Cellular heterogeneity within young and aged HSCs
To characterize the differences in HSC subpopulations, three datasets of HSC single cell expression profiles from young and aged mice were integrated together using Seurat [12, 13] to correct batch effect (dataset information is shown in Supplementary Table 1). Reduction of dimension and unsupervised clustering were performed using t-Distributed Stochastic Neighbor Embedding (t-SNE). A total of 5 clusters were identified (Fig. 1A, B), which were characterized further by identifying their markers using DEG analysis on the log-normalized data without any correction and by deducing their characteristics from gene set enrichment analysis (Fig. 1C, E). To assess the aging effect at the level of HSC populations, we compared the proportion of different clusters in young and aged HSCs (Fig. 1D). Interestingly, Cluster 2, a self-renewal-associated cluster marked by Ccnb1, decreased significantly upon aging. Besides, Cluster 3 (an inflammation-associated cluster marked by Il10ra) and Cluster 4 (an apoptosis-associated cluster marked by Tnfsf14) were increased upon aging. Tnfsf14 is a proapoptotic member of the TNF ligand family and promotes myeloid differentiation of HSCs [27]. As a whole, these results highlighted an exhaustion of HSCs being able to give rise to HSC itself without differentiation, which is a hallmark of HSC aging and indicated that Il10ra and Tnfsf14 were two markers for accumulation of inflammatory and apoptosis-biased state HSCs.
Identification of DEGs between young and aged HSCs
In order to elucidate the overall alterations of HSC aging, we next analyzed HSCs as a whole. According to the cut-off criterion (|Foldchange| > 1.5 and p < 0.05), 453, 614 and 297 DEGs between young and aged HSCs were identified from GSE100906, GSE70657 and GSE100426, respectively (Fig. 2A). Common DEGs were defined as genes that were significantly upregulated or downregulated in at least two datasets. By employing integrated bioinformatics analysis, 56 common upregulated genes and 51 common downregulated genes were identified (Fig. 2B, Supplementary Table 2). For instance, Birc5 and Kpna2 were downregulated, while Clu, Selp and Sdpr were upregulated in HSCs during aging. To confirm the results of common DEGs, the relative expression levels of top 30 DEGs were analyzed by qRT-PCR (Fig. 2C). We found that the PCR results for approximately 75% of the genes were consistent with our bioinformatics analysis (p < 0.05).
Functional enrichment analysis of DEGs
To further delineate the functional changes that occur during HSC aging, functional enrichment of the common DEGs was performed by using the DAVID gene annotation tool (Fig. 3A). For BP, the upregulated genes were mainly enriched in transcription, DNA-templated and regulation of transcription from RNA polymerase II promoter, and the downregulated genes were predominantly enriched in cell cycle, mitotic nuclear division, cell division and protein phosphorylation. This was consistent with the previous findings that cell cycle-related genes dominated the transcriptomic variability of aging and that aged HSCs underwent fewer cell divisions than young HSCs [28–30]. KEGG pathway enrichment analyses revealed that the upregulated DEGs were mainly involved in osteoclast differentiation and TNF signaling pathway, and the downregulated DEGs were mainly involved in cell cycle, oocyte meiosis and p53 signaling pathway. p53 is implicated in regulating HSC aging and quiescence [31] and regulating p53 can help to maintain hematopoietic cells by regulating intracellular ROS during oxidative stress [32]. These results highlighted the loss of self-renewal ability and quiescence in aged HSCs.
PPI network analysis of DEGs
To better understand the interactions among the common DEGs, a PPI network with 67 nodes (27 upregulated and 40 downregulated) and 413 edges was generated with the STRING tool (Fig. 3B). Two significant modules were also identified based on the degree of importance in Cytotype MCODE. Module 1 contained 25 nodes and 286 edges (Fig. 3C), and the expression of all the nodes was downregulated in aged HSCs. KEGG pathway enrichment analyses revealed that the DEGs in Module 1 were mainly enriched in the cell cycle. Module 2 contained 11 nodes and 17 edges (Fig. 3D), and the DEGs in Module 2 were mainly enriched in cell adhesion molecules and the hematopoietic cell lineage. In addition, highly connected nodes with a large number of edges in the network are likely to have significant functional importance and were defined as hub genes. By utilizing CytoHubba in Cytotype, we identified the top 5 upregulated hub genes (Jun, Aldh1a1, Egr1, Cd38 and Junb) and the top 5 downregulated hub genes (Aurka, Ccna2, Ccnb2, Cdk1 and Birc5). The alterations and potential functions of the hub genes were summarized in Table 1.
Table 1
Alterations and functions of hub genes during aging
Gene | Aliase | Alterations with Age | Functions |
Aurka | Aurora kinase A | Down | Contributed to the regulation of cell cycle progression. |
Ccna2 | Cyclin-A2 | Down | Controlled both the G1/S and the G2/M transition phases of the cell cycle. |
Ccnb2 | G2/mitotic-specific cyclin-B2 | Down | Essential for the control of the cell cycle at the G2/M transition. |
Cdk1 | Cyclin-dependent kinase 1 | Down | Essential for the control of the eukaryotic cell cycle, promoted G2-M transition, and regulated G1 progress and G1-S transition. |
Birc5 | Baculoviral IAP repeat-containing protein 5 | Down | Had dual roles in promoting cell proliferation and preventing apoptosis. |
Jun | Transcription factor AP-1 | Up | Regulated functional development of hematopoietic precursor cells into mature blood cells. |
Aldh1a1 | Retinal dehydrogenase 1 | Up | Regulated retinoic acid biosynthesis, the clearance of toxic byproducts of reactive oxygen species and HSC differentiation. |
Egr1 | Early growth response protein 1 | Up | Played a role in the regulation of cell survival, proliferation and cell death and in regulating the response to growth factors and DNA damage. |
Cd38 | Cluster of Differentiation 38 | Up | Cell adhesion and signal transduction. |
Junb | Transcription factor jun-B | Up | Involved in regulating gene activity following the primary growth factor response. |
Cell cycle analysis of HSCs
As the enrichment analysis of common DEGs and PPI network analysis suggested that the cell cycle was strongly associated with functional decline in aged HSCs, we next compared the expression levels of cell cycle-associated genes. Among 107 DEGs, 24 genes were associated with the cell cycle and most of them (22 genes), including Chek2, Kif20b and Clasp2, were downregulated in at least 2 datasets (Fig. 4A). Subsequently, the cell cycle phase of young and aged HSCs was identified by calculating the cell cycle phase score based on canonical markers in both young and aged mice (Fig. 4B). The frequency of cells in the G1 cluster significantly decreased in aged versus young cells (13.5% vs 6.8%; p < 0.05, shown in Fig. 4C).
To further identify the signaling pathways that drove HSCs to transit through G1 phase quickly, we evaluated genetic alterations in a cell cycle pathway diagram that was publicly available at https://www.genome.jp/kegg [33]. We found that the Skp2-induced signaling pathway (Skp2→Cip1→CycA/CDK2→DP-1) was significantly downregulated during the aging process; this change promoted the synthesis of mRNAs and proteins that are required for DNA synthesis during S phase (Fig. 4D).
Analysis of the alterations in bone marrow during the aging process
The PPI network analysis confirmed significant changes of adhesion molecules, highlighting an important role of bone marrow microenvironment in the aging process. Therefore, we reanalyzed two single cell transcriptome profiles of young and aged mouse bone marrow and investigated how the cellular composition of the bone marrow changed with age. The percentages of erythroblasts, granulocytopoietic cells and granulocytes increased with age, while the percentages of macrophages, late pro-B cells, monocytes and T cells decreased with age (Fig. 5A). Cell-cell communication mediated by receptor-ligand complexes is crucial for coordinating diverse biological processes, such as development, differentiation and responses to infection [24]. To assess changes in intercellular communication between HSCs and other cell types during aging, cellphoneDB software was used to predict interactions between the various cell types from the single-cell RNA-seq data [24, 25]. Significant predicted interactions were assessed separately for young and aged mice (Fig. 5B), and the differences were used to infer changes in intercellular interactions (Fig. 5C). The predicted interactions with hematopoietic stem/progenitor cells (HSPCs) were highest in monocytes and lowest in erythroblasts in both young and aged mice. The comparison between the young and aged groups indicated a significant increase in intercellular communication between HSPCs and proerythroblasts, and a decrease in intercellular communication between HSPCs and monocytes, granulocytopoietic cells and granulocytes (Fig. 5C).
As the surrounding cells of HSPCs are potential niche components, the ligands expressed by these populations and the receptors expressed by HSPCs were considered in the downstream analysis. Among a total of 68 heterologous ligand–receptor pairs detected between HSPCs and other cell types, immune response, inflammation response and signal transduction were the predominant biological processes involved (Fig. 5D). In addition, ligand–receptor pairs were involved in some canonical signaling pathways (such as the TNF signaling pathway, NF-kappa B signaling pathway and PI3K-Akt signaling pathway) and certain aging-associated diseases (such as Alzheimer's disease).
The ligand–receptor gene pairs exhibited differential expression patterns between young and aged populations when coupled with HSPCs. Interactions with HSPCs changed significantly in proerythroblasts, monocytes, granulocytopoietic cells and granulocytes, and the significant differentially expressed ligand–receptor gene pairs of these four groups of cells are shown in Fig. 5E. Adhesion complexes, such as aMb2 complex-Icam1 and aLb2 complex-Icam1, were differentially expressed during HSC aging. Moreover, inflammation-related ligand-receptor pairs, such as CXCL2-CXCR2, IL15-IL15 receptor, CCL2-CCR2 and CCL5-CCR5, were upregulated during HSC aging.
Analysis of inflammation levels in the bone marrow niche
The upregulated inflammation-related ligand-receptor pairs in aged HSCs suggested increasing inflammation levels in the bone marrow niche during aging. To test our hypothesis, a Cytoplex Assay was performed to measure the levels of inflammation-related cytokines and chemokines in the bone marrow. Most inflammation-related cytokines and chemokines, including TNF-α, IFN-β, IFN-γ, IL-1α, IL-1β, IL-6, IL-17α, IL-23, CCL2, CCL4 and CXCL1, were upregulated in aged bone marrow niche, while CCL20 and CXCL10 were downregulated in the aged bone marrow niche (Fig. 6A). To assess the effect of inflammatory cytokines on cellular senescence, HSCs were cultured in methylcellulose with either inflammatory cytokines or their inhibitors. TNF-α and IL-1β promoted myeloid-biased differentiation and inflammatory pathway blockade may rejuvenate aged HSC functions and increase B cell output (Fig. 6B).
To further determine which cell population accounted for the difference in inflammation-related cytokine levels, the outgoing communication patterns of secreting cells were analyzed (Fig. 6C). The secretion of HSPCs, erythroblasts and basophils contributed to high CCL levels in the aged bone marrow niche, while the secretion of T cells, promonocytes, macrophages, erythroblasts, and basophils contributed to high IFN-γ levels. In addition, the secretion of some interleukins and complement factors was upregulated in proerythroblasts and immature B cells. Interestingly, the levels of inflammation-related cytokines secreted by HSPCs, including CCL and IL2, were increased, highlighting the important role of autocrine signaling during aging process. Circle plots of the cell-cell communication network also revealed that inflammatory pathway networks (such as IL2, CCL and complement; Fig. 6D, Supplementary Fig. 1A) became more interconnected during aging, with some ligand-receptor communication enhanced significantly (such as IL2 - IL2RB/IL2RG and CCL5 - CCR1; Supplementary Fig. 1B).