Metabolite Patterns in Human Myeloid Hematopoiesis Result from Lineage-Dependent Active Metabolic Pathways

Assessment of hematotoxicity from environmental or xenobiotic compounds is of notable interest and is frequently assessed via the colony forming unit (CFU) assay. Identification of the mode of action of single compounds is of further interest, as this often enables transfer of results across different tissues and compounds. Metabolomics displays one promising approach for such identification, nevertheless, suitability with current protocols is restricted. Here, we combined a hematopoietic stem and progenitor cell (HSPC) expansion approach with distinct lineage differentiations, resulting in formation of erythrocytes, dendritic cells and neutrophils. We examined the unique combination of pathway activity in glycolysis, glutaminolysis, polyamine synthesis, fatty acid oxidation and synthesis, as well as glycerophospholipid and sphingolipid metabolism. We further assessed their interconnections and essentialness for each lineage formation. By this, we provide further insights into active metabolic pathways during the differentiation of HSPC into different lineages, enabling profound understanding of possible metabolic changes in each lineage caused by exogenous compounds.


Introduction
Assessment of toxic properties from environmental or xenobiotic compounds remains a major field of research. Mammalian laboratory animals are currently considered as the "gold standard" in toxicology [1]. However, as mammalian models are expensive and time-consuming, several in vitro models have been developed and applied [2][3][4]. Indeed, models based on primary human cells are currently believed to more accurately reflect in vivo responses towards selected compounds, even though their applicability is partially restricted by limited life span and the necessity for cell enables prediction of similar response patterns to other compounds with analogical modes of actions. Moreover, such relations may also enable prediction of response patterns across different tissues, leading to a better prediction of possible tissue-specific and toxic effects during drug development or the testing of xenobiotics.
Indeed, lineage-dependent regulatory involvement of single metabolic pathway activity during hematopoiesis is quite evident. Regulation of fatty acid oxidation (FAO) for instance, seems to be crucial for hematopoietic stem cell (HSC) maintenance, since blocking of FAO promotes HSC commitment [42]. However, autophagy-mediated generation of free fatty acids and subsequent degradation via FAO is crucial for neutrophil differentiation, indicating active FAO during differentiation of (at least) some lineages [43]. Furthermore, lymphocytes, neutrophils and macrophages utilize glutamine at high rates under catabolic conditions (e.g., sepsis), underlining the importance of glutaminolysis during HSPC differentiation [44]. Blocking glutaminolysis in erythropoietin (EPO)-stimulated HSPC, however, leads to a shift from erythroid commitment towards a myelomonocytic fate [45]. Therefore, modulation of glutaminolysis by xenobiotic compounds may also result in lineage-specific toxicity. Nevertheless, the assumption that glutaminolysis solely defines erythroid lineage commitment falls quite short, since it has been shown recently that blocking choline generation from phosphatidylcholine also impairs erythroid differentiation [46]. The role of phosphatidylcholine degradation within differentiation of other myeloid lineages, however, remains vague. In addition, several studies suggest a relation of polyamines with erythroid differentiation, their role in other lineages, however, again remains inconclusive [47][48][49]. Taken together, the essentialness of several different metabolic pathways during defined HSPC differentiation has already been shown for selected lineages. The general activities and interconnections between the different metabolic pathways, also within other lineages, however, still remains unclear. Therefore, a direct comparison of active metabolic pathways within different hematopoietic lineages is desirable in order to further elucidate the mode of action behind possible lineage-specific effects.
Here, we combined a known HSPC expansion approach with distinct lineage differentiations from the literature, resulting in formation of erythrocytes, dendritic cells (DC) and neutrophils. Due to the initial expansion step, large cell numbers can be generated with this approach, making it highly suitable for omics-based toxicity testing (e.g., demonstrated in [50]). Further assessment of metabolic and transcriptional changes during lineage formation resulted in unique and common metabolite sets, reflecting distinct metabolic changes in several interconnected pathways (namely glycolysis, glutaminolysis, polyamine synthesis, fatty acid oxidation and synthesis, as well as glycerophospholipid and sphingolipid metabolism). We further assessed the essentialness of glutaminolysis, polyamine synthesis and FAO for differentiation of each lineage, confirming the proposed activities. While several pathways were active in different lineages, interconnections between the distinct pathways were found to be unique for each lineage, while one of such interconnections was essential for erythrocytes. Taken together, we here established an HSPC differentiation model, suitable for metabolic toxicity screening and assessed the unique combination of active metabolic pathways, as well as their interconnections for each lineage, enabling the understanding of differing metabolic changes in myeloid hematopoiesis, caused by exogenous compounds.

Expansion Approach Preserves Restricted Myeloid-Lineage Potential
As MS-based quantification of metabolites typically requires a large amount of biological material, we applied an initial expansion step of CD34 + HSPC. The applied protocol led to a total expansion of alive cells by 286 ± 46-fold after 13 days ( Figure 1B) [51]. However, only around 45.65% ± 1.88% of total cells remained CD34 + positive (independent of CD38 expression) after expansion, also primitive markers as CD135 and CD45RA largely remained negative while a large part of the population indicated CD90 positive ( Figure 1A). Nevertheless, many typical lineage markers remained negative, only around 10.42% ± 2.04% were CD235a positive (Supplemental Figure S1A) [52]. Indeed, primitive HSC are often determined as CD34 + CD38 − , representing only a minor fraction of the here generated population [14]. However, as the CD34 + CD38 + population displayed a variety of further differentiated progenitors (e.g., common myeloid progenitors), still owning multilineage differentiation potential, the total amount of CD34 + cells (termed progenitor population) was assessed [14]. In brief, these results indicate that the differentiation of primitive progenitors occurs during expansion, a lack of most lineage markers, however, indicates a multilineage potential in the expanded population. population indicated CD90 positive ( Figure 1A). Nevertheless, many typical lineage markers remained negative, only around 10.42% ± 2.04% were CD235a positive (Supplemental Figure 1A) [52]. Indeed, primitive HSC are often determined as CD34 + CD38 -, representing only a minor fraction of the here generated population [14]. However, as the CD34 + CD38 + population displayed a variety of further differentiated progenitors (e.g., common myeloid progenitors), still owning multilineage differentiation potential, the total amount of CD34 + cells (termed progenitor population) was assessed [14]. In brief, these results indicate that the differentiation of primitive progenitors occurs during expansion, a lack of most lineage markers, however, indicates a multilineage potential in the expanded population. . Alive cell numbers were determined by trypan blue exclusion method, using a hemacytometer.
We therefore continued differentiation of the progenitor population into erythroid [53], megakaryocytic [54] and natural killer T cell lineages [55]. Erythroid differentiation could be confirmed ( Figure S1B) while lineage markers for megakaryocytic and natural killer T cell lineages remained negative ( Figure S1C,D). Assessment of further lineage markers, however, revealed that CD1a + dendritic cells ( Figure S1C) and CD66b + neutrophils ( Figure S1D) were formed as major populations. This observation is also supported by transcriptomic data, as several of the most extensively expressed genes could be matched to the afore mentioned lineages (Table S1). As HBB and TFRC are both erythroid, ELANE, S100-A8, PRTN3 and AZU1 are neutrophil markers, GPNMB is expressed by antigen-presenting cells (APC), MRC1 and C1QC by macrophages and DC, while EPX is expressed by eosinophils, the expression patterns indicate, that the main lineages formed are erythrocytes, DC as well as neutrophils [56][57][58][59][60]. Furthermore, gene ontology enrichment analysis of the generated populations also support this assumption, as most significantly upregulated biological processes include erythrocyte differentiation and homeostasis for the erythroid population, as well as biological processes related to immune response for the other populations (Table 1). We therefore continued differentiation of the progenitor population into erythroid [53], megakaryocytic [54] and natural killer T cell lineages [55]. Erythroid differentiation could be confirmed ( Figure S1B) while lineage markers for megakaryocytic and natural killer T cell lineages remained negative ( Figure S1C,D). Assessment of further lineage markers, however, revealed that CD1a + dendritic cells ( Figure S1C) and CD66b + neutrophils ( Figure S1D) were formed as major populations. This observation is also supported by transcriptomic data, as several of the most extensively expressed genes could be matched to the afore mentioned lineages (Table S1). As HBB and TFRC are both erythroid, ELANE, S100-A8, PRTN3 and AZU1 are neutrophil markers, GPNMB is expressed by antigen-presenting cells (APC), MRC1 and C1QC by macrophages and DC, while EPX is expressed by eosinophils, the expression patterns indicate, that the main lineages formed are erythrocytes, DC as well as neutrophils [56][57][58][59][60]. Furthermore, gene ontology enrichment analysis of the generated populations also support this assumption, as most significantly upregulated biological processes include erythrocyte differentiation and homeostasis for the erythroid population, as well as biological processes related to immune response for the other populations (Table 1). Cell activation involved in immune response 4.31 < 10 −9 Leukocyte activation 3.59 < 10 −9

Neutrophil Granulocytes
Defense response 4.09 < 10 −27 Immune response 3.47 < 10 −26 Immune system process 2.77 < 10 −24 Immune effector process 4.20 < 10 −22 Myeloid cell activation involved in immune response 5.98 < 10 −20 Taken together, these data demonstrate the multi-lineage potential of the generated progenitor population as it was possible to form erythroid, DC, neutrophil and eosinophil lineages. Since all populations reside in the myeloid branch of hematopoiesis, we assume that the progenitor population either consists of restricted myeloid progenitor cells or of a mixture of further differentiated progenitors. Indeed, transcriptomics data also indicate that other lineages are formed within each population, however, the main lineages formed are erythrocytes, DC as well as neutrophils, as verified via flow cytometry.

Myeloid Lineages Display Several Unique and Common Metabolic Patterns during Differentiation Accompanied by Changes in Pathway-Related Genes
Based on the successful identification of each main lineage formed, we performed global targeted metabolome profiling, testing for 514 metabolites in total. We totally identified 173 metabolites being altered from the progenitor population in the mature lineages ( Figure 2A). In this set, 81 metabolites were significant in erythrocytes, 99 metabolites in DC and 73 in neutrophils (Supplemental Tables S2-S4). While several metabolic changes seemed to be comparable between two or more populations, several unique changes were present (Figure 2A). By applying criteria for identification of the unique and common changes (as stated in the methods section), we found 17 unique metabolites changed in erythrocytes ( Figure 2A Table S6) and 46 unique metabolites in neutrophils ( Figure 2A highlighted with blue, Supplemental Table S7). When these unique metabolic changes were directly compared in all three lineages, a cluster for each lineage was clearly visible ( Figure 2B). Several metabolite changes in fact were found in common, with erythrocytes and DC displaying the largest overlap with 25 metabolites (Figure 2A highlighted with yellow, Supplemental Table S8). Common metabolites between the other combinations of populations, however, were rather rare ( Figure 2A highlighted with cyan, pink and black, Supplemental Tables S9-S11).  Table S13). (B) Unique metabolite patterns in each differentiated population. The mean log2 fold change, in comparison to the progenitor population, of the identified unique metabolites is displayed (n = 3). Biologically relevant, unique and common metabolites were identified as stated in the materials and methods section.
As we sought to identify the underlying changes on metabolic pathway levels, we analyzed the expression of enzymes, related to the relevant metabolites. Whole pathway activity is often regulated by a small number of key enzymes (e.g., hexokinase, phosphofructokinase and pyruvate kinase in glycolysis), therefore, whole pathway activity can often be derived from the expression of these in combination with metabolite concentrations [61]. Furthermore, as enzyme activity is tightly controlled via several mechanisms, including expression, we omitted FDR-correction of the p value in order to avoid an increase of the type II error (false negatives) [62]. By using this approach, we found several transcriptional changes, related to the corresponding metabolic pathways (Table 2). Table 2. Transcriptional changes in pathway-related genes in each population. Given is gene name, the corresponding metabolic pathway and the log2 fold change, compared to the progenitor population. Changes with a corresponding p value > 0.05 were considered not significant (n.s.).   Table S13). (B) Unique metabolite patterns in each differentiated population. The mean log 2 fold change, in comparison to the progenitor population, of the identified unique metabolites is displayed (n = 3). Biologically relevant, unique and common metabolites were identified as stated in the materials and methods section.

Gene
As we sought to identify the underlying changes on metabolic pathway levels, we analyzed the expression of enzymes, related to the relevant metabolites. Whole pathway activity is often regulated by a small number of key enzymes (e.g., hexokinase, phosphofructokinase and pyruvate kinase in glycolysis), therefore, whole pathway activity can often be derived from the expression of these in combination with metabolite concentrations [61]. Furthermore, as enzyme activity is tightly controlled via several mechanisms, including expression, we omitted FDR-correction of the p value in order to avoid an increase of the type II error (false negatives) [62]. By using this approach, we found several transcriptional changes, related to the corresponding metabolic pathways (Table 2). Table 2. Transcriptional changes in pathway-related genes in each population. Given is gene name, the corresponding metabolic pathway and the log 2 fold change, compared to the progenitor population. Changes with a corresponding p value > 0.05 were considered not significant (n.s.).

Gene
Metabolic

Myeloid Lineages Own a Higher Hexose Consumption with Different Fate
As the energy metabolism seems to play a crucial role during HSPC maintenance and differentiation, we focused on glycolysis, FAO and glutaminolysis, as these display the three major pathways for energy transformation [63]. Hexoses were found to be strongly reduced in all lineages (Supplemental Table S12), also expression of corresponding transmembrane transporters (SLC2A1, SLC2A5 and SLC2A6) was either upregulated or unchanged ( Table 2), indicating that lowered hexoses are related to raised catabolic reactions. This interpretation is further supported by the increased expression of HK3 in all lineages. Expression of PKLR, however, was upregulated in erythrocytes, while FBP1 was upregulated in DC and neutrophils. As PKLR catalyzes the formation from phosphoenolpyruvate to pyruvate, it can be concluded that erythrocytes own higher consumption of hexoses, subsequently used for the Krebs cycle. It has, however, been shown that nucleotide synthesis from glucose-6-phosphate (G6P) via the PPP is vital for erythroid differentiation [45]. However, we could not detect any difference in the expression of enzymes related to the PPP, indicating that erythrocytes use G6P under native conditions for further catabolic breakdown via glycolysis, as well as for nucleotide biosynthesis via PPP. FBP1 catalyzes the back conversion of fructose 1,6-bisphosphate to fructose 6-phosphate (F6P), acting as the rate-limiting enzyme in gluconeogenesis. As F6P readily can be converted to G6P, we assume that both lineages primarily use glucose for nucleotide biosynthesis and ROS production from NADPH via PPP, which is also supported by findings of other groups [64,65].

Combined Fatty Acid Generation and Respective Fate Is Unique for Each Myeloid Lineage
Several acylcarnitines were raised in erythrocytes and DC, 18 of which could be defined as changes common to both lineages (Supplemental Tables S3, S4 and S9). Only three acylcarnitines, however, were increased in the neutrophil lineage, with acetylcarnitine concentration being uniquely affected (Supplemental Tables S5 and S8). Acylcarnitines serve as a carnitine shuttle for fatty acids through the mitochondrial membranes, while even-chain acylcarnitines and hydroxylated species result from β-oxidation, odd-chain species from α-oxidation and dicarboxylated species from ω-oxidation [66][67][68][69]. It is well-known, that acylcarnitines can also play a role in maintaining free CoA availability, whilst also reflecting corresponding acyl-CoA concentrations within the cell [67]. As the activity of FAO was demonstrated in HSC and during neutrophil differentiation, it is reasonable to assume that both possess enhanced FAO [43,63]. We therefore assume, that raised acylcarnitines indicate a lower FAO, which is also supported by the fact that species, resulting from αand ω-oxidation, are raised as well. As αand ω-oxidation only take place at substantial amounts if β-oxidation is disturbed, this further indicates reduced FAO [70,71].
Conversely, we found expression of CPT1a lowered in the neutrophil lineage ( Table 2). As transesterification by CPT1a is one of the rate-limiting steps in FAO, this decreased expression seemed contradictory to our assumptions. However, usage of etomoxir as an irreversible inhibitor of CPT1 only affected neutrophil differentiation ( Figure 3A,D and Figure 4B,C), confirming our hypothesis of active FAO in neutrophils. Surprisingly, addition of etomoxir during neutrophil differentiation led to a higher expression of ELANE ( Figure 3D). As etomoxir effectively reduced the proportion of CD11b + cells during neutrophil differentiation, our results seemed rather conflicting [43]. However, ELANE and CD11b are both expressed at different stages during neutrophil maturation, while ELANE is expressed in promyelocytes and myelocytes; CD11b expression starts in metamyelocytes [72,73]. Furthermore, expression of ELANE is inversely correlated to the maturation of neutrophils [72,74]. Therefore, we concluded that the higher expression of ELANE in etomoxir-treated populations indicated an enrichment of promyelocytes and myelocytes. In line with this assumption, the expression of S100A8, correlating with neutrophil maturation, was lowered by increasing amounts of etomoxir ( Figure 3E) [74]. Thus, FAO seemed to be essential for neutrophil maturation, as already previously observed [43]. Nevertheless, formation of committed progenitors seemed to be independent of FAO. This interpretation, however, awaits further confirmation by future studies.
Interestingly, C16-OH and C16:1-OH acylcarnitines were uniquely raised in DC, but not in erythrocytes ( Figure 2B, Supplemental Table S8). In fact, intracellular fatty acids (FA) and corresponding acyl-CoA species can result from catabolic reactions via lipolysis, as well as from anabolic reactions via FA synthesis. We found expression of FASN and ELOVL6 uniquely stimulated in the erythrocyte lineage, while expression of LIPA was enhanced in DC and neutrophils. Furthermore, we found ACSL1 elevated in DC and THEM5 in neutrophils, also SCD was elevated in erythrocytes and DC.
Summarized, these results indicate that erythrocytes perform higher FA synthesis, while both, DC and neutrophils, perform higher lipolysis of cholesteryl esters (CE) and triglycerides (TG). FAs are further catabolized via FAO in neutrophils, while erythrocytes and DC seem to utilize FAs for anabolic processes. Erythrocytes, however, preferably elongate C16:0 and C16:1 FAs (Figure 5), which leads to diminished concentrations of these FAs, which, in turn, is reflected by corresponding acyl-CoAs and thus acylcarnitines. Interestingly, C16-OH and C16:1-OH acylcarnitines were uniquely raised in DC, but not in erythrocytes ( Figure 2B, Supplemental Table S8). In fact, intracellular fatty acids (FA) and corresponding acyl-CoA species can result from catabolic reactions via lipolysis, as well as from anabolic reactions via FA synthesis. We found expression of FASN and ELOVL6 uniquely stimulated in the erythrocyte lineage, while expression of LIPA was enhanced in DC and neutrophils. Furthermore, we found ACSL1 elevated in DC and THEM5 in neutrophils, also SCD was elevated in erythrocytes and DC.
Summarized, these results indicate that erythrocytes perform higher FA synthesis, while both, DC and neutrophils, perform higher lipolysis of cholesteryl esters (CE) and triglycerides (TG). FAs are further catabolized via FAO in neutrophils, while erythrocytes and DC seem to utilize FAs for anabolic processes. Erythrocytes, however, preferably elongate C16:0 and C16:1 FAs (Figure 5 (E) Effect of etomoxir in the presence of glutamine on S100A8/RPLP0 expression of neutrophils. Data are represented as mean ± SD (n = 3). Relative gene expression was calculated as stated in the methods section. Significant changes were assessed by one-way ANOVA (n.s. not significant; * p < 0.05; ** p < 0.01; *** p < 0.001).

Glutaminolysis Crosstalk with FAO and Its Interconnection to Polyamine Synthesis in Erythrocytes
In context to glutamine metabolism, we found uniquely lowered levels of the products Glu, Asn, Asp, Ala and Pro in neutrophils ( Figures 2B and 5). As it is known that glutaminolysis is critical to erythroid differentiation, we assume that both erythrocytes and DC perform glutaminolysis [45]. This assumption was further supported by the reduced GPNMB expression in DC upon glutamine depletion ( Figure 3C). In erythrocytes, however, glutamine depletion only reduced the cell expansion rate, HBB expression remained unaffected ( Figure 4H,I). Glutaminolysis, therefore, seems to be crucial for DC differentiation, however, in erythrocytes, it appears to be beneficial for cell expansion, but not for maturation. In neutrophils, however, we assume that high ATP production by FAO concomitantly leads to reduced free phosphate levels, which in turn lead to lowered glutaminase activity [75]. rate, HBB expression remained unaffected ( Figure 4H,I). Glutaminolysis, therefore, seems to be crucial for DC differentiation, however, in erythrocytes, it appears to be beneficial for cell expansion, but not for maturation. In neutrophils, however, we assume that high ATP production by FAO concomitantly leads to reduced free phosphate levels, which in turn lead to lowered glutaminase activity [75]. Interestingly, we found ME1 and IDH1 raised in DC, both resulting in NADPH generation from Krebs cycle intermediates (Table 2) [35]. Cross-regulation between glycolysis and FAO via NADPH and acetyl-CoA is well-known, furthermore, pyruvate is generated via ME1 [76]. We therefore assume, that in DC, FAO is mainly negatively regulated by glutaminolysis and not glycolysis ( Figure 5). 0.01; *** p < 0.001).
Interestingly, we found ME1 and IDH1 raised in DC, both resulting in NADPH generation from Krebs cycle intermediates (Table 2) [35]. Cross-regulation between glycolysis and FAO via NADPH and acetyl-CoA is well-known, furthermore, pyruvate is generated via ME1 [76]. We therefore assume, that in DC, FAO is mainly negatively regulated by glutaminolysis and not glycolysis ( Figure 5). Furthermore, we found putrescine concentrations uniquely increased in erythroid differentiation ( Figure 2B, Supplemental Table S6), while downstream metabolites spermidine and spermine remained unaffected. Orn was elevated in erythrocytes whilst not fulfilling the criteria as a unique metabolite, whereas Arg was lowered in all lineages ( Figure 2B, Supplemental Table S3,S8 and 12). We further found ODC expression reduced in DC and neutrophils, while SAT1 was raised ( Table 2). Polyamine levels in fact have been reported to be relatively high in erythrocytes, confirming our observations [78]. This pattern indicates that the anabolism of polyamines is stable during erythroid differentiation, resulting in raised putrescine levels, whilst being diminished during DC and neutrophil differentiation. Several reasons may account for this: it is well-known that polyamines are important in regulating gene expression, as they are mostly bound to RNA and are essential for cell cycle progression [49]. As such they are found at higher concentrations in high proliferating cells. Further, polyamines have been found to inhibit the trans-bilayer movement of phospholipids and reduce lipoperoxidation in erythrocytes [79][80][81]. The importance of polyamines during Furthermore, we found putrescine concentrations uniquely increased in erythroid differentiation ( Figure 2B, Supplemental Table S6), while downstream metabolites spermidine and spermine remained unaffected. Orn was elevated in erythrocytes whilst not fulfilling the criteria as a unique metabolite, whereas Arg was lowered in all lineages ( Figure 2B, Supplemental Tables S3, S8 and S12). We further found ODC expression reduced in DC and neutrophils, while SAT1 was raised ( Table 2). Polyamine levels in fact have been reported to be relatively high in erythrocytes, confirming our observations [78]. This pattern indicates that the anabolism of polyamines is stable during erythroid differentiation, resulting in raised putrescine levels, whilst being diminished during DC and neutrophil differentiation. Several reasons may account for this: it is well-known that polyamines are important in regulating gene expression, as they are mostly bound to RNA and are essential for cell cycle progression [49]. As such they are found at higher concentrations in high proliferating cells. Further, polyamines have been found to inhibit the trans-bilayer movement of phospholipids and reduce lipoperoxidation in erythrocytes [79][80][81]. The importance of polyamines during erythropoiesis was also previously demonstrated by other research groups [78,82]. In line with this, inhibition of polyamine synthesis by blockage of ODC via difluormethyl-ornithine (DFMO) reduced erythroid expansion rates and HBB expression ( Figure 4B,C). DC and neutrophils, in contrast, were not affected by ODC inhibition (Figure 3A,D). Therefore, polyamine synthesis seems solely crucial for erythropoiesis, at least in the myeloid branch of hematopoiesis. As yet, however, it is not exactly clear which of the stated functions of polyamines is eventually crucial for erythropoiesis; we assume that all of them might be crucial for different stages during erythropoiesis.
Interestingly, we found the expression of OAT solely increased during erythropoiesis (Table 2). OAT catalyzes the generation of Glu and glutamate-5-semialdehyde from Orn and aKG and vice versa [83]. Under most conditions, OAT uses Orn for anabolism of Glu and, therefore, belongs to the "glutamate crossway". Under certain conditions, however, OAT can switch to Orn synthesis. As Glu remained unchanged and Orn was raised during erythropoiesis, we, however, conclude that the favored reaction is Glu anabolism. In line with this, it has been shown that erythropoiesis is sustained during depletion of extracellular Gln, but is dependent on the intracellular synthesis from Glu under these conditions [45].
In order to evaluate the influence of OAT on erythropoiesis, as well as the direction of the conversion, we used 5-fluoromethylornithine (5-FMO), a specific inhibitor of OAT [84]. Surprisingly, addition of 5-FMO led to a reduced expression of HBB independent of Gln ( Figure 4E,G), whilst the expansion rate was only affected in the presence of Gln ( Figure 4D,F). We therefore propose that OAT works in both directions, depending on the conditions; in presence of Gln OAT catalyzes the conversion of Glu to Orn, subsequently fueling polyamine synthesis, whilst under conditions with low Gln, OAT prefers the conversion of Orn to Glu, leading to constant levels of Gln for nucleotide biosynthesis ( Figure 5) [45]. The homeostasis of polyamine synthesis and Gln levels is evident, as inhibition of ODC by DFMO leads, independent of Gln levels, to diminished HBB expression ( Figure 4C,G). Additionally, inhibition of OAT in the absence of Gln also reduced HBB expression to similar levels as DFMO, indicating that polyamine synthesis alone is not sufficient for erythropoiesis ( Figure 4G). Importantly, we found that inhibition of OAT or ODC in the absence of Gln did not influence GPNMB expression of DC ( Figure 3B), also expression of ELANE by neutrophils remained unaffected by 5-FMO ( Figure 3D). Therefore, OAT mediated homeostasis of polyamine synthesis and Gln seems to be solely crucial during erythropoiesis.

PLA2G15, PLA2G16 and Fatty Acid Levels Are Three Major Regulators of Unique Lipidomic Changes
As several uniquely changed metabolites were glycerophospholipids (GPL), we also focused on GLP metabolism. GLP need to be further subdivided in acyl-acyl species (acyl glycerophospholipids, aGPL) and acyl-alkyl species (ether lipids, EL), as the former are involved in glycerophospholipid metabolism and the latter are involved in ether lipid metabolism. Interestingly, the different GPL classes (e.g., phosphatidylcholine (PC), phosphatidylethanolamine (PE) and phosphatidylserine (PS)) followed the same trend in all lineages (not shown). However, erythrocytes displayed increased aGPL levels, DC showed lowered aGPL levels and higher EL levels, while neutrophils exhibited reduced aGPL levels, in comparison to the progenitor population (Figure 2 A,B, Supplemental Tables S5-S7). Furthermore, percentual content of EL was elevated in DC and neutrophils ( Figure 4A), as also reported previously [85].
We found several changes in the expression of enzymes, related to glycerophospholipid and ether lipid metabolism (Table 2); however, only PLA2G15, PLA2G16, LPCAT3, PISD, PHOSPHO1 and LPIN1 were affected distinctively in the corresponding populations. Furthermore, the activity of PLBD1, PLD3 and PISD has been reported as negligible, which was also reflected by the corresponding metabolic changes [86][87][88][89]. As we found enhanced de novo FA synthesis and elongation in erythrocytes, as well as stimulated lipolysis in DC and neutrophils, the unique metabolite patterns apparently partially reflected available FA levels, as longer aGPLs were raised in erythrocytes. In DC and neutrophils, aGPLs apparently are degraded by PLA2G15 and PLA2G16 [90,91]. The available FA levels are further reflected by ELs, as FAR1/2, the major regulator of EL biosynthesis, uses C16 and C18 acyl-CoA species as substrates [92,93]. Since C16 FA are further elongated in erythrocytes, this results in limited EL synthesis due to limited substrate availability and thus unchanged EL levels. In neutrophils, PLA2G16 additionally reduces EL biosynthesis by inducing peroxisomal dysfunction, which results in the lack of finding of uniquely raised EL, despite their higher percentual content [91].
In line with this, we found higher levels of C16, C18 and C18:1 ceramide in DC ( Figure 2B). Furthermore, expression of SPTSSB, well-known to stimulate serine palmitoyltransferase (SPT) activity, was raised in DC and erythrocytes [94]. As SPT uses C18-CoA substrates, we conclude that de novo ceramide synthesis is enhanced in DC, due to higher availability of sphingosine, resulting from higher C18-CoA availability.
Taken together, our data strongly indicate that parts of the cellular lipidome analyzed here are all regulated by FA levels, in combination with the expression of PLA2G15 or PLA2G16, resulting in the described unique metabolic signatures.

Discussion
Cellular metabolism is an essential feature of living organisms; various metabolites are involved in specialized cellular tasks, ultimately defining the function of each cell. In the context of hematopoietic stem cells, it is fully appreciated that the activity of metabolic pathways not only regulate quiescence, but also regulate commitment [63,95]. Inhibition of FAO, for example, leads to symmetric commitment of HSCs, however, neutrophil differentiation is also disturbed by inhibition of FAO. [42,43] The regulatory role of FAO, therefore, remains controversial to some extent. Furthermore, several studies have shown the importance of various metabolic pathways on differentiation of several lineages [45,46,78,96]. The lineage-specific activity of these metabolic pathways, however, remains vague, a direct metabolic comparison of distinct lineages is still missing.
The total cellular metabolism plays a fundamental role in the cellular response towards different stimuli. In the context of hematopoiesis, several different compounds are known to exert distinct effects on the different blood cell populations [15,40,[97][98][99]. The actual mode of action underlying this lineage-specific hematotoxicity, however, often remains inconclusive. Indeed, some known hematotoxic compounds (e.g., the benzene metabolite trans, trans-muconaldehyde) were shown to exert (some of) it's toxic effects by blockage of gap junction intercellular communication via cross-linking of connexin43, rendering these unidentifiable via metabolomic screening approaches [100]. Nevertheless, involvement of the total cellular metabolism in lineage-specific effects is quite likely, rendering fundamental knowledge of the different active and inactive metabolic pathways essential, in order to understand such lineage-specific effects.
We, therefore, established an HSPC differentiation model, suitable for omics-based screening approaches, and performed metabolic and transcriptomic comparison of three myeloid lineages, regarding to the initial progenitor population. Generated populations, in fact, were not purified further, potentially leading to overlapping metabolic signatures. However, by identifying distinctively altered metabolites, we found several lineage-related active metabolic pathways, some of which have also been identified previously (e.g., polyamine synthesis in erythrocytes [78] and FAO in neutrophils [43]). Presence of other lineages is even of advantage from a toxicological point of view, as toxic effects on these can also be identified simultaneously. The corresponding mode of action, then may be identified within the dominant lineage, using omics techniques. We found that conclusions made from metabolomic and transcriptomic data are in good agreement with the results of verifying inhibitor experiments, indicating the validity of the here described approach.
We could further demonstrate that not only FAO, but also the combination of fatty acid generation and their fate is unique for each lineage. Inhibition of FAO did impair neutrophil maturation, but not formation of committed progenitors, supplementing current knowledge about FAO dependence of HSC commitment. Of note, our data also suggest that glycolysis and glutaminolysis have a substantial impact on FAO, leading to differential usage of fatty acids in each lineage. We could further demonstrate a unique and essential connection of glutamine metabolism to polyamine synthesis in erythrocytes and identify the key players in lipid remodeling during myeloid lineage commitment, namely acyl-CoA availability, PLA2G15 and PLA2G16.
Our results thus connect several crucial active metabolic pathways, identified in single myeloid lineages, while also displaying their unique combination for each lineage, as well as their inactivity in the other lineages. Direct usability for hematotoxicity studies is evident, as several xenobiotic compounds are known to selectively interact with specific metabolic pathways [101][102][103]. Likewise, elucidation of effects by endogenous compounds on hematopoiesis can also be achieved by usage of the here presented model, which may ultimately result in new therapy approaches. Our results also enable improved interpretation of metabolic alterations, as observed by different compounds [104][105][106].
Taken together, the parent paper adds significant knowledge to the understanding of differences in metabolism of distinct myeloid lineages for future studies and enables transferability of results obtained with this model to other tissues. It will further contribute to an understanding of the differential impacts of environmental or xenobiotic compounds on hematopoiesis and thus adds essential knowledge for identifying phenotype-modulating metabolites.

Human CD34 + Cell Culture
CD34 + cells were thawed according to the vendors protocol and expanded in Stempro-34 SFM media containing 2 mM L-glutamine, 100 ng/mL SCF, 100 ng/mLFlt-3L, 50 ng/mL IL-6, 40 ng/mL TPO, 1 µM SR-1, 0.1 µM BML-210 and 0.2 mM valproic acid for 13 days [51]. Until day 8, each second day 5 mL fresh media was added to the culture. From day 9 to 13, cells were diluted 1:1 each day by addition of fresh media. Erythroid differentiation was performed in Stempro-34 SFM media containing 2 mM L-glutamine, 100 ng/mL SCF, 100 ng/mL Flt-3L, 50 ng/mL EPO and 20 ng/mL IL-3 for a total of 6 days [53]. Cells were diluted 2:3 on day 4 by addition of fresh media. DC differentiation was done in Stempro-34 SFM media containing 2 mM L-glutamine, 10 ng/mL SCF, 10 ng/mL TPO and 10 ng/mL IL-3 for a total of 11 days. Neutrophil differentiation was done in IMDM containing 10% FBS, 50 ng/mL SCF and 50 ng/mL IL-15 for a total of 10 days. Dendritic cells and neutrophils were diluted 2:3 on day 3 and 7 by addition of fresh media. Different inhibitors were added as indicated at the beginning of each differentiation and held constant during the experiment, the added volume did not exceed 0.1% (v/v).

Total RNA Isolation and RNA-Seq
Total RNA isolation was performed using the NucleoSpin RNA XS Kit (Macherey-Nagel GmbH & Co. KG). Purification of poly-A containing mRNA molecules followed by mRNA fragmentation, random primed cDNA synthesis and single read 50 bp sequencing, as well as transcriptome alignment and determination of differential expression levels were done by Eurofins GATC Biotech GmbH (Constance, Germany) [107].

LC/MS-based Metabolomics
Metabolite measurement was performed using a 4000 QTRAP mass spectrometer (ABI Sciex), connected to a NexteraXR HPLC (Shimadzu). Cell pellets were washed twice with ice-cold PBS and the pellets were resuspended at 1 × 10 7 cells/ml in −20 • C cold ethanol containing 15% (v/v) 10 mM K 2 PO 4 , pH 7.5, sonicated for 3 min and frozen in liquid nitrogen for 30 seconds. The sonication-freeze cycle was repeated twice, afterwards, samples were centrifuged at 31,500 rcf at 2 • C for 5 min and the supernatant was stored at −80 • C until measurement. The metabolome analyses were carried out using the AbsoluteIDQ p180 Kit (Biocrates Life Science AG), according to the manufacturer's instructions. Analysis of additional lipids, covering 162 glycerophospholipids, 33 sphingomyelins and 131 ceramides was done by Biocrates Life Sciences AG (Innsbruck, Austria).

Bioinformatic Analysis
Statistical analysis of metabolomics data using an empirical Bayes approach was done by using the statistical software R [108] and the bioconductor package limma [109]. Metabolite concentrations were log 2 transformed prior to statistical analysis, resulting "-Infinitive" values from metabolites below the LOD were removed from the data frame. Absolute log 2 fold changes (logFC) above 1, compared to the progenitor population, with a corresponding FDR-adjusted p value below 0.05 where considered relevant. From these, unique metabolites were defined by a ∆logFC > 1 to both other lineages. As ∆logFC, we defined the absolute difference in the logFCs from each lineage compared to the progenitor population. Common metabolites of two lineages were defined as owning a ∆logFC < 1, whilst both owning a ∆logFC compared to third lineage > 1, from the relevant set of metabolites. Common metabolites of all lineages were defined as ∆ logFC < 1 between all lineages from the relevant set of metabolites. Gene Ontology enrichment analysis (Overrepresentation Test) was done by using the PANTHER Software (http://pantherdb.org), using the Fisher's Exact test type in combination with Holm adjusted p value [110,111]. Significant changes were assessed by a one-way ANOVA, where a p value smaller than 0.05 was considered as significant. For metabolic pathway visualization, using KEGG pathways as template, the software PathViso [77,112] was used.

Data Availability
RNA sequencing datasets generated are available at GEO under the accession number GSE129993. A list of identified significant altered metabolites can be found in the supplementary materials.