A metabolic interplay coordinated by HLX regulates myeloid differentiation and AML through partly overlapping pathways

The H2.0-like homeobox transcription factor (HLX) regulates hematopoietic differentiation and is overexpressed in Acute Myeloid Leukemia (AML), but the mechanisms underlying these functions remain unclear. We demonstrate here that HLX overexpression leads to a myeloid differentiation block both in zebrafish and human hematopoietic stem and progenitor cells (HSPCs). We show that HLX overexpression leads to downregulation of genes encoding electron transport chain (ETC) components and upregulation of PPARδ gene expression in zebrafish and human HSPCs. HLX overexpression also results in AMPK activation. Pharmacological modulation of PPARδ signaling relieves the HLX-induced myeloid differentiation block and rescues HSPC loss upon HLX knockdown but it has no effect on AML cell lines. In contrast, AMPK inhibition results in reduced viability of AML cell lines, but minimally affects myeloid progenitors. This newly described role of HLX in regulating the metabolic state of hematopoietic cells may have important therapeutic implications.

L ong-term hematopoietic stem cells (LT-HSCs) are multipotent cells with self-renewal capacity primarily responsible for replenishing the entire hematopoietic system [1][2][3][4][5][6][7] . LT-HSC differentiation into mature blood and immune cells is a tightly regulated and multifaceted process. Transcription factors govern the mechanisms that maintain the balance between LT-HSC differentiation and self-renewal, or stemness [8][9][10] , and any perturbation in this process can ultimately lead to disease.
While it is well established that homeobox (HOX) transcription factors play a central role in hematopoietic development and disease, less is known about the function of non-clustered HOX factors in the hematopoietic system 11,12 . The non-clustered H2.0like homeobox transcription factor (HLX) has been recently identified as an important regulator of hematopoiesis. During development, HLX deficiency leads to a decrease in the colonyforming capacity of fetal liver cells [13][14][15][16] , and in adult hematopoiesis HLX regulates Th1/Th2 differentiation during T-cell development [17][18][19][20] . Recent evidence shows that HLX is essential for HSC maintenance and self-renewal [21][22][23] . Increased expression of HLX compromises self-renewal and eventually results in a myelomonocytic differentiation block concomitant with aberrant proliferation of myeloid progenitors 21 . Mechanistically, it has been suggested that this function of HLX in HSC maintenance and self-renewal is mediated by the p21-activated kinase PAK1. Indeed, it was demonstrated that inhibition of HLX or PAK1 induces differentiation and apoptosis of AML cells 21,22 . Consistent with this phenotype, HLX is overexpressed in 87% of AML patients and those presenting higher HLX expression have lower survival rates 21 . Recently, HLX has been shown to play a role in the browning of white adipose tissue, suggesting that this transcription factor is involved in the metabolic control of cell differentiation 24 .
Despite the pleiotropic functions of HLX and its critical regulatory role in multiple processes, particularly in hematopoiesis, only few direct downstream targets have been identified. Moreover, mechanistic insights into the function of HLX in hematopoiesis and myeloid differentiation are lacking. Thus, understanding the physiological roles of HLX in hematopoietic development and disease, including leukemia, remains a central issue in HSC biology.
Here, we use zebrafish, human hematopoietic stem and progenitor cells (HSPCs), and AML cell lines to explore the underlying mechanisms of HLX function during hematopoiesis. We show that HLX overexpression results in an aberrant proliferation of HSPCs and a myeloid differentiation block in both systems. We find that HLX exerts its biological function in hematopoiesis, at least in part, by direct control of electron transport chain (ETC) and PPARδ gene expression. Metabolic stress leads to an elevation of AMP-activated kinase (AMPK) levels and autophagy. Modulation of PPARδ signaling can rescue the hematopoietic phenotypes of HLX in both zebrafish and human cells, but has no obvious impact on AML cells. In contrast, AMPK inhibition reduces viability of AML cell lines, but minimally affects primary cells. This newly discovered link between HLX and metabolism could be a promising new avenue for treating hematological diseases.

HLX overexpression blocks zebrafish myeloid cell maturation.
To investigate the mechanisms underlying the role of HLX in promoting AML, we examined hematopoiesis in HLXoverexpressing zebrafish models. We crossed the Tg(fli1a: Gal4FF)ubs3 25 line with our Tg(UAS:HLX-GFP) to induce expression of human HLX (hHLX) in endothelial and hematopoietic cells and named these fish fli:hHLXOE. We chose to use human HLX in an effort to demonstrate conservation and translate our results into the human gene function. fli:hHLX overexpression led to increased specification of HSPCs at 36 h post fertilization (hpf) in the Aorta-Gonad-Mesonephros region as shown by runx1 whole-mount in situ hybridization (WISH) (Fig. 1a and Supplementary Fig. 1a). The increased number of HSPCs led to increased rag1 staining in the thymus at 96 hpf (Fig. 1b). WISH for the early myeloid marker pu.1 revealed that these transgenic fish presented an expansion of myeloid progenitors (Fig. 1c). We then asked whether HLX overexpression affects myeloid cell maturation. May-Grünwald-Giemsa staining showed that fli:hHLXOE embryos have a significantly larger proportion of immature myeloid cells (75.5%) when compared to their wild-type counterparts (35.3%) at 48 hpf (Fig. 1d). EdU staining revealed hyperproliferation of endothelial cells, which may be the underlying cause of the increased number of HSPCs (Fig. 1e). This enhanced proliferation does not induce apoptosis in fli:hHLXOE embryos, as shown by TUNEL assay (Supplementary Fig. 1b).
At 48 hpf most of the myeloid cells are derived from primitive/ prodefinitive and not definitive hematopoiesis. To verify that the differentiation block occurs in myeloid cells that arise from HSPCs we crossed Tg(Mmu.Runx1:GAL4) fish to Tg(UAS:HLX-GFP) and named the progeny Runx:hHLXOE. These fish express hHLX only in HSPCs. In this model we show that more HSPCs are specified at 26 hpf as indicated by runx1 staining, followed by modestly elevated c-myb staining and mRNA expression at 3 dpf (days post fertilization) ( Supplementary Fig. 1c, d). We then verified by May-Grünwald-Giemsa staining that at 5 dpf HLX overexpression in HSPCs leads to a strong myeloid differentiation block without affecting erythrocyte numbers (Fig. 1f). This result was verified by qPCR and WISH for a panel of mature myeloid markers and gata1 as a marker of erythroid differentiation ( Supplementary Fig. 1d, e).
Together, these results suggest that HLX overexpression results in increased numbers of HSPCs and blocks myeloid cell differentiation.
HLX is required for HSPC formation. To examine hematopoiesis in hlx1 knockdown animals, we generated hlx1 morphants (hlx1MO) using a previously published translational morpholino 26 . Inhibition of hlx1 translation in zebrafish embryos decreased the pool of HSPCs, as shown by runx1 and c-myb WISH at 36 hpf, respectively ( Fig. 1g and Supplementary Fig. 1a, f). To quantify the number of HSPCs in hlx1MO animals, we injected either the translational morpholino used in all experiments, or a splicing morpholino 26 in Tg(Runx:mCherry) fish 27 , a line with fluorescent HSPCs. The number of m-Cherry + cells (HSPCs) was significantly decreased in both types of morphants ( Supplementary Fig. 1g). WISH for rag1 showed that hlx1MO have fewer thymocytes at 96 hpf (Fig. 1g), when compared to the control embryos. To exclude the possibility that HSPC loss is due to arterial or vascular endothelial defects, we analyzed the expression of arterial (ephrinB2α) and endothelial (kdrl) markers in hlx1MO. In agreement with previous reports, deregulation of hlx1 affects the identity of stack cells, but does not seem to have severe effects on the cardinal vein ( Supplementary Fig. 1h) 26 . Additionally, TUNEL assays demonstrated that HSPC loss in hlx1MO is not caused by apoptosis, whereas EdU staining showed reduced proliferation of endothelial cells ( Supplementary Fig. 1b, and Fig. 1e, respectively).
Collectively, these data show that HLX regulates the formation of HSPCs.
HLX regulates genes involved in metabolism. To understand the mechanisms of HLX function in hematopoiesis, we performed RNA-Seq on FACS-sorted endothelial cells from fli:hHLXOE (fli: kaede) and hlx1MO (kdrl:GFP) embryos at 48 hpf. Compared with control embryos, we identified 2950 downregulated and 3419 upregulated genes that changed by more than two-fold (negative binomial test (NBT), P < 0.05) in fli:hHLXOE embryos (Fig. 2a, Supplementary Fig. 2a, and Supplementary Data 1). On the other hand, 942 genes were downregulated and 1162 genes upregulated over two-fold (NBT, P < 0.05) in hlx1MO animals (Fig. 2a, Supplementary Fig. 2a, and Supplementary Data 1). Seventy-nine (8%, hypergeometric test (hg.t), P < 0.01) of the downregulated genes in hlx1MO were inversely correlated in fli: hHLXOE. In total 869 genes were deregulated in both hlx1MO and fli:hHLXOE (hg.t, P < 2.27E-120). Next, to identify pathways 30   deregulated upon hHLX overexpression or hlx1 knockdown, we performed ingenuity pathway analysis (IPA), gene ontology, and gene set enrichment analysis (GSEA) and created networks using Cytoscape (Fig. 2b, Supplementary Fig. 2b, c, and Supplementary Data 1). These analyses showed that HLX is a pleiotropic transcription factor that regulates fundamental processes. Interestingly, the two canonical pathways most affected by hHLX overexpression were oxidative phosphorylation (OXPHOS) (right-tailed Fisher Exact Test (rtFET), P < 3.16E-23) and mitochondrial dysfunction (rtFET, P < 1.99E-21, Fig. 2b and Supplementary Data 1). GSEA also indicated that genes downregulated in fli:hHLXOE or deregulated in hlx1MO are associated with changes in mediators of OXPHOS ( Supplementary Fig. 2c). Multiple genes of the mitochondrial ETC belonged to the abovementioned categories and were downregulated in fli:hHLXOE embryos (Supplementary Data 1). Gene deregulation detected by RNA-Seq was confirmed by qPCR in fli:hHLXOE and Runx: hHLXOE or hlx1MO embryos ( Supplementary Fig. 2d-f). These results suggest that HLX regulates mitochondrial metabolic genes. This unexpected finding has important implications, as mitochondrial metabolism is essential for LT-HSCs stemness 28,29 and AML patients can present defects in mitochondrial metabolism 30 . Consistent with the transcriptional deregulation observed in the hlx1MO, we also detected differences in chromatin accessibility, by performing ATAC-Seq in endothelial (kdrl:GFP) cells sorted from control and hlx1MO embryos at 48 hpf. Using the MACS2 (version 2.1.0) bdgdiff command with the default settings we identified 16,409 peaks that were either lost or gained in hlx1MO, when compared to wild-type siblings ( Fig. 2c and Supplementary Data 2). Using Genomic Regions Enrichment of Annotations Tool (GREAT) analysis, we found that the majority of peaks were located between 5 and 500 kB of the transcription start site (TSS) and regulate a variety of processes (Supplementary Fig. 3a and Supplementary Data 2). We used a nominal cut-off of 25 kb from the TSS and assigned the differential peaks to 6431 genes (Supplementary Data 2). IPA analysis of these genes revealed that mitochondrial dysfunction (rtFET, P < 3.47E-09) and OXPHOS (rtFET, P < 8.13E-08) were among the upper enriched categories (Supplementary Data 2). To gain mechanistic insights we performed footprinting analysis 31 in control and hlx1MO datasets and obtained 54,588 and 50,471 footprints, respectively. Motif discovery revealed HOX motifs enriched in control-only footprints and AP-1 motifs in hlx1MO-only fooptrints ( Supplementary Fig. 3b). To determine whether these motifs were significantly differentially footprinted between the two datasets, we computed motif self-enrichments and cooccurrence enrichments from specific footprint populations over background occurrences computed from reciprocal datasets. This analysis revealed loss of HOX motifs and gain of AP-1 motifs in hlx1MO (co-occurrence enrichment computation z = 3.583 and z = 13.241, respectively) ( Supplementary Fig. 3c). Tn5 insertion profiles at these sites revealed diverging profiles at footprinted Hoxc9 and AP1 motifs (Student's t-test, Hoxc9 control specific P = 2.9884E-34, AP-1 hlx1MO specific P = 1.24262E-37) (Fig. 2d). Analysis of relative footprint occurrences to their cognate datasets also revealed that Hoxc9 and AP-1 motifs were more present in control and hlx1MO-specific footprints, respectively ( Supplementary Fig. 3d).
Together, these data demonstrate that Hlx1 regulates the expression of ETC and ppar genes in zebrafish endothelial cells and HSPCs.
PPARδ signaling rescues zebrafish hematopoietic phenotypes. The results above suggest that HLX regulates genes involved in metabolism. To investigate this further, we asked whether HLX modulation has any functional consequences for cell metabolism in vivo. We assessed OXPHOS by seahorse metabolic flux analysis, which measures oxygen consumption rate (OCR). These experiments revealed that there is a reduction in spare respiratory capacity in fli:hHLXOE embryos (Fig. 3a). Given the transcriptional deregulation of ETC genes upon hHLX overexpression, we measured mitochondrial membrane potential by using tetramethylrhodamine ethyl ester perchlorate (TMRM). TMRM is a dye that is sequestered in active mitochondria, and reflects the ability of cells to produce ATP. As expected, although the total mitochondrial mass was unchanged in fli:hHLXOE embryos, TMRM was significantly lower (Fig. 3b). In contrast, TMRM was higher in endothelial cells of hlx1MO, yet the total mitochondrial Fig. 1 hlx1 regulates hematopoietic stem cell formation and myeloid cell maturation in zebrafish. a-b Whole-mount in situ hybridization (WISH) for runx1 (a) and rag1 (b) in control or fli:hHLXOE zebrafish embryos at 36 or 96 hpf, respectively. Arrows indicate HSPCs. Numbers in the bottom right corner of panels indicate the number of zebrafish embryos with the indicated phenotype compared to the total number of zebrafish analyzed. Quantification of WISH was performed using FIJI software and statistical significance of three independent experiments in 12 zebrafish embryos was evaluated by Student's t-test, *P < 0.05, **P < 0.01 (mean + s.d.). c WISH for the early myeloid marker pu.1 in control or fli:hHLXOE zebrafish embryos at 48 hpf. Numbers and WISH quantification was performed as described above (n = 3, in total 12 fish, Student's t-test, *P < 0.05, mean + s.d.). d Zebrafish caudal hematopoietic tissue (CHT) smears stained with May-Grünwald-Giemsa stain in control or fli:hHLXOE embryos at 48 hpf. On the right, cell number counts of the indicated cell populations from 10 fish (n = 3, mean + s.d., Student's t-test, *P < 0.05). e EDU assay at 48 hpf in control, hlx1 morphants (hlx1MO) or fli:hHLXOE zebrafish cells (n = 2). f Zebrafish CHT smears stained with May-Grünwald-Giemsa stain in control or Runx:hHLXOE embryos at 5 dpf. On the right, cell number counts of the indicated cell populations from 10 fishes, in two independent experiments (mean + s.d., Student's t-test, **P < 0.01, ****P < 0.0001). g WISH for runx1 and rag1 in control or hlx1MO embryos at 36 and 96 hpf, respectively. h Representative images of Tg(Runx:mCherry) embryos at 72 hpf, injected or not with hlx1 morpholino and the indicated amounts of fli:hHLX construct. Numbers of mCherry-positive HSPCs from each embryo are represented in the graph (n = 10; mean + s.d., Student's t-test, *P < 0.05, ***P < 0.001) mass was slightly decreased (Fig. 3c). These data demonstrate that HLX overexpression affects mediators of OXPHOS and mitochondrial membrane potential in zebrafish in vivo.
As HLX regulates PPARδ gene transcription, a gene essential for lipid metabolism 33 , and in some systems mitochondrial biogenesis 34 , we assessed whether pharmacological modulation of PPARδ can rescue defects in mitochondrial membrane potential maintenance. Indeed, we were able to rescue TMRM levels by using a specific PPARδ antagonist (GSK3787) in fli:hHLXOE embryos, accompanied by an increase in the mitochondrial mass a  (Fig. 3d). This PPARδ antagonist also elevated the TMRM of the control cells, albeit to a lower extent. We then attempted to rescue the HLX-induced hematopoietic phenotypes by modulating the activity of PPARδ. First, we treated fli:hHLXOE zebrafish embryos with a PPARδ antagonist (GSK3787) and examined erythro-myeloid cells at 48 hpf. Pharmacological inhibition of PPARδ significantly reduced the number of immature myeloid cells and partially rescued the myeloid block (Fig. 3e). Conversely, a PPARδ agonist (L165,041) rescued HSPC loss in hlx1MO embryos (Fig. 3f). These results demonstrate that modulation of PPARδ rescues HLX-induced defects in HSPC formation and myeloid cell maturation.
HLX regulates ETC and PPAR genes in human cells. We next addressed the physiological relevance of the interplay between HLX and metabolic regulation in humans. From the database BloodSpot 35 and previously published data 36,37 , we confirmed that HLX is expressed in murine and human hematopoietic progenitor cells, but also in mature myeloid lineages (Supplementary Fig. 4a-c). Moreover, HLX is upregulated in AML patient samples, as it has been previously published 21 and according to information from cBioportal ( Supplementary  Fig. 4d). We next investigated whether AML patients show similar transcriptional signatures to those of hHLXOE zebrafish embryos. We obtained data generated by the TCGA Research Network (http://cancergenome.nih.gov/) from 193 AML patients. Normalized gene expression data of 156 AML samples were used for further analysis. We performed correlation analysis among all genes with significant expression (RSEM values >50) across all patients and the HLX gene. We then did GO and IPA analyses These genomic analyses in AML patients suggest that the role of HLX in metabolic regulation is conserved in humans. To identify genes that are directly regulated by HLX in human hematopoietic cells, we performed ChIP-Seq in two mammalian cell lines (chronic myelogenous leukemia CML: K562 and acute myeloid leukemia-AML HL60) overexpressing a FLAG-tagged version of hHLX ( Supplementary Fig. 5a). In both cell lines, the majority of HLX ChIP-Seq peaks fell in introns (48.7% in K562 and 50.2% in HL60) and 50-500 kb from the TSS ( Supplementary  Fig. 5b). After assigning the peaks located within 5 kb of the TSS to their corresponding genes, we found that HLX was bound to 2135 and 6838 genes in K562 and HL60 cells, respectively. 1689 genes were found in both cell lines (hg.t., P < 0) (Supplementary Fig. 5c and Supplementary Data 4). Notably, 421 (19%, hg.t., P < 7.044E-10) and 1431 (21%, hg.t., P < 1.088E-53) bound genes in K562 and HL60 cells, respectively, also showed differential ATAC-Seq peaks in zebrafish. GO, IPA, Cytoscape, and GREAT analyses revealed that, similar to zebrafish, HLX directly regulates basic cellular processes, including metabolic pathways (Supplementary Fig. 5d and Supplementary Data 4). IPA analyses in K562 and HL60 cells demonstrated that HLX regulates mitochondrial and PPAR/RXR pathways (Fig. 4a). Indeed, multiple ETC genes, but also PPARδ, were directly bound by HLX in either or both cell types on regions with characteristics of enhancers as indicated by H3K4me1 and other histone marks ( Fig. 4b and Supplementary Data 4). We confirmed these results by ChIP-qPCR on independent ChIP experiments with FLAGtagged or HA-tagged constructs (Fig. 4c). We also performed ChIP-qPCR for HLX target genes on K562 cells carrying an endogenous 3xTy tag on the HLX gene (Fig. 4d). Deletion of one of the bound regions in the vicinity of the ATP11b gene using CRISPR-Cas9 technology confirmed that HLX-bound regions can affect gene expression ( Supplementary Fig. 5e). Binding motifs for multiple transcription factors were identified in both K562 and HL60 cells ( Fig. 4e and Supplementary Data 4). Importantly, independent motif analysis uncovered motifs for homeobox containing factors such as HMBOX1 ( Fig. 4e and Supplementary Data 4). To unravel the chromatin landscape around the HLXbound genomic regions, we analyzed available K562 chromatin data from the ENCODE database 38 . HLX-bound genomic regions were located mostly on open chromatin and active enhancers, as indicated by co-localization with active histone marks (Fig. 4f).
Next, we asked whether HLX binding is associated with changes in gene expression in human hematopoietic cells. To this end, we knocked out HLX in K562 cells using CRISPR-Cas9 technology and performed RNA-Seq analysis ( Supplementary  Fig. 5f). We found that 1324 genes were downregulated and 600 genes were upregulated in HLX knockout cells (>2-fold change, NBT, P < 0.05, Fig. 4g and Supplementary Data 5). Two hundred eighty-four (hg.t., P < 9.874E-25) and 731 (hg.t., P < 4.460E-34) deregulated genes were directly bound by HLX in K562 and HL60 cells, respectively (Supplementary Data 4). Expression of some ETC genes bound by HLX was increased, whereas that of PPARδ was decreased, consistent with our results in zebrafish (Fig. 4h). Additionally, we performed qPCR for ETC and PPARδ genes in K562 cells stably overexpressing an inducible HLX construct. These experiments showed that increased HLX expression leads to high PPARδ expression with concomitant lower expression of ETC genes ( Supplementary Fig. 5g).
Here, we show that HLX directly regulates ETC and PPARδ gene transcription, and that this function is conserved from zebrafish to humans. These exciting results suggest that HLX controls myeloid differentiation, at least partly, through metabolic regulation. The combination of high PPARδ expression with low respiratory chain activity resembles the effects of pathways involved in controlling LT-HSC stemness 28,29,32 and could explain why HLX-overexpressing cells fail to terminally differentiate.
HLX overexpression leads to elevated AMPK and autophagy. HLX is particularly highly expressed in M4 and M5 AML leukemias 21 . We therefore selected THP1 cells deriving from a patient with AML (M5 subtype) to further explore the metabolic function of HLX. ChIP in HLX-overexpressing THP1 cells revealed that HLX binds to 5827 genes (Supplementary Data 4).
HLX was often found bound to intronic (42.24%) and intragenic (44.7%) regions, and specifically to ETC and PPARδ genes (Fig. 5a). HLX peaks were enriched for H3K27ac, a marker of active enhancers and promoters (Fig. 5b). Comparison of HLX ChIP in K562, HL60, and THP1 cells showed that 745 (12.8%, hg.t., P < 7.838E-46) and 1767 (30%, hg.t., P < 6.676E-18) of genes bound by HLX in THP1 cells were also bound in K562 and HL60 cells, respectively. To investigate whether the bound regions of HLX in THP1 cells represent open chromatin regions in HSCs and/or preleukemic and leukemic HSCs of AML patients, we used publicly available ATAC-Seq data 39    leukemic HSCs (Fig. 5c). Thus, it is possible that perturbation of HLX binding affects genes implicated in HSCs or leukemic transformation.
Consistent with our previous results, ETC genes were downregulated and PPARδ was upregulated at the protein level in THP1 cells (Fig. 5d). Moreover, maximal respiratory capacity was reduced, and lower levels of reactive oxygen species (ROS) were produced, in HLX-overexpressing THP1 cells (Fig. 5e, f)  THP1 cells cultured with 13 C-glucose. Indeed, we found increased incorporation of glucose-derived carbon in citric acid that can be used to produce fatty acids, which was also reflected in palmitic acid, fatty acid C18, and stearic acid (Supplementary Fig. 6).
To uncover the mechanisms downstream of ETC gene downregulation by HLX, we took a candidate approach and examined the AMPK pathway. It has recently been proposed that mitochondrial dysfunction can lead to AMPK activation 40 . Moreover, AMPK is a well-established sensor of metabolic stress and its activation results in elevated autophagy 41,42 . We found that AMPKα and phospho-AMPKα (p-AMPKα) are upregulated in HLX-overexpressing THP1 cells (Fig. 5d). Additionally, the protein levels of LC3-II, an autophagosome marker, are markedly increased in these cells upon chloroquine treatment (Fig. 5g).
Together, these results suggest that HLX overexpression in AML cells affects mitochondrial metabolism and fatty acid synthesis possibly through the upregulation of PPARδ gene expression. Additionally, HLX overexpression, possibly through downregulation of ETC genes, results in AMPK activation and autophagy.
HLX regulates the metabolic state of CD34 + human cells. Our findings could have important implications for patients with hematopoietic disorders, including AML. We therefore performed colony-forming unit (CFU) assays on human CD34 + HSPCs to measure the effects of HLX modulation in normal hematopoiesis. Consistent with our zebrafish results and with published data on mouse HSPCs 21 , HLX knockdown (sh-HLX) caused a significant reduction in the number of hematopoietic colonies, whereas HLX overexpression (CD34 + HLX) caused the opposite phenotype and large myeloid colonies (Fig. 6a, b and Supplementary Fig. 7a). Culturing CD34 + HLX in myeloid differentiation media resulted in a maturation block, as revealed by the accumulation of early granulocyte-monocyte progenitors (early GMPs) and the relatively reduced numbers of mature monocytes and granulocytes ( Fig. 6c and Supplementary Fig. 7b). To assess whether PPARδ and ETC genes are regulated by HLX in CD34 + cells, we performed RNA-Seq experiments upon HLX knockdown or overexpression. Due to the high variability in primary cells we considered all the genes that have at least 1.5fold change independently of P-value. The expression of many ETC genes was clearly upregulated upon sh-HLX and downregulated in CD34 + HLX cells ( Fig. 6d and Supplementary Data 5). Selected targets were validated with qPCR (Supplementary Fig. 7c). Thus, the function of HLX in hematopoiesis and its target genes are conserved in human primary hematopoietic cells.
To further determine whether the metabolic function of HLX is conserved in CD34 + cells, we measured OXPHOS. Similar to zebrafish, increased levels of HLX in human CD34 + cells led to a reduction in OCR, particularly spare respiratory capacity (Fig. 6e). Moreover, although HLX overexpression did not cause significant changes on the OXPHOS to extracellular acidification rate (ECAR, representative of glycolytic rate) ratio, HLX-overexpressing cells tended to have a lower ratio ( Supplementary Fig. 7d), suggesting a metabolic adaptation by engagement of glycolysis. TMRM staining shows that CD34 + HLX cells have decreased mitochondrial membrane potential, independently of mitochondrial mass (Fig. 6f). However, upon differentiation toward myeloid cells, CD34 + HLX cells exhibited significantly higher OXPHOS than control cells and a tendency to higher OCR/ECAR ratio ( Supplementary Fig. 7e, f).
Our results in THP1 cells showed that HLX overexpression is followed by AMPK activation. To determine whether AMPK activation affects myeloid maturation, we induced differentiation of human CD34 + HSPCs in the presence or absence of metformin, an AMPK activator that also blocks mitochondrial complex I thus mimicking the effect of HLX 43 . Notably, metformin induced a myeloid differentiation block in CD34 + cells (Fig. 6g). These results suggest that metabolic manipulation can indeed be the underlying reason for the hematopoietic phenotypes caused by HLX deregulation.
We next assessed whether pharmacological modulation of PPARδ activity rescues the myeloid differentiation phenotypes caused by HLX. Indeed, treatment with a PPARδ antagonist relieved the myeloid differentiation block in CD34 + HLX cells (Fig. 7a). Moreover, the inability of sh-HLX cells to form colonies in CFU assays was partially rescued by a PPARδ agonist (L165,041) (Fig. 7b). This agonist also rescued the increased mitochondrial membrane potential observed in sh-HLX cells (Fig. 7c). To investigate this further, we identified genes affected by HLX and potentially regulated by PPARδ using publicly available PPARδ-binding data in human macrophages 44 . We compared these data to our RNA-Seq from human CD34 + cells and found that 399 (hg.t., P < 0.006) deregulated genes upon HLX knockdown that can potentially be bound directly by PPARδ (Supplementary Data 6). IPA analysis on these genes showed involvement in FAO I (rtFET, P < 2.26 E-04) and AMPK signaling (rtFET, P < 5.24 E-03) (Supplementary Data 6).

AMPK inhibition causes reduced viability of AML cell lines.
To investigate the potential role of PPARδ and AMPK in promoting AML downstream of HLX, first, we analyzed the expression levels of HLX and PPARδ by qPCR in various AML cell lines and one CML cell line, K562. As expected, THP1, a M5 subtype leukemia, exhibited the highest levels of HLX expression 21 but also PPARδ expression (Fig. 8a). PPARδ protein was only detectable in THP1 cells (Fig. 8b). AMPKα and pAMPKα expression was detected in all cell lines without any noticeable differences (Fig. 8b). TMRM and autophagy were variable between cell lines (Fig. 8c, d). Next, we asked whether PPARδ and AMPK inhibitors could push the AML lines and/or K562 cells toward myeloid maturation or affect their viability. PPARδ antagonists (GSK3787) had no significant impact on either the viability or the differentiation of AML cell lines and K562 cells (Fig. 8e). However, AMPK inhibition with dorsomorphin significantly reduced the viability of all but one (HL60) AML cell lines tested and K562 cells (Fig. 8f). It is important to note that dorsomorphin had only a mild effect on the viability of CD34 + myeloid progenitor cell populations (Fig. 8f). Thus, we show here that AMPK inhibition decreases the viability of AML cell lines. Our study depicts HLX as a novel metabolic regulator in both normal and malignant cells (Fig. 8g).

Discussion
Recent evidence showing that the H2.0-like HLX is implicated in many malignancies highlights the importance of understanding the function and identifying the targets of this transcription factor 21,45,46 . In agreement with previous reports 21 , in the present study we demonstrate in zebrafish models that HLX affects myeloid differentiation, and we recapitulate these results in human HSPCs for the first time. We provide evidence that this regulation occurs, at least partly, through direct modulation of metabolic pathways by HLX. Specifically, we show that HLX directly regulates several metabolic genes, including PPARδ and genes of the mitochondrial ETC. In agreement with our results, it   was recently shown by Huang et al. that HLX controls a systematic switch from white to brown fat through metabolic gene regulation, including PPARs and genes that control mitochondrial biogenesis 24 . However, in that study HLX was found to positively regulate both mitochondrial biogenesis and PPARs, in contrast to our study where ETC genes are downregulated upon HLX overexpression. It will be interesting to study whether diverse HLX-interacting partners could account for these differences.
Recently, metabolism has emerged as a critical regulator of HSCs. LT-HSCs are quiescent and rely mostly on anaerobic glycolysis rather than OXPHOS 47,48 . A number of studies have shown that low mitochondrial activity is necessary to maintain the quiescent state and the self-renewal capacity of LT-HSCs and protect them from oxidative stress 28,[49][50][51][52][53] . The importance of reduced mitochondrial activity for HSC maintenance has also been demonstrated in human CD34 + HSPCs 54-56 . Antagonism of PPARγ signaling enhances glycolysis and leads to expansion of human HSPCs 57 . Moreover, recent evidence revealed that PPARδ, a regulator of fatty acid metabolism, plays an essential role in maintaining HSC stemness by regulating mitophagy and promoting HSC asymmetric cell divisions 32,58 . During differentiation, PPARδ signaling is downregulated leading to a gradual increase in mitochondrial mass and symmetric commitment of HSC daughter cells 32 . Thus, PPARδ, but also HLX that controls its expression, may constitute a metabolic switch for regulating HSC cell fate. It is interesting to speculate, based on our results, that HLX is a gatekeeper of HSC identity by maintaining their glycolytic state. We also show that HLX overexpression leads to low spare respiratory capacity, a characteristic of AML cells 59 . Both these results could be used in the future to better understand the implication of HLX in AML. However, since both HLX and PPARδ exert many functions besides metabolic regulation, we need to fully understand the precise mechanisms underlying our rescue experiments by PPARδ modulation. Interestingly, since PPARδ, like HLX, is overexpressed in a subset of M5 typemonoblastic AML cases 60 , it is conceivable that PPARδ inhibition could play a role in AML. Our results do not support this hypothesis, but further investigations should shed more light on a potential role of PPARδ in AML.
Also of note, we show that metformin, which blocks mitochondrial complex I and activates AMPK 43 , thereby mimicking the metabolic function of HLX, can affect myeloid differentiation. This finding proves that metabolic regulation can indeed be the direct mediator of HLX functions and, at least in part, causative for the phenotype. Metformin has been used as an anticancer therapy in many malignancies, including AML 61 . Based on our findings, it is pertinent to fully understand the role of metformin in physiological and pathological conditions. Finally, we found that HLX overexpression leads to activation of AMPK. AMPK does not seem to play a role in HSCs during homeostasis, transplantation, or under metabolic stress 62 , but protects leukemia initiating cells from metabolic stress 63 . This suggests AMPK as an ideal potential target for the treatment of leukemia without affecting normal cells. Indeed, we found that AMPK inhibition has a strong effect on the survival of AML cell lines. AMPK inhibition is also successful in eliminating MLLrearranged B-cell acute lymphoblastic leukemia 64 . However, other studies suggest that AMPK acts synergistically with mTORC1 and causes lethality in AML cells 65 . Further research with samples from human patients and specific mouse models are needed to clarify these discrepancies.
Our study points to differential requirements and regulatory mechanisms between normal and leukemic cells by the same transcription factor and identifies HLX as a new player in metabolic regulation in hematopoiesis.

Methods
Zebrafish maintenance. The zebrafish (D. rerio) strain Tübingen (Tü) and all zebrafish transgenic lines used in this study were maintained in the animal facility of the Max Planck Institute of Immunobiology and Epigenetics. The sample size for the animal experiments was chosen according to the following paper 66 . No animals were excluded from this study and no randomization was used. Only 1-5 dpf embryos were used in this study and sex was not determined at these stages. All animal experiments were performed in accordance with relevant guidelines and regulations, approved by the review committee of the Max Planck Institute of Immunobiology and Epigenetics and the Regierungspräsidium Freiburg, Germany (license Az 35-9185.81/G-14/95).
For the rescue experiments one-cell stage Tg(Runx:mCherry) 27 embryos were injected with 12.5 or 25 pg of fli:HLX or Runx:HLX constructs followed by injection of hlx1 morpholino as described above. Injected embryos were grown until 72 hpf, manually dechorionated and embedded in 1% low melting agarose, containing 0.04 mg/mL tricaine. Caudal hematopoietic tissue was imaged using Zeiss Apotome2 microscope at 10× magnification. Flow cytometry of mCherry-positive cells is described in the section "Preparation of zebrafish cells, flow cytometry, and cell sorting". WISH staining and analysis, constructs and generation of transgenic zebrafish lines are described in Supplementary Material and Methods.
Preparation of zebrafish cells, flow cytometry, and cell sorting. Embryos were incubated in 0.5 mg/mL Liberase TM (Roche) solution for 30 min at 37°C, then dissociated and resuspended in PBS-5% fetal bovine serum (FBS), and used for cell sorting, flow cytometry, seahorse assay, qPCR, and RNA-Seq experiments. Cell sorting was performed using Influx (BD Biosciences). For all experiments cellsorting purity was over 85%.
May-Grünwald-Giemsa staining of zebrafish blood. For CHT smears, 48 hpf (fli:hHLXOE) or 5 dpf (Runx:hHLXOE) embryos were placed in 0.9% NaCl with 0.04 mg/mL tricaine and the tails were isolated at the level of the cloaca/end of the yolk sac extension and incubated with Liberase TM (at 1:65 dilution in 0.9% NaCl, Roche) with 0.04 mg/mL tricaine for 20 min at 37°C. FBS was then added to a final concentration of 10%, to stop enzymatic digestion. The tails were triturated and then passed through a 40 µM mesh filter. The smears of dissociated cells were prepared by cytospin followed by May-Grünwald-Giemsa staining. Cells were imaged using Zeiss Axio Imager microscope with 100× objective. c Heatmap comparing HLX-bound regions in THP1 cells to ATAC-Seq regions of HSCs, pre-leukemic HSCs, and leukemic HSCs. d Western blot analysis of HA-HLX, PPARδ, AMPKα, and p-AMPKα, mitochondrial ETC complexes and Histone 3 (H3) as loading control. Cells were untreated or treated with doxycycline for 24 h to induce HLX expression. Representative immunoblots of at least three independent experiments. Quantifications were performed by FIJI software software and are shown below the blots as a ratio to H3. The five OXPHOS complexes are depicted as C-I to V. e Oxygen consumption rate (OCR) in control or HLX-overexpressing THP1 cells. Representative plot from three independent experiments (mean ± s.d.). f Bar graph depicting mitochondrial ROS production in control and THP1-HLX cells (n = 3, mean + s.d., Student's t-test, **P < 0.01). g Western blot analysis of LC3-II, HA, and H3 in control and THP1-HLX cells non-induced, induced with Doxycycline (Dox) with or without the addition of Chloroquine (CQ). Representative immunoblots of three independent experiments. Quantification was performed by Fiji software and is shown below the blot as a ratio to H3 TUNEL assay. Whole-mount TUNEL staining of developmentally staged control, morphant, and overexpression embryos was performed using the in situ cell death detection kit with fluorescein (Roche Applied Science, 11684795910). Embryos were then embedded in 1% low melting agarose and imaged with a Zeiss LSM780 confocal microscope and a 10× objective.
EdU labeling. Cell proliferation in zebrafish was assayed using the Click-iT EdU Alexa Fluor 647 Imaging Kit (Thermo Fisher Scientific). Briefly, zebrafish embryos at 48 hpf were dechorionated and incubated with Liberase TM as described in the section "Preparation of zebrafish cells, flow cytometry, and cell sorting".     Human CD34 + cells, isolated from the peripheral blood of granulocyte colonystimulating factor mobilized healthy volunteers, were purchased from the Fred Hutchinson Cancer Research Center. The cells were maintained as previously described 67 . Briefly cells were cultured in StemSpan SFEM (StemCell Technologies), supplemented with 2% P/S and cytokine mix: m-SCF 100 ng/mL, hFLT3 100 ng/mL, hIL-3 20 ng/mL, and hIL-6 20 ng/mL (expansion medium). Cells were kept at 1 × 10 5 -1 × 10 6 cells/mL densities. Lentiviral particle production. 293T cells were transfected with lentiviral plasmids (packaging vectors with GFP-HA or HLX-HA or sh-HLX or scrRNA) mix using the polyethylenimine (PEI) method. Briefly, 2 h before the transfection, fresh 2% FBS containing medium was added. For 9 cm culture dishes, 5 μg of total DNA mix (construct, pPAX2, and pMDG.2 were used at a ratio of 10:7.5:3, respectively) was diluted with plain DMEM (Gibco) up to 520 μL. 30 μL of 1 mg/mL PEI reagent was then added and the mix was vortexed shortly twice. The mix was incubated for 10 min at RT before adding drop wise to 293T cells. The medium was changed after 8 h. Supernatants containing lentiviral particles were collected at 48, 72, and 96 h post transfection and concentrated using Lenti-X concentrator (TAKARA). All lentiviral experiments were performed in S2 laboratory with permission from German authorities: Regierungspräsidium Tübingen (57-3/8817.40-020).
HLX knockdown: To generate shRNA against HLX (sh-HLX), we cloned annealed oligos (Supplementary Table 1) into the pLKO.1-TCR vector (a gift from David Root; Addgene plasmid #10878) digested with AgeI and EcoRI. As a control, a non-hairpin pLKO.1-TCR control (scrRNA) (a gift from David Root; Addgene plasmid #10879) 68 was used. For the generation of scrRNA and shHLX, human CD34 + cells were infected after 1-day expansion in expansion medium (see section "Cell line and primary cell maintenance"). Cells were placed on retronectin-coated (TAKARA) plates and transduced with concentrated virus at a multiplicity of infection of 5 in expansion medium. Cells (1 × 10 6 cells/mL; 1 mL per well in 6-well plates) were transduced three times using spinoculation (6 μg/mL polybrene, 800 × g, 90 min) at 6-12 h intervals for 2 consecutive days. Then cells were washed five times in PBS and transefered in expansion medium containing puromycin (1 μg/mL) for 3 more days.
CFU-C assays. The CFU-C assays were performed by plating 500 or 3000 CD34 + cells per plate in CFU-C media (R&D Systems, HSC003), according to the manufacturer's instructions. Colonies consisting of at least 40 cells were counted after 15 days at 37°C and 5% CO 2 . CFU-C colonies were counted blindly regarding control and experimental samples.  Detection of autophagic flux. Autophagy was measured using the Cyto-ID ® Autophagy detection kit (Enzo, ref. ENZ-51031), according to the manufacturer's instructions. Briefly, cells were grown at a density of 1 × 10 6 cells/mL for 16 h at 37°C, in the presence of DMSO or Chloroquine (60 µM). Cells were washed with PBS, pelleted at 1000 rpm for 5 min, and incubated with cyto-ID green reagent for 30 min at 37°C in the dark prior to flow cytometry analysis on LSRFortessa cell analyzer (BD Biosciences). Mean fluorescence intensity (MFI) was quantified using FlowJo software (Tree Star, Inc.).
Real-time qPCR. Total RNA was extracted from human or zebrafish cells using the RNA Clean & Concentrator-5 kit (Zymo Reasearch) kit or TRI Reagent, according to the manufacturer's instructions. cDNA was prepared with the SuperScript™ First-Strand Synthesis System for RT-PCR kit (Thermo Fisher Scientific). qPCR reactions were executed using Fast SYBR green Master Mix (Thermo Fisher Scientific) in a StepOnePlus Real-Time PCR machine (Applied Biosystem). Expression was plotted relative to RAB7 (for human cells) and cyyr1 (for zebrafish cells). All qPCR graphs in this study are representative of at least two independent experiments. qPCR primers can be found in Supplementary Table 2.
Metabolic assays. OCR and ECAR were measured using a 96-well XF or XFe extracellular flux analyzer (Seahorse Bioscience) in XF media (non-buffered RPMI medium 1640 containing 25 mM glucose, 2 mM glutamine, and 1 mM sodium pyruvate) containing the cytokine mix of expansion medium for CD34 + experiments or 10% FBS for zebrafish experiments. The measurements were performed under basal conditions and in response to 1 μM oligomycin, 1.5 μM FCCP, and 100 nM rotenone combined with 1 μM antimycin A.
Glucose tracing. For metabolic tracing THP1 cells were cultured in glucose-free RPMI media with dialyzed serum supplemented with 11 mM D-[U13C] glucose and 2% P/S for 24 h. Cells were washed with ice cold 0.9% w/v NaCl buffer and metabolites were extracted twice with hot ethanol (70%) and analyzed by GC mass spectrometry (GCMS). Fractional contribution from exogenous substrates was calculated as described previously 69 . Briefly, metabolite extracts were dried, resuspended in pyridine, and derivatized with methoxylamine and N-tert-Butyldimethylsilyl-N-methyltrifluoroacetamide with 1% tert-Butyldimethylchlorosilane before GCMS analysis on an Agilent 7890 GC with Agielnt 5977 MS. Peak areas of all possible labeling states were extracted for full-carbon-backbone fragments of selected metabolites to obtain mass distribution vectors (MDVs). MDVs were then corrected for natural abundance of heavy isotopes by constrained optimization of the linear equation where I denotes the measured fractional abundance of metabolite fragments of different labeling states, L denotes the correction matrix, and M denotes the corrected MDV 69 . The fractional contribution from exogenous substrates was then calculated as the weighted average of the MDV FC ¼ P n i¼0 i Á s i n ; where FC denotes the fractional contribution, i is the position in the MDV, s i is the value of the MDV at the position i, and n is the length of the MDV. All calculations were implemented in an in-house R script.
Mitochondrial membrane potential and mtDNA/nDNA measurement. Zebrafish cells or human cells were stained with tetramethylrhodamine, methyl ester (Image-iT™ TMRM Reagent, Invitrogen, ref. I34361; 50 nM) at 28 or 37°C, respectively, for 30 min, then washed with PBS-5% FBS and analyzed by flow cytometry with LSRFortessa or LSR II cell analyzers (BD Biosciences). MFI was quantified using FlowJo software (Tree Star, Inc.). To measure the ratio of mitochondrial and nuclear DNA (mtDNA/nDNA), genomic DNA was extracted from 10 5 cells using DNeasy Blood & Tissue Kit (Qiagen) and real-time qPCR was performed as described before 70 . Briefly, quantitative PCR reactions were assembled as follows: 2 μL of template DNA (3 ng/μL isolated DNA), 2 μL of mtDNA or nDNA target-specific primer pair (400 nM final concentration each), 12.5 μL SYBR Green PCR Master Mix, and 8.5 μL H 2 O in 1 well of the 96-well PCR plate. All qPCR experiments were performed in triplicate wells in three independent experiments. Primer sequences can be found in Supplementary Table 2.