PSAT1 Suppression Drives a Targetable Serine Dependence in AML

From antifolates to asparaginase to IDH inhibitors, targeting cancer metabolism has proven useful in heme malignancies. In solid tumors, a therapy with recent momentum is depriving cancer of serine. In vivo, this is done with a serine- and glycine-free (-SG) diet, which is tolerated by mice and can slow tumor growth (G is also depleted due to its conversion to S in vivo). This has led to a clinical trial of -SG medical food in pancreas cancer (NCT05078775). However, the impact of this therapy in heme malignancies is largely unknown. We thus sought to understand its potential in AML.

We tested the reliance of normal bone marrow (NBM) CD34+ cells and 29 human AML cell lines on external S. In -S media, the growth rate (GR) of NBM was modestly decreased (GR -S/+S = 0.66), but there was no cell death. Likewise, 66% (19/29) of AML cells had slower growth but not death (GR -S/+S = 0.71) and thus were serine non-auxotrophic (NA). However, 34% (10/29) of lines died in -S media (GR -S/+S = -0.2), indicating serine auxotrophy (SA) in a subset of AML cells.

To understand differences between NA and SA cells, we compared gene expression using RNA-seq from the CCLE. Two genes were severely suppressed only in SA cells: MGST1 and PSAT1 (log2 SA/NA = -4.9 and -5.1, respectively). PSAT1 is an enzyme in the serine synthesis pathway (SSP), and immunoblotting confirmed the protein was undetectable in all SA cells. Methylation-specific qPCR showed PSAT1 promoter methylation in SA, but not NA, cells (log2 meth/unmeth = 10.6 and -7.2, respectively). Isotope tracing of 13C-glucose showed serine synthesis in NA cells (M+3 S = 2.7%) that increased in -S (M+3 S = 10.3%), whereas no synthesis was detected in SA cells.

To test if PSAT1 mediates SA, we re-expressed wild type (WT) or K200A enzyme-dead PSAT1 using lentivirus in 5 SA lines. Growth in -S was rescued by WT, but not K200A, in 4 of 5 lines. The 5^th, HNT34, showed only partial rescue, but cells were fully rescued when PHGDH—an SSP enzyme we previously showed is suppressed by RNA missplicing from SF3B1 mutation in these cells—was also overexpressed. MGST1 overexpression did not affect SA. We further confirmed the need for PSAT1 in -S growth through CRISPR knockout in NA MOLM13 cells, which induced complete SA.

To test targetability of SA in vivo, we xenografted NSG mice with 5 SA lines and gave +SG or -SG diets. Increased overall survival (OS) with -SG (HRs: 0.036 – 0.065) occurred in 4 of 5 lines. As a control, OS of NA MOLM13 cells was unchanged with -SG, but it was increased in MOLM13-PSAT1-KO. Moreover, WT PSAT1-rescued SA P31FUJ cells—but not K200A controls—trended towards decreased OS with -SG.

For relevance to primary human AML, we noted 10-30% of TCGA/BEAT-AML samples had PSAT1 methylation with low RNA, and SSP-low AML was enriched for MECOM rearrangements and mutations in WT1, GATA2, CEBPA, and SF3B1. SSP-high AML was enriched for p53, IDH1, IDH2, and U2AF1 mutations. Moreover, we have, to date, found PSAT1 expression and NA for primary samples with p53 and IDH1 mutations, respectively, and we have found undetectable PSAT1 and SA in a MECOM-rearranged sample.

Lastly, we have data that a PEGylated, engineered variant of the human serine dehydratase enzyme (PEG-eSDH) degraded >90% of plasma S and decreased leukemia in mice xenografted with (SA) HNT34 cells.

In conclusion, we found a subset of AML cells with serine auxotrophy and deficient serine synthesis driven by methylation and suppression of PSAT1. These cells were targetable in vivo with a -SG diet, nominating PSAT1-null AML for trials of -SG diets. We also report this AML may be targetable by enzymatic serine depletion with PEG-eSDH.