Ascorbate regulates haematopoietic stem cell function and leukaemogenesis

Abstract

Stem-cell fate can be influenced by metabolite levels in culture, but it is not known whether physiological variations in metabolite levels in normal tissues regulate stem-cell function in vivo. Here we describe a metabolomics method for the analysis of rare cell populations isolated directly from tissues and use it to compare mouse haematopoietic stem cells (HSCs) to restricted haematopoietic progenitors. Each haematopoietic cell type had a distinct metabolic signature. Human and mouse HSCs had unusually high levels of ascorbate, which decreased with differentiation. Systemic ascorbate depletion in mice increased HSC frequency and function, in part by reducing the function of Tet2, a dioxygenase tumour suppressor. Ascorbate depletion cooperated with Flt3 internal tandem duplication (Flt3^ITD) leukaemic mutations to accelerate leukaemogenesis, through cell-autonomous and possibly non-cell-autonomous mechanisms, in a manner that was reversed by dietary ascorbate. Ascorbate acted cell-autonomously to negatively regulate HSC function and myelopoiesis through Tet2-dependent and Tet2-independent mechanisms. Ascorbate therefore accumulates within HSCs to promote Tet activity in vivo, limiting HSC frequency and suppressing leukaemogenesis.

Extended data figure 1. Stability of metabolites during cell isolation

a, Diagram of the isolation procedure. b-f, Fold changes in the levels of representative metabolites in 20,000 bone marrow cells before and after each step of the cell isolation procedure. Metabolites were extracted before and after bone marrow cells were kept on ice for 7 hours (b), or centrifuged at 4°C at 300×g for 5 minutes (c), or stained with the antibodies used for HSC isolation (d), or cells underwent positive selection with anti-CD45 and anti-Ter119 beads (e), or flow cytometrically sorted for CD45 and Ter119 (f). Although most bone marrow cells are CD45⁺ or Ter119⁺, unfractionated samples are somewhat different in cellular composition from samples after selection and sorting, perhaps contributing to the changes observed in some metabolites in panels e and f. Data were normalized to the median metabolite signal intensity of each sample. Statistical significance was assessed with t-tests performed on log2-transformed data. We accounted for multiple comparisons by controlling the false discovery rate (n=3 mice; *p<0.05, **p<0.01, ***p<0.001). g, Ascorbate levels were compared in whole bone marrow cells whose metabolites were extracted before the cell purification procedure, or after flow cytometric purification (a total of n=4 mice in 2 independent experiments). Statistical significance was assessed using a paired t-test. All data represent mean±SD

Extended data figure 2. Differences in metabolite levels among haematopoietic stem and progenitor cell populations in the bone marrow

a, Types of metabolites detected in 10,000 HSCs. b, Unsupervised clustering of metabolomic data from HSCs, MPPs, and CD45⁺ bone marrow hematopoietic cells isolated in 6 independent experiments. c, Metabolites that significantly differed between HSCs and CD45⁺ bone marrow cells (6 independent experiments with a total of 6 HSC samples and 16 CD45⁺ BM samples). Statistical significance was assessed using t-tests performed on log2 transformed data. We accounted for multiple comparisons by controlling the false discovery rate. All data represent mean±SD. d, The metabolites we measured in stem and progenitor cell populations. The display is autoscaled for each metabolite to illustrate changes across samples. 52 out of 64 metabolites show statistically significant changes among at least some cell populations (1 experiment is shown, representative of 3 independent experiments; *p<0.05, **p<0.01, ***p<0.001) assessed using a one-way ANOVA for each metabolite followed by Fisher’s LSD tests corrected for multiple comparisons by controlling the false discovery rate.

Extended data figure 3. The ascorbate content in haematopoietic stem and progenitor cells correlates with ascorbate transporter expression level and the phenotype of ascorbate-depleted Gulo^−/− mice

a, Ascorbate content versus slc23a2 expression in haematopoietic stem and progenitor cell populations. The plotted data are from Figures 1c and 1d. b-s, u, Analysis of ascorbate-depleted Gulo^−/− mice and littermate controls (a total of 5–11 mice per genotype per time point analysed in 3-6 independent experiments per time-point). t, The percentage of cells that incorporated a 3 day pulse of 5-bromodeoxyuridine (BrdU) at 7-8 weeks of age (a total of n=4-6 mice per genotype in 3 independent experiments). Statistical significance was assessed with a two way ANOVA followed by Fisher’s LSD tests (b-l), or Mann Whitney tests (p, u). We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001). All data represent mean±SD.

Extended data figure 4. Ascorbate depletion did not have detectable effects on global histone methylation, collagen levels, ROS levels, carnitine metabolism, or Tet1-3 expression in HSCs or other haematopoietic progenitors

a, Western blots with antibodies against the indicated histone modifications were performed using haematopoietic cells from 8 week old ascorbate-depleted Gulo^−/− mice and littermate controls (a total of n=3-6 mice per genotype in 3 independent experiments). Bar graphs show band intensity relative to the band intensity of the wild type sample for each cell type. We did not observe an increase in histone methylation for any of the modifications, as would be expected if ascorbate were promoting global histone demethylase activity. We observed reduced H3K27me3 in HPCs (note that data in Figure 3e show that the effects of ascorbate depletion on Flt3^ITD-driven myelopoiesis were mediated mainly by reduced Tet2 function). For gel source data, see Supplementary Figure 1. b, Histochemical staining of collagen with Sirius Red (brightfield or polarized light) or Masson’s trichrome (blue) in bone sections from 8–16 week old Gulo^−/− mice and littermate controls (photographs are representative of 6-8 mice per genotype in 2 independent experiments). c, Carnitine and acetylcarnitine levels were measured by LC-MS/MS in the indicated cell populations from 7-8 week old Gulo^−/− and littermate control mice (a total of n=4-7 mice per genotype in 4 independent experiments). Data show carnitine/acetylcarnitine levels relative to the average levels in wild type WBM samples. d, Reactive oxygen species (ROS) levels were measured in 12 week old Gulo^−/− and littermate control mice using CellRox Deep Red and Enzo Total ROS dyes (a total of n=3-4 mice per genotype in 3 independent experiments). Data show ROS levels relative to wild type WBM samples. e, Ascorbate levels were measured in the plasma and bone marrow of 4-6 month old Gulo^−/− mice or controls (a total of n=12-17 mice per genotype for plasma and n=5 mice per genotype for bone marrow, analysed in 3 independent experiments). f, Tet1-3 transcript levels did not change in Gulo^−/− versus littermate control mice, suggesting that the effects of ascorbate depletion on Tet activity was not due to reduced Tet1-3 transcription (n=3 mice). Statistical significance was assessed with two-way ANOVAs (a-d, f) followed by Fisher’s LSD tests or Welch’s tests (e), or Mann Whitney test (a- H3K36me2-HPC). We corrected for multiple comparisons by controlling the false discovery rate. All data represent mean±SD.

Extended data figure 5. Tet2 deletion phenocopies the increased HSC frequency and increased reconstituting potential of bone marrow cells from ascorbate depleted mice

a-g, Mx-1-Cre; Tet2^fl/fl mice and littermate controls were injected with poly I:C at 6–8 weeks of age. The frequencies of HSCs and other haematopoietic progenitors were analysed 3 weeks later (a total of n=10-24 mice per genotype in 9-14 independent experiments). h, Percentage of donor-derived haematopoietic cells after competitive transplantation of 200,000 donor Tet2^+/+ or Tet2^Δ/+ or Tet2^Δ/Δ bone marrow cells along with 500,000 wild-type recipient competitor cells into irradiated recipient mice (a total of 5 donors and 15-25 recipients per treatment in 5 independent experiments). All data represent mean±SD. Statistical significance was assessed with Kruskal-Wallis tests (a-g), or a non-parametric mixed model followed by a Kruskal-Wallis test for individual time points (h). We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001).

Extended data Figure 6. Ascorbate regulates HSC function in vivo and HSC differentiation in culture

a, Ascorbate levels in sorted cell populations from transplanted recipient mice (a total of n=2-4 mice per genotype from 2 independent experiments). b, Percentage of donor cells that are myeloid, B, or T cells in wild type or Gulo^−/− recipients 16 weeks after transplantation (a total of n=13-18 mice per condition analysed in 4 independent experiments). c-d, Competitive transplantation of 500,000 donor bone marrow cells from Tet2^fl/fl;Mx1Cre mice or littermate controls along with 1,500,000 competitor wild-type cells into irradiated wild-type (ascorbate replete) or Gulo^−/− (ascorbate depleted) mice (a total of 2 donor mice and 5–10 recipient mice per treatment in 2 independent experiments). e, Plasma ascorbate levels in wild type mice, Gulo^−/− mice, or Gulo^−/− mice fed an ascorbate supplemented diet (a total of n=6-17 mice per condition analysed in 4 independent experiments). f, Ascorbate content in bone marrow cells from wild type and Gulo^−/− mice fed normal mouse chow or an ascorbate supplemented diet (a total of n=3-11 mice per condition analysed in 3 independent experiments). g-m, Myeloid and erythroid differentiation were assessed 8 days after culturing HSCs in the presence or absence of ascorbate or its more stable derivative, 2-phospho-ascorbate (a total of n=48 wells for myeloid and 24 wells for erythroid differentiation in 2 independent experiments). All data represent mean±SD. Statistical significance was assessed with Welch’s tests (a-b), Kruskal-Wallis tests (e-m), or a non-parametric mixed model followed by a Kruskal-Wallis test for individual time points (d). We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001).

Extended data figure 7. Gene expression changes in HSCs/MPPs from Gulo^−/−, Tet2^Δ/Δ, and Tet2^Δ/Δ;Gulo^−/− mice relative to control HSCs/MPPs

a-c, HSCs/MPPs were isolated from 4–5 month old mice of the indicated genotypes. Mice in all treatments were maintained on low ascorbate water and treated with poly I:C to induce Tet2 recombination by Mx1-Cre two months before HSC/MPP isolation. a, Fold change values for all genes that significantly differed among any combination of genotypes (DESeq2 Likelihood Ratio Tests for differential expression, corrected for multiple comparisons by controlling the false discovery rate). The value for each gene in the control cells was set to 1. Red and green indicate increases and decreases in gene expression, respectively. The number of genes in Gulo^−/− and Tet2^Δ/Δ HSCs/MPPs that changed in expression in the same direction in both genotypes relative to control HSCs/MPPs was significantly more than would be expected by chance (p=0.01, binomial test). b, Fold change (log2) for HSCs/MPPs of each genotype relative to control HSCs/MPPs. Shown are all genes that changed in the same direction in all three genotypes relative to control and for which log2 fold change was >0.7 or <−0.7 and FPKM>1. Negative fold change values mean gene expression went down in the indicated genotypes compared to control and positive values mean it went up. c, Gene set enrichment analysis showed that of the top 20 gene sets (based on normalized enrichment score) that were downregulated in Gulo^−/− as compared to control HSCs/MPPs, 4 were also among the top 20 gene sets downregulated in Tet2^Δ/Δ as compared to control HSCs/MPPs (n=3 mice per genotype except for the Tet2 ^Δ/Δ treatment, for which there were n=2).

Extended data figure 8. Collaboration between Flt3^ITD and either Tet2 deficiency or ascorbate depletion

a-f, Analysis of the frequencies of haematopoietic stem and progenitor cell populations in the bone marrow of 10–12 week old Mx1-Cre;Tet2^fl/fl; Flt3^ITD mice and littermate controls 3 weeks after poly I:C treatment (12-17 independent experiments; the numbers of mice per treatment in a-c are shown across the top of panel a and for d-f across the top of panel d). g, Percentage of donor-derived haematopoietic cells after competitive transplantation of 300,000 donor bone marrow cells of the indicated genotypes along with 300,000 competing recipient cells into irradiated recipient mice (a total of 2-5 donor mice and 6-20 recipient mice per treatment in 5 independent experiments). h, Secondary transplantation of 5 million bone marrow cells from primary Gulo^−/− (ascorbate depleted) recipients of Flt3^ITD/+ cells into irradiated wild-type or ascorbate depleted Gulo^−/− recipient mice (a total of 3 donor mice and 10-18 recipient mice per treatment in 3 independent experiments). * indicates comparison to wild type and # to Flt3^ITD/+. i, Ascorbate levels in donor cells of the indicated genotypes sorted from transplant recipients (a total of n=4 mice per condition from 2 independent experiments). j, Frequencies of donor cells of the indicated genotypes in the blood of transplant recipients described in Fig. 3f–g. k, Analysis of HSC frequency in the fetal liver of embryonic day (E)17.5 mice of the indicated genotypes (5 independent experiments; the numbers of mice per treatment are shown across the top of the panel). All data represent mean±SD. Statistical significance was assessed with one way ANOVAs followed by Fisher’s LSD tests (d, k) or Kruskal Wallis tests (a-c, e). In other cases, we used two-way ANOVAs followed by Fisher’s LSD tests (i) or Kruskal-Wallis tests (j) or a non-parametric mixed model followed by Kruskal-Wallis tests for individual time-points (g, h). We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001).

Extended data figure 9. Analysis of Tet2^Δ/+;Flt3^ITD leukemias

a-g, Analysis of recipient mice described in Fig. 4a–d. Statistical significance was assessed with one-way ANOVAs followed by Fisher’s LSD tests (f-g) or Kruskal-Wallis tests (e). d shows representative images from the experiments quantified in Fig. 4a. All data represent mean±SD. We corrected for multiple comparisons by controlling the false discovery rate. *p<0.05, **p<0.01, ***p<0.001.

Extended data figure 10. Analysis of Tet2^Δ/+;Flt3^ITD and Tet2^Δ/Δ;Flt3^ITD leukemias

a-b, Analysis of recipient mice described in Fig. 4e–j. Statistical significance was assessed with one-way ANOVAs followed by Fisher’s LSD tests (b) or Kruskal-Wallis tests (a). c, Western blots with antibodies against the indicated histone modifications were performed using protein extracted from Tet2^Δ/+;Flt3^ITD/+ leukemia cells isolated by flow cytometry from wild type or ascorbate-depleted Gulo^−/− transplant recipients (results are representative of 2 independent experiments). d, Diff-Quik stained blood smears from Gulo^−/− recipients of Tet2^Δ/+;Flt3^ITD/+ cells fed with an ascorbate supplemented diet before and after the engraftment of leukemia cells (representative images from the experiments described in Fig. 5a and quantified in Fig. 5c). Cells with an immature blast-like morphology were more abundant in the blood of ascorbate-depleted Gulo^−/− recipients as compared to ascorbate-fed Gulo^−/− recipients. e-f, White blood cells in recipient mice from the experiment described in Figure 5a. The statistical significance of differences among treatments was assessed with Kruskal-Wallis tests (a, e, f) or a one-way ANOVA (b) or a two way ANOVA followed by Fisher’s LSD tests (c). All data represent mean±SD. Statistical significance was assessed with corrected for multiple comparisons by controlling the false discovery rate. (*p<0.05, **p<0.01, ***p<0.001).

Figure 1. HSCs have high ascorbate levels and ascorbate depletion increases HSC frequency

a, Unsupervised clustering of metabolomic data from haematopoietic stem and progenitor cell populations (see methods for the markers used to isolate each population; 1 experiment, representative of 4 total experiments). b-c. Ascorbate and Slc23a2 expression levels relative to CD45⁺ BM cells (b, n=6 mice from 2 independent experiments. c, n=3 mice from 2 independent experiments). d-e, HSC frequencies in Gulo^−/− ascorbate-depleted and littermate control mice at 6, 7, or 8 weeks of age (n=6-11 mice per genotype per time-point in 3–6 independent experiments per time-point). f, Percentage of donor derived haematopoietic cells after competitive transplantation of 500,000 donor Gulo^+/+ or Gulo^−/− bone marrow cells along with 500,000 competing wild-type recipient cells into irradiated recipient mice (a total of 3 donors and 14-15 recipients per genotype in 3 independent experiments). The exact number of mice analysed and the values obtained for each mouse are provided in the source data files for all figures. Statistical significance was assessed with t-tests (b-c) or two-way ANOVAs followed by Fisher’s LSD tests for individual time-points (d-f). All data represent mean±SD. We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001).

Figure 2. Ascorbate depletion reduces Tet2 activity in HSCs and progenitors in vivo

a-b, 5hmC:5mC and 5mC:C in sorted cell populations from 4–6 month old wild type (n=13 mice), Gulo^−/− (n=14 mice), Tet2^Δ/Δ (n=5 mice), and Tet2^Δ/Δ;Gulo^−/− mice (n=8 mice) from 10 independent experiments. c, Ascorbate levels in cells from 4 month old mice (a total of n=4 mice per condition from 2 independent experiments; * refers to comparisons between genotypes and # between cell types). d-e, 5hmC:5mC and 5mC:C in sorted cell populations from 4 month old mice with or without ascorbate feeding (a total of n=4 mice per condition from 3 independent experiments). f, HSC frequency in 8 week old mice (a total of n=8-11 mice per genotype in 6 independent experiments). g, Percentage of donor-derived haematopoietic cells after competitive transplantation of 500,000 donor bone marrow cells from Gulo^+/+ or Gulo^−/− mice, or Gulo^−/− mice supplemented with ascorbate along with 500,000 competitor wild-type cells into irradiated recipient mice (a total of 4 donors and 19-20 recipients per genotype in 4 independent experiments). Statistical significance was assessed with one-way (a-b, d-f) or two-way (c) ANOVAs, Kruskal-Wallis tests (for abnormally distributed data in a-b), or with a non-parametric mixed model followed by Kruskal-Wallis tests for individual time-points (g). All data represent mean±SD. We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001).

Figure 3. Low ascorbate levels cooperate with Flt3^ITD to promote myelopoiesis, partly by reducing Tet2 function, and cell-autonomously promote HSC function. a-d

, Competitive transplantation of 500,000 donor bone marrow cells from Flt3^ITD/+ mice or littermate controls along with 500,000 competing wild-type cells into irradiated wild-type (ascorbate replete) or Gulo^−/− (ascorbate depleted) recipient mice (a total of 4 donor mice and 13–20 recipient mice per treatment in 2 independent experiments). Panel a shows donor cell reconstitution levels in the blood and panel b shows the percentages of donor-derived cells in the bone marrow. c-d, Frequencies of donor and competitor-derived CD48⁺LSK HPC cells. e, Competitive transplantation of 500,000 donor bone marrow cells of the indicated genotypes along with 1,500,000 competitor wild-type cells into irradiated wild-type (ascorbate replete) or Gulo^−/− (ascorbate depleted) mice (a total of 4 donor mice and 15 recipient mice per treatment in 4 independent experiments). f-g, Competitive transplantation of 500,000 donor fetal liver cells from the indicated genotypes along with 2,000,000 competing wild-type bone marrow cells into irradiated wild-type recipient mice (for +/+ or Slc23a2^+/−, 10 donor and 50 recipient mice; for Slc23a2^−/−, 4 donors and 20 recipients; for Flt3^ITD/+ or Slc23a2^+/−;Flt3^ITD/+, 6 donors and 30 recipients; for Slc23a2^−/−;Flt3^ITD/+, 5 donors and 24 recipients, from 6 independent experiments). +/+ and Slc23a2^+/− genotypes were pooled as they showed no statistically significant differences. g, Percentage of donor-derived haematopoietic cells. * refers to comparisons to wild type (black line), # to Flt3^ITD/+ (blue line), and + to Slc23a2^−/− (red line). All data represent mean±SD. Statistical significance was assessed with a non-parametric mixed model followed by Kruskal-Wallis tests for individual time-points or cell types (a-b, e, g), one-way ANOVAs followed by Fisher’s LSD tests (between genotypes in c, d) and Mann-Whitney tests (between donor and competitor in c, d). We corrected for multiple comparisons by controlling the false discovery rate (*p<0.05, **p<0.01, ***p<0.001). * indicates comparisons to wild type while # or + indicate comparisons to other conditions as indicated by the brackets.

Figure 4. Low ascorbate levels accelerate leukaemogenesis

a-d, Eight million bone marrow cells from Tet2^Δ/+;Flt3^ITD or wild-type donor mice were transplanted into irradiated wild type or ascorbate-depleted Gulo^−/− recipient mice and analysed 4–6 weeks after transplantation (a total of 3 donors with 14-15 recipients per treatment in 3 independent experiments). a, Frequency of myeloblasts in the blood (a total of n=2-7 recipients per treatment). b, Kaplan-Meier survival curve of transplant recipients of Tet2^Δ/+;Flt3^ITD cells (3 independent experiments, Mantel-Cox log-rank test). c-d, Analysis of the blood or spleen of transplant recipients (a total of n=4-15 recipients per treatment). e-j, Eight million bone marrow cells from donor mice of the indicated genotypes were transplanted into irradiated wild type or ascorbate-depleted Gulo^−/− recipient mice and analysed 6-8 weeks after transplantation (a total of 4 donors and 8-19 recipients per treatment in 4 independent experiments). e-f, 5hmC:5mC and 5mC:C ratios in donor cells from the bone marrow of transplant recipients (a total of n=6 mice per treatment from 6 independent experiments). g, Kaplan-Meier survival curve of transplant recipients (4 independent experiments, Mantel-Cox log-rank test). h-j, Analysis of the blood or spleen of transplant recipients. Statistical significance was assessed with one-way ANOVAs followed by Fisher’s LSD tests (c-f, i-j) or Kruskal-Wallis tests (a, h). All data represent mean±SD. We corrected for multiple comparisons by controlling the false discovery rate. *p<0.05, **p<0.01, ***p<0.001.

Figure 5. The effect of ascorbate depletion on leukemogenesis is reversible and ascorbate levels are high in human HSCs

a-e, Eight million bone marrow cells from Tet2^Δ/+;Flt3^ITD or wild-type donor mice were transplanted into irradiated wild type or Gulo^−/− recipient mice. Some Gulo^−/− recipients were fed with ascorbate before (pre-leukemia) or 6–7 weeks after transplantation (post leukemia) (a total of n=3 donors with 5-13 recipients per treatment in 4 independent experiments). Mice were analysed 10-11 weeks after transplantation. b, Kaplan-Meier survival curve of transplant recipients of Tet2^Δ/+;Flt3^ITD cells treated (6 independent experiments, Mantel-Cox log-rank test). f-g, Ascorbate levels and SLC23A2 expression in human bone marrow haematopoietic cells (see methods for the markers used to isolate each cell population. f, n=7 samples analysed in 7 independent experiments. g, n=5 samples analysed in 5 independent experiments). Statistical significance was assessed with t-tests (g), one-way ANOVAs followed by Fisher’s LSD tests (d, f) or Kruskal-Wallis tests (c, e). We corrected for multiple comparisons by controlling the false discovery rate. All data represent mean±SD (*p<0.05, **p<0.01, ***p<0.001).