Aging Human Hematopoietic Stem Cells Manifest Profound Epigenetic Reprogramming of Enhancers That May Predispose to Leukemia

Abstract

Aging is associated with functional decline of hematopoietic stem cells (HSC) as well as an increased risk of myeloid malignancies. We performed an integrative characterization of epigenomic and transcriptomic changes, including single-cell RNA sequencing, during normal human aging. Lineage^-CD34⁺CD38^- cells [HSC-enriched (HSCe)] undergo age-associated epigenetic reprogramming consisting of redistribution of DNA methylation and reductions in H3K27ac, H3K4me1, and H3K4me3. This reprogramming of aged HSCe globally targets developmental and cancer pathways that are comparably altered in acute myeloid leukemia (AML) of all ages, encompassing loss of 4,646 active enhancers, 3,091 bivalent promoters, and deregulation of several epigenetic modifiers and key hematopoietic transcription factors, such as KLF6, BCL6, and RUNX3. Notably, in vitro downregulation of KLF6 results in impaired differentiation, increased colony-forming potential, and changes in expression that recapitulate aging and leukemia signatures. Thus, age-associated epigenetic reprogramming may form a predisposing condition for the development of age-related AML. SIGNIFICANCE: AML, which is more frequent in the elderly, is characterized by epigenetic deregulation. We demonstrate that epigenetic reprogramming of human HSCs occurs with age, affecting cancer and developmental pathways. Downregulation of genes epigenetically altered with age leads to impairment in differentiation and partially recapitulates aging phenotypes.This article is highlighted in the In This Issue feature, p. 983.

Fig. 1:. Focal histone and DNA methylation alterations with HSCe aging.

(A) Heatmap representation of regions with either loss or gain (log₁₀ likelihood ratio > 3, absolute fold-change >1.5) of H3K4me1, H3K27ac, H3K4me3, or H3K27me3 signal in aged HSCe compared to young. The log₂(IP/Input) signal is plotted for each replicate, centered on the differential peak +/− 5 kb. Each column is representative of an individual donor. (B) Functional annotation using ChIP-enrich Gene Ontology Biological Processes for genes annotated to peaks that have reduced H3K4me1, H3K27ac, H3K4me3, or H3K27me3 signal in aged HSCe compared to young. Select significant (FDR<0.05) annotations are shown. (C) Row-scaled heatmap of the percent methylation for the 529 differentially methylated regions (DMRs) (FDR < 0.05, absolute methylation difference ≥ 20%) in aged HSCe versus young HSCe. Each row corresponds to a unique region with differential methylation and each column corresponds to one donor. (D) Bar plot representation of significant (FDR <0.05) pathways associated with genes that are differentially methylated in aged HSCe.

Fig. 2:. Decreased H3K27ac at immune and cancer associated enhancers with age.

(A) Heatmap depicting the 12 clusters identified using k-means clustering of regions with significant (LLR > 3 and absolute fold-change > 1.5) changes of H3K4me1, H3K4me3, H3K27me3, or H3K27ac with age (n=37,058 peaks). The fold-change(aged/young) signal is plotted for each histone modification for each peak. Annotation to active and poised enhancers as well as bivalent promoters identified in young HSCe is also shown. (B) Heatmap of H3K4me1 and H3K27ac signal at the enhancer enriched clusters J-L. The log₂(Pooled IP/Pooled Input) signal is plotted for each age group, centered on the differential peak +/− 5kb. Select genes within each cluster are denoted on the right of the heatmap. (C) Bubble plot representation of select KEGG pathways that are enriched in clusters J-L. The size of each bubble corresponds to the number of genes within the cluster that are within the given gene set. Highly significant (FDR <0.05) categories are colored in yellow. (D) Example of transcription factor binding DNA motifs enriched in clusters J-L. Motifs with significant enrichment (q-value < 0.05) are denoted with a white circle. (E) UCSC genome browser track examples of active enhancers within clusters J-L that overlap transcription factor ChIP-Seq peaks. Tracks are of pooled replicates for each age group, normalized to reads per million and to the corresponding Input. Lines below tracks represent transcription factor peaks annotated in publicly available datasets of CD34+ cells.

Fig. 3:. Loss of activating histone modifications at promoter regions with age.

(A) Heatmap of H3K4me1, H3K27ac, H3K4me3 and H3K27me3 signals at age-associated clusters that are enriched for active promoters (top and middle) and bivalent promoters (bottom). The log₂(Pooled IP/Pooled Input) signal is plotted for each age group, centered on the differential peak +/− 5kb. (B) Density plots of the log₂(Pooled IP/Pooled Input) signal for the characteristic histone marks for the peaks within the Active TSS I (top), Active TSS II (middle) and Bivalent promoter (bottom) categories. (C) UCSC genome browser tracks of genes with altered promoters from the Active TSS I cluster (GFI1 and NFATC4) and Active TSS II cluster (CARM1 and PER1). Tracks are of pooled replicates for each age group, normalized to reads per million and to the corresponding Input for ChIP-seq. Light green bars below tracks represent the differential promoter regions identified from the cluster analysis. (D) Bubble plot representation of select KEGG pathways that are enriched in the active promoter clusters (clusters A-D). The size of each bubble corresponds to the number of genes within the cluster that are within the given gene set. Highly significant (FDR <0.05) categories are colored in yellow. (E) Bubble plot representation of select KEGG pathways that are enriched in clusters E and F. The size of each bubble corresponds to the number of genes within the cluster that are within the given gene set. Highly significant (FDR <0.05) categories are colored in yellow. (F) UCSC genome browser tracks of the HOXC cluster, which contains bivalent promoters within clusters E-F that are altered with age. Tracks are of pooled replicates for each age group, normalized to reads per million and to the corresponding Input for ChIP-seq. Teal bars below tracks represent the differential bivalent promoter regions identified from the cluster analysis. (G) Example of transcription factor binding DNA motifs enriched in clusters E and F. Motifs with significant enrichment (q-value < 0.05) are denoted with a white circle.

Fig. 4:. Differential gene expression of transcription factors and epigenetic modifiers in aged HSCe.

(A) Volcano plot of the log₂ fold-change (aged/young) gene expression versus the −log₁₀(p-adj). Significant downregulated and upregulated genes in aged HSCe compared to young are colored in blue and red, respectively (FDR < 0.05 and fold-change< −1.5 or fold-change > 1.5, respectively). Select differentially expressed genes are labeled. (B) Bar plot depicting the number of age-associated differential genes that also epigenetically deregulated with age at the different genomic regions identified in Figure 2. The numbers above the bars denote the number of genes that are both differentially expressed and within the epigenomic category. (C) UCSC tracks of pooled replicates for young and aged HSCe at the KLF6 and LMNA loci. Colored bars below the tracks denote differential epigenetic regions as identified in Figure 2A. (D) Bar plot representation of the normalized enrichment score (NES) for select gene set enrichment pathways that are up or downregulated with HSCe aging (n=10 donors per age group; FDR< 0.05).

Fig. 5:. Age-associated changes in epigenetic landscapes are driven by reprogramming.

(A) Schematic of the analysis used to identify and classify aged HSC using scRNAseq data from young (24-37 yo) and aged (64-71 yo) HSCe. (B) Heatmap of young and aged nc-HSC. Expression centroids were calculated from the 3 clusters (C1-Young in green, C2-Aged in purple, and C3-Aged in red) identified using the genes differentially expressed at the single-cell level (shown in Figure S5B) and used to cluster nc-HSC. Each column corresponds to an individual cell, and age group of the donor is depicted in the bottom bar.

Fig. 6:. Age-associated epigenetic changes may predispose for AML.

(A & B) Boxplots of the percent methylation for k-means clusters of regions that are (A) hypermethylated (k=8) or (B) hypomethylated (k=3) in aged HSCe compared to young. Percent methylation was calculated for each DMR within the clusters generated by k-means clustering. Arrows denote clusters of DMRs that may be predisposing to AML (p< 0.05, Mann-Whitney rank sum test, and corrected for multiple testing). (C) Heatmap of H3K27ac signal at peaks within the enhancer category that is altered with HSCe aging. Rows are ordered using k-means clustering (k=10) performed using young and aged HSCe and AML blasts (n=71 patients). Bars above heatmap denote donor age and cell type. (D) Boxplots of the log₂(H3K27ac/Input) enrichment for peaks (n=4,931) within clusters 1-7 (C) which showed consistent changes in H3K27ac in aged HSCe and AML compared to young HSCe (p < 0.05, Mann-Whitney rank-sum test, and corrected for multiple testing). (E) Select significant (FDR <0.05) gene ontology biological processes associated with peaks in clusters 1–7 (C). (F) UCSC genome browser tracks of the read-normalized fold-enrichment (H3K27ac/Input) signal at MEIS1, a gene predicted by k-means clustering to have reduced H3K27ac at its intergenic enhancer. Each track is derived from 1 donor and 2 biological replicates are shown for each condition.

Fig. 7:. Loss of KLF6 impairs differentiation and leads to expression changes reminiscent of leukemia.

(A) Colony-forming unit assays of CD34+ cells with CRISPR-Cas9 knockout of KLF6. Normalized colony numbers per 500 CD34+ cells plated are plotted for total colony number, granulocyte-macrophage (GM), granulocyte-erythrocyte-macrophage-megakaryocyte (GEMM) and burst-forming unit erythroid (BFU-E). Colony numbers for each biological replicate (n=5) are normalized to the total colony number for that replicate. (B) Representative flow cytometry histograms for 2 donors and contour plots from 1 donor for CD34+ cells transfected with sgCTRL or sgKLF6 and cultured in myeloid (top and middle) or erythroid (bottom) promoting conditions for 7 days. Dot plot representation of the percentage of CD34+ cells (top) CD34− CD11b+ (middle) and CD235a− CD71+ (bottom) cells is also shown (n=5). Cells transfected with sgCTRL were gated on KLF6+, while cells transfected with sgKLF6 were gated on KLF6−. (C) GSEA leading edge plots showing the enrichment of the gene sets for genes up- or down-regulated with HSCe aging in CD34+ cells with sgKLF6 knockout (n=4 replicates). GSEA was ran using a list pre-ranked by the Wald-statistic (DESeq2), with the weighted enrichment score. (D) Bar plot representation of the normalized enrichment score (NES) for the topmost gene set enrichment pathways that are up- or down-regulated with sgKLF6 knockout (FDR< 0.05). p-values from paired one-tailed t-test, calculated with Prism, p>0.05=ns, p ≤ 0.05=*, p≤0.01=**, p≤0.001=***, p≤0.0001=****.