2-Hydroxyglutarate destabilizes chromatin regulatory landscape and lineage fidelity to promote cellular heterogeneity

Abstract

The epigenome delineates lineage-specific transcriptional programs and restricts cell plasticity to prevent non-physiological cell fate transitions. Although cell diversification fosters tumor evolution and therapy resistance, upstream mechanisms that regulate the stability and plasticity of the cancer epigenome remain elusive. Here we show that 2-hydroxyglutarate (2HG) not only suppresses DNA repair but also mediates the high-plasticity chromatin landscape. A combination of single-cell epigenomics and multi-omics approaches demonstrates that 2HG disarranges otherwise well-preserved stable nucleosome positioning and promotes cell-to-cell variability. 2HG induces loss of motif accessibility to the luminal-defining transcriptional factors FOXA1, FOXP1, and GATA3 and a shift from luminal to basal-like gene expression. Breast tumors with high 2HG exhibit enhanced heterogeneity with undifferentiated epigenomic signatures linked to adverse prognosis. Further, ascorbate-2-phosphate (A2P) eradicates heterogeneity and impairs growth of high 2HG-producing breast cancer cells. These findings suggest 2HG as a key determinant of cancer plasticity and provide a rational strategy to counteract tumor cell evolution.

Keywords: 2-hydroxyglutarate; DNA repair; breast cancer; cancer cell plasticity; chromatin CyTOF; epigenome fluctuation; lineage fidelity; luminal-to-basal transition; oncometabolite; single-cell epigenomics.

Figure 1.. 2HG-mediated remodeling of the chromatin-transcriptional landscape is restored by A2P in the presence of 2HG

(A) 5hmC and 5mC levels at long interspersed element-1 (LINE-1) sequences. HMECs were treated for 72 h with 1 mM of D2HG, L2HG, or dimethyloxalylglycine (DMOG) and subjected to TET-assisted bisulfite (TAB) pyrosequencing to differentiate 5hmC from 5mC (see STAR Methods for further details). DMOG is an analog of αKG and competes for binding at the active center of the enzyme. (B) Global levels of 5hmC and H3K27me3 detected by immunostaining. ***p < 0.001 versus control by one-way ANOVA with Dunnett’s multiple comparison test (n = 8 images per condition). ns, not significant. At least 60 nuclei were examined, and signal intensities were normalized to DAPI nuclear counterstain. Mean ± SEM is presented. (C) Fold change in different types of histone methylation. *p < 0.05, **p < 0.01, ***p < 0.001 versus control (n = 3 independent replicates). (D) Cell cycle analysis of HMECs exposed for 72 h to 100 μM of D2HG or L2HG. No statistical significance was observed using one-way ANOVA. Results are from three independent experiments. (E) Hierarchical clustering of RNA-seq profiles. Cells were treated for 72 h with 100 μM of D2HG (D), L2HG (L), D2HG plus L2HG (D + L) or L2HG in the presence of 1 mM A2P (L + A2P), or received no treatment (C). L2HG treatment was followed by 4-day withdrawal (LW). (F) Dot plot of GSEA hallmark pathways. Color indicates normalized enrichment scores (NESs) of positively and negatively enriched gene sets relative to control, and circle size corresponds to false discovery rate (FDR). (G) Heatmap depicting changes in expression of DDR genes. The corresponding DNA repair pathways are shown on the right. (H) Pie chart representing the proportion of DDR pathway genes. Shaded fractions correspond to genes that are downregulated by 2HG. (I) Expression levels of DDR proteins detected by capillary-based immunoassay. (J) GSEA plots evaluating enrichment for methylated genes in breast cancer cells. See also Figure S1 and Table S1.

Figure 2.. 2HG enhances molecular heterogeneity in gene regulatory dynamics

(A) Genome tracks showing open chromatin regions detected by scATAC-seq. scATAC-seq profiling is highly consistent with DNase hypersensitive sites (DHSs) detected by bulk DNaseI-seq (GSE29692). (B) PCA projection showing distinct cell subpopulations based on the single-cell chromatin accessibility landscape. (C) Proportions of cell subpopulations identified by model-based clustering. (D) PCA plots of five distinct cell clusters identified by model-based clustering (top). Bottom plots indicate variability in chromatin accessibility at consensus TF binding motifs (n = 386) across single cells in each cluster. (E) Representative DNA accessibility at TF binding motifs. (F) Chromatin accessibility at lineage-specific TF binding motifs. *p < 0.05, **p < 0.01, ***p < 0.001 versus cluster 1. See also Figure S2.

Figure 3.. 2HG initiates cell-level fluctuations in the mammary epithelial epigenome

(A) ATAC-seq signal enrichment across ChromHMM-annotated genomic regions. Txn, transcription; CNV, copy number variation. *p < 0.05, **p < 0.01, ***p < 0.001 versus control. (B and C) Heatmaps showing enrichment of scATAC-seq signals at typical enhancers (B) and super-enhancers (C). Each row represents one individual element, and color represents the intensity of chromatin accessibility in the left panels. The corresponding density plots are shown on the right. (D–G) Single-cell accessibility landscape across the genome centered on TSSs (D), enhancers (E), and weakly transcribed (F) and heterochromatin regions (G). Colors represent individual cells. Solid lines indicate mean values, and semi-transparent shade around the mean curve shows SEM across the region. See also Figure S3.

Figure 4.. Chromatin remodeling involves tumor-associated promoter hypermethylation

(A) Heatmap depicting mRNA expression of highly methylated genes (n = 150) whose promoter regions showed diminished chromatin accessibility following 2HG exposure. The rectangle outlined in white represents genes that are downregulated in basal-like tumors. (B) Box-and-whisker plot showing mRNA expression of genes that are indicated with white outline in (A). ***p < 0.001 versus adjacent normal tissue. (C) Gene Ontology (GO) biological processes identified by DAVID pathway enrichment analysis of 150 genes that are hypermethylated in breast tumors. (D) Heatmap showing chromatin accessibility of hypermethylated genes (n = 150) in breast tumors. The rectangle outlined in white represents genes exhibiting decreased expression in (A). (E) Box-and-whisker plot showing chromatin accessibility of genes that are indicated with white outline in (D). ***p < 0.001 versus basal-like tumors. (F) Genome tracks showing diminished chromatin accessibility at gene promoters following 2HG exposure. ChromHMM chromatin states (GSE38163), DNaseI-seq (GSE29692), H2AFZ chromatin immunoprecipitation sequencing (ChIP-seq) (GSE29611), and WGBS profiles (GSE86732) from HMECs are shown. See also Figure S4 and Table S2.

Figure 5.. 2HG induces epigenetic discordancy and enhances phenotypic diversity

(A) Heatmaps of selected markers used for chromatin CyTOF, which simultaneously detects epigenetic modification levels as well as marker expression in single cells. HMECs were treated for 72 h with 100 μM of D2HG (D) or L2HG (L), or received no treatment (C). L2HG treatment was followed by 4-day withdrawal (LW). Each data point on the t-SNE maps represents an individual cell, and color corresponds to cellular levels of each marker. (B) t-SNE projection of epigenetically distinct cell subsets defined by consensus hierarchical clustering. (C) Heatmaps depicting changes in histone modifications. Normalized median values of signal intensities are shown for each cluster (right). Pie charts (left) indicate the proportion of cells from different experimental groups in each cluster. (D) Expression levels of DDR genes in cluster 1. **p < 0.01, ***p < 0.001 versus control. (E) Heatmaps of pairwise Spearman correlations between marker levels in single cells. Black and blue arrowheads indicate reduced concordancy in gene expression and epigenetic modifications, respectively. (F) Projection of minimum spanning trees (MSTs) obtained by SPADE analysis. The size and color of nodes represent the number of cells in each cluster, allowing visualization of the extent of cellular heterogeneity in treatment groups. (G) Quantification of cellular heterogeneity using Simpson and Shannon entropy indices. ***p < 0.001 versus control. (H) Correlation between cell-to-cell variance in H3K9me3 and marker expression levels in each node identified by SPADE analysis. See also Figure S5 and Table S3.

Figure 6.. Enhanced cellular diversity and undifferentiated stem-like signatures correlate with 2HG production

(A) Pairwise correlation matrix of breast cancer cell lines. Spearman’s correlation coefficients between samples were assessed based on DNA methylation profiles of 3,000 CpG islands (left), RNA-seq profiles of 18,319 protein-coding transcripts (center), and reverse-phase protein microarray (RPPA) profiles of 213 proteins (right). Green area plots indicate intracellular levels of 2HG. (B) Heatmap depicting expression of luminal (n = 25) and myoepithelial (n = 25) genes in breast cancer cell lines (n = 47). (C) Variance in mean expression levels between luminal and myoepithelial gene sets in 2HG-low (n = 12) and 2HG-high (n = 12) cell lines. ***p < 0.001 by unpaired Student’s t test. (D) Scatterplot showing the relationship between 2HG accumulation and genomic intratumor heterogeneity (gITH) in breast cancer cell lines (n = 47). ns, not significant. (E and F) Pairwise correlation matrix of TCGA primary tumors. (G and H) Scatterplots showing the relationship of 2HG accumulation with transcriptomic intratumor heterogeneity (tITH) and genomic intratumor heterogeneity (gITH) in primary breast tumors (n = 20). (I and J) Mature luminal and undifferentiated stem cell-like scores obtained by single-sample scoring analysis using luminal mature (n = 50) and embryonic stem (n = 40) gene sets. Hexagonal density plots represent all breast tumor samples available in the TCGA cohort (n = 1,100). *p < 0.05, **p < 0.01, ***p < 0.001 by unpaired Student’s t test. (K and L) Prognostic significance of undifferentiated stem-like signatures in individuals listed in the TCGA. Progression-free survival was evaluated using Kaplan-Meier analysis based on mRNA expression (K) and DNA methylation (L) of embryonic stem (n = 40) and luminal mature (n = 50) gene sets. See also Figure S6.

Figure 7.. 2HG induces DNA damage accumulation, and A2P eradicates cancer cell heterogeneity

(A) Immunofluorescence staining of γH2AX. MCF-7 cells were treated with 2HG under the same conditions (100 μM for 72 h) as those used in the RNA-seq and ATAC-seq experiments. Scale bar, 10 μm. (B) Quantification of cells with γH2AX focus-positive nuclei. For each treatment, 200–300 cells were examined, and cells with at least four γH2AX foci in nuclei were counted as positive. ***p < 0.001 versus control. (C) CyTOF analysis of γH2AX levels. HMECs were treated for 72 h with 100 μM of D2HG (D) or L2HG (L) or received no treatment (C). L2HG treatment was followed by 4-day withdrawal (LW). (D) HRD scores in TCGA breast tumors. *p < 0.05 by unpaired Student’s t test. (E) t-SNE visualization of CyTOF profiling. MDA-MB-231 cells were treated with 1 mM A2P or 100 μM αKG for 72 h or received no treatment. (F) Projection of SPADE trees. The size and color of nodes represent the number of cells in each cluster, allowing visualization of the extent of cellular heterogeneity in treatment groups. (G) Quantification of cellular heterogeneity using Simpson and Shannon diversity indices. *p < 0.05, **p < 0.01, ***p < 0.001 versus control. (H) Heatmaps of pairwise Spearman correlations between marker levels in single cells. Arrowheads indicate reduced concordancy in gene expression. See also Figure S7 and Table S3.