IDH1-Mutant Preleukemic Hematopoietic Stem Cells Can Be Eliminated by Inhibition of Oxidative Phosphorylation

Preleukemic hematopoietic stem cells with an IDH1-mutation are epigenetically, transcriptomically, and metabolically distinct from wild-type hematopoietic stem cells and IDH2-mutant cells and can be specifically targeted using complex I inhibitors.

Abstract

Rare preleukemic hematopoietic stem cells (pHSC) harboring only the initiating mutations can be detected at the time of acute myeloid leukemia (AML) diagnosis. pHSCs are the origin of leukemia and a potential reservoir for relapse. Using primary human samples and gene editing to model isocitrate dehydrogenase 1 (IDH1) mutant pHSCs, we show epigenetic, transcriptional, and metabolic differences between pHSCs and healthy hematopoietic stem cells (HSC). We confirm that IDH1-driven clonal hematopoiesis is associated with cytopenia, suggesting an inherent defect to fully reconstitute hematopoiesis. Despite giving rise to multilineage engraftment, IDH1-mutant pHSCs exhibited reduced proliferation, blocked differentiation, downregulation of MHC class II genes, and reprogramming of oxidative phosphorylation metabolism. Critically, inhibition of oxidative phosphorylation resulted in the complete eradication of IDH1-mutant pHSCs but not IDH2-mutant pHSCs or wild-type HSCs. Our results indicate that IDH1-mutant preleukemic clones can be targeted with complex I inhibitors, offering a potential strategy to prevent the development and relapse of leukemia.

Significance:

A high burden of pHSCs is associated with worse overall survival in AML. Using single-cell sequencing, metabolic assessment, and gene-edited human models, we find human pHSCs with IDH1 mutations to be metabolically vulnerable and sensitive to eradication by complex I inhibition.

See related commentary by Steensma, p. 83.

This article is featured in Selected Articles from This Issue, p. 80

INTRODUCTION

Acute myeloid leukemia (AML) arises due to a stepwise acquisition of somatic mutations, with the initiating lesion often occurring in genes related to epigenetic regulation (1–3). The most common founding mutations in AML are also the most frequently observed mutations in clonal hematopoiesis of indeterminate potential (CHIP) which include DNMT3A, TET2, and ASXL1 (4–6). Mutations in the metabolic enzymes isocitrate dehydrogenase 1 (IDH1) and 2 (IDH2) are rarely observed in CHIP (7, 8). However, IDH1 and IDH2 mutations are found in 10% to 15% of AML and are often early, initiating mutations as they can be found in rare preleukemic hematopoietic stem cells (pHSC) detected at the time of diagnosis (2, 4). These pHSCs harbor only the initiating but not the subsequent mutations present in the bulk leukemia and immunophenotypically closely resemble healthy hematopoietic stem cells (HSC). It is likely that these pHSCs not only constitute the origin of the leukemia but also remain a potential reservoir for relapse. Furthermore, it has been shown that the frequency of pHSCs at the time of AML diagnosis correlates to overall survival, suggesting that these evolutionary remnants from the initiation of leukemia are of therapeutic relevance (4). However, because pHSCs are rare and share the immunophenotype of healthy HSCs, they have remained challenging to study and no specific biomarkers or biological properties suitable for specific targeting have been described to date.

Recurring hotspot mutations in IDH1 and IDH2 cause the aberrant production of the oncometabolite (R)-2-hydroxyglutarate (2HG), which in turn inhibits alpha-ketoglutarate dependent epigenetic modifiers such as TET2, resulting in increased DNA methylation (9, 10). IDH1-mutant cells have previously been shown to be sensitive to targeting with inhibitors of both BCL2 and metabolic pathways; however, these observations are limited to the blast compartment and do not inform on the potential sensitivity of pHSCs (11–13).

The clinically available IDH1-mutant-specific inhibitor ivosidenib has shown promising results as monotherapy and in combination with a hypomethylating agent in AML, but a majority of these patients will relapse and succumb to disease, indicating a need for a better understanding of the pathophysiology as well as new therapeutic options (14). Here, we combined single-cell full transcriptome RNA sequencing of primary pHSCs with a new high-fidelity genotyping assay, enabling a direct comparison of pHSCs to wild-type (WT) HSCs in patients with AML. We also modeled IDH1-mutant pHSCs by genetically engineering human hematopoietic stem and progenitor cells (HSPC) and showed that IDH1-mutant pHSCs are specifically sensitive to inhibition of oxidative phosphorylation but not ivosidenib. Moreover, IDH2-mutant pHSCs remain insensitive to complex I inhibition, constituting a targetable metabolic difference between IDH1-mutant and IDH2-mutant pHSCs, results in direct clinical implications in selecting patients who might benefit from novel inhibitors of oxidative phosphorylation.

RESULTS

IDH1 Is Mutated in Preleukemic Stem Cells in Patients with AML and Clonal Cytopenia

IDH1 mutations are frequently observed in AML. To evaluate the fractional abundance of IDH1-mutant cells in different cellular compartments, we genotyped sorted cell populations using a digital droplet PCR (ddPCR) assay specific for IDH1^R132X mutations (Fig. 1A; Supplementary Fig. S1A). We found a variant allele frequency (VAF) of close to 0.5 for both the SSC^dimCD45^dim leukemic blasts and the immunophenotypic leukemic stem cells (LSC)-enriched CD3⁻CD19⁻CD20⁻CD34⁺CD38^lowCD99⁺TIM3⁺ population, whereas the CD3⁻CD19⁻CD20⁻CD34⁺CD38^lowCD99^negTIM3^neg residual population (rHSC) showed a VAF between 0 and 0.5, indicating that this compartment contained a mixture of IDH1-wild-type HSC and IDH1-mutant pHSC (Fig. 1B). In general, T cells (SSC^lowCD45⁺CD3⁺) and B cells (SSC^lowCD45⁺CD19/CD20⁺) showed low mutational VAF in these AML samples. Sorting the rare rHSCs and performing colony-forming assays followed by genotyping of individual colonies revealed that the immunophenotypically defined rHSC population does indeed contain a mixture of WT HSCs and IDH1-mutant pHSC, both of which can form colonies (Fig. 1C–E).

Figure 1.

IDH1 is mutated in preleukemic HSCs in relapse-causing clones in AML. A, Gating schema for FACS isolation of AML blasts (SSC^dimCD45^dim), leukemic stem cells (LSC; CD3⁻CD19⁻CD20⁻CD34⁺CD38^lowCD99⁺TIM3⁺), and residual HSCs (rHSC; CD3⁻CD19⁻CD20⁻CD34⁺CD38^lowCD99⁻TIM3⁻) from AML patient bone marrow aspirates at time of diagnosis. B, Variant allele frequencies (VAF) of IDH1 mutations in sorted blasts, T cells (SSC^lowCD45⁺CD3⁺), B cells (SSC^lowCD45⁺CD19/CD20⁺), LSCs, and rHSCs in patients with primary AML (n = 9) as determined by digital droplet PCR (ddPCR). The IDH1-mutant VAF is significantly lower in the rHSC compared with the LSC compartment as determined by Student t test. C, Frequency of FACS isolated rHSCs per million mononuclear cells (MNC) in patients with AML (n = 15). D, A representative well of colonies imaged two weeks after plating rHSCs from the AML sample SU582 showing IDH1-mutant pHSC-derived colonies in red and WT HSC-derived colonies in gray as determined by ddPCR. E, Mutant pHSC frequency in the immunophenotypically defined rHSC compartment in patients with AML (n = 8) as determined by ddPCR. F, Mutational frequencies of DNMT3A, TET2, and ASXL1 in a combined cohort of 19,194 individuals with CHIP, 92 clonal cytopenia of undetermined significance (CCUS) patients, and 187 patients with AML. G, Lower frequencies of IDH1 mutations in CHIP (0.1%) than CCUS (6.5%) and AML (7.7%), statistical significance determined by Fisher exact test. H,IDH1 mutations are retained to a higher degree (100%) in patients with AML with IDH1 mutated at diagnosis (n = 20) than other mutations (81%), statistical significance determined by Fisher exact test. **, P < 0.01; ****, P < 0.0001.

Because mutations in IDH1 often occur early during leukemogenesis, we wanted to investigate their prevalence in the precursor conditions CHIP and clonal cytopenia of undetermined significance (CCUS). Using publicly available data on 19,194 individuals with CHIP (6, 8, 15, 16), 92 CCUS patients (17), and 187 patients with AML diagnosed at Stanford University Hospital, mutations in DNMT3A, TET2, and ASXL1 (DTA mutations) showed an expected distribution in all three conditions (Fig. 1F). In contrast, IDH1 mutations were exceedingly rare in CHIP (0.1%), but relatively frequently observed in both patients with CCUS and AML (6.5% and 7.7%, respectively), indicating a difference in disease initiation and progression compared with patients with common CHIP mutations and suggesting that IDH1-mutant clonal hematopoiesis cannot sustain effective normal hematopoiesis (Fig. 1G).

AML is known to experience clonal evolution from diagnosis to relapse (18, 19). We, therefore, compared IDH1-mutant VAF in matched diagnosis and relapse AML samples in publicly available data (18, 20–25) and found that the IDH1 mutation was retained at relapse in all samples (Fig. 1H). We conclude that IDH1 mutations are early leukemic events in relapse-causing AML clones and are frequently found in patients with cytopenia, indicating a clinically relevant and potentially actionable mutation for leukemia and relapse prevention. Similar results were seen for IDH2 mutations (Supplementary Fig. S1B–S1D).

Distinct Transcriptomic Profiles of IDH1-Mutant Preleukemic Stem Cells

To further study these clinically relevant but rare pHSCs, we developed a protocol for single-cell transcriptome sequencing combined with high-fidelity genotyping (Fig. 2A). By building on the Smart-seq and TARGET-seq protocols (26, 27), adding optimized primers for specific capture and amplification of transcribed mRNA and genomic IDH1 mutations, and a three-primer ddPCR genotyping assay, we were able to improve the frequency of correctly genotyped cells 44-fold for LSCs and 2-fold for HSPCs, leading to successful genotyping of 98% of cells (Supplementary Fig. S2A–S2D). Using this method, we sequenced 356 single rHSCs from two AML bone marrow samples, comprising of 58% IDH1 WT HSCs and 41% IDH1-mutant pHSCs. In 95% of the mutant cells, we could detect both the WT and mutant alleles of the heterozygous IDH1 mutation; less than 2% of cells had no genotyping information and were thus excluded from downstream analyses (Fig. 2B). Additionally, 182 immunophenotypic LSCs were sequenced as controls. Using index sorting data and genotyping information, we confirmed overlapping immunophenotypes of HSCs and pHSCs (Supplementary Fig. S2E and S2F). The mean number of genes expressed per cell was 2,800 and mitochondrial reads were low with more than 99% of cells having lower than 18% mitochondrial reads (Supplementary Fig. S2G).

Figure 2.

Combined single-cell RNA sequencing and genotyping reveal distinct transcriptomic profiles of IDH1-mutant pHSCs in AML. A, Schematic overview of the single-cell RNA sequencing workflow with specific amplification of mRNA and gDNA for parallel genotyping of IDH1 mutations using ddPCR. Ficoll separated mononuclear cells from two patients with AML bone marrow aspirates at the time of diagnosis were FACS isolated into 96-well plates containing lysis buffer. The mixed population of HSCs and pHSCs was defined as CD19⁻CD20⁻CD34⁺CD38^lowCD99⁻TIM3⁻ and separated based on subsequent genotyping, the LSC population was defined as CD19⁻CD20⁻CD34⁺CD38^lowCD99⁺TIM3⁺. B, Identification of pHSCs and HSC based on single-cell IDH1 genotyping by ddPCR, a total of 356 single cells were genotyped. C, UMAP of transcriptomic single-cell data from sorted LSC and rHSC form five Leiden clusters. D, UMAP of transcriptomic single-cell data within the CD19⁻CD20⁻CD34⁺CD38^low compartment, annotated as HSC (CD99⁻TIM3⁻IDH1^wt), pHSC (CD99⁻TIM3⁻IDH1^R132H), and LSC (CD99⁺TIM3⁺) based on immunophenotype and genotype. E, Cell state frequencies in the five Leiden clusters. F, Top differentially expressed genes in the three cell types. G, Three of the most overexpressed genes in pHSCs compared with HSCs. H, Three of the most downregulated genes related to MHC class II antigen presentation on pHSCs compared with HSCs. Mut, mutant; UMAP, Uniform Manifold Approximation and Projection for Dimension Reduction.

Combining the RNA sequencing data with mutational status of the individual cells, we observed distinct transcriptomic profiles of HSCs, pHSCs, and LSCs, with some pHSCs clustering together with HSCs and some more closely resembling LSCs, reflecting an evolutionary trajectory from HSC via pHSC to LSC (Fig. 2C and D; Supplementary Fig. S2H and S2I). Comparing cell type frequencies in the five Leiden clusters generated by recursively merging cell populations into single nodes, showed that clusters 0 and 4 contained primarily HSCs, clusters 2 and 3 contained primarily LSCs, whereas pHSCs were found in all clusters but constituted the majority of cells in cluster 1 (Fig. 2E). Genes that were significantly differently expressed between the cell types include the G-protein–coupled receptor P2RY14 in HSCs, the ribosomal subunit RPS3A in pHSCs, and cell-surface glycoprotein CD99 in LSCs (Fig. 2F). Focusing on pHSCs, we observed differential expression of many genes compared with HSCs, the most upregulated were SH3BGRL3, TPSD1, and MT-CO2 (Fig. 2G). RPS3A has been described to regulate mitochondrial function and MT-CO2 encodes for part of the electron transport chain, suggesting an altered metabolism in these cells (28). SH3BGRL3 has been shown to be upregulated in AML and suggested to play a role in regulating leukemia progression but the role of either SH3BGRL3 or TPSD1 is not known in the preleukemic setting (29). Previous literature has suggested that AML cells downregulate MHC class II as part of immune evasion in the relapse setting (30–32). Interestingly, we see the significant downregulation of multiple genes related to MHC-II in pHSCs, including CD74, HLA-DPA1, and HLA-DRA (Fig. 2H), as well as HLA-DOA, HLA-DRB5, HLA-DPB1, HLA-DQB1, and HLA-DQA1 (adjusted P < 0.01 and log fold change < −2), possibly priming these cells for immune evasion. Our data thus show clear transcriptional changes associated with the founding IDH1 mutation in pHSCs from primary AML samples.

Human IDH1-Mutant CD34⁺ HSPCs Show Impaired Proliferation and Differentiation

To confirm that the observed changes were driven solely by the somatic mutation and to allow further transcriptomic and epigenetic studies, we next developed a model of human IDH1-mutant pHSCs. Using CRISPR/Cas9 and recombinant adeno-associated virus 6 (rAAV6) to provide template for homology-directed repair (HDR), we introduced IDH1^R132H, IDH2^R140Q, their WT counterparts, or green fluorescent protein (GFP; AAVS1) as control into the human AAVS1 locus (Fig. 3A). Successful integration could be tracked by GFP, allowing for FACS isolation of edited cells (Fig. 3B). We confirmed that IDH1^R132H and IDH2^R140Q induced 2HG production, which was completely inhibited by the mutant-specific inhibitors ivosidenib and enasidenib, respectively; genomic integration was confirmed using in-out PCR (Fig. 3C–E). The edited cells secreted 2HG into the culture supernatant, and the levels of 2HG were similar to those seen in primary AML blasts, confirming the pathophysiologic relevance of this model (Supplementary Fig. S3A and S3B).

Figure 3.

Modeling human pHSCs using CRISPR/Cas9 with HDR shows reduced proliferation and blocked differentiation caused by IDH1 mutations. A, Schematic overview of gene editing of human CD34⁺ HSPCs. B, Representative flow plot showing editing efficiency as measured by GFP expression. C, 2HG production induced by IDH1^R132H is completely inhibited by ivosidenib 10 μmol/L, here showing the mean of three technical replicates, statistical significance determined by ANOVA. D, 2HG production induced by IDH2^R140Q is completely inhibited by enasidenib 10 μmol/L, here showing the mean of three technical replicates, statistical significance determined by ANOVA. E, Representative gel electrophoresis showing genomic integration of construct. F, Proliferative capacity is significantly reduced in IDH1^R132H and IDH2^R140Q cells as compared with AAVS1, IDH1^wt, and IDH2^wt controls, here showing results from three cord blood donors, statistical significance determined by Student t test. G, Colony-forming capacity is significantly reduced in IDH1^R132H and IDH2^R140Q cells as compared with AAVS1, IDH1^wt, and IDH2^wt controls, here summarizing results from four cord blood donors, statistical significance determined by ANOVA.H,IDH1^R132H blocks erythroid differentiation in liquid culture, left showing flow cytometry data 10 days after plating, right showing summary statistics from six cord blood (circles) and three adult donors (triangles). I,IDH1^R132H blocks megakaryocytic differentiation in liquid culture, left showing flow cytometry data 10 days after plating, right showing summary statistics from four cord blood donors. J,IDH1^R132H blocks granulocyte differentiation in liquid culture, left showing flow cytometry data 10 days after plating, right showing summary statistics from four cord blood donors. K, Bone marrow aspirates 9 weeks after engraftment in NSG mice show significantly lower levels of human IDH1^R132H cells compared with IDH1^wt. L, Engraftment analyses of whole crushed bone marrow 17 weeks after transplantation into NSG mice show multilineage engraftment potential of edited human IDH1^R132H cells. For H–K, statistical significance was determined by Student t test. ExP, External PAM site; LHA, left homology arm; pA, poly-adenosine; P2A, 2A self-cleaving peptide; RHA, right homology arm. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant.

Interestingly, IDH1^R132H and IDH2^R140Q cells showed impaired proliferation as measured by reduction in GFP frequency compared with controls (Fig. 3F; Supplementary Fig. S3C). Both IDH1^R132H and IDH2^R140Q also conferred a reduction in colony-forming capacity (Fig. 3G; Supplementary Fig. S3D and S3E), in concordance with phenotypes observed in mouse models (33–35). Unlike for TET2-mutant cells (36), we observed no increase in replating capacity (Supplementary Fig. S3F). Differentiation was impaired across multiple hematopoietic lineages in IDH1^R132H cells with a significant decrease in mature CD71⁺GPA⁺ erythroblasts (Fig. 3H), CD41⁺CD61⁺ megakaryocytes (Fig. 3I), and CD15⁺ granulocytes (Fig. 3J) in liquid culture assays. Furthermore, introducing IDH1^R132H into cord blood–derived human HSPCs decreased engraftment in NSG mice compared with IDH1^Wt controls (Fig. 3K). A similar competitive disadvantage for IDH1^R132H cells was confirmed with genetically engineered adult human HSPCs (Supplementary Fig. S3G and S3H). In keeping with their preleukemic state, IDH1^R132H cells still gave rise to multilineage engraftment in both NSG and NSGS models (Fig. 3L; Supplementary Fig. S3I–S3K).

To confirm the observed phenotype, we introduced IDH1^R132H at the endogenous IDH1 locus in human HSPCs. Because the neomorphic function of the mutant IDH1 protein requires the formation of a heterodimer with the WT protein, IDH1 mutations in patients with AML are consistently heterozygous (37, 38). Thus, we used an sgRNA targeting exon 4 of IDH1 and designed two different rAAV6 viruses for HDR, allowing for the isolation of GFP⁺/BFP⁺ double-positive cells that carried a heterozygous IDH1^R132H mutation (Supplementary Fig. S3L; ref. 39). Although at low efficiency, correct integration was confirmed by PCR and FACS isolated heterozygous cells were shown to produce 2HG and exhibited a clear reduction in colony-forming potential (Supplementary Fig. S3M–S3P).

Overall, our data confirm that oncogenic IDH1^R132H mutations confer a defect in hematopoietic proliferation, colony-forming ability, and differentiation of multiple hematopoietic lineages including erythroid, megakaryocytic, and granulocytic. This phenotype is consistent with the multilineage cytopenia observed in CCUS patients with IDH1 mutations and suggests that, despite being a common founder mutation in AML, IDH1-mutant clones have a growth disadvantage at the preleukemic stage.

IDH1 Mutations Downregulate 5-Hydroxymethylcytosine and Alter the Transcriptome

IDH1 mutations have been shown to mediate epigenetic effects that promote leukemia via 2HG-dependent inhibition of TET enzymes, which canonically converts 5-methylcytosine to 5-hydroxymethylcytosine (5hmC; ref. 10). We therefore used our model of IDH1-mutant pHSCs to study 5hmC changes as a potential regulator of our observed phenotype. Our recent work has shown that the distribution of 5hmC in normal hematopoiesis is cell-type specific and distinctly altered in cells carrying TET2 mutations (36). Using a 5hmC-directed pulldown and sequencing assay, we compared 5hmC patterns in gene bodies as an epigenetic mark known to promote gene expression (40). As expected, we detected distinct epigenetic alterations in both IDH1^R132H and IDH2^R140Q cells compared with AAVS1 and WT controls (Fig. 4A; Supplementary Fig. S4A–S4D). Overall, we observed a global reduction of 5hmC in IDH1^R132H and IDH2^R140Q cells compared with WT counterparts, measuring mean peak height and peaks per million (Fig. 4B; Supplementary Fig. S4E), consistent with 2HG-mediated inhibition of TET enzyme function. We then used gene set enrichment analysis (GSEA) to identify key pathways exhibiting loss of 5hmC in IDH1^R132H cells. We noted 5hmC downregulation in genes involved in cell growth (MTORC1 signaling, E2F targets, adipogenesis; FDR values <0.05), cell proliferation (G₂–M checkpoint, mitotic spindle; FDR values <0.05), and various metabolic pathways (fatty acid metabolism, oxidative phosphorylation, heme metabolism, cholesterol homeostasis; FDR values <0.05; Fig. 4C). Most but not all gene sets were also downregulated in IDH2^R140Q samples with the exception of MYC targets and glycolysis, only downregulated in IDH2^R140Q samples (Fig. 4C). Interestingly, we observed consistent and striking downregulation across the gene body of Formyl Peptide Receptor 2 (FPR2, Fig. 4D), a protein known to have a role in metabolic regulation (41) and bone marrow microenvironment (42).

Figure 4.

IDH1 mutations reduce 5-hydroxymethylcytosine in gene bodies. Human CD34⁺ HSPCs from three cord blood donors were genetically engineered and FACS isolated based on GFP expression and kept in culture for four days to allow for a phenotype to establish. DNA was harvested and a whole-genome and a 5hmC-enriched library were constructed and sequenced. A, PCA plot of 5hmC levels in gene bodies of edited human HSPCs show distinct clustering of IDH1^R132H as well as IDH2^R140Q compared with AAVS1, IDH1^wt, and IDH2^wt controls. B, Estimated global 5hmC levels within gene bodies are reduced in IDH1^R132H and IDH2^R140Q compared with WT controls, here showing the mean of three cord blood donors, statistical significance determined by ANOVA. C, GSEA of hallmark gene sets show distinct alterations of 5hmC levels in many pathways in both IDH1^R132H and IDH2^R140Q compared WT controls. D, Gene coverage tracks for FPR2 show lower levels of 5hmC in IDH1-mutant cells compared with WT controls in genetically engineered cells from three cord blood donors. *, P < 0.05; **, P < 0.01.

To study changes in the transcriptome that may be linked to 5hmC downregulation, we performed RNA sequencing. Global projection of RNA sequencing data in an unsupervised PCA plot showed a clear separation between IDH1^R132H and IDH2^R140Q samples as well as the controls (Fig. 5A). In a heat map of the 1,000 most variable genes, the transcriptomic changes induced by IDH1^R132H were distinct from the AAVS1, IDH1^Wt, and IDH2^Wt control samples but largely overlapping with IDH2^R140Q, consistent with our 5hmC results (Fig. 5B). Comparing statistically significant differential gene expression between IDH1^R132H and IDH1^Wt revealed 639 genes downregulated and 390 upregulated genes (Fig. 5C) that were involved in proliferation and metabolism (MTORC1 signaling, E2F targets, adipogenesis, oxidate phosphorylation; FDR-values <0.05; Fig. 5D). Similar results were seen when comparing IDH2^R140Q and IDH2^Wt (Supplementary Fig. S5A) but with important differences. IDH2^R140Q samples did not show downregulation of certain metabolic gene sets including oxidative phosphorylation and fatty acid metabolism (Supplementary Fig. S5B–S5F). Integration of 5hmC and RNA sequencing data revealed concordant 5hmC and transcript levels for many genes (Fig. 5E) with both data modalities showing downregulation of gene sets involved in proliferation and metabolism (MTORC1 signaling, hypoxia, oxidative phosphorylation, E2F targets, fatty acid metabolism, adipogenesis, and DNA repair; FDR < 0.05; Fig. 5F).

Figure 5.

IDH1 mutations alter metabolism in pHSCs. A, PCA plot of RNA expression in edited human HSPCs from four cord blood donors show distinct clustering of IDH1^R132H as well as IDH2^R140Q compared with AAVS1, IDH1^wt, and IDH2^wt controls. B, Heat map showing the top 1,000 variable genes with unsupervised clustering of samples. C, Volcano plot of differentially expressed genes show significant downregulation of 639 genes and upregulation of 390 genes in IDH1^R132H compared with WT controls with the fold change cutoff −0.5 to 0.5 and FDR < 0.01. D, GSEA of hallmark gene sets highlights similarities as well as differences between IDH1^R132H and IDH2^R140Q compared with their WT controls. E, Dot plot combining 5hmC within gene bodies and RNA sequencing data of differentially expressed genes in IDH1^R132H samples compared with WT controls. Each dot represents the mean fold change of 5hmC (n = 3) and mRNA (n = 4) between IDH1^R132H and IDH1^wt cells for a specific gene. F, Venn diagram of overlapping, significantly downregulated hallmark gene sets in IDH1^R132H cells compared with WT controls using both data modalities highlight changes related to proliferation and metabolism as well as DNA repair.G, Metabolic assay showing reduced metabolism in IDH1^R132H compared with both IDH1^wt, AAVS1, and IDH2^R140Q samples. H, Increased mitochondrial dependency in IDH1^R132H compared with WT controls. I, No significant differences in mitochondrial mass of IDH1^R132H compared with controls (n = 4). J, No significant differences in mitochondrial membrane potential of IDH1^R132H compared with controls (n = 5). For G, I, and J, statistical significance was determined by ANOVA, and for H, statistical significance was determined by Student t test. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant.

Overall, these results show marked epigenetic effects in global 5hmC induced by both IDH1 and IDH2 mutations in a model of human preleukemic stem cells, prior to onset of frank AML. These epigenetic changes are likely to exert negative effects on cell proliferation and cell growth in keeping with our observed loss-of-fitness biological phenotype. Notably, important differences were also observed between IDH1^R132H and IDH2^R140Q pHSCs predominantly in transcriptional programs involved in metabolic pathways, consistent with other recent reports (13, 43).

IDH1 Mutations Alter Cell Metabolism in pHSCs

To validate the targetable nature of the metabolic changes observed in IDH1^R132H pHSCs, we performed flow-based assays to measure the basal metabolism and mitochondrial dependency using Single-Cell Energetic Metabolism by Profiling Translation Inhibition, known as SCENITH, which quantifies metabolic dependencies at the single-cell level using protein translation as a readout (44). Interestingly, IDH1^R132H pHSCs were observed to have a reduced basal metabolic activity compared with AAVS1 and WT controls, as well as IDH2^R140Q pHSCs (median MFI for IDH1^R132H 13 491 compared with 20 100, 18 529, and 16 998 for AAVS1, WT, and IDH2^R140Q, respectively, P < 0.001 using ANOVA, Fig. 5G). These data indicate a metabolic vulnerability impacting protein translation to a greater extent in IDH1^R132 compared with IDH2^R140Q pHSCs. The IDH1^R132H pHSCs were shown to be sensitive to blockade of mitochondrial function by oligomycin (Fig. 5H), indicating a mitochondrial dependency that was not primarily due to a decrease in mitochondrial mass (Fig. 5I; Supplementary Fig. S5G), similar to previous reports (13), or mitochondrial membrane potential (Fig. 5J; Supplementary Fig. S5H). These data are consistent with a major perturbation of basal and mitochondrial-dependent metabolism induced by IDH1^R132H in pHSCs, observed to a greater extent in IDH1 compared with IDH2-mutant cells.

IDH1-Mutant pHSCs Are Sensitive to Inhibition of Oxidative Phosphorylation

The metabolic changes induced by IDH1^R132H suggest that these cells may be susceptible to therapeutic targeting of metabolic pathways. We and others have shown that AML cell lines and AML blasts have a synthetic lethal relationship between IDH1 mutations and targeting of metabolic pathways, including BCL-2, fatty acid synthesis, and oxidative phosphorylation (11–13). However, the susceptibility of IDH1-mutant pHSCs to targeted inhibition of metabolic pathways has not been explored. We therefore treated engineered IDH1^R132H pHSCs with ivosidenib, the specific inhibitor of oxidative phosphorylation IACS-010759, or DMSO control and plated cells for colony formation. As expected, the introduction of IDH1^R132H induced a reduction of colony formation which was reversed by treatment with ivosidenib, leading to outgrowth of IDH1^R132H mutant colonies. However, treatment with IACS-010759 effectively inhibited IDH1^R132H colony formation (Fig. 6A and B). This effect was seen in pHSCs engineered from CD34⁺ HSPCs derived from adult peripheral blood as well as from cord blood (Supplementary Fig. S6A and S6B). Of note, IACS-010759-induced reduction of colony output from IDH1^Wt cells was seen in cord blood (Supplementary Fig. S6A and S6B) but not adult HSPCs (Fig. 6A and B), indicating that the therapeutic window could potentially be larger in adult patients.

Figure 6.

IDH1-mutant pHSCs are sensitive to inhibition of oxidative phosphorylation. A, Representative example of erythroid and myeloid colony formation of adult CD34⁺ HSPCs genetically edited to express IDH1^R132H or WT controls after treatment with DMSO, ivosdenib 10 μmol/L or the inhibitor of oxidative phosphorylation IACS-010759 at 5 μmol/L (IACS). B, Significant reduction of IDH1-mutant colony formation after treatment with IACS-010759 compared with ivosidenib and DMSO control, here summarizing data from three adult donors, statistical significance determined by ANOVA. C, Schematic overview of FACS separation of the mixed rHSC population from primary AML samples, treatment with ivosidenib 10 μmol/L, enasidenib 10 μmol/L, IACS-010759 at 5 μmol/L or DMSO control, plating for colony formation, scoring of colony formation, and individual genotyping of colonies to distinguish pHSC and HSC-derived colonies. D–F, Selective eradication of IDH1-mutant pHSCs after IACS-010759 treatment in three different primary AML samples, to the left erythroid and myeloid scoring of colonies, to the right information on the mutational status of the colonies. G–I, No significant effect of IACS-010759 treatment on IDH2-mutant pHSCs from three different primary AML samples, to the left erythroid and myeloid scoring of colonies, to the right information on the mutational status of the colonies. **, P < 0.01; ****, P < 0.0001; ns, not significant.

To confirm these findings, we sorted the mixed rHSC populations from IDH1-mutant AML patients at the time of diagnosis and subjected the cells to treatment with DMSO, ivosidenib, enasidenib, or IACS-010759 before plating for colony formation, and subsequently genotyped each colony individually (Fig. 6C). The results show specific elimination of IDH1-mutant pHSCs with the outgrowth of only WT HSCs after a short treatment with IACS-010759 (Fig. 6D–F). Ivosidenib did not alter the frequency of HSCs or pHSCs. No IDH1-mutant colony from any patient in this study was detected following IACS-010759 treatment. To verify the preleukemic status of the cells, we further genotyped all colonies for additional mutations found in frank leukemia and confirmed that colonies only contained the IDH1 mutation, consistent with a pHSC population (Supplementary Fig. S6C–S6H). Only three colonies from one ivosidenib-treated sample showed the presence of additional mutations which serendipitously proved the order of mutational acquisition for this patient to be IDH1^R132C, followed by NPM1^L287ins and then NRAS^G13D (Supplementary Fig. S6H). Interestingly, IACS-010759 did not exhibit specific targeting of IDH2-mutant pHSCs, indicating that the potential therapeutic relevance might be limited to patients harboring an IDH1 mutation (Fig. 6G–I; Supplementary Fig. S6I–S6K). Finally, we also observed a selective killing of IDH1^R132H pHSCs using the metformin-derivative IM156 and the complex I inhibitor rotenone, suggesting that alternative and novel inhibitors of oxidative phosphorylation could have clinical value (Supplementary Fig. S6L–S6O; refs. 45, 46).

We conclude that IDH1 mutations confer a metabolic vulnerability and that pHSCs can be specifically targeted by inhibiting oxidative phosphorylation, principally through targeting electron transport chain complex I, here using IACS-010759, rotenone, and IM156. This effect is specific for IDH1 mutations, in that it is not observed in IDH2-mutant pHSCs, indicating that the epigenetic, transcriptomic, and metabolic differences exerted by IDH1 and IDH2 mutations in pHSCs are of clinical importance.

DISCUSSION

pHSCs are remnants of the initiation of leukemia with a potential role as a reservoir for relapse. Their abundance at AML diagnosis correlates to overall survival, making them both biologically interesting and clinically relevant (3, 4). The study of these cells has, however, been limited by their scarcity at diagnosis, and much of our knowledge is derived from mouse models with limited study of primary human cells (33–35, 47). Here, we investigated primary IDH1-mutant pHSCs from patients with AML and genetically engineered primary human HSPCs to further our understanding of pHSCs and facilitate additional studies in a relevant cell context.

IDH1 mutations are often the first mutation in AML and sometimes the only detectable mutation in the pHSCs. The exact mechanism by which IDH1 mutations initiate the development of myeloid neoplasms remains elusive. It is clear that 2HG production and block of differentiation are key aspects of pathogenesis, but the relative lack of IDH1 mutations in CHIP, despite the fact that it is often the founder mutation in AML, is counterintuitive. Combining single-cell RNA sequencing with genotyping of primary samples revealed distinct transcriptomic changes caused by the founding mutation in pHSCs compared with immunophenotypically identical WT HSCs. These pHSCs are further distinct from the TIM3⁺CD99⁺ LSC from the same samples, indicating an intermediate transcriptional state reflecting the presumed trajectory from HSC via pHSC to LSC. We observed that these pHSCs downregulate MHC class II-related genes, a recently described immune evasive mechanism seen in leukemia (31, 32). This immune evasion could potentially counter the reduced proliferation and increased metabolic vulnerability observed in this study, allowing the IDH1-mutant pHSCs to persist in CCUS patients and transform to AML upon acquisition of additional genetic aberrations (48). Additionally, the 2HG produced by the IDH1-mutant clone might suppress residual healthy hematopoiesis, potentially explaining the paucity of IDH1 mutations in CHIP with retained hematopoietic output compared with cytopenic CCUS. A similar 2HG-mediated suppression has been observed in healthy T cells and IDH1–WT leukemia (49–51).

To complement the single-cell analysis, we genetically engineered human pHSC to express the most common IDH1 mutation IDH1^R132H. We saw that CD34⁺ HSPCs derived from both umbilical cord blood and adult donors could be engineered to produce clinically relevant levels of 2HG that conferred a reduced proliferation and a block in differentiation rather than a CHIP phenotype. This recapitulates previous data from mouse models of IDH1-mutant pHSCs and TF-1 cells expressing mutant IDH1 (33–35, 47, 52). However, we also saw a reduced competitive advantage of our human IDH1-mutant pHSCs, potentially explaining the paucity of IDH1-mutant CHIP and a feature not reflected in mouse models or cell lines.

Previous work from us and others has highlighted the altered metabolic state of IDH1-mutant AML blasts and cell lines (11–13, 53). Using inducible IDH1-mutant THP1 cells, these features have been studied in detail, and IDH1-mutant AML blasts have been shown to be susceptible to the targeting of the BCL-2 pathway, mitochondrial metabolism, and most recently, fatty acid metabolism (11–13, 43). Here, we study these aspects of pHSCs in an unbiased way using 5hmC and RNA sequencing, as well as direct metabolic assays. This revealed that IDH1-mutant pHSCs, in contrast to both their WT HSC counterparts and IDH2-mutant pHSCs, are metabolically vulnerable to targeting using inhibitors of oxidative phosphorylation, primarily through electron transport chain complex I. IDH1-mutant pHSCs consistently demonstrate perturbation of metabolic programs with decreased 5hmC across gene bodies and downregulation of key transcripts involved in cell growth, concomitant with a decreased basal metabolic activity that is dependent on an intact mitochondrial function to a greater degree than IDH2-mutant pHSCs or WT HSCs. Differences in intracellular localization, with IDH1 primarily in the cytosol and IDH2 in the mitochondria, could potentially also contribute to the different sensitivity to inhibition of oxidative phosphorylation (54). Further, our recent report showed that IDH1-mutant AML blasts have a greater dependency on fatty acid oxidation and a greater NADPH deficit than IDH2-mutant cells (43). Efficient β-oxidation of fatty acids to restore NADH requires an intact complex I. Here we show that IDH1-mutant pHSCs are particularly sensitive to complex I inhibition, suggesting that WT IDH2 protein cannot adequately compensate for the deficit in human HSPCs. Future work should examine differences in protein expression in mitochondria and peroxisomes, as well as α- and β-oxidation flux perturbations, in IDH1-mutant pHSCs.

There is increasing recognition that various mutations associated with clonal hematopoiesis differ substantially in terms of competitive fitness, hematopoietic cell production, and response to external stressors (55). Our data add to this concept from the perspective of IDH1 and suggest that, whereas ultimately leukemogenic, IDH1-mutant clonal hematopoiesis is associated with a short-term loss of fitness, giving rise to defective rather than normal hematopoiesis, as well as a metabolic vulnerability that is targetable at the pHSC stage. This is somewhat similar to TP53-mutant clonal hematopoiesis, which does not show a strong competitive phenotype in the absence of further cytotoxic therapy, yet in some patients gives rise to therapy-related myeloid neoplasm if left unchecked (15, 56). In contrast, clonal hematopoiesis with TET2 mutations is coupled to a strong competitive advantage and increased self-renewal with a gain of fitness phenotype conferred within days of gene editing and is present as a dominant clone in healthy persons with normal blood counts (36). We can thus discern two major subgroups of clonal hematopoiesis: clones with mutations such as DNMT3A, ASXL1, and TET2 associated with robust hematopoietic production, short-term gain of fitness with competitive advantage, and clones driven by mutations such as IDH1, IDH2, and possibly TP53, associated with defective hematopoietic production and a short-term loss of fitness. Recent data have shown that DNMT3A-NPM1-mutant AML patients that, at the time of diagnosis, still harbor HSPCs capable of multilineage engraftment in NSG mice, have a longer overall survival (57). Although the difference between WT HSC and DNMT3A-mutant pHSC-derived engraftment remains to be further delineated, this suggests that the initiating mutation may affect clinically relevant outcomes. Further supporting this notion is the lack of prognostic significance of residual DTA mutations post allogeneic stem cell transplantation (58). It will be important to determine which other recurrent mutations give rise to clonal hematopoiesis with a loss of fitness that could be vulnerable to targeted therapy.

Therapeutically targeting pHSCs in a specific manner has not yet been feasible but is of great interest due to the associated risk of developing leukemia as well as the potential capacity of pHSCs to serve as a reservoir for relapse. A notable example of cell state–specific drug sensitivity within the leukemic hierarchy is the insensitivity of chronic myeloid leukemia stem cells to tyrosine kinase inhibitors (59). Similarly, our data indicate that specific IDH1-mutant inhibitor ivosidenib, although effective in IDH1-mutant AML blasts, might allow for a normalization of the competitive disadvantage conferred by the IDH1 mutation in pHSCs, potentially causing an increase in clone size rather than clearance of the mutant cells. The potential therapeutic efficacy of both mutation-specific inhibitors and metabolically active drugs is thus cell type–dependent and differs based on mutational and metabolic states. Recent data summarizing the clinical challenges of inhibiting oxidative phosphorylation, with primarily neurologic toxicities, further highlight the need to select patient subgroups where the potential clinical benefit outweighs the associated toxicities (60). Additionally, we suggest IM156 as an alternative inhibitor of oxidative phosphorylation, currently in clinical development for solid tumors (45). Our data show that inhibition of oxidative phosphorylation can target IDH1-mutant pHSCs but not IDH2-mutant pHSCs, a striking finding with potential clinical relevance for patient selection for both existing and novel molecules with improved tolerability when treating AML and IDH1-mutant preleukemic conditions such as CCUS.

METHODS

Study Design

The objective of this study was to better understand pHSCs and potential vulnerabilities for targeted therapies. Because pHSCs are exceedingly rare, we complement the data derived from primary patient bone marrow aspirates with data generated using a genetically engineered human pHSC model. Results from the epigenetic, transcriptional, and metabolic assays lead to the hypothesis of a targetable vulnerability in the IDH1-mutant pHSCs that was validated in the model system and in primary patient material. For information on the research subjects of investigation, see the sections “Primary AML samples and healthy human HSPCs” and “In vivo engraftment assay” below.

Primary AML Samples and Healthy Human HSPCs

Bone marrow and peripheral blood from patients with AML (Supplementary Table S1) were collected after written informed consent according to Stanford University guidelines [Stanford University Institutional Review Board (IRB) No. 6453]. Healthy donor CD34⁺ umbilical cord blood cells were obtained via the Binns Program for Cord Blood Research (Stanford IRB #33818) after written informed consent from the mothers admitted at Lucile Packard Children's Hospital or purchased from AllCells. Healthy adult peripheral blood-derived CD34⁺ HSPCs were purchased from AllCells. Mononuclear cells from patients with AML were isolated from peripheral blood or bone marrow by ficoll separation and when applicable enriched for CD34 expressing cells using magnetic CD34 MicroBeads (Miltenyi Biotec).

Flow Cytometry and Fluorescent-Activated Cell Sorting

For liquid differentiation assays, engraftment analyses, and mitochondrial assays, a BD FACSymphony A5 Cell Analyzer was used. For FACS isolation of edited cells, an upgraded four-laser BD Aria II was used. In general, cells were stained in PBS buffer containing 2% FBS and 2 mmol/L EDTA, protected from light for 20 minutes at 4°C, and subsequently washed and resuspended in the same buffer supplemented with propidium iodide (PI) at a final concentration of 0.5 μg/mL or DAPI at 0.1 μg/mL. Antibodies used are detailed in Supplementary Table S2. For all gene-editing experiments where positive selection using a fluorescent marker was used, post sort purities of >99% were required. For single-cell sequencing experiments, a 100-μm nozzle was used, index sorting data were collected, and a preceding test-sort of GFP expressing cells into 96-well plates with subsequent manual scoring of wells by microscope was performed to confirm <2% of wells containing more than one cell or cell fragment.

Colony Assays and Genotyping of Colonies

Primary human rHSCs were FACS isolated based on immuno­phenotype and CRISPR/Cas9 HDR edited cells based on GFP expression. Cells were plated in MethoCult H4435 (STEMCELL Technologies) at 1,000 cell/mL in triplicates when possible. Colonies were scored as myeloid or erythroid (CFU-G, CFU-M, CFU-GM, and CFU/BFU-E, respectively) after 2 weeks of culture at 37°C with 5% CO₂. Individual colonies were manually picked and resuspended in 15 μL QuickExtract DNA Extraction Solution (Lucigen) for DNA extraction. Colonies were genotyped on a QX200 Droplet Digital PCR System (Bio-Rad Laboratories, Inc.) using specific SNP assays (Thermo Fisher Scientific; assay ID ANKA6UT for IDH1^R132H with primers 5′-CCAACATGACTTACTTGATCCCCATA-3′ and 5′-CTTGTGAGTGGATGGGTAAAACCTA-3′, and probes 5′-CATCATAGGTCGTCATGC-3′ and 5′-ATCATAGGTCATCATGC-3′; assay ID AH1SAKK for IDH1^R132C with primers 5′-CACATTATTGCCAACATGACTTACTTGAT-3′ and 5′-CTTGTGAGTGGATGGGTAAAACCTA-3′, and probes 5′-AAGCATGACGACCTATG-3′ and 5′-AAGCATGACAACCTATG-3′; assay ID ANPRNJJ for IDH2^R140Q with primers 5′-CTGCAGTGGGACCACTATTAT-3′ and 5′-GACTAGGCGTGGGATGTTT-3′, and probes 5′-TATCCGGAACATC-3′ and 5′-ACTATCCAGAACAT-3′; assay ID AHWR63Z for Type A NPM1^W288fs with primers 5′-GATGTCTATGAAGTGTTGTGGTTCCT-3′ and 5′-ATTTTCTTAAAGAGACTTCCTCCAC-3′, and probes 5′-CCAGACAGAGATCTT-3′ and 5′-TGCCAGAGATCTT-3′; Bio-Rad Laboratories; assay ID dHsaMDS962731646 for NRAS^G13D). FLT3-ITD was detected by PCR using primers 5′-TTCCAATGGAAAAGAAATGCTGCA-3′ and 5′-AACTGCCTATTCCTAACTGACTCA-3′.

Single-Cell RNA Sequencing and Genotyping

Single-cell RNA sequencing was performed using an optimized SMART-seq protocol (26). Single cells were sorted into 96-well plates containing lysis buffer supplemented with 1.07 AU/mL Protease (QIAGEN) and an optimized primer at 35 nmol/L (5′-GCCAACCCTTAGACAGAGCC-3′), designed to specifically bind close to the IDH1^R132H and IDH1^R132C mutations in the IDH1 transcript. Annealing and simultaneous protease inactivation were done at 72°C for 15 minutes followed by reverse transcription to cDNA using SMARTScribe Reverse Transcriptase (TakaraBio). Before preamplification, two IDH1 specific primers, 5′-TGATGCCACCAACGACCAAG-3′ and 5′-GTGTTGAGATGGACGCCTATTTG-3′, were added to amplify both gDNA and cDNA containing the IDH1 mutation. The preamplification PCR was performed with KAPA HiFi HotStart ReadyMix (Roche) for 24 PCR cycles. Before bead cleanup using AMPure XP beads (Beckman Coulter), 2 μL aliquots were removed for genotyping. Concentrations of cDNA were measured using a Fragment Analyzer High-Sensitivity NGS 1-6000 kit (Agilent Technologies) and input amounts were normalized. Libraries were prepared using Nextera XT DNA Library Preparation Kit (Illumina) with dual 10-nucleotide indexes (Integrated DNA Technologies) and sequenced on a NovaSeq 6000 (Illumina) using SP flow cells. Genotyping was done by ddPCR using custom three primer assays detecting both mRNA derived and gDNA derived IDH1 sequences, assay ID ANKCHVZ for IDH1^R132H and ANH6PA3 for IDH1^R132C (Thermo Fisher Scientific), primer 5′-CCAACATGACTTACTTGATCCCCATA-3′ (Integrated DNA Technologies) was added manually to the reaction mixture. RNA sequencing reads were trimmed from adaptors and based on phred quality scores using skewer (v.0.2.2; ref. 61). Reads were aligned to the GRCh38p.13 reference genome using STAR (v.2.7.10a) for first-pass mapping with the extraction of novel splice junctions and a second-pass mapping after rebuilding the genome index. RSEM was used for quantifying transcript abundances (62) and data normalization and visualization were done in scanpy (v.1.9.1).

CRISPR/Cas9 Genome Editing with HDR

To generate IDH1-mutant pHSCs, we precultured human CD34⁺ cells in StemSpan SFEM II (STEMCELL Technologies) supplemented with 100 ng/mL of TPO, SCF, FLT3L, IL6 (PeproTech) and the small-molecule UM-171 (35 nmol/L) at 250,000 cells/mL in low oxygen conditions (5% O2) for 48 hours before gene editing. To introduce specific double-strand breaks, cells were electroporated with preassembled ribonucleoprotein particles consisting of sgRNA and Alt-R HiFi CRISPR-Cas9 V3 (Integrated DNA Technologies) at a 1:2.5 molar ratio using the program DZ100 on a Lonza Nucleofector 4D in P3 buffer (Lonza) as previously described (63). To facilitate HDR, rAAV6 was used to deliver templates for repair. The rAAV6 vectors were either produced in HEK293 cells after transfection with 6 μg ITR-containing plasmid and 22 μg pDGM6 plasmid and harvested using the AAVpro Purification Kit (Takara Bio) or purchased from Charles River Laboratories. Plasmids with 400-bp homology arms were designed in SnapGene and cloned using NEBuilder HiFi DNA Assembly (New England Biolabs). To avoid ITR-related concatemer formation and tandem insertions, external PAM sites were introduced adjacent to the ITRs (F. Suchy; submitted for publication). Following electroporation, cells were immediately plated with rAAV6 at an MOI of 1:20,000 vector genomes per cell. Cells were washed after 6 hours and plated at 250,000 cells/mL. Edited cells were FACS isolated according to GFP expression after 72 hours in culture and genomic integration was confirmed with “In-Out PCR” using primers complementary to the genome flanking the cut site and in the insert. For the safe harbor locus AAVS1, the sgRNA 5′-GGGGCCACTAGGGACAGGATTGG-3 was used with an HDR template consisting of the external PAM site, a 400 bp left homology arm, the promoter SFFV, the fluorescent reporter GFP, a P2A self-cleaving peptide, IDH1^R132H, IDH2^R140Q or their respective WT counterpart, the BGH poly-A sequence, a 400 bp right homology arm, and the external PAM site. For the AAVS1 control, the same sgRNA was used and the HDR template consisted of the external PAM site, the left homology arm, the promoter UBC, the fluorescent reporter GFP, the BGH poly-A sequence, the right homology arm, and the external PAM site. For editing of the endogenous IDH1 locus, we used the sgRNA 5′-GCCACCCAGAATATTTCGTATGG-3′ and a two-vector system for HDR to introduce a heterozygous mutation. The constructs consisted of the external PAM site, a 400 bp left homology arm, the codon-optimized exon sequence of mutant or WT IDH1 downstream of the cut site, the SV40 poly-A sequence, the promoter UBC, the fluorescent reporter GFP or BFP, the BGH poly-A sequensce, a 400 bp right homology arm, and the external PAM site.

In Vitro Differentiation Assay

To assess the erythroid differentiation capacity of IDH1^R132H pHSCs, GFP⁺ cells were FACS isolated 72 hours after CRISPR/Cas9 HDR editing and plated in StemSpan SFEM II supplemented with Erythroid Expansion Supplement according to the manufacturer's instructions (STEMCELL Technologies). To assess megakaryocyte differentiation capacity, StemSpan SFEM II media were supplemented with human LDL and Megakaryocyte Expansion Supplement (STEMCELL Technologies). To assess myeloid differentiation capacity, cells were plated in Myelocult H5100 (STEMCELL Technologies) and supplemented with SCF, FLT3L, IL3, IL6, GM-CSF, G-CSF at 20 ng/mL (PeproTech), and 0.5 μg/mL hydrocortisone. All cultures were supplemented with 0.5% penicillin–streptomycin (Gibco). Media were changed after 4 days, and flow cytometry readout was performed between 7 and 10 days after plating.

In Vivo Engraftment Assay

To establish the engraftment potential of our edited pHSCs, 6- to 8-week-old male or female NSG or NSGS mice (The Jackson Laboratory) were sublethally irradiated with 200 rad 2 to 24 hours before intrafemoral transplantation of at least 100,000 FACS isolated GFP⁺ HSPCs 72 hours after CRISPR/Cas9 HDR editing. Mice were housed in microisolator cages in a specific pathogen-free animal facility, the experimental protocol was approved by Stanford University's Administrative Panel on Lab Animal Care (APLAC; #22264). Engraftment analyses were performed on bone marrow aspirates or whole crushed bone marrow using flow cytometry.

2-Hydroxyglutarate Assay

To measure (R)-2-hydroxyglutarate production, cells were washed in PBS, pelleted by centrifugation, and lysed using CelLytic M (Sigma-Aldrich). For detection in the culture supernatant, cells were kept in phenol-red-free media and the supernatant was directly harvested from culture plates. Samples were deproteinized using a Deproteinizing Sample Preparation Kit (Biovision) and 2HG was detected using D-2-Hydroxyglutarate (D2HG) Assay Kit (Sigma-Aldrich) with standards made fresh at the time of analyses. After 1 hour of incubation at 37°C, plates were analyzed on a fluorescent plate reader set to 540 nm excitation and 590 nm emission detection.

5hmC Assay

To examine the effects on the 5-hydroxymethylcytosine landscape caused by IDH1^R132H and IDH2^R140Q, we subjected our edited pHSCs to 5hmC-directed pulldown sequencing as previously described (36, 64, 65). In brief, edited GFP⁺ cells were FACS sorted and kept in culture for four days to allow for a phenotype to be established. DNA was subsequently harvested and a whole-genome and a 5hmC enriched library were constructed from each sample. The 5hmC bases were biotinylated and enriched by binding to Dynabeads M270 Streptavidin (Thermo Fisher Scientific). Libraries were quantified with a Qubit dsDNA High-Sensitivity Assay (Thermo Fisher Scientific) and normalized before sequencing. Fragment size profiles were determined using a Bioanalyzer dsDNA High-Sensitivity assay (Agilent Technologies). An Illumina NextSeq550 was used to sequence with 75 bp paired-end reads using version 2 reagent chemistry according to the manufacturer's instructions (Illumina). Reads were aligned to the GRCh38 reference genome using the default parameters of BWA-MEM. Duplicate reads were marked and removed from the aligned BAM files using Picard MarkDuplicates. Peak calling was done in MACS2 using standard parameters (66). To estimate changes in total gene body 5hmC patterns, peak height was multiplied with peaks per million and compared between samples. Differential gene 5hmC patterns were visualized using principal component analyses plots (Qlucore) and GSEA was used to assay hallmark gene sets differentially enriched between sample types.

RNA Sequencing

To examine the effects on the transcriptome caused by IDH1^R132H and IDH2^R140Q, edited GFP⁺ cells were FACS sorted and kept in culture for four days to allow for a phenotype to be established. Cells were washed in PBS and resuspended in RNA/DNA Shield (Zymo Research) for transport to the Novogene commercial sequencing facility. The RNA was extracted using poly-T oligo-attached magnetic beads and first-strand cDNA was synthesized using random hexamer primers with subsequent second-strand synthesis, end repair, A-tailing, adapter ligation, size selection, amplification, and purification. Novogene also provided Bioanalyzer quality control, Illumina-based massive parallel sequencing, and initial quality control. All samples had a RIN value of 10 and >94% of bases had phred scores >30, indicating excellent sequencing quality. Reads were trimmed using skewer (v.0.2.2; ref. 61), and reads were aligned to the GRCh38p.13 reference genome using STAR (v.2.7.10a) for first-pass mapping with the extraction of novel splice junctions and a second-pass mapping after rebuilding the genome index. RSEM was used for quantifying transcript abundances (62). Data normalization and visualization were done using DESeq2 (v1.38.3) and Qlucore.

Metabolic Assay

Genetically edited cells were subjected to the metabolic assay SCENITH where puromycin integration is used as a marker for protein synthesis and ATP consumption during inhibition of specific metabolic pathways (44). Briefly, edited GFP⁺ cells were sorted and kept in culture for four days to allow for a phenotype to be established. Cells were then washed and plated in culture media alone or supplemented with 100 mmol/L 2-deoxyglucose, 1 μmol/L oligomycin A, or a combination of both compounds and kept at 37°C for 30 minutes before adding O-propargyl-puromycin for a final concentration of 10 μg/mL, followed by 30 minutes incubation at 37°C. Cells were washed and stained on ice with Ghost Dye Red 780 (Cytek Biosciences) for 20 minutes before washing and subsequent fixation and permeabilization using 2% PFA and 1% saponin. To quantify the puromycin integration, a click chemistry approach was used. Cells were resuspended in PBS with 0.5 mmol/L copper sulfate, 2 μmol/L AF647 Azide dye (Thermo Fisher Scientific), and 5 mmol/L sodium ascorbate and then washed and resuspended in PBS before analyzes on a BD FACSymphony A5 Cell Analyzer. Technical triplicates from three biological replicates were used.

MitoTracker and Mitochondrial Membrane Potential

Cells edited to express IDH1^R132H or IDH2^R140Q as well as IDH1^Wt and AAVS1 controls were sorted based on GFP expression and kept in culture for four days to allow for a phenotype to be established. Mitochondrial mass was assessed by incubating cells with MitoTracker Red CMXRos (Thermo Fisher) at 50 nmol/L for 20 minutes at 37°C, washed twice, and analyzed by flow cytometry. The mitochondrial membrane potential was assessed by incubating cells with MitoProbe TMRM (Thermo Fisher) at 20 nmol/L for 30 minutes at 37°C, washed twice, and analyzed by flow cytometry, cells treated with CCCP (carbonyl cyanide 3-chlorophenylhydrazone) were used as controls for membrane depolarization.

In Vitro Drug Treatment

Primary human pHSCs/HSCs from patients with AML were FACS isolated based on immunophenotype and CRISPR/Cas9 HDR-edited cells based on GFP expression. Cells were treated with DMSO control, IACS-010759 at 5 μmol/L, ivosidenib 10 μmol/L, enasidenib 10 μmol/L, IM156 at 50 μmol/L, or rotenone 10 μmol/L for 24 hours and subsequently collected and resuspended in MethoCult H4435 (STEMCELL Technologies) for plating. Colonies were scored after 2 weeks of culture at 37°C with 5% CO₂ and individually picked and resuspended in 15 μL QuickExtract DNA Extraction Solution (Lucigen) for DNA extraction. Colonies were genotyped using ddPCR as described above.

Statistical Analyses

Statistical analyses were performed using GraphPad Prism (v 9.5.1). In experiments using primary patient material biological triplicates or more were used if not otherwise stated. In experiments using the genetically engineered model system, biological (different HSPC donors) and/or technical triplicates or more were used. Two-sided Student t test was used to compare two groups with normally distributed data. Fisher exact test was used to compare the distribution between two groups. One-way ANOVA was used to compare more than two groups. False discovery rate thresholds of 0.05 or 0.01 were used as specified in the manuscript.

Data Availability

The data generated in this study are available within the article and its supplementary data, sequencing data are available via NCBI's Gene-Expression Omnibus (67), using accession number GSE245669.

Supplementary Material

Supplementary Tables 1 and 2

Supplementary Table 1. Clinical characteristics of the 17 IDH-mutant AML patients assessed. Supplementary Table 2. Summary of antibodies and flow cytometry reagents.

Supplementary Figures 1-6

Supplementary Figure 1. IDH2 is mutated in preleukemic hematopoietic stem cells in relapse causing clones in AML. Supplementary Figure 2. Single-cell RNA sequencing and combined genotyping reveal distinct transcriptomic profiles of IDH1-mutant pHSCs in AML Supplementary Figure 3. Modeling human pHSCs using CRISPR/Cas9 with HDR show reduced proliferation and blocked differentiation caused by IDH1 mutations Supplementary Figure 4. IDH1 mutations reduce 5-hydroxymethylcytosine in gene bodies Supplementary Figure 5. IDH1 mutations drives transcriptomic alterations Supplementary Figure 6. IDH1-mutant pHSCs are sensitive to inhibition of oxidative phosphorylation

Authors’ Disclosures

T. Köhnke reports personal fees from TenSixteen Bio outside the submitted work and is a special fellow of The Leukemia and Lymphoma Society. M.H. Linde reports personal fees from Scribe Biosciences, Inc. outside the submitted work. R. Majeti reports grants from NIH, Stanford Ludwig Center for Cancer Stem Cell Research, Leukemia and Lymphoma Society, The Mark Foundation for Cancer Research, The Paul G. Allen Frontiers Group, Snowdome Foundation, and Leukaemia Foundation during the conduct of the study, personal fees and other support from Kodikaz Therapeutic Solutions, Orbital Therapeutics, and Pheast Therapeutics, personal fees from 858 Therapeutics, and other support from MyeloGene and Gilead Sciences outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

N. Landberg: Conceptualization, data curation, formal analysis, funding acquisition, visualization, methodology, writing–original draft, writing–review and editing. T. Köhnke: Formal analysis, writing–review and editing. Y. Feng: Formal analysis, writing–review and editing. Y. Nakauchi: Formal analysis, writing–review and editing. A.C. Fan: Formal analysis, writing–review and editing. M.H. Linde: Formal analysis, writing–review and editing. D. Karigane: Formal analysis, writing–review and editing. K. Lim: Formal analysis, writing–review and editing. R. Sinha: Formal analysis, writing–review and editing. L. Malcovati: Data curation, writing–review and editing. D. Thomas: Conceptualization, formal analysis, supervision, writing–original draft, writing–review and editing. R. Majeti: Conceptualization, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing.