Distinct and opposite effects of leukemogenic Idh and Tet2 mutations in hematopoietic stem and progenitor cells

Significance

In myeloid neoplasms, treatment outcome is associated with the presence of specific combinations of driver mutations. Delineating their molecular effects is important to understand the basis of disease heterogeneity and to help design better therapies. We discovered that two mutually exclusive leukemogenic mutations, Idh2^R172K and Tet2 loss-of-function, unexpectedly cause opposite molecular alterations in hematopoietic stem and progenitor cells. These results could pave the way for the development of more effective and patient-specific treatments.

Abstract

Mutations in IDH1, IDH2, and TET2 are recurrently observed in myeloid neoplasms. IDH1 and IDH2 encode isocitrate dehydrogenase isoforms, which normally catalyze the conversion of isocitrate to α-ketoglutarate (α-KG). Oncogenic IDH1/2 mutations confer neomorphic activity, leading to the production of D-2-hydroxyglutarate (D-2-HG), a potent inhibitor of α-KG-dependent enzymes which include the TET methylcytosine dioxygenases. Given their mutual exclusivity in myeloid neoplasms, IDH1, IDH2, and TET2 mutations may converge on a common oncogenic mechanism. Contrary to this expectation, we observed that they have distinct, and even opposite, effects on hematopoietic stem and progenitor cells in genetically engineered mice. Epigenetic and single-cell transcriptomic analyses revealed that Idh2^R172K and Tet2 loss-of-function have divergent consequences on the expression and activity of key hematopoietic and leukemogenic regulators. Notably, chromatin accessibility and transcriptional deregulation in Idh2^R172K cells were partially disconnected from DNA methylation alterations. These results highlight unanticipated divergent effects of IDH1/2 and TET2 mutations, providing support for the optimization of genotype-specific therapies.

Myeloid neoplasms are characterized by the altered proliferation and aberrant differentiation of immature myeloid cells (1). Within this disease spectrum, acute myeloid leukemia (AML) has the worst prognosis, which is heavily influenced by the underlying genetics (1–4). In many cases, AML emerges from myelodysplastic syndrome (MDS) or myeloproliferative neoplasm (MPN) transformation, occasionally preceded by clonal hematopoiesis (CH) (1, 5).

Alterations in epigenetic regulators are found in a large proportion of myeloid neoplasms and AML (6). Among these, missense mutations in the genes encoding isocitrate dehydrogenase-1 and -2 (IDH1 and IDH2) are observed in around 20% of AML and 5 to 10% of MDS and MPN (3, 4, 6, 7). These mutations cause substitutions at arginine 132 (R132) in IDH1, at the equivalent residue (R172) in IDH2, or at R140 in IDH2 (3, 4, 8–11). Canonically, wild-type IDH1/2 catalyzes the reversible conversion of isocitrate into α-ketoglutarate (α-KG) (8). Mutant IDH1/2 gains neomorphic enzymatic activity, reducing α-KG to D-2-hydoxyglutarate (D-2-HG) (11, 12), a potent inhibitor of several dioxygenases that normally use α-KG as a cofactor (13, 14). Among those enzymes are the ten-eleven translocation (TET) family of methylcytosine dioxygenases, which hydroxylate 5-methylcytosine (5-mC) bases in the DNA to generate 5-hydroxymethylcytosine (5-hmC) (13, 15).

TET2 is mutated in 15 to 30% of MDS or MPN and 10 to 20% of AML (3, 4, 15–20). TET2 and IDH1/2 lesions are mutually exclusive and display an overlapping hypermethylated phenotype (3, 4, 13, 21). In the rare instances where both occur in the same tumor, they are most often in distinct subclones (22, 23). Thus, the oncogenic effect of D-2-HG in hematopoietic cells could be mainly attributable to the inhibition of TET2. However, several observations argue against a simple equivalence of IDH1/2 and TET2 mutations. First, D-2-HG can alter the activity of multiple α-KG-dependent dioxygenases beyond TET family members, including several histone demethylases and prolyl hydroxylases, thus potentially affecting a larger spectrum of cellular processes (8, 14, 24). Second, TET2 mutations are frequent in CH, whereas IDH1/2 lesions are much rarer, suggesting distinct effects on hematopoietic stem cell (HSC) fitness or differences in time to progression to more overt disease (5, 25, 26). Third, while TET2 and IDH2 mutations are mutually exclusive in AML, they frequently co-occur in a distinct malignancy, angioimmunoblastic T cell lymphoma (AITL), pointing to cell context-dependent interplay between them (27, 28). Fourth, divergent patterns of co-occurring lesions are observed in AML with TET2 versus IDH1/2 mutations (3, 4). Nevertheless, it remains unclear whether these differences reflect non-overlapping or altogether opposite molecular effects of IDH1/2 and TET2 lesions. Adding further complexity, there are apparent differences between IDH1/2 mutations (3, 4, 29), possibly related to the level of D-2-HG production (30) or to isoform-specific cellular localization (IDH1 in cytoplasm versus IDH2 in mitochondria).

IDH inhibitors have shown success in the treatment of AML (31–33), but the description of resistance mechanisms highlights the need to better understand the oncogenic mechanism of action of IDH1/2 mutations and to develop improved genotype-specific treatments (34–37). Toward this goal, we set out to delineate the cellular effects of distinct IDH1 and IDH2 mutations and how they may differ from TET2 loss-of-function, by performing phenotypic, transcriptomic, and epigenomic analyses of hematopoietic stem and progenitor cells in genetically engineered mice.

Results

Graded Hematopoietic Effects of Idh1/2 Mutations in Knock-in Mice.

To directly compare the effects of Idh1^R132H, Idh2^R140Q, and Idh2^R172K on hematopoiesis, we crossed mice carrying the Cre-inducible Idh1^LSL-R132H, Idh2^LSL-R140Q, and Idh2^LSL-R172K knock-in alleles with Vav-Cre transgenic animals, thus activating the mutations in hematopoietic stem and progenitor cells. The recombination efficiency of all three alleles was similar (SI Appendix, Fig. S1A). Mice expressing the knock-in alleles displayed elevated levels of D-2-HG in the circulation and in lineage-negative (Lin⁻) bone marrow cells (Fig. 1 A and B), which were much higher in Idh2^R172K mice than in Idh1^R132H and Idh2^R140Q animals. These differences, which are in line with previously reported data from engineered cell lines (30), are unlikely to be due to technical variations between the similarly constructed mutant alleles. Indeed, while Idh2 was modestly upregulated in Idh2^R172K cells, this was primarily attributable to increased relative expression of the wild-type allele, probably due to homeostatic feedback (SI Appendix, Fig. S1B). Consistently, total IDH2 protein levels were elevated in Idh2^R172K cells (SI Appendix, Fig. S1C).

Fig. 1.

(A and B) Level of D-2-HG in the blood (A) and lineage-negative cells (B) isolated from 3 to 9-mo-old mice with the indicated genotypes. In panel A, n = 68 to 272 per genotype. In panel B, n = 4 per genotype. (C) Kaplan–Meier curves depicting the survival of mice with the indicated genotypes. (D) Disease classification of Idh2^R172K mice at humane endpoint, based on flow cytometry analysis. (E) Representative spleen (Left), H&E-stained spleen histological section (Center) and peripheral blood smear (Right) of Idh2^R172K and control mice at endpoint. (F) Counts of circulating white blood cells and red blood cells in Idh2^R172K mice exhibiting myeloid neoplasms, and in their control littermates, at humane endpoint. Each dot represents an individual animal. (G) Representative flow cytometry plots and quantification of CD11b⁺Gr1⁺ myeloid cells, CD71⁺Ter119⁺ erythroid cells, and Lin^-cKit⁺Sca1⁺ hematopoietic progenitors in the bone marrow and spleen of Idh2^R172K mice exhibiting myeloid neoplasms and of their control littermates, at humane endpoint. Each dot represents an individual animal. Data were analyzed using ANOVA with Tukey post hoc tests (A and B) and t tests (F and G). **P < 0.01; ***P < 0.001.

Idh2^R172K mice succumbed to lethal hematopoietic disease with a median latency of 324 d (Fig. 1C). Tet2^−/− mice showed a milder phenotype, with a median survival of 448 d matching previous reports (Fig. 1C) (38–41). Most (~60%) of the diseased Idh2^R172K mice exhibited overt MDS/MPN, and a smaller proportion developed lymphoid malignancies (Fig. 1D and SI Appendix, Fig. S2 A and B). Animals with myeloid disease displayed splenomegaly, peripheral leukocytosis, anemia, and blasts in the circulating blood (Fig. 1 E and F). In the bone marrow, CD11b⁺Gr1⁺ myeloid cells were expanded, while CD71⁺Ter119⁺ erythroid progenitors were depleted (Fig. 1G). Concomitantly, the enlarged spleens exhibited disrupted architecture and contained a high proportion of myeloid and erythroid cells (Fig. 1 E and G). The spleens also harbored a substantial population of hematopoietic Lin⁻Sca1⁺cKit⁺ (LSK) progenitors (Fig. 1G), which included some LSK CD150⁺CD48⁻ long-term hematopoietic stem cells (LT-HSCs), indicative of extramedullary hematopoiesis (SI Appendix, Fig. S2C). LSK cell expansion was also noted in the bone marrow (Fig. 1G). Similar myeloproliferative diseases were occasionally observed in aging Idh1^R132H and Idh2^R140Q mice (SI Appendix, Fig. S2D). Targeted sequencing using the M-IMPACT assay detected very few additional variants, of uncertain significance, in Idh2^R172K MDS/MPN (SI Appendix, Fig. S2E and Dataset S1).

Complete blood count analysis indicated neutropenia, lymphopenia, and anemia as early as 3 mo of age in Idh2^R172K animals (SI Appendix, Fig. S3A). Erythrocyte mean corpuscular volume (MCV) was also increased in Idh2^R172K mice, suggesting dysplasia. Idh1^R132H and Idh2^R140Q mice were largely normal across all the measured parameters, although both exhibited mild increases in erythrocyte MCV (SI Appendix, Fig. S3A). In addition, older Idh2^R140Q mice showed moderate lymphopenia (SI Appendix, Fig. S3A). Overall, these results indicate that under otherwise identical genetic conditions, distinct Idh1/2 mutations differ substantially in their phenotypic consequences: Idh2^R172K causes progressive hematopoietic defects, evolving to overt myeloid or lymphoid neoplasms, while Idh2^R132H and Idh2^R140Q have much milder effects.

Idh2^R172K and Tet2 Loss of Function Differentially Affect Hematopoietic Stem and Progenitor Cells.

To better understand the effects of Idh1/2 mutations on early hematopoiesis, we examined the stem and progenitor cell compartments in the bone marrow of 3 to 9-mo-old mice. Idh2^R172K mice displayed increased percentages of Lin⁻cKit⁺Sca1⁻ (LK) erythromyeloid progenitor, as well as Lin⁻cKit⁺Sca1⁺ (LSK) stem and early progenitor cells (Fig. 2 A and B and SI Appendix, Fig. S3B). Within the LSK compartment, Idh2^R172K mice had normal abundance of CD150⁻CD48⁻ multipotent progenitors (MPPs), but increased percentages of slightly more committed CD150⁻CD48⁺ HPC-1 and CD150⁺CD48⁺ HPC-2 progenitors and decreased CD150⁺CD48⁻ LT-HSCs (Fig. 2C and SI Appendix, Fig. S3B). This was confirmed using alternative markers, CD34 and Flt3 (Fig. 2D). Within the LK compartment, Idh2^R172K animals had elevated percentages of CD34⁺CD16/32^mid common myeloid progenitors (CMP) and CD34⁺CD16/32⁺ granulocyte/monocyte progenitors (GMPs), whereas CD34⁻CD16/32⁻ megakaryocyte/erythroid progenitors (MEPs) were decreased, indicating myeloid skewing (Fig. 2E). Overall, Idh2^R172K animals somewhat differed from Tet2-deficient mice, which have increased LSK cells and progenitors but normal numbers of LT-HSCs, as previously reported (39, 40, 42, 43) and confirmed here (SI Appendix, Fig. S3C). Idh1^R132Hand Idh2^R140Q mice did not substantially differ from control littermates (Fig. 2 B–E and SI Appendix, Fig. S3B).

Fig. 2.

(A) Representative flow cytometry plots depicting the gating strategy to identify Lin⁻cKit⁺Sca1⁺ (LSK) hematopoietic progenitors and their subsets of long-term hematopoietic stem cells (LT-HSCs), multipotent progenitors (MPPs), and hematopoietic progenitor cells (HPC-1 and HPC-2) in the bone marrow of Idh2^R172K and control mice. (B) Quantification of LSK cells in the bone marrow of 3 to 9-mo-old mice with the indicated genotype. Each dot represents an individual mouse. (C) Quantification of LT-HSC, HPC-1, HPC-2, and MPP cells in the bone marrow of 3 to 9-mo-old mice with the indicated genotype. Each dot represents an individual mouse. (D) Representative flow cytometry plots and quantification of LSK subsets divided based on Flt3 and CD34 immunoreactivity to identify LT-HSCs, short-term hematopoietic stem cells (ST-HSCs), and MPPs in the bone marrow of 3 to 9-mo-old mice with the indicated genotypes. Each dot represents an individual animal. (E) Representative flow cytometry plots and quantification of Lin⁻cKit⁺ (LK) erythromyeloid progenitor subsets (common myeloid progenitors, CMP; granulocyte/monotype progenitors, GMP; megakaryocyte/erythroid progenitor, MEP) in the bone marrow of 3 to 9-mo-old mice with the indicated genotypes. Each dot represents an individual animal. (F) Proportion of LSK cells immunoreactive for Flt3, of Flt3⁺LSK cells immunoreactive for IL7R, and of HPC-1 cells negative for Flt3, in the bone marrow of mice with the indicated genotype, assessed by CYTOF. Each dot represents an individual animal. (G) Proportion of Flt3⁻HPC-1 cells positive for iododeoxyuridine (IdU) in the bone marrow of mice with the indicated genotype, assessed by CyTOF. Each dot represents an individual animal. (H) Schematic summarizing hematopoietic progenitor alterations in Idh2^R172K mice. Thicker and dashed lines represent differences from wild type. (I) Number of colonies generated in primary plating of LT-HSCs sorted from mice with the indicated genotype (Left), and in subsequent serial replating (Right), in M3434 methylcellulose medium. N = 3 to 5 mice per genotype. Data were analyzed with ANOVA and Tukey (B–E and I) or Dunnett (F and G, with +/+;Vav-Cre mice being the control group) post hoc tests. *P < 0.05; **P < 0.01; ***P < 0.001.

To confirm and extend our flow cytometric assessment, we examined stem and progenitor cells in the bone marrow of healthy 4-mo-old mutant and control mice using time-of-flight mass cytometry (CyTOF) (SI Appendix, Fig. S4 A–D). Results for the major stem and progenitor subsets across the LK and LSK compartments were consistent with the data obtained by fluorescent flow cytometry (SI Appendix, Fig. S4 A and B). We further dissected the progenitor subsets by separating the LSK cells based on the expression of Flt3, which distinguishes lymphoid-biased (Flt3⁺) from myeloid-biased (Flt3⁻) progenitors. Flt3⁺ LSK cells were increased in Idh2^R172K mice (Fig. 2F and SI Appendix, Fig. S4A). This population can be divided between IL7R-α⁻ lymphoid-biased progenitors and slightly more differentiated IL7R-α⁺ committed lymphoid precursors (CLP). CyTOF analyses indicated increased Flt3⁺ IL7R-α⁻ LSK progenitors, but decreased CLPs in Idh2^R172K animals, suggesting a differentiation block during lymphoid commitment (Fig. 2F). Consistent with the lower number of MEPs observed in Idh2^R172K mice, Ter119⁺CD71⁺ and CD71⁺CD45^lo erythroblasts were significantly reduced (SI Appendix, Fig. S4C). Idh1^R132H and Idh2^R140Q mice were largely noral across all the parameters evaluated (SI Appendix, Fig. S4 B and C). To assess whether changes in stem and progenitor subsets in Idh2^R172K mice could be explained by abnormal proliferation, we pulse-labeled cells in vitro with Iodo-deoxy-uridine (IdU) prior to CyTOF analysis. The frequency of IdU⁺ cells was either unchanged (in LT-HSCs, MPPs, and erythroblasts) or decreased (in Flt3^- HPC-1s and HPC-2s, and in erythromyeloid progenitors) (Fig. 2G and SI Appendix, Fig. S4 D and E). Collectively, these results indicated that alterations in progenitor numbers in Idh2^R172K mice were not accompanied by corresponding changes in proliferation and likely reflect impaired differentiation (Fig. 2H).

To evaluate whether stem and progenitor alterations affected their self-renewal potential, we measured the colony-forming ability of bone marrow cells in serial replating assays in methylcellulose medium containing cytokines (IL-3, IL-6, SCF, and EPO). While control bone marrow cells exhibited a much-reduced capacity to form new colonies by the fourth plating, Idh2^R172K cells retained colony-forming potential through at least eight replatings (SI Appendix, Fig. S4F). Idh1^R132H and Idh2^R140Q cells preserved some ability to yield colonies through five and six platings, respectively (SI Appendix, Fig. S4F). Given that Idh2^R172K, but not Tet2^−/−, causes a depletion of the LT-HSCs in vivo (SI Appendix, Fig. S3C and Refs. 39, 40, and 42), we asked whether the two mutations differentially affect HSC functionality in vitro. In contrast to Tet2^−/−, sorted Idh2^R172K LT-HSCs were profoundly impaired in their initial colony-forming potential (Fig. 2I). However, Tet2^−/− and Idh2^R172K cells showed similarly enhanced serial replating ability through at least six passages (Fig. 2I). These data suggest that Idh2^R172K and Tet2^−/− both enhance progenitor self-renewal but have distinct effects on HSC fitness.

Transcriptomic Effects of Idh2^R172K Revealed by scRNA-seq.

To investigate the molecular basis for differences between the Idh1/2 and Tet2 mutants on hematopoietic stem and progenitor cells, we analyzed the transcriptome of LSK cells from young adults (before any overt disease onset) +/+;Vav-Cre (WT), Idh1^R132H, Idh2^R140Q, Idh2^R172K, and Tet2^−/− mice. t-SNE analysis indicated that Idh2^R172K and Tet2^−/− cells robustly clustered away from controls and from each other (Fig. 3A). By contrast, Idh1^R132H and Idh2^R140Q samples intermingled with controls (Fig. 3A). Most of the differentially expressed genes in Idh2^R172K and Tet2^−/− cells did not overlap (Fig. 3B), with many showing opposite trends compared with WT (SI Appendix, Fig. S5A). In agreement with our flow cytometry analyses, Gene Set Enrichment Analysis (GSEA) indicated that Idh2^R172K and Tet2^−/− cells downregulated markers of hematopoietic progenitor differentiation and proliferation but showed opposite alterations of genes associated with HSCs and early progenitors (SI Appendix, Fig. S5B).

Fig. 3.

(A) t-SNE plot depicting the similarity in overall transcriptional profile of bone marrow LSK cells isolated from mice with the indicated genotypes, based on RNA-sequencing analysis. Each dot represents an independent sample (biological replicate), each derived from cells pooled from two mice. (B) Plot depicting significantly upregulated and downregulated genes in LSK cells with the indicated genotypes, compared with +/+;Vav-Cre (WT) cells. (C) t-SNE and UMAP representations of bone marrow LSK cells from mice with the indicated genotypes, clustered based on the similarity of their transcriptome determined by scRNA-seq. Cells are colored according to their Seurat-identified cluster number and genotype. (D) SCENIC analysis of transcription factor regulon activity in the Seurat-identified clusters depicted in panel C. The Z score of regulon activity in each cluster, calculated relative to the average regulon activity in all the clusters, is represented by the size and color scale of the dots. (E) Slingshot-predicted differentiation trajectories, starting from HSCs, overlaid on top of the UMAP clustering of LSK cells shown in panel C. (F) UMAP representation of bone marrow cells from mice with the indicated genotypes, clustered based on the similarity of their transcriptome determined by scRNA-seq. WT (+/+;Vav-Cre) and Idh2^R172K cells are from the LSK samples shown in panel C. WT (+/+;Mx1-Cre) and Tet2^−/− (Tet2^flox/flox;Mx1-Cre) cells are from a previously published dataset of Lin⁻ bone marrow cells. Clustering was performed following iterative rounds of data integration as described in the Methods section. Clusters of interest are emphasized. (G) SCENIC analysis of transcription factor regulon activity in the Seurat-identified clusters depicted in the inset in panel F. The Z score of regulon activity in each cluster, calculated relative to the average regulon activity in all the clusters, is represented by the size and color scale of the dots.

To examine the transcriptomic changes driven by mutant IDH1/2 with more granularity, we performed single-cell RNA-seq on LSK cells isolated from Idh1/2 mutant mice and their WT controls (SI Appendix, Fig. S6A). Clustering using Seurat, and visualization by t-SNE and UMAP, revealed that Idh2^R172K cells robustly segregated from the other genotypes (Fig. 3C). Idh1^R132H, Idh2^R140Q, and WT cells remained intermingled even when reclustered in the absence of Idh2^R172K cells (SI Appendix, Fig. S6 B and C), consistent with another Idh2^R140Q mouse model (44). Eleven “normal” clusters comprising mostly Idh1^R132H, Idh2^R140Q, and WT cells were functionally annotated using marker genes and SCENIC (45) to predict transcription factor activity. These analyses identified HSCs/early progenitors (Gata2, Hlf, Meis1, and Procr), erythroid-biased progenitors (Gata1, Pbx1, Stat5, and Tal1), myeloid-biased progenitors (Cebpb, Ctsg, Mpo, and Irf8), lymphoid-biased progenitors (Dntt, Il7r, Ikzf1, and Klf3), transitory MPPs, and proliferative (activated) cells (E2f3, E2f8, mKi67, and Top2a) (Fig. 3D and SI Appendix, Fig. S6D).

We next examined the four clusters that contained almost exclusively Idh2^R172K cells. The largest (Idh2^R172K Prog.1) expressed high levels of the HSC-enriched genes Gata2 and Kit (SI Appendix, Fig. S6E). SCENIC indicated that these cells had features of HSCs and early progenitors but aberrantly upregulated Runx1- and Sox4-driven gene expression programs (Fig. 3D). This was also seen in another Idh2^R172K-specific cluster (Idh2^R172K Prog.2) (Fig. 3D), which further upregulated modulators of self-renewal and differentiation (e.g., Cdk6, Pim1, Notch2, and Kdm6b) (SI Appendix, Fig. S6 D and E). Accordingly, this cluster uniquely combined transcription factor activities seen in HSCs and activated progenitors (Fig. 3D). Thus, Idh2^R172K Prog.1, and Idh2^R172K Prog.2 likely represent quiescent and activated/self-renewing abnormal early progenitors, respectively. A third Idh2^R172K-specific cluster (Idh2^R172K Prog.3) segregated from other cell populations by t-SNE analysis but expressed few unique markers (SI Appendix, Fig. S6 D and E). An “Idh2^R172K lymphoid” cluster highly expressed markers of lymphoid-primed progenitors (e.g., Dntt, Il12a, Il7r, and Flt3) (SI Appendix, Fig. S6 D and E) and aberrantly upregulated Runx1- and Sox4-driven transcriptional programs (Fig. 3D).

The above results suggested that Idh2^R172K disrupts hematopoietic stem and progenitor differentiation. To investigate this possibility further, we used Slingshot (46) to predict cell differentiation trajectories, using HSCs as the starting point. This analysis inferred lineage paths flowing through early and transitory progenitors and branching off to terminate in lymphoid-biased (Lineage 2), myeloid-biased (Lineage 5), erythroid-biased (Lineage 3), and activated cell clusters (Lineage 1) (Fig. 3E and SI Appendix, Fig. S7A). Two additional trajectories comprising mostly Idh2^R172K cells were observed: one terminated within the activated/self-renewing Idh2^R172K Prog.2/3 clusters (Lineage 4) and the other within the large population of abnormal quiescent cells forming the Idh2^R172K Prog.1 cluster (Lineage 6) (Fig. 3E and SI Appendix, Fig. S7A). These data, together with the marker and transcription factor activity analyses described above, are consistent with impaired differentiation of early Idh2^R172K progenitors.

Tet2-Deficient and Idh2^R172K Abnormal Early Progenitors Are Transcriptionally Different.

To understand in more detail how Tet2^−/− and Idh2^R172K LSK cells transcriptionally differ, we combined our single-cell transcriptomic results with a published dataset from Lin⁻ Tet2^flox/flox;Mx1-Cre (hereafter, Tet2^−/−) bone marrow cells (44). Data integration using Seurat (47), and iterative clustering, indicated that WT cells from the two studies largely intermingled, whereas Idh2^R172K cells clustered separately, as expected (Fig. 3F). Interestingly, around 40% of the Tet2^−/− cells formed a population that segregated from both control and Idh2^R172K cells (cluster 7 in Fig. 3F). Marker and SCENIC analyses indicated that these cells represented early progenitors, downstream of HSCs (Fig. 3G). Differential gene expression analysis between this cell population and WT early progenitors indicated relatively modest differences but was notable for upregulation of the myeloid oncogene and hematopoietic self-renewal driver Myb (48–50) and downregulation of the HSC/progenitor regulator Lmo2 and of the AP-1 family transcription factors Fos and Junb in Tet2^−/− cells (SI Appendix, Fig. S7B). Overall, bulk and single-cell studies indicated distinct effects of Tet2 and Idh2^R172K mutations on the transcriptome of LSK cells.

Distinct and Opposite Chromatin Accessibility Changes Induced by Idh2^R172K and Tet2^−/−.

Given that D-2-HG inhibits histone lysine demethylases and TET methylcytosine dioxygenases, we next examined epigenetic alterations in Idh1/2 and Tet2 mutated cells. We first profiled chromatin accessibility in LSK cells by ATAC-seq. As seen with RNA-seq, Idh2^R172K samples profoundly differed from the other genotypes, corresponding to a large number of differentially accessible loci (Fig. 4 A and B and SI Appendix, Fig. S8). Tet2^−/− samples also diverged from WT but showed fewer significant changes (Fig. 4 A and B and SI Appendix, Fig. S8). Idh1^R132H and Idh2^R140Q cells did not display robust alterations (Fig. 4B and SI Appendix, Fig. S8). In Idh2^R172K and Tet2^−/− cells, differences in gene expression correlated with changes in chromatin accessibility near gene promoters (SI Appendix, Fig. S9A). We next examined previously identified active LSK enhancers, marked by H3 acetylated on lysine 27 (H3K27ac) and methylated on lysine 4 (H3K4me1) (51). Nearly all the significantly altered LSK enhancers in Tet2^−/− cells were closing (109/110; 99%— Fig. 4B and Dataset S2). In striking contrast, Idh2^R172K cells harbored a mixture of opening (691/2,461; 28%) and closing (1,770/2,461; 72%) enhancers (Fig. 4B and Dataset S2). For example, Idh2^R172K-specific upregulation of Gata2 and Meis1 was associated with increased accessibility of known regulatory elements (52–56) (Fig. 4 C–E). By contrast, downregulation of the HSC modulator Prdm16 (57–59) in Idh2^R172K cells was accompanied by chromatin closing near the promoter and within an intronic enhancer (51) (Fig. 4F).

Fig. 4.

(A) t-SNE plot depicting the similarity in overall chromatin accessibility profile of LSK cells isolated from mice with the indicated genotypes, based on ATAC-seq analysis. Each dot represents an independent sample (biological replicate), each derived from cells pooled from two mice. (B) Top: Plot depicting loci with significantly higher or lower chromatin accessibility in LSK cells with the indicated genotypes, compared with +/+;Vav-Cre (WT) cells. Bottom: proportion of significantly altered LSK enhancers that are opening and closing in Idh2^R172K and Tet2^−/− cells. Data were analyzed by a chi-square test with Yates correction. (C) ATAC-seq tracks representing the normalized average chromatin accessibility near the Gata2 gene in LSK cells isolated from mice with the indicated genotypes. Tracks depicting the levels of H3K27ac and H3K4me1, derived from published chromatin immunoprecipitation experiments in LSK cells, were aligned to the ATAC-seq tracks. Differentially accessible peaks of interest are boxed. (D) Relative expression of the indicated genes, determined by bulk RNA-seq analysis, in LSK cells isolated from mice with the indicated genotypes. Each dot represents an individual sample. (E) ATAC-seq tracks representing the normalized average chromatin accessibility near the Meis1 gene in LSK cells isolated from mice with the indicated genotypes. Tracks depicting MEIS1, H3K27ac, and H3K4me1 occupancy at an autoregulatory enhancer (boxed), derived from published chromatin immunoprecipitation experiments in LSK and HPC-7 cells, were aligned to the ATAC-seq tracks. (F) ATAC-seq tracks representing the normalized average chromatin accessibility near the Prdm16 gene in LSK cells isolated from mice with the indicated genotypes. Tracks depicting the levels of H3K27ac and H3K4me1, derived from published chromatin immunoprecipitation experiments in LSK cells, were aligned to the ATAC-seq tracks. A differentially accessible peak of interest is boxed. (G) Enrichment of the indicated transcription factor binding motifs at loci gaining (opening) or losing (closing) chromatin accessibility in LSK cells isolated from mice with the indicated genotype. P values are indicated using a color scale, which was truncated at log P = −25 for clarity. (H) Relative expression of the indicated genes, determined by bulk RNA-seq analysis, in LSK cells isolated from mice with the indicated genotypes. Each dot represents an individual sample. (I) ATAC-seq tracks representing the normalized average chromatin accessibility near the Tgfbr1 gene in LSK cells isolated from mice with the indicated genotypes. (J) GSEA plot depicting the relative expression of TGFβ signaling pathway genes in Idh2^R172K and Tet2^−/− LSK bone marrow cells, compared to their wild-type counterparts, based on RNA-sequencing data.

To identify potential regulatory networks affected in Idh2^R172K and Tet2^–/– cells, we probed the differentially accessible chromatin regions for known transcription factor binding sites. Gata2, Runx1, Meis1, and Sox motifs were enriched at chromatin opening loci in Idh2^R172K cells (Fig. 4G), matching the SCENIC analysis of single-cell transcriptomic data (Fig. 3D). In contrast, motifs for positive regulators of myeloid maturation, such as Spi1/Pu.1, Irf8, and Cebp, were enriched at sites of chromatin closing (Fig. 4G), as seen near the promoter of the mature myeloid marker Fcer1g (SI Appendix, Fig. S9B). As in Idh2^R172K cells, Runx1 and Meis1 motifs significantly opened in Tet2^–/– cells, albeit more modestly (Fig. 4G). In striking opposition to Idh2^R172K cells, Gata2 motifs were enriched in closing chromatin sites in Tet2^–/– cells (Fig. 4G), accompanied by a mild downregulation of Gata2 expression (Fig. 4D). Hox binding sites gained accessibility in Tet2^–/– cells compared with Idh2^R172K cells, tracking with Tet2^–/–-specific upregulation of several Hox family transcription factors (Fig. 4 G and H). Potentially relevant for these observations, NPM1 mutations, which are strongly associated with Hox gene upregulation, co-occur with Tet2, Idh1^R132, and Idh2^R140 but are mutually exclusive with Idh2^R172 mutations in AML (3, 60). Conversely, AP-1 and Smad-bound loci had decreased accessibility in Tet2^–/– cells (Fig. 4G). For AP-1, this was consistent with the lower transcriptional activity of FOS and JUN in early Tet2^–/– progenitors in single-cell data (Fig. 3G). Closing of binding sites for SMAD2/4, the canonical effectors of TGFβ signaling, was associated with Tet2^–/–-specific down-regulation of Tgfbr1, decreased accessibility of the Tgfbr1 promoter, and increased activity of the SMAD2 transcriptional corepressors, Tgif1/2, predicted by SCENIC in single Tet2^–/–progenitors (Figs. 3G and 4 H and I). Accordingly, GSEA indicated that TGFβ signaling was downregulated in Tet2^–/– compared with WT and Idh2^R172K cells (Fig. 4J). Given the well-described effects of TGFβ on HSC self-renewal and differentiation (61, 62), these results raise the possibility that altered TGFβ signaling may contribute to the effects of Tet2 loss-of-function. Overall, these analyses reveal distinct, and sometimes opposite, effects of Idh2^R172K and Tet2^-/-on chromatin accessibility in LSK cells.

Discordant Changes in DNA Methylation and Chromatin Accessibility in Idh2^R172K Cells.

To identify potential changes in DNA methylation, we performed oxidative reduced representation bisulfite-sequencing (oxBS-seq) on the same cell samples described above. t-SNE analysis indicated that similar to the transcriptome and chromatin accessibility profiles, Idh2^R172K and Tet2^–/– cells clustered away from each other and from WT cells, with Idh2^R172K having a much stronger effect (Fig. 5A). Idh1^R132H and Idh2^R140Q cells slightly differed from WT cells, associated with several differentially methylated loci (Fig. 5B and SI Appendix, Fig. S9C). Consistently, dot blot analysis of Lin^– cells revealed a significant global loss of 5-hmC in Idh1^R132H and Idh2^R140Q cells, which was milder than in their Idh2^R172K and Tet2^–/– counterparts (Fig. 5C). To evaluate which changes may be functionally relevant, we examined differentially methylated CpGs within active LSK enhancers (51). Most altered enhancers were hypermethylated, consistent with data from human AML with IDH1/2 or TET2 mutations (21). Changes in Idh1^R132H and Idh2^R140Q cells were modest but affected some enhancers located within genes encoding key HSC regulators (Fli1, Hlf, and Prmd16) (SI Appendix, Fig. S9 D–F and Dataset S3).

Fig. 5.

(A) t-SNE plot depicting the similarity in overall DNA methylation profile of LSK cells isolated from mice with the indicated genotypes, based on oxBS-seq analysis. Each dot represents an independent sample (biological replicate), each derived from cells pooled from two mice. (B) Number of significantly hyper- and hypo-methylated CpGs in LSK cells isolated from mice with the indicated genotypes, compared to wild-type controls. (C) Representative dot blot and quantification of three independent biological replicates showing the levels of 5-hmc in Lin^– cells isolated from mice with the indicated genotypes. For quantification, the 5-hmc signal was normalized to total DNA content measured by methylene blue staining of the same membrane. (D) Percent methylation changes over genomic area that are opening, closing, or unchanged (q < 0.05), determined by the ATAC-seq analyses shown in Fig. 4, in Idh2^R172K and Tet2^–/– LSK cells compared with wild-type controls. Box plots represent the median and interquartile range. (E) Methylation levels over genomic regions binned in accessibility quartiles in LSK cells with the indicated genotypes. Box plots represent the median and interquartile range. (F) Distribution of hypomethylated, hypermethylated, or stable loci within all significantly opening chromatin regions in Idh2^R172K and Tet2^–/– cells and within LSK enhancers in Idh2^R172K cells. Data were analyzed using ANOVA with Tukey post hoc tests (C), Mann–Whitney tests with Benjamini–Hochberg FDR correction (D), Wilcoxon tests with Benjamini–Hochberg FDR correction (E), or chi-square test (F). *P < 0.05; **P < 0.01; ***P < 0.001.

To investigate whether changes in chromatin accessibility correlate with altered methylation, we focused on Idh2^R172K and Tet2^–/– cells. As expected, closing loci were hypermethylated in both mutants (Fig. 5D). Importantly, methylation levels differed within opening chromatin regions in Idh2^R172K versus Tet2^–/– cells (Fig. 5D). To probe this in more detail, we binned genomic regions into quartiles based on their level of chromatin accessibility. Regardless of genotype, regions with increased accessibility tended to be less methylated (Fig. 5E). However, Idh2^R172K cells consistently displayed hypermethylation compared to the other genotypes across all the accessibility quartiles (Fig. 5E). A substantial number of loci were hypermethylated within opening chromatin regions in Idh2^R172K cells, including at active LSK enhancers (e.g., Gata2, Fli1, Pdgfrb), which never occurred in Tet2^–/– cells (Fig. 5F and Dataset S4). This was true when considering the entire enhancers or only the opening chromatin peaks within them (Fig. 5F). Together, these data indicate divergent relationships between chromatin accessibility and DNA methylation changes in Idh2^R172K versus Tet2^–/– cells, particularly at opening loci.

We next investigated whether genotype-specific epigenetic changes could be linked to transcriptomic alterations. Given that activating PDGFRB rearrangements are oncogenic drivers in myeloid neoplasms (63–65), we further examined the Pdgfrb hematopoietic enhancer that selectively gained accessibility despite adjacent hypermethylation in Idh2^R172K cells (Fig. 6A). Integration with ChIP-seq data from LSK cells (51) and HPC-7 mouse progenitors (66, 67) indicated that this site is marked by H3K27ac and H3K4me1 and bound by several hematopoietic regulators that are predicted to have elevated activity in Idh2^R172K cells, including GATA2, RUNX1, and MEIS1 (Fig. 6A). Accordingly, Pdgfrb expression and PDGFRB pathway activity were strongly upregulated in Idh2^R172K cells, whereas the opposite was seen in Tet2^–/– cells (Fig. 6 B–D and SI Appendix, Fig. S9G). PDGFRB signaling gene sets were also positively enriched in Idh1^R132H and Idh2^R140Q mutants (SI Appendix, Fig. S9H). Pdgfrb upregulation in single Idh2^R172K cells was associated with their abnormal differentiation based on Slingshot analysis (SI Appendix, Fig. S7A). PDGFRB activates RAS-dependent signaling, a known tumorigenic pathway in AML (68, 69). Consistently, GSEA indicated positive enrichment of RAS pathway components in Idh2^R172K LSK cells (SI Appendix, Fig. S9G), associated with upregulation of Kras, Gab1/2, Sos2, Fyn, Src, and the PDGFRB ligand Pdgfb (Fig. 6B). By contrast, the expression of these genes was either unchanged or downregulated in Tet2^–/– cells (Fig. 6B). As another example, we noted an Idh2^R172K-specific opening of a chromatin region 3′ to the Sox4 gene, with features similar to the aforementioned Pdgfrb enhancer (Fig. 6E). SOX4 is a marker of the malignant progenitor state in human AML (70), can drive self-renewal and differentiation arrest in hematopoietic progenitors and AML cells (71), and predicts poor prognosis in AML patients (72). In our single-cell analyses, Slingshot identified Sox4 as a main component of the transcriptional program associated with abnormal Idh2^R172K progenitor differentiation (SI Appendix, Fig. S7A). SCENIC predicted elevated Sox4 transcriptional activity in Idh2^R172K cells (Fig. 3D), tracking with increased Sox4 expression and with the opening of Sox-bound chromatin loci (Figs. 4G and 6 B and F). None of these were affected in Tet2^–/– cells. Overall, these data indicate that mutation-specific epigenetic alterations are associated with aberrant expression and activity of putative drivers of myeloproliferation and impaired differentiation.

Fig. 6.

(A) Tracks representing the normalized average chromatin accessibility (determined by ATAC-seq) and DNA methylation (determined by oxBS-seq) near the Pdgfrb gene in LSK cells isolated from mice with the indicated genotypes. Tracks depicting the binding of selected transcription factors or histone marks, derived from published chromatin immunoprecipitation experiments in HPC-7 or LSK cells, were aligned to the ATAC-seq and oxBS-seq tracks. A differentially accessible peak of interest is boxed. (B) Relative expression of the indicated genes, determined by RNA-seq analysis, in LSK cells isolated from mice with the indicated genotypes. Each dot represents an individual sample. (C) GSEA plots depicting the relative expression of PDGFRB signaling pathway genes in Idh2^R172K and Tet2^−/− LSK bone marrow cells, compared to their wild-type counterparts, based on RNA-sequencing data. (D) Relative expression (z-scored) of Pdgfrb in single LSK bone marrow cells, overlaid on the t-SNE plot presented in Fig. 3C and reproduced here to highlight single-cell genotypes. (E) Tracks representing the normalized average chromatin accessibility (determined by ATAC-seq) and DNA methylation (determined by oxBS-seq) near the Sox4 gene in LSK cells isolated from mice with the indicated genotypes. Tracks depicting the binding of selected transcription factors or histone marks, derived from published chromatin immunoprecipitation experiments in HPC-7 or LSK cells, were aligned to the ATAC-seq and oxBS-seq tracks. A differentially accessible peak of interest is boxed. (F) Relative expression (z-scored) of Sox4 in single LSK bone marrow cells, overlaid on the t-SNE plot presented in Fig. 3 and reproduced in panel C to highlight single-cell genotypes.

Discussion

Our data showing distinct and opposite effects of Idh2^R172K and Tet2^–/– on the transcriptome and epigenome of hematopoietic stem and progenitor cells (SI Appendix, Fig. S10) may help to explain some of the clinical distinctions between IDH1/2 and TET2 mutated myeloid neoplasms (3, 4). The results raise the intriguing possibility that mutual exclusivity between IDH1/2 and TET2 lesions in AML (22, 23) could be partially attributable to their antagonistic effects on transcription factor activity and gene expression. Experimentally testing this possibility would be facilitated by the development of disease models that faithfully recapitulate the mutation order and clonal heterogeneity observed in human patients (73). Some of the differences between Idh2^R172K and Tet2^–/– could be due to the concerted inhibition of multiple α-KG-dependent enzymes by D-2-HG, including histone demethylases (8). Indeed, the occurrences of increased gene expression and chromatin accessibility despite elevated DNA methylation in Idh2^R172K cells hint at a potential dominant effect of altered histone regulation, which will be important to examine in further studies. We note that some epigenetic and transcriptomic changes observed in bulk LSK samples may reflect an underlying altered distribution of cell states. However, we clearly observed cases (e.g., Pdgfrb and Sox4) where expression changes could not readily be linked to the expansion or contraction of specific hematopoietic compartments.

It remains to be clarified whether all the IDH1/2 mutations have a similar role in mediating oncogenic transformation. Specific IDH1/2 lesions may have prognostic significance in patients, although this is controversial (74), differs greatly between studies (3, 4, 9, 29, 75–80), and may be influenced by several factors such as co-occurring mutations and patient age (9, 78, 79). Differences in isoform localization (IDH1 in the cytoplasm versus IDH2 in the mitochondria) may result in mutation-specific alterations in metabolism that will be important to address in future studies (81–83). Nevertheless, given that Idh2^R172K and Idh2^R140Q mice are genetically identical except for their respective point mutation, it is clear that the higher level of D-2-HG in Idh2^R172K cells is a key determinant of phenotype severity. This could explain the unique genomic profile of IDH2^R172 AMLs, characterized by fewer mutations compared to IDH1^R132 and IDH2^R140 diseases (4, 29, 31, 84). Higher D-2-HG production by IDH2^R172 could activate oncogenic pathways that normally require second-hit mutations in IDH1^R132 and IDH2^R140 AMLs, such as RAS signaling as we observe here (29).

Mutant IDH inhibitors have demonstrated substantial clinical efficacy (31–33). Importantly, these drugs induce leukemic cell differentiation, rather than cell death or proliferation arrest (37, 84). In some patients, this can be associated with the development of clinically significant “differentiation syndrome” (85, 86). In addition, resistance to IDH inhibitors can emerge from leukemic clones that retain IDH1/2 mutations and acquire new lesions (37). Therefore, it may be beneficial to target the growth or maintenance of IDH1/2-mutated leukemic cells in combination with differentiation induction. Given that RAS/RTK pathway mutations have been associated with resistance to IDH inhibitors (84, 87), it was interesting to identify PDGF/RAS signaling as a possible contributor to myeloproliferation in Idh2^R172K mice, paralleling prior observations in IDH-mutated AML (88). Future studies could investigate the effects of cotargeting mutant IDH and PDGF/RAS signaling. This and other approaches informed by a better understanding of genotype-specific oncogenic mechanisms may help to develop improved therapies.

Materials and Methods

A detailed description of Materials and Methods is provided in SI Appendix, Supplementary Materials and Methods.

Mice.

The Idh1^LSL-R132H, Idh2^LSL-R140Q, and Idh2^LSL-R172K alleles were generated by inserting a loxP-flanked STOP cassette in the third intron and point mutations in exon 4 (89). Tet2^−/− mice (JAX #23359) and Vav-Cre (JAX #008610) have been described previously (43, 90). All mouse strains were backcrossed to a C57BL/6J background for >10 generations. Animal experiments were performed in accordance with institutional and federal guidelines and were approved by an animal care committee (University Health Network, protocol #985).

Cell Isolation and Flow Cytometry.

Bone marrow cells were isolated by flushing in cold MACS buffer (phosphate-buffered saline with 0.5% bovine serum albumin and 2 mM EDTA pH8.0). The spleen and thymus were crushed through a 70-µm nylon mesh in a cold buffer. Following red blood cell lysis, cells were counted and resuspended at the appropriate dilution for further processing. Colony-forming assays were performed in MethoCult GF M3434 methylcellulose medium (STEMCELL Technologies) following the manufacturer’s protocol. Lineage-negative cells were enriched using the Mouse Lineage Cell Depletion Kit (Miltenyi Biotec #130-090-858), according to the manufacturer’s instructions and using an AutoMACS Pro instrument (Miltenyi Biotec). For CYTOF, lineage-negative cells were stained with metal-tagged antibodies and 500 nM IdU to label newly synthesized DNA as previously described (91). All antibodies and staining conditions are described in SI Appendix, Supplementary Methods.

RNA-seq, ATAC-seq, and RRoxBS-seq Analyses.

For multiomic profiling, RNA and DNA were extracted from 1 × 10⁵ LSK cells (two animals pooled per biological replicate) using the AllPrep DNA/RNA Micro Kit (Qiagen #80284). For ATAC-seq, the OMNI-ATAC method was used on 5 × 10⁴ cells (92). Detailed procedures are provided in SI Appendix, Supplementary Methods. Library preparation, sequencing, and data acquisition were performed at the Weill Cornell Medicine Epigenomics Core as previously described (93, 94). The bioinformatics pipelines to analyze RNA-seq (94), ATAC-seq (95), and RRoxBS-seq (96) followed previously described approaches and are detailed in SI Appendix, Supplementary Methods. For single-cell RNA-seq, LSK cells were loaded on a 10× single cell A chip and processed using Single Cell 3′ Reagent Kits v2 for single-cell capturing and mRNA barcoding (10x Genomics). cDNA libraries were prepared using the Chromium Single-Cell 3′ Library Kit and Gel Bead Kit v2 and i7 Multiplex Kit (10X Genomics) and sequenced on an Illumina HiSeq 2500 system in 2 × 150 bp paired-end mode. The raw sequencing reads were first processed and mapped to mouse genome build GRCm38 using the CellRanger software (v2.1.0, 10X Genomics), followed by analysis using Seurat (47, 97), SCENIC (45), and Slingshot (46) as detailed in SI Appendix, Supplementary Methods. Data are available in Gene Expression Omnibus (Bulk RNA-seq, ATAC-seq, and RRoxBS-seq: GSE204941(98). Single-cell RNA-seq: GSE202696(99)).

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Data, Materials, and Software Availability

Data are available in Gene Expression Omnibus (Bulk RNA-seq, ATAC-seq, and RRoxBS-seq: GSE204941. Single-cell RNA-seq: GSE202696).