Serine Biosynthesis Is a Metabolic Vulnerability in FLT3-ITD–Driven Acute Myeloid Leukemia

Recent large-scale genome sequencing studies have revealed the remarkable heterogeneity that underpins acute myeloid leukemia (AML; refs. 1–4). Although new therapies for AML have recently been approved (5), the backbone of induction therapy in newly diagnosed patients remains a high-dose chemotherapy regimen of the cytosine analogue cytarabine, in conjunction with an anthracycline such as daunorubicin. Long-term survival is relatively poor, with a five-year survival rate of approximately 30% (6).

Activating mutations in receptor tyrosine kinases (RTK), including FMS-like tyrosine kinase 3 (FLT3), are frequently observed in AML and are associated with dismal prognosis (7, 8). FLT3 is primarily expressed on hematopoietic progenitor cells, and during early hematopoiesis coordinates a ligand-dependent signaling cascade that regulates the proliferation and maturation of the progenitor pool (9). FLT3 internal tandem duplication (ITD), the most common type of FLT3 mutation, promotes constitutive FLT3 tyrosine kinase activity and hyperactivation of downstream signaling pathways, including STAT5, PI3K/mTOR, and MAPK pathways (10–13). A number of FLT3 inhibitors have been developed and are in clinical trials. First-generation inhibitors, such as midostaurin (currently approved as first-line therapy in patients with AML with FLT3-ITD mutations), sorafenib, and lestaurtinib, were developed as broad-spectrum kinase inhibitors. The inability of these agents to induce durable clinical responses resulted in the development of highly specific second-generation inhibitors (quizartinib and gilteritinib) with potent binding affinity for FLT3. Unfortunately, the efficacy of FLT3 inhibitors is limited by the rapid development of acquired resistance (14).

Extensive reprogramming of cellular metabolism is required to fulfill the increased energetic and biosynthetic demands associated with unrestrained proliferation (15). Metabolic reprogramming events can also introduce new vulnerabilities that can be exploited for cancer therapy (16, 17). In line with this notion, recent studies have demonstrated that FLT3-mutant AML cells are dependent on glycolysis and the pentose–phosphate pathway (18, 19). Moreover, glutamine metabolism has been shown to promote resistance to FLT3-targeted therapy (20, 21). Emerging evidence also suggests that many cancer cells are highly dependent on the nonessential amino acid serine (22, 23). Serine is a major source of one-carbon units to the folate cycle and therefore supports numerous metabolic pathways including nucleotide synthesis (24). The functional importance of serine metabolism in FLT3-mutant AML has not been defined.

Herein, we developed a physiologically relevant mouse model of MLL-rearranged, FLT3-ITD–driven AML and used a multiomics approach to identify exploitable vulnerabilities in this disease. We demonstrate that FLT3-ITD reprograms serine biosynthesis and uptake in an mTORC1-ATF4–dependent manner and show that FLT3-ITD AML cells are selectively sensitive to inhibition of serine biosynthesis. Importantly, this metabolic dependency can be exploited to sensitize FLT3-ITD AML cells to cytotoxic chemotherapy in vitro and in vivo. Collectively, these data reveal key insights into metabolic reprogramming events driven by FLT3 mutations in AML and reveal a novel combinatorial therapeutic strategy to enhance the efficacy of standard-of-care chemotherapy in this aggressive subtype of AML.

Although FLT3-ITD mutations play a critical role in the pathogenesis of AML, FLT3-ITD alone is insufficient to induce leukemic transformation (25–27). We therefore generated a genetically engineered mouse model of doxycycline-inducible FLT3-ITD expression in MLL-AF9–rearranged AML (ref. 28; Fig. 1A). Mice injected with hematopoietic progenitor cells coexpressing MLL-AF9 and doxycycline-inducible FLT3-ITD (denoted as MLL-AF9/iFLT3-ITD) developed fully penetrant disease, detectable seven days after transplantation via luciferase bioluminescence imaging (Fig. 1B and C). In contrast, detectable disease was observed only 35 days after mice were transplanted with MLL-AF9–expressing hemopoietic progenitor cells, although all transplanted mice did eventually succumb to disease (Fig. 1B and C). MLL-AF9/iFLT3-ITD cells expressing the dsRed reporter were harvested from bone marrow of mice at endpoint and used for subsequent experiments. We confirmed harvested cells faithfully recapitulated AML via immunophenotyping and histologic analysis (Supplementary Fig. S1A).

To investigate the consequences of FLT3-ITD loss in vitro, MLL-AF9/iFLT3-ITD cells were propagated in doxycycline-free culture medium for 24 hours (denoted MLL-AF9/iFLT3-ITD–OFF). Doxycycline withdrawal rapidly abolished FLT3-ITD expression and resulted in decreased phosphorylation of the canonical FLT3 target STAT5 (Fig. 1D). Importantly, FLT3-ITD depletion induced a time-dependent decline in cell viability, as determined by Annexin V staining (Fig. 1E; Supplementary Fig. S1B), immunoblot analysis of PARP/caspase-3 cleavage (Fig. 1F), and reduction in dsRed expression (Supplementary Fig. S1C). To determine if loss of FLT3-ITD expression impairs in vivo disease progression, secondary NSG recipient mice (with no prior irradiation) were injected with MLL-AF9/iFLT3-ITD cells and placed on doxycycline water and chow for three days prior to removal (Supplementary Fig. S1D). Doxycycline withdrawal (and thus depletion of FLT3-ITD) dramatically reduced tumor burden (Fig. 1G) and enhanced overall survival (Fig. 1H). Depletion of FLT3-ITD following removal of doxycycline from leukemia-bearing mice was confirmed in bone marrow, spleen, and peripheral blood using dsRed as a marker of transgene expression (Supplementary Fig. S1E). Collectively, these data demonstrate that MLL-AF9 and FLT3-ITD cooperate to drive onset and progression of AML, and genetic depletion of FLT3-ITD alone is sufficient to induce apoptosis of MLL-AF9/iFLT3-ITD cells and extend overall survival in mice transplanted with these leukemias.

To elucidate the pathways involved in FLT3-ITD signaling in AML, we performed 3′ RNA sequencing (RNA-seq) comparing MLL-AF9/iFLT3-ITD–OFF cells cultured for 24 or 48 hours in the absence of doxycycline with MLL-AF9/iFLT3-ITD–ON cells maintained in doxycycline-supplemented medium (Fig. 2A). We identified a total of 1,009 differentially expressed genes (DEG), of which 575 were downregulated and 434 upregulated (P < 0.05 and –1 > logFC > 1) 24 hours post FLT3-ITD depletion (Supplementary Table S1). We further identified a total of 1,972 DEGs (1,057 downregulated, 915 upregulated, P < 0.05 and –1 > logFC > 1) 48 hours post FLT3-ITD depletion (Supplementary Fig. S2A; Supplementary Table S2). Interestingly, among the most suppressed genes 24 hours post FLT3-ITD depletion were genes encoding components of the de novo serine biosynthesis pathway (Phgdh and Psat1), neutral amino acid transporters involved in serine and glycine uptake (Slc1a4, Slc1a5, and Slc6a9), the mitochondrial serine transporter (Sfxn1), and one-carbon metabolism (Mthfd2, Shmt1, and Shmt2; Fig. 2B). Indeed, terms that defined gene sets involved in serine biosynthesis, one-carbon metabolism, and nucleotide metabolism were significantly suppressed following depletion of FLT3-ITD in MLL-AF9/iFLT3-ITD cells (Fig. 2C), and gene set enrichment analysis (GSEA) further showed marked suppression of the KEGG pathway gene sets “glycine, serine, and threonine metabolism,” and “one-carbon pool by folate,” consistent with Gene Ontology (GO) term analysis and individual gene changes observed (Fig. 2D). Examination of gene-expression changes 48 hours post FLT3-ITD depletion generally showed exacerbation of loss of mRNA in all of the genes examined (Supplementary Fig. S2B).

To confirm that the transcriptional response to FLT3-ITD depletion was not secondary to changes in cell viability, FLT3-ITD was depleted for 24 hours in MLL-AF9/iFLT3-ITD cells cultured in the presence or absence of the pan-caspase inhibitor QVD-OPh. QVD-OPh treatment reduced apoptosis (Supplementary Fig. S2C) but did not affect changes in Phgdh, Psat1, Slc1a4, and Slc1a5 expression induced upon FLT3-ITD depletion, confirming repression of these genes to not be a consequence of apoptosis that occurs following depletion of FLT3-ITD (Supplementary Fig. S2D). Consistent with our RNA-seq data (Fig. 2E), immunoblot analysis confirmed loss of Phgdh and Psat1 expression 24 hours following FLT3-ITD depletion (Fig. 2F). It has been acknowledged that the use of tetracyclines, including doxycycline, can affect mitochondrial translation and cellular metabolism (29, 30). To discount the possibility that differential gene expression was a confounding result of doxycycline withdrawal, we performed RNA-seq in our previously published Tet-inducible MLL-AF9/NRAS^G12D AML leukemias (31). Analysis of RNA-seq data at 24 hours post MLL-AF9 depletion revealed minimal overlap with our FLT3-ITD depletion data. Importantly, many of the genes involved in serine biosynthesis and one-carbon metabolism that respond to FLT3-ITD depletion were not differentially expressed in this system (Supplementary Fig. S2E; Supplementary Table S3). A complementary leukemia model, wherein doxycycline withdrawal results in depletion of MLL-AF9 (referred to as iMLL-AF9/FLT3-ITD), revealed that loss of MLL-AF9 did not affect the expression of Phgdh, Psat1, Slc1a4, or Slc1a5 (Supplementary Fig. S2F). These data provide important insights into the divergent molecular pathways regulated by FLT3-ITD and MLL-AF9. Our data indicate a role for FLT3-ITD in the transcriptional regulation of important metabolic processes, including serine and one-carbon/folate metabolism, whereas MLL-AF9 transcriptionally regulates a well-characterized network of genes involved in cellular differentiation (32). These observations are consistent with the divergent biological consequences of FLT3-ITD and MLL-AF9 depletion, namely, induction of apoptosis (Fig. 1E) and myeloid differentiation, respectively (33).

To determine if our in vitro results were conserved in vivo, mice transplanted with MLL-AF9/iFLT3-ITD leukemias were divided into two cohorts, one with and one without doxycycline supplementation for 48 hours (Fig. 2G). Leukemias were subsequently harvested from bone marrow for 3′-RNA-seq. Global transcriptome analysis showed a high degree of correlation between DEGs derived from 24-hour in vitro and 48-hour in vivo FLT3-ITD depletion (Fig. 2H; Supplementary Table S4), with strong overlap in the direction and magnitude of DEGs observed (Fig. 2I). Finally, to determine if acute FLT3-ITD depletion in human AML cell lines phenocopies genetic FLT3-ITD depletion in our murine model, we treated MV4-11 cells (which harbor FLT3-ITD and an MLL translocation) with quizartinib, a potent FLT3 inhibitor (34, 35), for 24 hours and performed 3′ RNA-seq. Strikingly, PHGDH, PSAT1, SLC1A4, and SLC1A5 were among the most highly suppressed genes upon FLT3-ITD inhibition in this human leukemia cell line (Supplementary Fig. S2G; Supplementary Table S5).

Although cellular serine requirements are largely fulfilled via uptake of dietary serine, cells retain the capacity to generate serine endogenously via the redirection of glycolytic intermediates into the de novo serine biosynthesis pathway (36). PHGDH, the first and rate-limiting enzyme of the de novo serine biosynthesis pathway, catalyzes oxidation of the glycolytic intermediate 3-phosphoglycerate (3-PG) to 3-phosphohydroxypyruvate (3-PHP) in a NAD⁺-dependent manner (Fig. 2J). 3-PHP is transaminated to 3-phosphoserine (3-PS) by PSAT1 in a reaction that utilizes glutamate-derived nitrogen and produces α-ketoglutarate (α-KG). Finally, 3-PS undergoes phosphate ester hydrolysis by PSPH to yield serine. Serine is essential for the biosynthesis of proteins and other metabolites that sustain cellular proliferation, including nucleotides (23). Consistent with the observed changes in the expression of genes involved in serine/glycine uptake and de novo serine biosynthesis, a dramatic reduction in intracellular serine and glycine abundance was observed when MV4-11 cells were treated with quizartinib (Fig. 2K). To specifically demonstrate that de novo serine biosynthesis was suppressed upon FLT3-ITD inhibition, we examined the fate of glucose-derived serine in MV4-11 cells and FLT3 wild-type (FLT3-WT) OCI-AML3 cells treated with quizartinib for 24 hours. FLT3-ITD inhibition dramatically reduced the incorporation of U-¹³C-glucose into serine (Fig. 2L) and glycine (Fig. 2M) in MV4-11 cells. This subsequently reduced ¹³C incorporation, via serine, into the antioxidant glutathione (GSH; Fig. 2N) and ATP (Fig. 2O). Importantly, quizartinib treatment of FLT3-WT OCI-AML3 cells did not alter ¹³C incorporation into any of these metabolites (Supplementary Fig. S2H–S2K). Collectively, these data suggest that FLT3-ITD regulates serine uptake, de novo serine biosynthesis, and one-carbon metabolism.

To determine the functional role of serine metabolism in FLT3-ITD–driven human AML, we generated MV4-11, MOLM-13 (MLL-rearranged, FLT3-ITD expressing), and OCI-AML3 AML cell lines in which PHGDH was deleted using CRISPR/Cas9 gene editing (Supplementary Fig. S3A). A competition assay in which PHGDH-deleted cells were cocultured with WT cells and proliferation was monitored over 14 days was then performed (Fig. 3A). Deletion of PHGDH in MV4-11 and MOLM-13 cells, using two independent single-guide RNAs (sgRNA), resulted in a loss of representation of these cells over time in vitro. In contrast, deletion of PHGDH in FLT3-WT OCI-AML3 cells had minimal effect (Fig. 3B). To address the broader implications of these findings, we performed an in vivo competition assay, in which a fixed 50:50 ratio of parental Cas9-mCherry cells was cotransplanted with cells expressing Cas9-mCherry and GFP-tagged sgRNAs targeting PHGDH into NSG recipients. The percentage of splenic GFP-positive cells was subsequently assessed at the ethical endpoint (Fig. 3C; Supplementary Fig. S3B). Depletion of PHGDH in MV4-11 and MOLM-13 cells resulted in marked loss of representation in vivo relative to nontargeting scramble controls (Fig. 3D). Importantly, PHGDH depletion in FLT3-WT OCI-AML3 cells did not affect representation relative to nontargeting controls (Fig. 3D). Collectively, these data provide strong evidence that inactivation of the de novo serine biosynthesis pathway in vitro and in vivo is selectively detrimental to the maintenance of FLT3-ITD–mutant but not FLT3-WT AMLs, even in contexts where exogenous serine/glycine is available.

Given our results showing that transporters involved in serine and glycine uptake are regulated by FLT3-ITD (Fig. 2B), we next sought to determine the contribution of exogenous serine/glycine to AML maintenance. Withdrawal of exogenous serine and glycine from RPMI medium suppressed the proliferation of leukemia cells expressing WT FLT3 or FLT3-ITD with minimal effects on cell viability (Supplementary Fig. S3C). However, serine/glycine withdrawal triggered apoptosis in PHGDH-depleted FLT3-ITD mutant cell lines (MV4-11 and MOLM-13; Fig. 3E; Supplementary Fig. S3D). Although PHGDH deletion exacerbated the antiproliferative impact of serine/glycine withdrawal in FLT3-WT cells (OCI-AML3), the apoptotic effect in these cells was significantly less than that observed in MV4-11 and MOLM-13 cells (Supplementary Fig. S3E). Previous studies have shown that serine, but not glycine, supports one-carbon metabolism and proliferation of cancer cells (37). Therefore, a rescue experiment was performed to determine if serine alone could reverse the loss of viability observed in MV4-11 and MOLM-13 cells following PHGDH depletion and serine/glycine withdrawal. Serine supplementation at three concentrations (∼286 μmol/L, the concentration of serine in RPMI culture medium; ∼150 μmol/L, the concentration of serine found in human serum; ref. 38; and 71.5 μmol/L, to mimic a low serine state) was sufficient to reverse the apoptotic phenotype (Fig. 3F), suggesting serine is essential for the survival of FLT3-ITD mutant AMLs. Collectively, these experiments illustrate that, while exogenous serine is uniformly required for the maximal proliferation of AML cells, depletion of de novo and exogenous serine is exquisitely lethal to FLT3-ITD mutant AMLs.

To determine the effect of pharmacologic inhibition of PHGDH in the context of AML, an allosteric small-molecule inhibitor, WQ-2101 (39), was applied across a panel of FLT3-ITD mutant and FLT3 WT human AML cell lines. Potent induction of apoptosis was observed in cell lines harboring FLT3-ITD mutations (MV4–11, MOLM-13, and PL-21) at low micromolar ranges of WQ-2101, compared with FLT3 WT cell lines (OCI-AML3, Kasumi-1, HL-60; Fig. 3G). WQ-2101 did not induce apoptosis in PHGDH-depleted MOLM-13 cells (Supplementary Fig. S3F), thereby confirming the on-target specificity of the compound. Additionally, a profound antiproliferative effect was observed in FLT3-ITD mutant cells exposed to sublethal concentrations of WQ-2101 (Fig. 3H; Supplementary Fig. S3G). Collectively, these data provide genetic and pharmacologic evidence that disruption of the de novo serine biosynthesis pathway is detrimental to the maintenance of FLT3-ITD–mutant AMLs in vitro and in vivo.

We next sought to identify the molecular pathways downstream of FLT3-ITD that regulate expression of genes involved in the de novo serine biosynthesis pathway and amino acid transporters involved in serine/glycine uptake. Differentially repressed genes from quizartinib-treated MV4-11 cells, from the RNA-seq data set described in Fig. 2, were compared with differentially repressed genes from MLL-AF9/iFLT3-ITD cells wherein iFLT3-ITD had been depleted for 24 hours (Fig. 4A). A total of 84 genes overlapped in both data sets and included PHGDH, PSAT1, SLC1A4, SLC1A5, SLC6A9, SHMT2, and MTHFD2 (Supplementary Table S6). Consistent with the observation that genetic depletion of FLT3-ITD from MLL-AF9/iFLT3-ITD cells resulted in reduced expression of PHGDH (Fig. 2E), a similar reduction in PHGDH protein expression was observed following quizartinib treatment of MV4-11 cells (Supplementary Fig. S4A). To identify potential transcription factors mediating FLT3-ITD–dependent gene expression, the i-cisTarget motif analysis platform was used to interrogate the 84 overlapping genes (40). Strikingly, an ATF4 binding motif was identified as the most highly enriched (Fig. 4B). Indeed, ATF4 and several canonical ATF4 target genes (ASNS, DDIT3, and CHAC1) were transcriptionally repressed following pharmacologic inhibition of FLT3-ITD in MV4-11 cells (Fig. 4C), as well as in MLL-AF9/iFLT3-ITD cells 24 hours post FLT3-ITD depletion (Supplementary Fig. S4B) with complete loss of ATF4 expression observed at the protein level (Fig. 4D). ATF4 is a stress-induced master transcriptional regulator of amino acid metabolism and has previously been linked with regulation of serine metabolism (41). Importantly, ATF4 knockdown in MV4-11 cells resulted in a reduction in PHGDH and PSAT1 expression, indicating ATF4 regulates the de novo serine biosynthesis pathway in FLT3-ITD mutant AML (Fig. 4E). GSEA of the transcriptional data set from MV4-11 cells treated with quizartinib revealed mTORC1 signaling to be among the most suppressed gene sets (Fig. 4F), a finding that was recapitulated in MLL-AF9/iFLT3-ITD cells 24 hours post FLT3-ITD depletion (Supplementary Fig. S4C). The link between FLT3-ITD and mTORC1 has been previously studied (42), and mTORC1 has been shown to regulate both the stability of ATF4 mRNA transcripts and ATF4 translation through three upstream open reading frames (uORF) in the 5′ UTR of ATF4 (43). Consistent with these observations, genetic or pharmacologic inhibition of FLT3-ITD reduced phosphorylation of ribosomal protein S6 (RPS6), a canonical downstream target of mTORC1 (Supplementary Fig. S4D). Moreover, treatment of MV4-11 cells with the mTORC1-specific inhibitor everolimus reduced ATF4 and PHGDH expression (Fig. 4G).

To determine how ATF4 modulates de novo serine biosynthesis and serine uptake, chromatin immunoprecipitation sequencing (ChIP-seq) was performed in MV4-11 and OCI-AML3 cells using an anti-ATF4 antibody. A 24-hour treatment of MV4-11 cells with quizartinib resulted in global loss of ATF4 occupancy across the genome (Fig. 4H and I), whereas ATF4 occupancy was not substantially altered in quizartinib-treated FLT3-WT OCI-AML3 cells (Fig. 4H and I). Peak calling analysis of ATF4 ChIP-seq data and subsequent annotation of peaks to genes in DMSO-treated MV4-11 cells identified 11,869 unique genes (5 kb either side of TSS), while in quizartinib-treated MV4-11 cells this was reduced to 367 genes. Both conditions maintained enrichment for the ATF4 binding motif (Supplementary Fig. S4E). Ranking of all genes according to their ATF4 occupancy in DMSO-treated MV4-11 cells revealed that ATF4 binding was enriched at key components of the de novo serine biosynthesis pathway (PHGDH and PSAT1) and serine transporters (SLC1A4 and SLC1A5), suggesting these genes are particularly strongly bound by ATF4 (Fig. 4J). The ATF4 peaks for PHGDH, PSAT1, SLC1A4, and SLC1A5 were almost completely eliminated in quizartinib-treated MV4-11 cells (Supplementary Fig. S4F). Quizartinib treatment had negligible effects on ATF4 expression in OCI-AML3 cells relative to MV4-11 cells (Supplementary Fig. S4G).

To determine target genes directly transcriptionally regulated by ATF4, the RNA-seq data from Fig. 4A were integrated with the ChIP-seq data from Fig. 4H and I. This analysis yielded 64 genes that were most highly enriched for ATF4 binding and were significantly repressed following treatment of MV4-11 cells with quizartinib (Fig. 4K; Supplementary Table S7). GO term analysis revealed a recurrent amino acid transmembrane transporter activity signature that included the L-serine pathway, as well as a purine nucleotide binding signature (Fig. 4L). To confirm that FLT3-ITD functionally promotes transcriptional activity at genes involved in serine biosynthesis and serine uptake, ChIP-seq using an anti-RNA polymerase II (Pol II) antibody was performed in MV4-11 and OCI-AML3 cells cultured in the presence or absence of quizartinib for 24 hours. In all instances, MV4-11 cells demonstrated greater Pol II occupancy in transcribed regions of PHGDH, PSAT1, SLC1A4, and SLC1A5 genes than corresponding regions in OCI-AML3 cells (Fig. 4M). Strikingly, quizartinib treatment almost completely ablated Pol II occupancy at these regions in MV4-11 cells, suggesting FLT3-ITD inhibition effectively silences active transcription of these genes. Importantly, quizartinib treatment did not induce global reduction in Pol II occupancy (Supplementary Fig. S4H). Finally, to directly link ATF4 to transcriptional regulation of PHGDH, PSAT1, SLC1A4, and SLC1A5, we performed RT-qPCR analysis on ATF4-depleted MV4-11 cells and confirmed reduced mRNA expression (Supplementary Fig. S4I). Collectively, these data suggest that FLT3-ITD regulates de novo serine biosynthesis and serine uptake in an ATF4-dependent manner.

We next sought to investigate the downstream metabolic consequences of both FLT3-ITD and PHGDH inhibition. Although quizartinib treatment of MV4-11 cells greatly reduced intracellular serine and glycine pools (Fig. 2K), inhibition of de novo serine biosynthesis with WQ-2101 had little impact on global serine/glycine abundance (Fig. 5A). Critically, however, serine/glycine levels in cells exposed to WQ-2101 and deprived of exogenous serine/glycine largely phenocopied the results seen following treatment of MV4-11 cells with quizartinib, providing further evidence that FLT3-ITD inhibition suppresses both de novo serine biosynthesis and serine uptake (Fig. 5A). Importantly, the abundance of other nonessential (Supplementary Fig. S5A) and essential (Supplementary Fig. S5B) amino acids remained largely unaltered, demonstrating that pharmacologic inhibition of FLT3-ITD or PHGDH did not induce widespread changes in amino acid abundance. Quizartinib treatment induced changes in metabolites associated with glycolysis, namely, increased glucose-6-phosphate and fructose-6-phosphate, coupled with decreased fructose-1,6-bisphosphate (Supplementary Fig. S5C), suggesting a block in glycolytic metabolism. Although this phenomenon is not necessarily related to specific effects on serine metabolism, it is indicative of disruption of central carbon metabolism as previously described (18, 19). Furthermore, FLT3-ITD inhibition with quizartinib or inhibition of PHGDH with WQ-2101 resulted in dramatic alterations in the steady-state abundance of TCA cycle intermediates (Supplementary Fig. S5D). Of note, PHGDH inhibition significantly increased the abundance of citrate/isocitrate and succinate, consistent with previous reports (44). To functionally examine serine uptake dynamics in AML cells, MV4-11, MOLM-13, and OCI-AML3 cells were pretreated with quizartinib or WQ-2101 for 24 hours before the uptake of tritium-radiolabeled L-[³H(G)]-serine was measured. FLT3-ITD inhibition potently reduced serine uptake in FLT3-ITD–driven MV4-11 and MOLM-13 cells, whereas no change in serine uptake was observed in FLT3-WT OCI-AML3 cells (Fig. 5B).

To examine the relative importance of metabolic pathways perturbed upon FLT3-ITD inhibition on AML survival, a series of metabolite rescue experiments were performed. Supplementation with the glycolytic intermediates pyruvate and lactate had little impact on quizartinib-induced changes in MV4-11 cell viability (Fig. 5C). In contrast, supplementation with glycine or the one-carbon/folate intermediate formate conferred a small but significant protection against quizartinib treatment (Fig. 5C). Supplementation with serine or hypoxanthine, a purine nucleobase derivative that undergoes conversion into the purine nucleotide inosine monophosphate (IMP), was sufficient to significantly reduce the apoptotic effects of quizartinib treatment on MV4-11 cells (Fig. 5C). Indeed, the steady-state abundance of the purine nucleotide guanosine triphosphate (GTP) was significantly reduced upon quizartinib treatment, whereas, interestingly, the abundance of the pyrimidine nucleotide cytosine triphosphate (CTP) increased in response to quizartinib treatment (Fig. 5D). Given that nucleotide imbalance is genotoxic to proliferating cells (45, 46), we hypothesized that suppression of one-carbon/folate cycle activity and thus purine biosynthesis might contribute to the cytotoxicity-associated with FLT3-ITD inhibition. Supplementation of purine nucleosides, but not pyrimidine nucleosides or a cocktail of purine and pyrimidine nucleosides, significantly reversed the apoptotic effects of quizartinib treatment (Fig. 5E). Of note, increasing the exogenous supply of serine (from 286 to 572 μmol/L) largely phenocopied the effects of purine supplementation (Fig. 5E), further supporting the idea that serine contributes to purine biosynthesis in FLT3-ITD AML. Given that purine biosynthesis in particular has been shown to be heavily dependent on the mTORC1–ATF4 axis (47, 48), we examined the impact of the mTORC1-specific inhibitor everolimus on U-¹³C-glucose incorporation into purine nucleotide species. Everolimus reduced ¹³C incorporation into ATP in MV4-11 cells, but not OCI-AML3 cells, supporting the notion that downstream of FLT3-ITD, mTORC1 directly contributes to purine biosynthesis (Supplementary Fig. S5E).

To determine if our findings could be recapitulated in primary human samples, a genetic signature encompassing genes involved in de novo serine biosynthesis, the one-carbon/folate cycle, and key rate-limiting enzymes of the de novo purine synthesis pathway was devised (ref. 47; see Supplementary Materials and Methods). As FLT3-mutant AMLs have been demonstrated to upregulate FLT3 mRNA expression (49), stratification of The Cancer Genome Atlas (TCGA) LAML data set by FLT3 mutation status showed a greater, and significant, positive correlation between this gene signature and FLT3 mRNA expression in FLT3-mutant patient samples (Fig. 5F). Furthermore, examination of the BEAT-AML data set of 473 FLT3-ITD versus FLT3 WT leukemias (4) verified PHGDH is more highly expressed in FLT3-ITD–mutant tumor cells (Fig. 5G). Collectively, these data are consistent with the notion that FLT3-ITD inhibition reduces global serine availability and induces nucleotide insufficiency by suppressing both de novo serine biosynthesis pathway activity and serine uptake.

Having demonstrated that PHGDH inhibition is a metabolic vulnerability in FLT3-ITD–mutant AML, strategies by which this could be exploited for the treatment of AML were explored. Cytarabine, the standard-of-care chemotherapeutic agent used as first-line therapy in patients with AML, is a deoxycytosine analogue that interferes with DNA synthesis by damaging DNA during the S-phase of replication. Given our findings demonstrating purine nucleotide insufficiency upon FLT3-ITD inhibition (Fig. 5D and E), we hypothesized that simultaneous treatment of FLT3-ITD–positive AMLs with cytarabine and a PHGDH inhibitor would exacerbate DNA damage by restricting serine, and thus one-carbon unit incorporation for purine nucleotide synthesis and subsequent DNA-repair processes.

Three FLT3-ITD–positive AML cell lines—MV4-11, MOLM-13, and PL-21—were used for these studies. Of note, although MV4-11 and MOLM-13 cells display an MLL-translocated karyotype, PL-21 cells display a complex non-MLL karyotype (50). Exposure of all three cell lines to increasing concentrations of WQ-2101 in combination with cytarabine revealed positive synergistic interactions (highlighted in red), as determined using the SynergyFinder computational package and the Bliss synergy index (ref. 51; Fig. 6A; Supplementary Fig. S6A). The combinatorial effect of cytarabine and WQ-2101 in MV4-11 cells was confirmed by Annexin V staining (Supplementary Fig. S6B). Importantly, the combination of cytarabine and WQ-2101 did not induce synergistic cell death in FLT3-WT cell lines (HL-60 and KASUMI-1) as determined by a negative Bliss synergy index (Supplementary Fig. S6C). Consistent with the hypothesis that PHGDH inhibition would limit the availability of purine nucleotide species for DNA repair, WQ-2101 exacerbated cytarabine-induced phosphorylation of the DNA damage marker histone variant H2A.X in all three FLT3-ITD–mutant AML cell lines (Fig. 6B). To determine if this combinatorial strategy has potential clinical implications, two FLT3-ITD–mutant primary AMLs obtained from patient bone marrow aspirates, termed AML-01-226-2014 and AML-01-309-2014, were sourced (Fig. 6C, clinical data available in Supplementary Fig. S6D). Of note, AML-01-226-2014 was refractory to standard-of-care 7+3 therapy with cytarabine and daunorubicin. Consistent with data shown using FLT3-ITD–mutant human AML cell lines, a combinatorial effect between cytarabine and WQ-2101 was evident in both primary AML samples (Fig. 6D; Supplementary Fig. S6E).

We next sought to determine the in vivo efficacy of combining cytarabine with WQ-2101. Initially, the effects of PHGDH inhibition on steady-state hematopoiesis in non–tumor-bearing PTPRCA mice treated with two doses of WQ-2101 (5 and 20 mg/kg; ref. 39) for 14 days was assessed. WQ-2101 had negligible effects on hematopoietic progenitor and more differentiated cells of the myeloid and lymphoid lineages in both the bone marrow and spleen of treated mice (Supplementary Fig. S6F). In addition, WQ-2101 treatment did not alter gross peripheral white and red blood cell counts or hemoglobin and platelet counts (Supplementary Fig. S6G). We subsequently transplanted MV4-11 cells into NSG recipient mice and treated them with single-agent cytarabine, WQ-2101, or a combination of both (Fig. 6E). Single-agent WQ-2101 or cytarabine provided limited survival advantage to leukemic mice relative to vehicle treatment, whereas the combinatorial arm displayed significant survival advantage relative to vehicle and both WQ-2101 and cytarabine single-agent arms (Fig. 6F). These in vitro data in human AML cell lines and primary patient specimens, in conjunction with evidence of in vivo activity in an aggressive disseminated model of AML, provide a strong rationale for targeting dysregulated serine biosynthesis in conjunction with standard-of-care chemotherapy in FLT3-ITD mutant AML (Fig. 6G).

The recent approval of the first-generation pan-RTK inhibitor midostaurin, as first-line therapy for patients harboring FLT3-ITD mutations, has resulted in significant interest in targeting mutant FLT3 in the context of AML (7). Although FLT3 inhibitor monotherapy initially yielded encouraging results, the development of acquired resistance has reduced the efficacy of this approach. Here, we reveal key insights into metabolic reprogramming events driven by FLT3 mutations and identify novel therapeutic strategies to treat this aggressive subtype of AML.

A key finding from this study is that FLT3-ITD modulates serine metabolism by upregulating mRNA expression of genes involved in serine synthesis (including PHGDH and PSAT1) and serine uptake (including SLC1A4 and SLC1A5). Although the cellular demand for serine is often met by uptake of exogenous serine, genomic and non-genomic events can divert glucose-derived carbons to the de novo serine biosynthesis pathway. Indeed, genomic analyses in primary breast cancer, melanoma, and prostate cancers identified copy-number amplification in enzymes of the de novo serine biosynthesis pathway, most notably PHGDH, which encodes the first and only rate-limiting enzyme of this pathway (52–55). Importantly, transcriptional mechanisms that mediate activation of the de novo serine biosynthesis pathway have also been discovered. ATF4, a transcription factor that responds to cellular stressors such as amino acid starvation and hypoxia, and in this study was identified as central to FLT3-ITD–mediated regulation of the serine biosynthesis pathway, has been previously demonstrated to activate key enzymes of the de novo serine biosynthesis pathway in other disease contexts (41, 56). Other transcription factors such as MYC (57, 58) and NRF2 via ATF4 (59) have also been demonstrated to induce the de novo serine biosynthesis pathway. Interestingly, RAS pathway activation has been shown to enhance serine biosynthesis. KRAS activation and concurrent loss of the tumor suppressor gene LKB1 enhances proliferative capacity in pancreatic cancer models, primarily via stimulation of serine biosynthesis (60). A second study demonstrated lack of sensitivity to restriction of dietary serine restriction in pancreatic and intestinal tumors, due to the inherent ability of KRAS mutations to activate de novo serine biosynthesis (61). Thus, the work herein raises the possibility that concurrent inhibition of PHGDH and dietary modulation of serine/glycine may prove to be particularly detrimental to the maintenance of FLT3-ITD–positive AML. Indeed, the modulation of amino acid pools via diet is an emerging area of interest in the field (62–65). Whether FLT3-ITD inhibition would prove synergistic with a serine- and glycine-deprived diet remains an open question, particularly as our data suggest loss of FLT3-ITD hampers both the de novo generation of serine and exogenous serine uptake.

The data from our study and others show that the mTORC1–ATF4 axis is pivotal to the induction of serine biosynthesis. The link between FLT3-ITD and mTORC1 has been previously defined (42), with mTORC1 modulating ATF4 via two distinct mechanisms: by regulating the stability of ATF4 mRNA transcripts and via regulation of ATF4 translation through three upstream uORFs in the 5′ UTR of ATF4 (43). In the context of FLT3-ITD mutations in AML, we provide a model whereby mTORC1 activates ATF4 to drive transcriptional regulation of both the de novo serine biosynthesis pathway and several key amino acid transporters associated with serine/glycine import into the cell. Previous studies have demonstrated that FLT3-ITD–driven AMLs are preferentially sensitive to mTOR/PI3K inhibitors (66–69), and it is possible that at least part of their efficacy can be attributed to suppression of the mTORC1/ATF4 axis and disruption of downstream metabolic pathways such as the de novo serine biosynthesis pathway.

There has been considerable interest in the development of small-molecule inhibitors to target the de novo serine biosynthesis pathway, and small-molecule PHGDH inhibitors have been investigated in the context of solid malignancies such as breast and prostate cancers (39, 70–73). Our analysis provides evidence that PHGDH inhibitors can be applied to exploit the dependency of FLT3-ITD–driven AML on sustained serine biosynthesis. Why tumors can exhibit sensitivity to PHGDH inhibition even when exogenous serine is available is poorly understood. One possible explanation is that de novo–generated serine may ensure availability of one-carbon units for nucleotide biosynthesis by reducing cytoplasmic SHMT1 activity (71). Inhibition of PHGDH, and subsequent activation of SHMT1, even in contexts in which exogenous serine is abundant, reduces the incorporation of one-carbon units into nucleotides which may result in the inability of exogenous serine to rescue the loss of PHGDH. Another study investigating WQ-2101 discovered that PHGDH inhibition affects nucleotide synthesis in a manner that is independent of serine utilization and instead involves defects in central carbon metabolism, the pentose–phosphate pathway, and TCA cycle (44). As we have demonstrated, FLT3-ITD inhibition triggers purine insufficiency. The de novo serine biosynthesis pathway donates carbon units to the folate cycle, which produces cytosolic one-carbon formyl units that are critical for purine ring assembly (47, 48). Inhibition of purine metabolism has long been explored as a therapeutic modality in leukemia (74), and more recently several groups have investigated the therapeutic implications of perturbing enzymes in the de novo purine synthesis pathway (such as PAICS) in the context of AML (75). We showed that inhibition of PHGDH sensitized FLT3-ITD AML to cytarabine, and posit that this is through enhanced DNA damage resulting from nucleotide insufficiency. Our biochemical studies demonstrating greatly enhanced phosphorylation of γH2A.X signal following combined cytarabine and WQ-2101 treatment is consistent with this hypothesis. Moreover, a very recent study in the context of brain metastasis has shown that serine deprivation, resulting from PHGDH inhibition and the low serine/glycine state of cerebrospinal fluid, is in itself sufficient to reduce purine nucleotide pools and induce DNA damage (76).

A potential limitation of this study is the difficulty in performing systemic investigation of metabolic changes induced by FLT3-ITD or PHGDH inhibition in vivo. Although our competitive proliferation experiments provide important functional evidence that PHGDH is required for the maintenance of FLT3-ITD–driven AML blasts in vivo, the abundance of metabolites in the bone marrow and spleen (two physiologic niches relevant to AML) in response to FLT3-ITD or PHGDH inhibition was not interrogated in this study. The process of isolating AML blasts or interstitial fluid from these niches, without altering metabolite abundances during purification, is technically challenging but warrants further investigation.

Collectively, our study reveals new mechanistic insights into FLT3-ITD–induced metabolic reprogramming events in AML and reveals serine biosynthesis as a unique vulnerability in this disease. Our work also highlights the therapeutic potential of combining small-molecule PHGDH inhibitors with standard-of-care chemotherapy for the treatment of FLT3-ITD AML.

MLL-AF9/iFLT3-ITD murine cells were generated via retroviral transduction of E13.5-day fetal liver cells, and cultured in Anne Kelso Modified Dulbecco's Modified Eagle Medium (AK-DMEM—low-glucose DMEM; Invitrogen Life Technologies, 4 g/L glucose, 36 mg/mL folic acid, 116 mg/L L-arginine HCl, 216 mg/L L-glutamine, 2 g/L NaHCO₃ and 10 mmol/L HEPES), supplemented with 20% FBS (Invitrogen Life Technologies), 1% penicillin/streptomycin (PenStrep; Invitrogen Life Technologies), and 1% L-asparagine (Sigma-Aldrich) at 37°C and 10% CO₂. Cells at baseline were cultured in 1 μg/mL of doxycycline (Sigma-Aldrich) to maintain transgene expression. MV4-11, Kasumi-1, HL-60, and HEK-293T cells were purchased from the ATCC. MOLM-13, OCI-AML3, and PL-21 cells were purchased from the DSMZ. MV4-11, MOLM-13, and PL-21 cells were cultured in RPMI-1640 (Gibco) supplemented with 10% FBS, 1% PenStrep, and 1% Glutamax (Thermo Fisher). Kasumi-1, HL-60, and OCI-AML3 cells were cultured in the same conditions, with the exception of 20% FBS content. HEK-293T cells were cultured in DMEM (Gibco), supplemented with 10% FBS, 1% PenStrep, and 1% Glutamax. All human lines were cultured at 37°C and at 5% CO₂. For serine/glycine deprivation experiments, we synthesized RPMI-1640 and omitted inclusion of serine and glycine during preparation of medium. Exact composition of media is supplied in Supplementary Materials and Methods. All cells used in this study were tested regularly for Mycoplasma via PCR validation. All cell lines were validated via short-tandem repeat (STR) profiling prior to use.

MLL-AF9/iFLT3-ITD murine cell lines were generated via retroviral transduction of E13.5 fetal liver cells from a mouse endogenously expressing the CAG-rtTA3 transgene as described in Supplementary Materials and Methods. For lentiviral transductions for CRISPR/Cas9–mediated depletion of PHGDH, and lentiviral transductions for shRNA-mediated silencing of ATF4, see Supplementary Materials and Methods.

All in vivo experiments performed in this study were approved by the Peter MacCallum Cancer Centre (PMCC) Animal Ethics Committee under ethics approval numbers E555 and E627. Ptprca, NSG, and C57BL/6J mice endogenously expressing the CAG-rtTA3 transgene were bred in-house (PMCC). For primary Ptprca recipient transplants, 4- to 6-week-old females were sublethally irradiated with 6.5 Gy whole-body gamma-ray irradiation, using a Gammacell 1000 Cs-137 irradiator (AECL). 1 × 10⁶ retrovirally transduced fetal liver cells were then intravenously (i.v.) tail-vein injected, and mice were placed on doxycycline-containing chow (Specialty Foods) and water [0.2% doxycycline hyclate (Sigma-Aldrich), 2% sucrose] to maintain transgene expression until ethical endpoint (hunched, ruffled, impaired movements). For secondary recipients, 4- to 6-week-old female NSG mice were i.v. injected with MLL-AF9/iFLT3-ITD cells as described above. At endpoint, blast cells were harvested from femoral bone. For in vivo bioluminescence imaging, mice were injected with 50 mg/kg D-luciferin and imaged in an IVIS bioluminescence imager (Perkin-Elmer) to assess disease burden. Analysis and normalization of in vivo luminescence was performed using Living Image v3.2 and v4.2 software.

Primary patient samples used for in vitro synergy studies were obtained from the Alfred Hospital (Melbourne, Australia) after obtaining written informed consent under Alfred Health Ethics Committee–approved guidelines. These studies were performed under Alfred Health Ethics Approval Number 42/20. All use of human tissue was performed in accordance with the Declaration of Helsinki. Primary AML blasts were cultured in StemSpan SFEM (STEMCELL Technologies) supplemented with 10 ng/mL IL3, 10 ng/mL IL6, 50 ng/mL stem cell factor, and 50 ng/mL FLT3 L (all PeproTech) and 35 nmol/L UM171 and 500 nmol/L stemreginin (STEMCELL Technologies). No patient data other than age, sex, and genomic profile were made available to nonclinical/authorized authors in this article.

Quizartinib, WQ-2101, and everolimus small-molecule inhibitors were purchased from Selleckchem and reconstituted in DMSO for in vitro use. Fresh cytarabine in saline was acquired from the PMCC cytotoxic suite. In vivo quantities of WQ-2101 were purchased from Glixx Labs and reconstituted in 10% DMSO, 20% Kolliphor HS-15, and 70% sterile water supplemented with 0.1% Tween 80 as previously described (39). For in vivo dosing, WQ-2101 was reconstituted daily. Cytarabine was diluted in PBS as required.

For the assessment of cell viability via Annexin V staining, cell suspension was stained with 1:100 dilution Annexin V conjugated to an APC fluorophore (BD Pharmingen; cat. #550475) in binding buffer (10 mmol/L HEPES, 5 mmol/L CaCl₂, 140 mmol/L NaCl₂) prior to flow cytometry analysis. For assessment of cell viability in response to WQ-2101, 8,000 AML cells were plated into 384-well plates and exposed to escalating doses of WQ-2101 for 48 hours. At endpoint, 1:2 dilution CellTiter-Glo was added to each well, and cells were lysed for 10 minutes prior to assessment of absorbance with a Cytation 3 spectrophotometer (BioTek).

For in vitro and in vivo competition assay protocols, see Supplementary Materials and Methods. For determination of proliferation in the presence/absence of serine and WQ-2101, 50,000 cells were seeded in 6-well plates and cultured for 3 or 9 days, respectively. Cells were counted using a cellometer, and viability was determined by the trypan-exclusion assay. For CellTrace Violet analysis of cell proliferation in response to WQ-2101 or OCI-AML3 cells in response to serine/PHGDH depravation, 1 × 10⁷ cells were washed in PBS, and stained for 20 minutes at 37°C with agitation with CellTrace Violet label (Thermo Fisher). A narrow peak of cells was then sorted using a FACSAria Fusion 5 (BD) to synchronize cell-cycle phase and enable defined peak analysis. Cells were then cultured for 72 hours in 3 μmol/L WQ-2101 and proliferation assessed via flow cytometry. For OCI-AML3 cells in response to serine depravation, sgSCR or sgPHGDH-transduced cells were cultured for 48 hours in either full serine/glycine or serine/glycine-deficient RPMI, and proliferation was assessed via flow cytometry.

Details of 3′-RNA-seq RNA preparation, sequencing, and analysis pipelines can be found in Supplementary Materials and Methods.

Details of ChIP-seq protocols, library preparation, sequencing, and analysis pipelines can be found in Supplementary Materials and Methods.

MV4-11 and/or OCI-AML3 cells were treated as described, harvested, and processed for steady-state metabolomics and ¹³C-glucose labeling analysis as described in Supplementary Materials and Methods.

Detailed methods for serine rescue experiments in sgPHGDH cells and metabolite rescue experiments in response to FLT3-ITD inhibition with quizartinib can be found in Supplementary Materials and Methods.

Detailed methods for the assessment of exogenous serine uptake post FLT3-ITD and PHGDH inhibition with quizartinib and WQ-2101 can be found in Supplementary Materials and Methods.

Experimental protocols and analysis for combination therapy with WQ-2101 and cytarabine in AML cell lines, primary patient samples, and in vivo, can be found in Supplementary Materials and Methods.

GraphPad Prism v8.0 and R version 3.6.1 software were used for statistical analysis. Statistical tests performed for each experiment are highlighted in the figure legends. Unless otherwise specified, error bars are indicative of ± SD, and threshold for significance was P < 0.05.

RNA-seq and ChIP-seq data were deposited in the Gene Expression Omnibus under accession number GSE163932. Outputs from RNA-seq DEG analysis and other relevant data are included in the Supplementary Data Tables. Any other relevant raw data are available from the corresponding authors at reasonable request.

L. Kats reports personal fees and other support from Agios Pharmaceuticals and Celgene outside the submitted work. A. Wei reports grants and personal fees from Novartis, AbbVie, Servier, Celgene/BMS, Amgen, and AstraZeneca and personal fees from Astellas, Pfizer, Macrogenics, Genentech, and Janssen outside the submitted work; in addition, A. Wei has a patent for Venetoclax with royalties paid. G.P. Gregory reports personal fees from Roche, Novartis/Sandoz, and Gilead, grants and personal fees from Janssen, grants from AbbVie, BeiGene, and MSD outside the submitted work. R.W. Johnstone reports grants from AstraZeneca during the conduct of the study; grants from Roche and BMS, grants and personal fees from MecRx outside the submitted work. No other disclosures were reported.

S. Bjelosevic: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. E. Gruber: Resources, data curation, software, formal analysis, investigation, and visualization. A. Newbold: Formal analysis and investigation. C. Shembrey: Conceptualization, resources, software, formal analysis, investigation, and visualization. J.R. Devlin: Resources, formal analysis, and investigation. S.J. Hogg: Resources and investigation. L. Kats: Conceptualization, resources, formal analysis, investigation, and methodology. I. Todorovski: Resources, data curation, software, formal analysis, and investigation. Z. Fan: Resources and investigation. T.C. Abrehart: Investigation. G. Pomilio: Resources, formal analysis, and investigation. A. Wei: Resources, supervision, writing–original draft, writing–review and editing. G.P. Gregory: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing. S.J. Vervoort: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, writing–original draft. K.K. Brown: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. R.W. Johnstone: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–originaldraft, project administration, writing–review and editing.

S. Bjelosevic was supported by an Australian Government Research Training Program Scholarship, the Peter MacCallum Cancer Foundation, and the Picchi Brothers Foundation; E. Gruber, C. Shembrey, I. Todorovski, and Z. Fan by an Australian Government Research Training Program Scholarship; J.R. Devlin by a Victorian Cancer Agency Early Career Seed Grant (ECSG17018); S.J. Hogg by a National Health and Medical Research Council of Australia (NHMRC) Investigator Grant; L. Kats by a Victorian Cancer Agency (VCA) Mid-Career Research Fellowship and grants from the Cancer Council of Victoria (CCV) and National Health and Medical Research Council (NHMRC); S.J. Vervoort was supported by a Rubicon Fellowship from the Netherlands Organization for Scientific Research (NWO; 019.161LW.017), an NHMRC EL1 Fellowship (GNT1178339), and a Peter MacCallum Cancer Foundation Grant; K.K. Brown by an NHMRC Project Grant (GNT1146642), a VCA Mid-Career Research Fellowship (MCRF17020), and a Susan G. Komen Career Catalyst Research Grant (CCR18548354); R.W. Johnstone by a Project Grant from CCV, a Project Grant and Program Grant (grant 454569) from the NHMRC, an NHMRC Senior Principal Research Fellowship, and a grant from The Kids' Cancer Project (to R.W. Johnstone and S.J. Vervoort). We acknowledge support from the Peter MacCallum Cancer Centre Foundation and the Australian Cancer Research Foundation. This research used National Collaborative Research Infrastructure Strategy (NCRIS)–enabled Metabolomics Australia infrastructure at the University of Melbourne, funded through BioPlatforms Australia. We also extend our thanks to the Peter MacCallum Cancer Centre Core Facilities and their staff who provided support for this work, namely, the Flow Cytometry, Molecular Genomics, and Animal Core Facilities. We especially thank Eva Vidacs, Lauren Matthews, Lena Ly, Mark Sheehan, Michael Durrant, and Stephanie Le for their invaluable technical expertise for in vivo experiments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.