Abstract
Histone H3 lysine 4 methylation (H3K4Me) is proximally associated with chromatin activation, and therefore removing H3K4 methyl groups is normally coincident with gene repression. H3K4Me demethylase KDM1a/LSD1 is a potential therapeutic target for multiple diseases, including for the treatment of the β-globinopathies (sickle cell disease and β-thalassemia) since it is a component of multiple γ-globin repressor complexes, and its inactivation leads to robust induction of the fetal globin genes. However, the effects of LSD1 inhibition in definitive erythroid cells are not well characterized. Here we examined the consequences of erythroid-specific conditional inactivation of Lsd1 in vivo using a new Gata1creERT2 bacterial artificial chromosome (BAC) transgene. Conditional loss of Lsd1 in adult mice led to a differentiation block in erythroid progenitor cells and the surprising expansion of a GMP-like cell pool, apparently converting hematopoietic differentiation potential from an erythroid to a myeloid fate. The analogous phenotype was also observed in human cells: inactivation of LSD1 in hematopoietic stem and progenitor cells (HSPC) also blocked erythroid differentiation, coincident with robust induction of myeloid transcription factor genes (e.g. Pu.1 and Cebpa). Remarkably, blocking the activity of PU.1 or RUNX1 (a transcriptional activator of Pu.1) at the same time as blocking LSD1 activity reverted the myeloid lineage conversion back to an erythroid phenotype. Taken together, the data show that LSD1 maintains erythropoiesis by reversibly repressing a myeloid cell fate in adult erythroid cell precursors, and that inhibition of the myeloid differentiation pathway can reverse the negative effects of LSD1 inactivation on erythroid differentiation.
Introduction
Sickle cell disease (SCD) is a genetic disorder arising from a single nucleotide transversion, resulting in a Glu to Val amino acid change in the human β-globin protein, βs. βs subunits comprise equal parts, with α-globin subunits, of the tetrameric human hemoglobin, HbS [α2β2S]. HbS polymerizes in hypoxic red blood cells (RBC) causing RBC sickling, increased fragility and subsequent destruction, thereby leading to pathological organ damage, episodic acute pain and early death in affected individuals1, 2. Clinical studies showed that the pathophysiological severity of SCD negatively correlates with increased levels of fetal hemoglobin (HbF; α2γ2)2. β-thalassaemia major (TM; aka Cooley’s Anemia, CA) can arise by biallelic inheritance of any of hundreds of different recessive genetic mutations leading to diminished or altered expression of the adult β-globin gene3. However, co-inheritance of hereditary persistence of fetal hemoglobin (HPFH) alleles, normally leading to a benign hematopoietic condition characterized by elevated HbF levels in adults, significantly ameliorates SCD and TM disease pathology, making HbF re-activation in adults a focal therapeutic target for treating the β-globinopathies4–8. The beneficial effect of increased HbF levels on RBC longevity and stability in SCD has been confirmed in cell culture studies as well as in multiple animal models9.
The human β-globin locus spans more than 70 kb on chromosome 11 and is composed of five genes that are sequentially transcribed during development10. Spatially located from closest to farthest from the locus control region (LCR)11, 12, the five genes are arranged in their temporal order of developmental expression: ε- [embryonic], Gγ- and Aγ- [fetal], and δ- and β-globin [adult]. The switch from ε- to γ-globin transcription occurs as definitive erythropoiesis initiates in the fetal liver. At around the time of birth, γ-globin is gradually silenced and active transcription “switches” predominantly to the adult β-globin gene13. The competition model was proposed as the earliest explanation for a mechanism that might control globin gene switching14, and suggested that each of the globin genes competes for the trans-activating activity of the LCR super-enhancer13. According to this hypothesis, the globin gene closest to the LCR would have a greater chance to interact with the LCR and therefore are actively transcribed, unless a linearly closer gene is repressed by gene-autonomous silencing15–20. Thus, a more comprehensive mechanistic understanding for how the γ-globin genes are regulated in adult erythroid cells could provide crucial conceptual insights into how HbF might be reactivated, by either genetic or pharmacological means, for therapeutic benefit in the erythroid cells of TM and SCD patients.
Extensive effort has focused on the mechanisms that regulate γ-globin autonomous repression/gene silencing in adult red blood cells. Most notably, mutations in the binding sites for multiple transcription factors (TFs) in the vicinity of the γ-globin gene result in functional hereditary persistence of fetal hemoglobin (HPFH)9. These TFs include Bcl11a, LRF, GATA1 and an orphan nuclear receptor heterodimer composed of TR2 and TR421, 22,23, 24. We previously reported that TR2/TR4 directly binds to the γ-globin promoter to recruit a multimeric Direct Repeat Erythroid-Definitive (DRED) multi-subunit protein complex to repress γ-globin transcription in adult erythroid cells21, 22. The core co-repressor subunits of both the DRED complex and BCL11a include a scaffold protein, NCoR1, which recruits the epigenetic chromatin modifying enzymes DNMT1 and LSD1, among several others25–27.
LSD1/KDM1A is a flavin adenine dinucleotide-dependent histone demethylase that is specific for methylated histone H3 lysine 4 and 9 (H3K4 and H3K9)28, 29. Pharmacological inhibition of LSD1 de-represses HbF synthesis by up to 30% in cell cultures of erythroid differentiated human HSPC27, which is a level widely acknowledged to be of significant potential therapeutic benefit to SCD or TM patients. In vivo administration of one LSD1 inhibitor, RN-1, to baboons, an animal model with a globin gene expression profile that closely resembles human developmental expression, robustly re-activated HbF synthesis 30, 31. Additionally, RN-1 administration to humanized SCD model mice was shown to re-activate HbF synthesis and significantly ameliorate the pathological characteristics of SCD32. Taken together, the data indicate that safe, efficient LSD1 inhibition would be efficacious for HbF induction in humans, and as a widely accessible treatment for SCD and TM by millions of affected individuals worldwide.
Given the promising effect of LSD1 inhibitors in HbF induction in vitro and in vivo, it would be important to understand in preclinical studies any adverse effects of LSD1 inhibition by drug candidates in adult erythroid cells. Pan-hematopoietic Lsd1 deletion using Mx1Cre33 or inducible RNAi-based LSD1 mRNA knockdown34 in adult mice impaired erythropoiesis coupled to an expansion of HSPC, suggesting that LSD1 plays an active role in multiple hematopoietic lineages, and not solely in erythropoiesis. Therefore, a detailed mechanistic understanding of how LSD1 inhibition affects adult hematopoiesis would be critical to address the potential therapeutic and adverse effects of any candidate LSD1 inhibitor for clinical use.
In this study, we generated an erythroid-inducible Cre mouse line by modifying a mouse Gata1 BAC that would allow us to conditionally investigate functions of LSD1 in adult murine erythroid cells. During the analysis of erythroid-specific Lsd1 conditional loss-of-function (LOF) mice, we discovered that this epigenetic modifying enzyme plays a key role in repressing myeloid gene expression. Erythroid-specific Lsd1 inactivation in vivo led to a marked expansion of granulocyte-monocyte progenitor (GMP)-like cells in the bone marrow, and surprisingly these novel GMPs arise from erythroid progenitor (EP) cells. Furthermore, PU.1, a key transcription factor required for myeloid differentiation, was significantly induced in colony forming unit-erythroid (CFU-E) cells recovered from Lsd1-CKO mouse bone marrow. In a similar manner, pharmacological inactivation of LSD1 in human CD34+ HSPC that were induced to undergo erythroid differentiation significantly re-activated myeloid genes (Pu.1, Cebpa and Runx1) to severely impair erythroid differentiation. Finally, co-inactivation of Lsd1 and either Pu.1 or Runx1 partially rescued the erythroid cytotoxicity associated with LSD1 inhibition. Taken together, the data indicate that LSD1 functional loss impairs erythropoiesis through a reversible mechanism that induces the activation of a myeloid gene expression program in erythroid progenitors leading to an erythroid to myeloid cell fate conversion. Furthermore, the data show that inhibition of the RUNX1/PU.1 myeloid differentiation axis allows fetal hemoglobin induction in the absence of the LSD1 LOF-induced block in erythroid differentiation.
Results
LSD1 inhibitors induce HbF but block erythroid differentiation
Current LSD1 inhibitors, exemplified by tranylcypromine (TCP), are all small chemical compounds that initially showed promising potential for use as SCD therapeutics because of their ability to stimulate fetal γ-globin expression to high levels27. However, all of the commercial inhibitors that we have tested to date exhibit one or more properties that prohibits their use as promising drug candidates. One novel LSD1 inhibitor that we recently synthesized, CCG50, is a small organic molecule that was designed to bind reversibly to the flavin-dependent catalytic center of LSD1 (data not shown), thus differing from the majority of current (irreversible) commercial inhibitors. We showed that CCG50 (referred to interchangeably as LSD1i) efficiently and specifically inhibited LSD1 function with an IC50 of 115 nM (Supplemental Fig. 1A).
To test whether LSD1i administration would, like TCP, also induce abundant γ-globin expression, we tested erythroid differentiation in human CD34+ HSPC using a standard three-phase culture system (Supplemental Fig. 1B)35, 36. To score erythroid differentiation progression, we monitored CD71 (transferrin receptor) and CD235a (glycophorin A) expression beginning on d7 (prior to LSD1i addition) and again on d11, d14 and d18 by flow cytometry. In this culture system, proerythroblasts first acquire CD71 followed by CD235a surface antigens and subsequently lose CD71 expression to finally display only CD235a as mature erythroid cells36, 37. A minority of such HSPC remained double negative (15%) while most cells had acquired one (14% CD71) or both (70% CD71 + CD235a) erythroid differentiation antigens by d7 (Supplemental Fig. 1C). After differentiation induction on d7 (following the removal of hydrocortisone and human IL3), over 90% of control cells treated with DMSO alone expressed both of these erythroid cell surface markers by d11 and d14, while by d18, 20% of the cells had lost CD71 expression to finally become CD235a singly-positive mature erythroid cells (DMSO panels, Fig. 1A). At the lowest LSD1i concentration tested (120 nM), the flow cytometric pattern of cells at d11 (last column, Fig. 1A) resembled that of d7 controls (Supplemental Fig. 1C) with the majority (65%) of cells remaining double-positive erythroid precursors. From d14 through d18, the majority (95%) of cells remained CD71+ CD235a+ with only 3-4% converting to more mature CD71-CD235a single positive cells. In contrast, at the highest concentration of LSD1i tested (1.1 μM), most of the cells (>70%) remained double negative even by d18 (Fig. 1A). Hence, the novel, reversible LSD1 inhibitor CCG50 exhibits the same concentration-dependent inhibitory effect on erythroid differentiation that was also observed with other LSD1 inhibitors, including RN-1 and TCP (data not shown).
Next, we assessed HbF synthesis by HPLC analysis of the hemoglobins produced in d18 differentiated erythroid cells cultured in the presence or absence of LSD1i. At all three concentrations tested (120 nM, 370 nM or 1.1 μM), LSD1i induced γ-globin synthesis and HbF above the baseline level of 2.5% (in DMSO mock-treated control cells) to between 10.8% and 24.4% (Fig. 1B). Globin mRNA analysis indicated that although LSD1i enhanced the γ/(γ+β) ratio in a dose-dependent manner (Fig. 1C), total β-like globin (γ+β) mRNA significantly declined and negatively correlated with increasing LSD1i concentration (Fig. 1D). As anticipated, expression of several key erythroid transcription factors (GATA1, KLF1 and TAL1) was also reduced after LSD1i administration (Fig. 1E). Taken together, the data indicate that while low to moderate LSD1 inhibition (at 120 or 370 nM CCG50) robustly induced HbF, while higher concentrations significantly impaired erythroid differentiation accompanied by increased cytotoxicity in ex vivo expanded human CD34+ HSPC.
Generation of conditional, erythroid lineage-specific deleter transgenic mice
We next investigated whether genetic Lsd1 LOF in vivo would similarly inhibit adult erythroid differentiation in a mouse model. While Mx1-Cre38, EpoR-Cre39, and Vav1-Cre40 deleter strains have been utilized in the past to achieve pan-hematopoietic or erythroid-specific recombination of loxP-flanked genes in vivo, none were ideal for generating conditional LOF alleles (CKO) in adult murine red blood cells, as Cre activity is either not inducible (e.g. in EpoR-Cre or Vav1-Cre animals), or is not restricted solely to the erythroid lineage (Vav1-Cre and Mx1-Cre). To circumvent this issue, we generated and characterized erythroid lineage-specific, tamoxifen (Tx)-inducible lines of Cre transgenic mice to determine the in vivo effect(s) of Lsd1 LOF during definitive erythropoiesis in adult animals.
We utilized a 196 kb mouse Gata1 Bacterial Artificial Chromosome (abbreviated G1B) as the Cre delivery vehicle, which we had previously shown to faithfully recapitulate endogenous mouse Gata1 gene expression41–43. A tamoxifen (Tx)-inducible CreERT2 fusion gene44, encoding a Cre recombinase whose activity is dependent on the binding of the Tx ligand, was inserted at the ATG start codon of the Gata1 BAC by gene editing (Supplemental Fig. 2). Correctly targeted G1BCreERT2-Neo+ clones were verified by PCR (not shown). The frt-flanked Neo selection marker was subsequently excised from the BAC by activating L-arabinose-induced Flp recombinase expression in EL250 cells41. The resultant modified G1BCreERT2 BAC clones were verified by PCR and finally by Sanger sequencing (not shown). The BAC was then microinjected into C57Bl/6J zygotes to generate transgenic mice, from which two G1BCreERT2 transgenic lines, L245 and L259, were recovered.
To determine the tissue specificity and activity of Cre recombinase in the two G1BCreERT2 mouse lines, we crossed them to the Rosa26-loxP-Stop-loxP-TdTomato (R26T) reporter mouse to generate compound R26T:G1BCreERT2 mutant mice, in which TdTomato (TdT) expression is activated upon Cre-mediated deletion of the loxP site-flanked transcriptional stop cassette45. R26T:G1BCreERT2 mice were injected with Tx every other day five consecutive times; untreated R26T:G1BCreERT2 mice were included as negative controls. Two weeks after the first Tx administration, total bone marrow (BM) cells were collected to analyze TdT epifluorescence in cells of the hematopoietic hierarchy. Total BM cells from control and Tx-treated mice were separately co-stained with anti-CD71 and - Ter119 antibodies to monitor erythroid differentiation.
As erythroid cells mature, they first acquire CD71 expression (gate II; Fig. 2A) before co-expressing both Ter119 and CD71 (gate III) and then gradually lose CD71 immunopositivity (gate IV) to express Ter119 alone (gate V)46. No TdT signal was detected in the BM of either L245 or L259 mice that were not administered Tx, indicating that CreERT2 activity was not “leaky” (blue peaks, Fig. 2B). Since CreERT2 was under the transcriptional control of Gata1 cis-regulatory elements contained within the 196 kb BAC, we expected that upon Tx administration, only cells with erythroid developmental potential would express TdT. Indeed, fraction I, which primarily represents BM non-erythroid cells, were almost exclusively TdT-negative, whereas > 90% of the cells in fraction II, which are committed to erythroid differentiation, were labeled by TdT (Fig. 2B, red peaks). Notably, the TdT intensity diminished as cells differentiated (fractions III-V), likely due to some combination of global erythroid nuclear condensation, TdT turnover and dilution (Fig. 2B). We also stained whole BM cells with a cocktail of antibodies recognizing non-erythroid hematopoietic cell surface antigens to determine whether CreERT2 expression was expressed in B (B220), T (CD3e) or myeloid (Gr1, CD11b) cell lineages. Flow cytometry indicated that very few TdT+ cells were present in the B220/Gr1/CD11b/CD3e+ population (Supplemental Fig. 3, red peak), indicating that the G1BCreERT2 transgene was active essentially only in erythroid lineage cells. Both transgenic lines L245 and L259 shared similar TdT expression profiles (Fig. 2B and Supplemental Fig. 3), and therefore only line L259 was used for all subsequent studies.
Since the mouse Gata1 gene is known to be expressed in megakaryocytes (Mk), and other myeloid cells47, we examined Cre activity in Mk progenitors (MkPs), Mk and mast cells in R26T:G1BCreERT2 mice (Supplemental Figs. 4 and 5). Only 10% of MkP (defined as Lin-cKit+Sca1-CD41+CD150+) (Supplemental Figs. 4A, 4B and 5) and 25% of mature CD41+CD61+ Mk cells were TdT+ (Supplemental Figs. 4C, 4D). Of the FcRIa+c-Kit+ mast cells, 15% were TdT+ (Supplemental Figs. 4E, 4F). These data suggested that either the G1BAC does not contain the Gata1 regulatory element(s) that are required for expression in Mk and mast cells or that endogenous GATA1 is expressed in only a subset of those lineages. Taken together, the data indicate that the G1BCreERT2 mouse is a novel, inducible genetic tool that is faithfully expressed in the erythroid lineage.
Hematopoietic lineage expression of G1BCreERT2
We next investigated the expression of G1BCreERT2 within the early hematopoietic hierarchy. Total BM cells were harvested from R26T:G1BCreERT2 mice that were either injected with Tx or corn oil (control) and then stained with antibodies for Lin-Sca1+c-Kit+ (LSK cells, murine hematopoietic stem and progenitor cells), common myeloid progenitors (CMP: Lin-c-Kit+Sca1-CD34+CD16/32-), granulocyte-monocyte progenitors (GMP: Lin-c-Kit+Sca1-CD34+CD16/32+), megakaryocyte-erythrocyte progenitors (MEP: Lin-c-Kit+Sca1-CD34-CD16/32-) or colony forming unit-erythroid cells (CFU-E: Lin-c-Kit+Sca1-CD41-CD16/32-CD150-CD105+) using conventional cell surface markers to identify each subpopulation48 (Supplemental Figs. 5-7, Fig. 2C). Of the LSK cells, approximately 5% were TdT+ after Tx-induced CreERT2 activation, which may represent an erythroid- or Mk-primed progenitor cell population (Fig. 2C, red peak)49. As anticipated, no CMP or GMP cells were labeled after Tx administration (Fig. 2C). In contrast, 85% of MEP and 90% of CFU-E were TdT+ (Fig. 2C). In summary, the data indicate that the Cre activity of G1BCreERT2 mice can be detected as early as the MEP stage of erythroid developmental commitment as well as at all subsequent stages of erythroid differentiation.
Erythroid progenitor deficiencies in Lsd1 CKO mice
In previous studies using EpoR Cre to facilitate erythroid lineage-specific Lsd1 ablation, Kerenyi et al.33 reported prenatal lethality as a result of defective erythropoiesis. To investigate the effects of Lsd1 conditional erythroid ablation in adult animals, we intercrossed G1BCreERT2 L259 mice with Lsd1 homozygous floxed (Lsd1f/f) mice50, all in a congenic C57Bl/6 background, to generate compound mutants. Lsd1f/f:G1BCreERT2 (Lsd1 CKO) or Lsd1+/+:G1BCreERT2 (control) mice were administered Tx seven times every other day for two weeks. Mice were sacrificed for hematological analysis 12 hours after the final Tx injection, unless otherwise noted. This strategy increased the possibility that erythroid progenitor cells, when continuously exposed to Tx, might more likely delete Lsd1 in progenitors that had escaped earlier Cre-mediated inactivation.
Peripheral blood differential analysis indicated that the Lsd1 CKO mice suffered from anemia, as all erythroid parameters were significantly affected (Table 1). In contrast, the WBC count was indistinguishable between the Lsd1 CKO and control groups. Somewhat surprisingly, platelet counts were significantly higher in the Lsd1 CKO mice (Table 1), consistent with the increase in the number of CD41+ Mks in the BM of Lsd1 CKO mice (data not shown).
To investigate the source of the anemia observed in the Lsd1 CKO mice in greater detail, we performed in vitro colony assays of total BM cells from Tx-treated Lsd1 CKO and control mice. The data indicated that colony-forming unit-granulocyte, erythroid, monocyte, megakaryocyte (GEMM or CFU-GEMM) numbers were indistinguishable between the Lsd1 CKO and control animals (Fig. 3A). Interestingly, there were slightly, but significantly, more colony-forming unit-granulocyte, monocyte (GM or CFU-GM) colonies in Lsd1 CKO BM (Fig. 3A). Among committed erythroid progenitor populations, both burst-forming unit-erythroid (BFU-E) and CFU-E were quantitatively (by 50% or 90%, respectively; Fig. 3A) and qualitatively reduced (Fig. 3B) compared to control BM progenitors, thus indicating a severe depletion in the erythroid progenitor populations as a result of Lsd1 ablation. Hence, the anemia observed in Lsd1 CKO mice stemmed from a severe reduction in the number of erythroid progenitor cells.
We did not observe splenomegaly in Lsd1 CKO mice, suggesting that erythroid progenitors responsible for extramedullary hematopoiesis were also affected (Fig. 3C). In agreement with the results of the CFU colony assays, flow cytometry showed that the absolute number of CFU-E in Lsd1 CKO mice was significantly reduced, by approximately 70%, when compared to controls (Fig. 3D, E). Based on AnnexinV staining results51, this reduction was not attributable to increased programmed cell death of CFU-E (Supplemental Fig. 8). Taken together, the data indicate that Lsd1 LOF leads to a severe block in generating a normal number of erythroid progenitor cells, most markedly at the CFU-E stage, and consequentially to fewer circulating mature red blood cells, and that decrease is not due to increased erythroid cell death.
In the Lsd1 CKO BM, the number of CD71+Ter119+ erythroid precursor cells was significantly lower than in control BM (6.5% vs 24%; Fig. 3F). CD71+ erythroid precursor cells can be further fractionated into progressively more mature basophilic erythroblasts (BasoE), polychromatic erythroblasts (PolyE) and orthochromatic erythroblasts (OrthoE) using a CD44+ vs. forward scatter (FSC) flow cytometry gating strategy (Fig. 3F)46. The absolute cell number of BasoE, PolyE and OrthoE were all significantly reduced in the Lsd1 CKO CD71+Ter119+ BM (Fig. 3F, G), and the >70% reduction in erythroid precursor cells was likely due to continuous differentiation dysfunction that can first be detected in BFU-E cells (Fig. 3A, E).
To test the possibility that the residual erythroid precursor cells in the treated animals were not simply cells that had escaped Lsd1 conditional deletion, we quantified Cre-mediated deletion efficiency in flow-sorted BasoE. A primer pair that was designed to amplify the Lsd1 genomic region predicted to be lost after excision between the loxP sites (P1), and the product was normalized a the control primer amplicon (P2) that amplified intact genomic DNA lying adjacent to the loxP sites (Fig. 3H). The data show that >80% of FACS-sorted BasoE had deleted the genomic DNA between the two loxP sites (Fig. 3H). Hence, G1BCreERT2 achieved highly efficient conditional Lsd1 inactivation in adult erythroid lineage cells.
During erythroid cell maturation, there is normally an exuberant increase in cell numbers as erythroid precursor cells transit from basophilic to polychromatic to orthochromatic erythroblasts, as observed in the control mice (Fig. 3G, gray bars). However, in Lsd1 CKO mice, the cell numbers increased only marginally between the basophilic and orthochromatic cell stages (Fig. 3G, black bars). This difference in differentiation potential between Lsd1 CKO and control mice indicates that LSD1 plays an important role in maturation at multiple sequential stages during erythroid cell differentiation.
We next asked whether increased cell death might account for the observed deficiency in erythroid precursor cells in Lsd1 CKO mice. Early and late apoptotic cells (AnnexinV+DAPI- and AnnexinV+DAPI+, respectively41) were analyzed in BasoE, PolyE and OrthoE cells in Lsd1 CKO and control mice. Surprisingly, the late apoptotic frequency of both BasoE and PolyE in Lsd1 CKO mice was significantly reduced (to approximately half), whereas cell death of OrthoE in Lsd1 CKO mice was unchanged (Supplemental Fig. 9). Furthermore, the cell cycle in the CD71+Ter119+ precursor population was unaffected when comparing Lsd1 CKO and control mice (Supplemental Fig. 10). Taken together, the data indicate that Lsd1 LOF blocks the differentiation of erythroid precursor cells in a manner that is independent of programmed cell death or alterations in the cell cycle.
Lsd1 CKO erythroid progenitors acquire myeloid characteristics
The data presented thus far indicates that Lsd1 conditional LOF results in marked depletion of definitive erythroid progenitor and precursor cells that is independent of cell cycle and cell death. Curiously, data from the colony assays showed that the GM colony number was slightly but significantly increased (Fig. 3A). To potentially explain this curious observation we hypothesized that erythroid progenitor cells might have acquired myeloid features through a lineage switch that reciprocally impaired erythroid potential. To test this hypothesis, we examined the LSK, CMP, MEP and GMP progenitor cell populations in Lsd1 CKO and control mice (Fig. 4A-B and Supplemental Fig. 7); the data show that the absolute numbers of LSK, CMP and MEP were comparable among the groups. However, consistent with the hypothesis that erythroid progenitors were being diverted into other hematopoietic lineages, the GMP population significantly increased in the Lsd1 CKO BM (Fig. 4A, 4B). Curiously, we also detected prominent expansion of a new Lin-c-Kit+Sca1-CD16/32+CD34- cell population in the Lsd1 CKO bone marrow when compared to control mice (Fig. 4A).
To test whether the expanded GMP population in Lsd1 CKO mouse BM might arise from erythroid lineage origin, we generated Lsd1f/f:G1BCreERT2:R26T triple mutant mice (hereafter referred to as Lsd1 CKOT) in order to lineage trace TdT fluorescence in the expanded GMP population discovered in the Lsd1 CKO mice. Since the erythroid cells, but not GMP cells, in G1BCreERT2:R26T mice were labeled by TdT (Fig. 2C), we reasoned that the surfeit GMP cell population in Lsd1 CKOT mice would be TdT+ if they were in fact derived from erythroid precursor cells. As expected, LSK or MEP populations in either Tx-treated control (R26T:G1BCreERT2) versus Lsd1 CKOT mice were TdT- or TdT+, respectively (Fig. 4C), consistent with earlier observations (Fig. 2C). In Tx-treated R26T:G1BCreERT2 control mice, GMP cells were not labeled by TdT (Fig. 2C). Notably, and consistent with the observed three-fold increase in GMP cell number (Fig. 4A, 4B), 60% of GMP cells in Lsd1 CKOT mice were TdT+ (Fig. 4C). Based on detailed characterization of hematopoietic TdT expression (Fig. 2 and Supplemental Figs. 3 and 4), we conclude that these LSD1-deficient TdT+ GMP cells must originate from an erythroid progenitor population.
We next asked whether the GMP population in Lsd1 CKO mice (Fig. 4B) might be differentially susceptible to programmed cell death. While the early apoptotic populations of GMP were comparable in Lsd1 CKO and control groups, late apoptosis was reduced by 50% in the Lsd1 CKO GMP fraction (Supplemental Fig. 11), suggesting that the ectopic TdT+ GMPs are not differentially susceptible to apoptotic cell death and may simply differentiate into more mature myeloid cells. To test this hypothesis, we flow sorted the CFU-E, Pre-CFU-E and Pre-MegE from control and Lsd1-CKO BM and then seeded all three erythroid progenitor populations as a mixture into MethoCult semi-solid media to test their differentiation potential (Fig. 4D). The majority of these erythroid progenitors in the control mice gave rise to BFU-E, Meg and MegE colonies as well as very few GM colonies (possibly contaminants contributed by slight gating differences during cell sorting). In contrast, erythroid progenitors from Lsd1-CKO mice largely shifted differentiation potential from BFU-E to GM lineages, as the majority of colonies generated from Lsd1 CKO erythroid progenitors acquired GM characteristics (Fig. 4E). To determine whether the CFU-GM colonies derived from Lsd1 CKO erythroid progenitors are indeed myeloid cells, we picked individual colonies from Lsd1 CKO-seeded media and subjected the cells to flow cytometry, examining cell surface markers Gr1, CD11b and Ter119. The majority of these individual GM colonies were composed of cells that were Gr1+ or CD11b+; most significantly, none of the cells expressed the erythroid lineage Ter119+ marker (Fig. 4F), indicating that these myeloid colonies arose from early erythroid (CFU-E or Pre-CFU-E) progenitors in Tx-treated Lsd1 CKO mice.
To test the hypothesis that LSD1 inhibition leads to an erythroid to myeloid cell fate switch (EMS) using an orthogonal strategy, we next isolated untreated CFU-E, Pre-CFU-E and Pre-MegE from wild type mice by cell sorting. The cells were treated with either DMSO or increasing concentrations of LSD1i in the MethoCult semi-solid media. Consistent with the data from genetic Lsd1 LOF (Fig. 4E), LSD1i treatment of these cultures induces EMS in a dose-dependent manner (Fig. 4G). Notably, 3 μM LSD1i treatment completely switches the erythroid to a myeloid lineage progenitor differentiation potential. Taken together, these data strongly support the hypothesis that both Lsd1 genetic LOF or LSD1 pharmacological inhibition in erythroid progenitor cells leads to an erythroid to myeloid cell fate conversion.
Pu.1 is activated at the CFU-E stage in Lsd1 CKO mice
To better understand the molecular basis for why Lsd1 LOF in erythroid progenitors leads to gain-of GMP features, we flow sorted CFU-E cells from Lsd1 CKO and control mice (Fig. 3D). When we compared gene expression profiles in the two populations, LSD1 mRNA was reduced (as anticipated) by more than 90% in the CKO cells. PU.1 is a well characterized transcription factor that is critical for myeloid differentiation52–54, and transcriptional activation of Pu.1 is regulated by RUNX1 in HSPC55. Consistent with the observed increase in GMP activity in Lsd1 CKO mice, PU.1 expression levels were 7-fold higher in CFU-E recovered from their bone marrow. In the same cells, GATA1 mRNA was slightly but significantly reduced, another indication of impaired erythroid differentiation, whereas RUNX1 levels were unchanged (Fig. 5A).
Erythroid LSD1 inhibition activates ectopic Pu.1 expression
To further explore whether LSD1 inhibition in erythroid progenitor cells de-represses the expression of common myeloid markers, we performed qRT-PCR on human CD34+ cell-derived erythroid cells after 14 days of in vitro differentiation culture in the absence or presence of either TCP or CCG50 LSD1 inhibitors (Supplemental Fig. 1B). Consistent with the in vivo data, both inhibitors significantly activated two myeloid regulatory genes, Pu.1 and Cebpa, whereas expression of the key erythroid regulator GATA1 was significantly impaired (Fig. 5B). Notably, treatment with either LSD1 inhibitor also significantly induced RUNX1 expression in vitro (Fig. 5B) even though Lsd1 genetic LOF does not affect in vivo RUNX1 expression in murine CFU-E (Fig. 5A).
To test whether LSD1 directly regulates Pu.1 expression, we performed LSD1 ChIP assays in HUDEP2 cells, a human cell line that most closely corresponds to human adult definitive erythroid cells56. The results show that LSD1 is bound at the transcriptional start site of the human Pu.1 gene (Fig. 5C), an observation consistent with LSD1 ChIP-seq data reported for K562 myeloerythroid lineage cells (UCSC genome browser).
To determine whether LSD1i also induces EMS in primary human CD34+ cells, we performed flow cytometry at d11 of erythroid differentiation after four days of LSD1i or vehicle (DMSO) treatment (Supplemental Fig. 1B). LSD1i addition enhanced expression of myeloid cell surface marker CD11b in these erythroid differentiation cultures, up to 4.1% of total, in a dose dependent manner (Fig. 5D). The efficiency of EMS in the CD34+ cell differentiation cultures was less pronounced than in semi-solid culture media (Fig. 4E-G), perhaps as a consequence of missing myeloid cytokines in the erythroid differentiation media.
In summary, these data indicate that LSD1 normally acts to repress Pu.1 expression in order to maintain erythroid lineage potential in hematopoietic progenitor stages. Once Lsd1 is deleted or LSD1 protein is inactivated by chemical inhibitors, Pu.1 is induced in erythroid progenitors, likely through increased activity of RUNX155, which initiates the GMP cascade and impairs normal erythroid differentiation (Fig. 5E).
Pu.1 or Runx1 LOF rescues the LSD1i-mediated block in erythroid differentiation
To test whether or not the RUNX1/PU.1 axis plays a functional role in blocking erythroid differentiation that is induced by LSD1 genetic or pharmacological inactivation, we generated Pu.1 or Runx1 homozygous knockout HUDEP2 cell clones to ask whether ablation of either gene could rescue the erythroid differentiation blocking effects of LSD1 inhibition. For each gene, four clones were generated using two different sgRNAs. Properly targeted mutation of all clones (except one) was confirmed by Sanger sequencing (Supplemental Figs. 12, 13). Three different non-targeting sgRNAs (NT1, NT4 and NT3) containing scrambled guide sequences were used to generate negative control HUDEP2 clones.
LSD1i treatment at 300 nM significantly reduced the erythroid fractional percentage of CD71+CD235a+ cells (by 50%) in control HUDEP2 cells (Fig. 6A and Supplemental Figs. 14A, 14B), further confirming that LSD1i treatment impairs erythroid differentiation. The percentage reduction in CD71+CD235a+ cells was rescued by either Pu.1 or Runx1 knockout (Fig. 6A and Supplemental Fig. 14). Consistent with the erythroid differentiation block visualized by flow cytometry, LSD1i treatment of control HUDEP2 cells repressed γ- globin, β-globin and GATA1 expression, and activated Pu.1 expression, effects which could be rescued in Pu.1 or Runx1 knockout HUDEP2 cell clones (Fig. 6B). Taken together, the data indicate that RUNX1/PU.1 activation plays a key role in blocking erythroid lineage differentiation that is induced by LSD1 inhibition.
RUNX1 inhibitor treatment rescues the LSD1i-mediated block in erythroid differentiation
As LSD1 inhibitors have been proposed as potential therapeutic agents for SCD clinical treatment, we next ask if pharmacological inhibition of the RUNX1/PU.1 pathway could rescue the erythroid differentiation blocking effects of LSD1 inhibitor treatment. As no effective PU.1 inhibitor is commercially available, we tested Ro5-3335, a published RUNX1 inhibitor (RUNX1i) 57. To investigate whether co-treatment with a genuine and well charcterized RUNX1i plus LSD1i could rescue the erythroid differentiation defect caused by LSD1i treatment, we also added RUNX1i AI-10-10458 to the CD34+ erythroid differentiation cultures.
Either RUNX1i was added to CD34+ erythroid differentiation media at the same time as LSD1i, and erythroid cell maturation was examined at d11 by flow cytometry (Supplemental Fig. 1C). Treatment with either LSD1i alone inhibited erythroid differentiation as indicated by the reduced production of CD71+CD235a+ cells (Fig. 7A). However, co-treatment of cultures with either LSD1i plus RUNX1i rescued the erythroid differentiation block in a dose dependent manner (Fig. 7A). Although RUNX1i co-treatment rescued the erythroid differentiation block induced by LSD1i, significant cell death was also observed at d11 after Ro5-3335 addition (data not shown). To exclude the effect of RUNX1i-induced cell death during erythroid differentiation, we analyzed cells at d9, just two days after treatment with LSD1i, RUNX1i or both (Supplemental Fig. 15A). We found that co-treatment with either RUNX1 inhibitor partially rescued the LSD1i-mediated erythroid differentiation block (Fig. S15B).
Finally, we examined γ- and β-globin expression after treating the CD34+ erythroid differentiation cultures with RUNX1i, LSD1i or both, at d11 (Fig. 7B). RUNX1i co-treatment with LSD1i significantly rescued globin gene expression that was repressed by LSD1i treatment alone. Surprisingly, γ-globin expression was induced 4-fold by RUNX1i treatment alone, suggesting that RUNX1 may play a previously uncharacterized but LSD1-independent role in fetal globin gene repression in adult erythroid cells (Fig. 7B). Taken together, the data show that pharmacological inhibition of the RUNX1/PU.1 axis partially rescues the differentiation block induced by LSD1i treatment of erythroid precursor cells.
Discussion
In this study, we set out to investigate the effects of LSD1 inhibition on erythropoiesis in vivo; this required the generation of an erythroid lineage-specific, inducible Cre deleter line to investigate the effects of conditional Lsd1 loss-of-function in murine adult definitive erythroid cells. Previously, Takai et al. showed that a 196 kb mouse Gata1-GFP knock-in BAC exhibited prominent GFP expression in MEP and that this fluorescence continued to peak as erythropoiesis (at the CFU-E and proerythroblast stages) progressed, and then gradually weakened in basophilic and polychromatic erythroblasts43. Interestingly, they also reported that the first (weak) hematopoietic GFP fluorescence was detected in CMP; however, in R26T:G1BCreERT2 mice that were treated with tamoxifen, no CMP cells labeled with TdTomato. This could be due to the abundance of the CreERT2 fusion protein in CMP of G1BCreERT2 mice that somehow does not approach a threshold required for efficient loxP recombination, thus coincidentally causing the G1BCreERT2 transgene to be highly erythroid-specific. In the R26T:G1BCreERT2 mice, we observed that about 5% of Tx-treated LSK cells were labeled by TdTomato. However, none of the downstream myeloid progeny (such as GMP) were TdTomato+, indicating that the GATA1+ LSK cells in these mice do not differentiate into myeloid lineage cells.
Here we report that the loss of LSD1 activity either via pharmacological inhibition in human CD34+ HSPC or through genetic deletion in mice leads to disrupted adult erythropoiesis. In human erythroid HSPC differentiation cultures, a novel LSD1 inhibitor (CCG50) blocks erythropoiesis in a dose-dependent manner during the early stages of culture (from days 7 to 11) when hSCF is included in the culture media. After withdrawal of hSCF, erythroid differentiation temporarily recovers (fully or partially, depending on inhibitor concentration) by day 14, suggesting that the first effect of LSD1 inhibition is hSCF-dependent, which very likely occurs during progenitor cell stages. Furthermore, the production of reticulocytes is impaired, indicating that a second effect of LSD1 inhibition occurs at a late erythroid maturation stage. In agreement with the latter hypothesis, we also observed deficiencies at both the erythroid progenitor and late erythroblast stages in the Lsd1 conditional loss-of-function mice. Global transcriptomic analyses of the affected cell stages may provide additional molecular mechanistic insights into how loss of LSD1 affects erythroid differentiation and clues for how to eliminate or minimize any toxic effects of LSD1 inhibition in the future.
The in vivo role of LSD1 within the murine hematopoietic compartment has been explored previously. Using the pan-hematopoietic Mx1Cre transgene to facilitate conditional inactivation of Lsd1, Kerenyi et al. reported that LSD1 deficiency impacted HSC self renewal33. Furthermore, they showed that EpoRCre-mediated Lsd1 deletion in vivo resulted in prenatal lethality due to anemia. Using the newly generated G1BCreERT2 deleter line reported here, we were able to examine the consequences of LSD1 ablation in adult erythropoiesis for the first time. We found that definitive erythropoiesis was compromised as a consequence of drastically reduced erythroid progenitor and precursor cell numbers.
GATA1 was previously reported to transcriptionally repress PU.1 expression by directly binding to GATA consensus sites at the TSS of Pu.1 in erythroblasts59. We confirmed LSD1 binding at these same sites in the human erythroid HUDEP2 cell line by ChIP assays (Fig. 5C). It has been reported that a GATA2/TAL1/LSD1 complex binds to GATA sites to repress GATA1 transcription during erythroid differentiation60. Similarly, perhaps a GATA1/LSD1 complex may cooperatively repress Pu.1 transcription in erythroid cells. Interestingly, PU.1 forced overexpression impairs erythroid differentiation and leads to erythroleukemia in mice61. Mice that express GATA1 at 5% of wild type levels display a propensity for erythroleukemia47. Curiously, we detected the expansion of a GMP-like population as well as a new population that was immunophenotyped as Lin-cKit+Sca1-CD34-CD16/32+ in LSD1 erythroid-restricted knockout mice (Fig. 4A). Whether these unusual cells might represent harbingers of a pre-leukemic state or a previously undiscovered myeloid progenitor state will required further investigation. In this regard, long term monitoring of possible leukemogenesis in the Lsd1 CKO mice will certainly be interesting to investigate further.
Increased levels of LSD1 have been detected in different neoplasms, and LSD1 was shown to play a key role in carcinogenesis, highlighting the potential therapeutic effect of LSD1 inhibitors in cancer and myelodysplastic syndrome treatment62. However, LSD1 inhibitors have multiple side effects, thus potentially lessening their utility for targeted therapeutic development. One phase-I clinical study of LSD1 inhibitor TCP to treat AML and MDS patients showed that 13% of the patients developed anemia63, a profound negative effect of this earliest LSD1 inhibitor that was originally applied to treat neurological disorders.
In this regard, the mechanisms of LSD1 action revealed in this study may illuminate a path to further development of safe and effective therapeutics for the treatment of the β-globinopathies, SCD and β-thalassaemia. From the data presented here, one could envision a safe and effective binary therapeutic compound that both induces fetal hemoglobin (by blocking LSD1 activity) and at the same time promotes normal erythroid differentiation by blocking the induction of PU.1 and/or RUNX1 in HSPC. The identical treatment could be equally applicable to the treatment of AML and/or MDS patients suffering from malignant or pre-malignant neoplasms.
Materials and Methods
Erythroid differentiation of human CD34+ cells
Human CD34+ HSPCs from healthy donors were purchased from Fred Hutchinson Cancer Research Center. The CD34+ cells were cultured and differentiated into mature erythroid cells using a standard three-phase culture system36. The basic media was composed of Iscove’s Modified Dulbecco’s Medium (IMDM) supplemented with 5% pooled human AB plasma (Octaplas, Rhode Island Blood Center), 1x penicillin-streptomycin (Gibco), 1% L-glutamine, 330 µg/ml holo-transferrin (T4132, Sigma), 2 IU/ml heparin (H3149, Sigma), and 10 μg/ml human insulin (91077C, Sigma).
From day (d) 0-7 of the expansion culture period, the basic medium was supplemented with 1 μM hydrocortisone (07904, Stem Cell Technologies), 5 ng/ml human IL-3 (203-IL, R&D), 100 ng/ml human SCF (255-SC-050, R&D), and 3 IU/ml erythropoietin (Epoetin Alfa, Amgen). Hydrocortisone and IL-3 were withdrawn from the media beginning on d8 while SCF was omitted after d12. On differentiation day (d) 7, CD34+ HSPC-derived erythroid cells were counted and reseeded at 105 cells/ml in the presence of increasing concentrations of LSD1i (0, 120 nM, 370 nM or 1.1 DM) for an additional 11 days.
The novel LSD1 inhibitor described in this manuscript (CCG50; LSD1i) was designed, synthesized and characterized by the Center for Chemical Genomics at the University of Michigan, and will be described in a separate study (in preparation). At d7 of erythroid differentiation, cells (105/ml) were cultured in the presence of DMSO or 3-fold dilutions of LSD1i (1.1 μM, 370 nM or 120 nM; Fig. S1B). Thereafter, the medium was changed at d11 and d14 with freshly added LSD1i. At the indicated time points, aliquots of cells were removed for flow cytometric, RNA and/or HPLC analyses. Experiments were repeated using CD34+ cells from at least two separate donors.
In vitro LSD1 inhibition assay
A fluorescent in vitro LSD1 inhibition assay (Item No. 700120, Cayman Chemicals) was performed according to the manufacturer’s instructions.
HPLC
On the day erythroid differentiation terminated, cells were washed with PBS and the cell pellets were lysed; the contents were then subjected to HPLC analyses using a BioRad CDM System CDM5.1 VII Instrument according to the manufacturer’s instructions.
Total RNA isolation and qRT-PCR analyses
Total RNA was isolated from cells using Trizol (ThermoFisher Scientific) or RNeasy Micro Kit (Qiagen) according to manufacturer’s instructions. cDNA was synthesized using SuperScript III Reverse Transcriptase (ThermoFisher Scientific). qRT-PCR was performed using FastSybr Green Mastermix on an ABI Step One Plus instrument (ThermoFisher Scientific). Sequences of all primers are listed in Supplementary Table 1.
BAC recombination
BAC recombination and the 196 kbp mouse Gata1 BAC clone (RP23-443E19) have been described previously43, 64. Briefly, the targeting DNA fragment was constructed by inserting a CreERT2-polyA-Neo cassette at the translational start site of mouse Gata1 (in the 2nd exon) between 5’ and 3’ homologous recombination arms. The resultant targeting DNA fragment was excised from the cloning vector and transfected into transiently heat-activated EL250 bacteria (to induce recombinase expression) containing BAC RP23-443E19. Recombinant bacterial colonies (with newly acquired Kanamycin resistance) were individually screened for homologous recombination at the Gata1 ATG by PCR using two primer sets (Supplemental Table 1). Clones that were determined to have successfully recombined (CreERT2 Neo+) were then cultured in the presence of L-arabinose to induce flp recombinase expression to promote excision of the frt-flanked Neo selection marker. Finally, the BAC DNA was purified by NucleoBond BAC100 and submitted for microinjection into fertilized egg of C57BL6 background mice (Transgenic Animal Model core, University of Michigan). Primers are listed in Supplementary Table 1.
Mice
G1BCreERT2 transgenic founders were intercrossed with wild type C57BL6 mice for three generations to eliminate possible mosacism and to confirm germline transmission. Two lines, L245 and L259, were interbred with Cre reporter mice R26 tdTomato 45 (Jackson Stock# 007914) or homozygous Lsd1f/f mutant mice50. To induce G1BCreERT2 activity, mice (from 8-12 weeks old) were injected intraperitoneally with tamoxifen (Tx; 2 mg per injection on alternate days for 2 weeks). All animal procedures were approved by the Institutional Animal Care and Use Committee of the University of Michigan (IUCAC Protocol PRO00009778). Primer sequence for genotyping are listed in Supplemental Table 1.
Differential peripheral blood analyses
Peripheral blood was collected from mice by facial vein bleed into EDTA-treated tubes and analyzed on a Hemavet according to the manufacturer’s instructions.
Genomic DNA qPCR
Genomic DNAs were purified from FACS-sorted mouse basophilic erythroblasts. To quantify Lsd1 deletion efficiency, qPCR was performed using FastSybr Green Mastermix (Applied Biosystems) and Lsd1 primer pairs that specifically annealed to genomic sequences within (P1) or outside of (P2) the two LoxP sites flanking Lsd1 exon 6 50(Fig. 3H). Deletion efficiency was calculated from the Ct ratio of P1/P2 as described previously50, 65. The primers used for qPCR analysis are listed in Supplemental Table 1.
Flow cytometry
Total bone marrow (BM) isolation and subsequent antibody staining protocols were described previously66. AnnexinV staining was performed as described previously 41. Flow analysis was performed on a BD Fortessa while flow sorting was performed on a BD Aria III. Antibodies used for flow cytometry are listed in Supplemental Table 2.
ChIP
ChIP assays were conducted as described previously26.
Colony-Forming Unit (CFU) assay
Isolated total BM cells or purified, sorted cell populations were counted and seeded into Methocult M3434 or M3334 (Stemcell Technologies) according to manufacturer’s instructions for enumeration of BFU-E, CFU-GM, CFU-GEMM or CFU-E colonies.
CRISPR knockout of Pu.1 or Runx1 genes
Genome edited Pu.1 or Runx1 loss of function strategy was essentially as described in a previous study26. Individual HUDEP2 Pu.1 or Runx1 targeted clones were initially identified by PCR (sequences shown in Supplemental Table 1) and verified by Sanger sequencing. One clone (P2-4) exhibited no amplification of the guide-targeted exon, suggesting that this clone bore a large genomic DNA deletion (data not shown). The guide sequences used for gene editing are also listed in supplemental Table 1.
Supplemental Information
Acknowledgements
We gratefully acknowledge the assistance and insights provided by multiple colleagues at the University of Michigan (Sojin An, Uhnsoo Cho, Susan Hagen, Pil Li and Mathivanan Packiarajan). We gratefully acknowledge continued support from the NHLBI (awards U01 HL117658 and P01 HL146372; JDE and AW), from the Cooley’s Anemia Foundation (LY) as well as a center of excellence award from the NIDDK (U54 DK106829) to the Fred Hutchinson Cancer Center to support the isolation and distribution of normal human hematopoietic stem and progenitor (CD34+) cells for the scientific community.