD+ AML patients are treated with TK inhibitors (TKIs) in combination with chemotherapy. Although the addition of TKIs to chemotherapy confers a clinical advantage compared to chemotherapy alone, it is not curative in the majority of cases, suggesting treatment resistance arising over time [10].

Methods
An extended description of the methods is in the Additional file 1.

Human samples
Normal peripheral blood (PB) and BM samples were obtained from healthy donors at the City of Hope National Medical Center (COHNMC). AML samples were obtained from patients from the COHNMC (Additional file 1: Table S1). Mononuclear cells (MNCs) were isolated using Ficoll separation. When necessary, CD34 + cells were then isolated using a positive magnetic bead selection protocol (Miltenyi Biotech, Germany). Sample acquisition was approved by the Institutional Review Boards at the COHNMC, in accordance with an assurance filed with and approved by the Department of Health and Human Services and met all requirements of the Declaration of Helsinki. Healthy donors and AML patients were consented on the IRB # 06229 and IRB# 18067 protocols, respectively.

Immunofluorescent staining and 3D confocal imaging of long bones
Long bones (femurs and/or tibias) from the mice were processed, sectioned and imaged as described previously [16] with ad-hoc modifications (see Additional file 1 for details).

Intravital imaging
Intravital confocal microscopy was used to image the calvarium BM vasculature to study the vascular permeability, as previously described [17] (see Additional file 1 for details).

Statistical analysis
Comparison between groups was performed by twotailed, paired or unpaired Student's t-test, adjusting for multiple comparisons as appropriate. The log-rank test was used to assess significant differences between survival curves. All statistical analyses were performed using Prism version 8.0 software (GraphPad Software). Sample sizes chosen are indicated in the individual figure legends and were not based on formal power calculations to detect prespecified effect sizes but were based on previous experience with similar models. All of the in vitro experiments were performed 3-6 times using biologically independent samples; the in vivo experiments were performed using 6-16 mice in each group. p values < 0.05 were considered significant. Results shown represent mean ± SEM. * p ≤ 0.05, * * p < 0.01, * * * p < 0.001, * * * * p < 0.0001.
To study changes of the vascular compartment of the BM niche during leukemia growth, we then utilized the Mll PTD/WT /Flt3 ITD/ITD mouse, a model that recapitulates phenotypic, cytogenetic, molecular and pathological features of human FLT3-ITD+ AML, a relatively frequent molecular subset of the disease [11]. Of note, we will refer hereafter to "wt" or "Mll PTD/WT /Flt3 ITD/ITD " to indicate the mouse genotype and to "normal" or "leukemic" to indicate the disease status.
Firstly, we noticed a significant decrease in CD31 + Sca-1 high EC lined vessels (i.e., arterioles) in the leukemic Mll PTD/WT /Flt3 ITD/ITD mice compared with normal wt mice ( Fig. 1a; Additional file 1: Fig. S2a). These results were corroborated by a flow cytometry analysis showing an overall increase in total BM ECs (CD45 − Ter119 − CD31 + ) in the leukemic mice (Additional file 1: Fig. S2b, c) but with a lower frequency of CD31 + Sca-1 high ECs and a higher frequency of CD31 + Sca-1 low ECs compared with normal wt mice ( Fig. 1b; Additional file 1: Fig. S2d). Similar results were also obtained when normal wt mice were engrafted with BM MNCs from congenic leukemic Mll PTD/WT /Flt3 ITD/ ITD donors and compared with controls engrafted with BM MNCs from congenic normal wt donors (Fig. 1c, d; Additional file 1: Fig. S2e).
Next, to assess the relevance of these changes to the human disease, we transplanted human primary FLT3-ITD+ AML blasts into NSG-SGM3 (NSGS) mice and generated a patient-derived xenograft (PDX) model. Similar to the murine AML models, the FLT3-ITD+ PDX showed fewer BM CD31 + Sca1 high ECs and arterioles as compared with NSGS mice engrafted with normal CB CD34 + cells (Fig. 1g, h; Additional file 1: Fig. S3).
Thus, using different AML models and imaging techniques, flow cytometric analyses and permeability studies of the BM niche, we showed that FLT3-ITD+ AML growth led to a decrease of CD31 + Sca-1 high vessels (i.e., arterioles) in the BM niche.
Taken altogether, these results support a TNFαinduced depletion of CD31 + Sca-1 high vessels in the leukemic BM niche during FLT3-ITD+ AML growth.
Taken altogether, these results support a role of miR-126 in determining enrichment of the Sca-1 high EC fraction and in turn arteriolar density in the BM niche and a role of TNFα-dependent miR-126 downregulation in the loss of arterioles during leukemia growth.
While the molecular mechanisms through which TNFα induces endothelial miR-126 downregulation are likely to be multifaceted, we focused on GATA2 since this protein is a verified miR-126 transcription factor [34][35][36] and is reportedly downregulated by TNFα [34,37]. We first noticed that Gata2 levels were reduced in BM ECs from AML mice compared with normal mice (Additional file 1: Fig. S11a) and that GATA2 KD by siRNA in human umbilical vein ECs (HUVECs) decreased miR-126 levels (Additional file 1: Fig. S11b, c). We then demonstrated that in vitro, TNFα treatment (1 ng/ml) reduced Gata2 levels in both murine BM ECs and HUVECs (Additional file 1: Fig. S11d-f ) as compared with vehicle-treated controls. Using chromatin immunoprecipitation assay, we showed a reduced GATA2 enrichment on the EGFL7/ miR-126 promoter [34] in HUVECs exposed to TNFα (Additional file 1: Fig. S11g, h).
Taken altogether, these results support the notion that loss of CD31 + Sca-1 high ECs and associated vessels (i.e., arterioles) observed during AML growth is at least partly mediated by EC miR-126 downregulation via TNFαdependent decrease of GATA2 transcriptional activity.

Antileukemic treatment restores CD31 + Sca-1 high vessels that safeguard LSCs
Having demonstrated that TNFα secreted by the AML blasts contributes to loss of CD31 + Sca-1 high ECs, we then reasoned that cytoreductive therapy that eliminates TNFα-secreting blasts could reverse the depletion of CD31 + Sca-1 high EC-lined vessels (i.e., arterioles) as observed in the leukemic BM niche. To this end, we transplanted BM MNCs from Mll PTD/wt /Flt3 ITD/ITD AML mice (CD45.2) into congenic B6 mice (CD45.1) and generated a cohort of AML mice that developed disease at a similar time. We elected to treat these mice with TKIs rather than chemotherapy to restrict the observed BM niche changes to a direct blast cytoreduction rather than to non-specific chemotherapy cytotoxicity to other nonhematopoietic cells.
Upon confirmation of AML development at 4 weeks after transplantation, we treated these mice with the TKI AC220 [38] to target FLT3-ITD and cytoreduce AML blasts (20 mg/kg/day, oral gavage, Fig. 5a). After 3 weeks treatment, we observed leukemic cytoreduction (Fig. 5b) and decreased BM TNFα levels (Fig. 5c) in AC220treated AML mice compared with vehicle-treated controls. In AC220-treated mice, we noticed a gain in BM CD31 + Sca-1 high ECs and vessels (i.e., arterioles) (Fig. 5d,   Fig. 4 TNFα induced vascular remodeling of the leukemic BM niche via EC miR-126 downregulation. a miR-126 levels in BM ECs from normal wt and Mll PTD/WT /Flt3 ITD/ITD AML mice by Q-RT-PCR (n = 5 mice per group). b miR-126 levels in BM ECs from wt recipient mice engrafted with BM cells from normal wt or Mll PTD/WT /Flt3 ITD/ITD AML mice by Q-RT-PCR (n = 5 mice per group). c Levels of pri/pre-and mature miR-126 by Q-RT-PCR (n = 5 mice per group; left) and mature miR-126 expression by miRNA staining (right; one of three independent experiments with similar results is shown) in BM ECs from normal wt mice exposed in vitro to mrTNFα (1 ng/ml) or vehicle for 24 h. Scale bars represent a size of 10 µm. d Levels of Vcam1 and Spred1 in BM ECs from normal wt mice exposed in vitro to mrTNFα (1 ng/ml) or vehicle for 24 h by Q-RT-PCR (n = 5 mice per group). e Levels of pri/ pre-(left) and mature miR-126 (right), as analyzed by Q-RT-PCR, in BM ECs from normal wt mice treated in vivo with mrTNFα (1 µg/day, ip, 3 weeks). f CD31 (FITC) and Sca-1 (PE) IF staining of tibias (left) and flow cytometry analysis of BM EC Sca-1 high and Sca-1 low subfractions (right) from normal wt mice treated in vivo with vehicle, mrTNFα (1 µg/day, ip), or mrTNFα+ miR-126 mimics (20 mg/kg/day, iv, 3 weeks; n = 3 mice per group). Yellow arrows indicate CD31 + Sca-1 high EC lined-vessels; white arrows indicate CD31 + Sca-1 low EC-lined vessels. Scale bars represent a size of 100 µm. One of three independent experiments with similar results is shown. g CD31 (FITC) and Sca-1 (PE) IF staining of tibias from normal wt recipient mice engrafted with BM MNCs from congenic normal mice (normal to wt) or leukemic mice (AML to wt) and of tibias from normal Spred1 ECΔ/Δ recipient mice engrafted with BM MNCs from congenic leukemic mice (AML to Spred1 ECΔ/Δ ). Yellow arrows indicate CD31 + Sca-1 high EC lined-vessels; white arrows indicate CD31 + Sca-1 low EC-lined vessels. One of three independent experiments with similar results is shown. Results represent mean ± SEM. Significance values: *p < 0.05; **p < 0.01; ****p < 0.0001 e), an increase of Gata2 levels in ECs (Fig. 5f ) but not in LSKs (Additional file 1: Fig. S12a), an increase of miR-126 levels in both ECs and LSCs (i.e., AML LSK; Fig. 5g, Additional file 1: Fig. S12b), and a higher frequency of quiescent LSK cells (LSK G0 , Fig. 5h; Additional file 1: Fig.  S12c) compared with vehicle-treated controls.
In Mll PTD/WT /Flt3 ITD/ITD AML mice, BM LSKs are the subpopulation mostly enriched in leukemia-initiating cells (LICs; hereafter called LSCs) as shown by limiting-dilution transplantation of immunophenotypically defined BM cell subpopulations (Additional file 1: Fig.  S12d). To this end, we showed that while AC220 effectively cytoreduced AML blasts and increased survival of the primary treated mice (Fig. 5i), it also unexpectedly increased LSC burden and/or activity. In fact, in secondary (2nd) transplant experiments, recipients of BM MNCs from AC220-treated donors had a higher disease burden (Additional file 1: Fig. S12e) and shorter survival (Fig. 5j) than recipients of BM MNCs from vehicletreated donors, indicating not only persistence but also an expansion of LSCs.
Similar results were obtained by pharmacologic miR-126 downregulation with miRisten. A cohort of Mll PTD/WT /Flt3 ITD/ITD leukemic mice were generated as described above and then treated with miRisten (20 mg/kg, iv, daily), SCR, miRisten+ AC220 (10 mg/kg, oral gavage, daily), or SCR+ AC220 for 3 weeks, followed by assessment of donor AML cell engraftment in PB, BM and spleen (Additional file 1: (See figure on next page.) Fig. 5 Treatment-induced CD31 + Sca-1 high re-vascularization mediates LSC resistance. a Schematic design of the experiments. BM MNCs from diseased Mll PTD/wt /Flt3 ITD/ITD mouse (CD45.1/CD45.2) were transplanted into normal wt recipient mice (CD45.1, 6 Gy) to generate a cohort of leukemic mice with a similar disease onset time. At day 14 post transplantation, the mice were treated with AC220 (20 mg/kg/day, oral gavage) or vehicle for 3 weeks. After completion of treatment, a cohort of treated mice (n = 10 mice per group) were monitored for survival and another cohort of mice (n = 9 mice per group) were euthanized and assessed for BM vascular changes and LSC burden by 2nd transplantation. b Frequency and number of AML cells in BM and spleen from AC220-treated versus vehicle-treated AML mice. c TNFα mRNA expression in BM MNCs by Q-RT-PCR (left) and TNFα protein concentrations in BM plasma by Luminex assay (right) in AC220-treated versus vehicle-treated AML mice. d and e Long bones (femurs and tibias) from AML mice treated with vehicle or AC220 for 3 weeks were evaluated for: BM EC Sca-1 high and Sca-1 low subfractions (d left, representative plots; right, aggregate results) by flow cytometry analysis (n = 4 mice per group), and CD31 + Sca-1 high EC-lined vessels (i.e., arterioles) by CD31 (FITC) and Sca-1 (PE) IF staining (e left) and quantification (e right) (n = 3 mice per group). For e: Yellow arrows indicate CD31 + Sca-1 high EC-lined vessels; white arrows indicate CD31 + Sca-1 low EC-lined vessels; scale bars represent a size of 50 µm. f Gata2 levels in BM ECs from AML mice treated with AC220 or vehicle for 3 weeks, analyzed by Q-RT-PCR (n = 4 mice per group). g miR-126 levels in BM ECs (left) and LSKs (right) from AML mice treated with vehicle or AC220 for 3 weeks, analyzed by Q-RT-PCR. h Representative plots of cell cycling of BM LSKs (left) and frequency of quiescent BM LSKs (i.e., LSKs in G0 phase, LSC G0 , right) from AML mice treated with vehicle or AC220 for 3 weeks, analyzed by Ki-67 and DAPi staining and flow cytometry analysis. i and j Survival of AML mice treated with vehicle or AC220 for 3 weeks (primary, i) and survival of 2nd recipient mice (2nd survival, j) receiving BM cells from AC220-treated or vehicle-treated AML donors. Two of three independent experiments with similar results were shown. Results represent mean ± SEM. Significance values: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.  S14a). We confirmed miR-126 KD in ECs from miRisten-treated mice compared with SCR-treated mice (Additional file 1: Fig. S14b). Mice receiving the combination of miRisten+ AC220 had a significant reduction in the percentage of AML cells in PB, BM and spleen (Additional file 1: Fig. S14c, d), a significant decrease in the frequency of BM AML LSKs (Additional file 1: Fig. S14e) and increased survival (Additional file 1: Fig. S14f ) compared to mice treated with SCR+ AC220. Similar to the mrTNFα+ AC220 combination, the miRisten+ AC220 combination also significantly prolonged survival in 2nd transplant experiments (Additional file 1: Fig. S14g; median survival: 37 days vs. not reached at day 65 post transplantation, p < 0.0001), suggesting reduced post-treatment LSC burden. These results were corroborated in NSGS mice transplanted with FLT3-ITD+ AML blasts and then treated with miRisten+ AC220 or SCR+ AC220 for 3 weeks (Additional file 1: Fig. S15a). Treatment with miRisten+ AC220 significantly reduced human CD45 + CD33 + cell engraftment (Additional file 1: Fig. S15b) and prolonged survival (Additional file 1: Fig. S15c, left) in the primary treated mice. Recipient mice receiving BM cells from miRisten+ AC220-treated donors lived significantly longer than the mice receiving BM cells from SCR+ AC220-treated donors (Additional file 1: Fig. S15c, right; median survival: 167 vs. 126 days, p < 0.0001), suggesting that the combination of miRisten+ AC220 had decreased the LSC burden.

Discussion
AML cells reportedly produce inflammatory cytokines that profoundly remodel the BM vascular niche and create a microenvironment supportive of competitive leukemia growth over normal hematopoiesis [31,39,40]. However, how the leukemic BM niche adapts to the changing conditions caused by treatment and ultimately impacts on clinical outcome remains to be fully elucidated. Herein, we utilized FLT3-ITD+ AML murine and PDX models to study the vascular changes of the leukemic BM niche that occur upon molecular targeting of the AML blasts. During AML growth, in the leukemic BM niche, we observed a loss in CD31 + Sca-1 high ECs, which line mainly non-permeable arterioles, and a gain in CD31 + Sca-1 low ECs, which line mainly fenestrated, permeable sinusoids. These vascular changes were caused partly by high levels of TNFα produced by the AML blasts, causing downregulation of miR-126 in CD31 + Sca-1 high ECs, which became depleted while an enrichment in CD31 + Sca-1 low ECs was observed.
TNFα has been intensively studied for its role in normal and malignant hematopoiesis [41]. Yamashita and Passegue recently reported on the complex role of TNFα in normal and clonal hematopoiesis, showing that while inducing myeloid progenitor apoptosis, TNFα promotes HSC survival [42]. AML blasts express high levels of TNFα [43], and likely hijack TNFα-driven mechanisms of normal hematopoiesis to support leukemia growth [42]. Herein, we show a novel pro-leukemogenic role of TNFα that is extrinsic to AML cells and involves downregulation of miR-126 in the vascular compartment of the BM niche. While TNFα has been implicated in the remodeling of blood vessels and shown to promote angiogenesis during inflammation [44,45], to our knowledge the Fig. 6 Preventing treatment-induced CD31 + Sca-1 high re-vascularization enhances LSC sensitivity to TKI. a Schematic design of the experiments. BM MNCs from diseased Mll PTD/wt /Flt3 ITD/ITD AML mice (CD45.1/CD45.2) were transplanted into normal wt recipient mice (CD45.1, 6 Gy) to generate a cohort of mice with a similar disease onset time. At day 14 post transplantation, the leukemic mice were treated with AC220 (20 mg/kg/day, oral gavage)+ mrTNFα (1 µg/day, ip) or AC220+ vehicle for 3 weeks. After completion of treatment, a cohort of treated mice (n = 12 mice per group) were monitored for survival and another cohort of mice (n = 7 mice per group) were euthanized and assessed for BM vascular changes and LSC burden by 2nd transplantation. b and c Long bones (femurs and tibias) from the above leukemic mice treated with AC220+ mrTNFα or AC220+ vehicle for 3 weeks (n = 3 mice per group) were evaluated for: BM EC Sca-1 high and Sca-1 low subfractions (b left, representative plots; right, aggregate results) by flow cytometry analysis, and CD31 + Sca-1 high EC-lined vessels (i.e., arterioles) by CD31 (FITC) and Sca-1 (PE) IF staining (c left) and quantification (c, right). For c: Yellow arrows indicate CD31 + Sca-1 high EC-lined vessels; white arrows indicate CD31 + Sca-1 low EC-lined vessels; scale bars represent a size of 50 µm. d Representative plots (left) and aggregate results of frequency and number of LSCs (i.e., CD45.1/CD45.2 LSKs, middle), and ratio of LSCs/HSCs (i.e., CD45.1/CD45.2 AML LSKs: CD45.1 normal LSKs, right) in the BM from AC220+ mrTNFα-treated and AC220+ vehicle-treated leukemic mice. e Survival of treated primary AML mice (left) and 2nd recipient mice (right) receiving BM cells from AC220+ mrTNFα-treated or AC220+ vehicle-treated AML donors. f Schematic design of the experiments. BM MNCs from diseased Mll PTD/wt /Flt3 ITD/ITD mice (CD45.1/CD45.2) were transplanted into wt and EC miR-126 KO (miR-126 ECΔ/Δ ) recipient mice (CD45.2, 6 Gy) and evaluated for response to AC220 (20 mg/kg/day, oral gavage) given for 3 weeks. g Circulating leukemia burden (left) and frequency of BM LSCs (CD45.1/CD45.2 LSKs, middle) by flow cytometry analysis and miR-126 levels in LSCs by Q-RT-PCR (right) from leukemic miR-126 wt and miR-126 ECΔ/Δ mice treated with AC220 for 3 weeks. h Survival of AC220-treated leukemic miR-126 wt and miR-126 ECΔ/Δ primary mice (left) and 2nd recipients (right, n = 10 mice per group) of BM cells from the AC220-treated leukemic miR-126 wt and miR-126 ECΔ/Δ mice. Results represent mean ± SEM. Significance values: *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001 TNFα-induced switch from Sca-1 high ECs to Sca-1 low ECs, and in turn from an arteriole-to a sinusoid-enriched BM niche as observed during leukemia growth has not been previously reported. As these changes were a phenocopy of genetic (EC miR-126 KO) and pharmacologic (i.e., miRisten) EC miR-126 deprivation, we postulated and proved that loss of CD31 + Sca-1 high ECs and arterioles were due to TNFα-induced miR-126 downregulation.
MiR-126 is one of the most highly expressed micro-RNAs in ECs, where it acts as a master regulator of angiogenesis [12,28,29]. In developmental vasculogenesis, miR-126 supports differentiation of embryonic stem cells into endothelial precursor cells and mature ECs [46]. In mature ECs, miR-126 contributes to the maintenance of quiescence and vascular integrity and inhibition of endothelial permeability and apoptosis [12,28,29]. Furthermore, miR-126 enhances the activity of angiopoietin-1 (Ang-1), a glycoprotein that regulates vessel stabilization, maturation and permeability [47,48]. Of note, the molecular mechanisms through which TNFα induces endothelial miR-126 downregulation are likely to be multifaceted and remain to be fully elucidated. Herein, we showed that they likely involve GATA2 [34,35], a miR-126 transcription factor, which is reportedly silenced by TNFα [37]. But how do the TNFαinduced vascular changes in the leukemic BM niche ultimately promote leukemia growth? We previously showed that CD31 + Sca-1 high ECs expressed the highest levels of miR-126 in the BM niche and supply miR-126 to maintain LSC quiescence [31]. Thus, by inducing loss of CD31 + Sca-1 high ECs, TNFα decreases the endothelial supply of miR-126 to LSCs and enables them to enter the cell cycle and partially differentiate into proliferating "bulk" AML blasts [31,49].
Of note, these observations may have direct translational and clinical relevance. In fact, under these conditions, therapeutic cytoreduction of AML blasts can cause a drop in the BM levels of TNFα and in turn lead to a post-treatment enrichment of BM CD31 + Sca-1 high ECs and arterioles with a consequent increase in the endothelial miR-126 supply to LSCs, which, once enriched in miR-126, are more resistant to therapy [31,49]. We proved this model in FLT3-ITD+ AML murine and PDX mice treated with TKIs. We showed that after TKI treatment, an increase in CD31 + Sca-1 high ECs and arterioles occurred and LSCs persisted as demonstrated using secondary transplant experiments. Of note, LSC persistence could be prevented by blocking the gain in CD31 + Sca-1 high ECs and the arteriolar "re-vascularization" of the BM niche with administration of recombinant TNFα or with deprivation of endothelial miR-126 via genetic EC miR-126 KO or pharmacological treatment with the anti-miR-126 miRisten.
In comparing our results with recent reports on the BM vascular changes occurring during AML growth and treatment, we found interesting similarities between these studies and our work [17,39]. While Passaro et al. showed that engraftment of AML blasts in mouse BM increased CD31 + Sca-1 + ECs [17], in line with our results, Duarte et al. reported that AML growth caused loss of CD31 + Sca-1 + ECs in the BM niche [39]. Similar to our observation, both studies demonstrated an increase in BM vascular permeability during leukemia growth, which associates with CD31 + Sca-1 low vessels. While these studies showed that the post-treatment changes rescued AML-induced BM permeability and enhanced sensitivity of the proliferating blasts to chemotherapy, they did not test the effects of these changes on LSC activity as done in our study. To our knowledge, we are the first to show that, in the BM niche, therapeutic cytoreduction of proliferating blasts decreases TNFα levels and leads to enrichment in CD31 + Sca-1 high arterioles that safeguard LSCs. Of note, using PDX model transplanted with human FLT3-ITD+ AML cells, Maifrede et al. showed that AC220 (10 mg/kg/day, 7 days) prolonged survival of primary treated mice, but did not change survival of 2nd recipient mice [50]. Herein, using FLT3-ITD+ AML murine model, we showed that, treatment with AC220 (20 mg/kg/day, 21 days) prolonged survival of primary treated mice and resulted in a more severe disease in 2nd recipient mice (i.e., shorter survival). These results are not in conflict as in both studies, TKI treatment failed to reduce LSC burden. The difference in survival of secondary transplanted mice observed in these two studies could be related to technical differences including the used models and the schedule (7 vs. 21 days) and dosage (10 vs. 20 mg/kg/day) of the drug administration.
Thus, taken altogether, these results provide the evidence for non-genetic, LSC extrinsic mechanisms of treatment resistance that depend on the vascular plasticity of the leukemic BM niche and likely exemplify a treatment-related Janus phenomenon in AML [51]. Janus was an ancient Roman god with two faces looking in opposite directions. The two opposing "faces" of the TKI treatments in FLT3-ITD+ AML models as described here are the initially beneficial cytoreduction of proliferating blasts followed by post-treatment CD31 + Sca-1 high EC gain and arteriolar re-vascularization of the BM niche that safeguard LSCs thereby creating the condition for disease relapse. Of note, it is likely that the "Janus" phenomenon that we report here for TKIs may also apply to other antileukemic therapies that target proliferating blasts but do not kill LSCs. To this end, prevention of post-treatment CD31 + Sca-1 high EC expansion, arteriolar re-vascularization and LSC protection is achievable with pharmacological deprivation of BM endothelial miR-126 (i.e., miRisten), which may represent a novel strategy to overcome non-genetically driven, extrinsic mechanisms of LSC resistance in AML.

Conclusions
The TNFα-miR-126 axis plays a key role in the BM vascular changes induced by antileukemic treatments and mediates non-genetic, extrinsic mechanisms of LSC treatment-resistance that can be overcome with preemptive therapeutic deprivation of EC miR-126.