Introduction

Aging is accompanied by physiological and biochemical dysfunctions, imbalances in homeostatic and degenerative diseases1,2. Frailty, initially defined as a non-specific clinical condition representing a manifestation of aging, leads to the decline of physiological reserves in elderly people, increased vulnerability of the body and the decreased ability to resist stress3,4. Although more convincing evidence is needed, frailty is widely accepted as a preventable and reversible process5,6,7. Thus, the timely and effective prevention and reversal of frailty can play a key role in achieving healthy aging.

The incidence of frailty is approximately 15–40% in elderly populations8, and age is also typically considered an independent risk factor for frailty9,10. However, some elderly individuals maintain robustness and vitality, indicating that the connection between frailty and age might be intricate or non-linear9,11,12. Thus, age-independent factors also contribute to the occurrence and progression of frailty. However, most preclinical research studies on frailty have mainly focused on age-associated factors, and the detailed mechanisms and differences between age-associated and age-independent frailties remain elusive.

Frailty is modulated by multiple organs and systems, primarily including skeletal muscles13, the brain14, the immune system15,16, and the endocrine system17. Physical frailty is a subset of frailty, encompassing unintentional weight loss, exhaustion, weakness, slow walking speed, and low physical activity levels18. A key driver of frailty is sarcopenia, a progressive and generalized skeletal muscle disorder marked by the accelerated loss of muscle mass and function, particularly in elderly people13,19. Sarcopenia underlies approximately 70% of the frailty cases and is strongly associated with adverse frailty-related outcomes13,19,20. Therefore, sarcopenia is often regarded as the biological substrate of physical frailty21,22. Previous studies characterized physical frailty as a decline in muscle function that contributes to cognitive impairment14,23,24. Thus, studies focusing on muscles and physical frailty are of vital importance to comprehensively understand the concept of frailty. However, the molecular evidence and mechanisms involved in physical frailty remain unclear.

Single-cell RNA sequencing (scRNA-seq) technology is an advanced method used to analyze the heterogeneity and complexity of RNA transcripts in individual cells, as well as the composition of specific cell types in highly organized tissues. Previous single-cell transcriptomic analyses revealed aging-related alterations in gene expression in 23 types of tissues in mice25,26. In the context of frailty, Luo et al. used scRNA-seq and the TCR sequencing of human peripheral blood to reveal age-dependent transcriptomic heterogeneity of immune cells (B cells, T cells, monocytes and NK cells) in frailty, with frailty-specific monocyte subsets, highlighting immune cell characteristics as potential biomarkers and intervention targets for frailty16. Although studies have focused on uncovering the mechanisms of frailty27,28,29,30, the transcriptional heterogeneity in frail muscles has not been specifically investigated using single-cell transcriptional analysis.

Thus, performing scRNA-seq on male mice hind limb muscles, this study investigated the overall transcriptional heterogeneity in different cell types during relatively healthy aging (RHA), age-associated (AAF) or age-independent frailty (AIF) by evaluating cell type composition, cell-specific molecular programs, regulatory transcription factors (TFs), and cell-cell communication networks. We found that RHA, AAF or AIF group each showed one major expression pattern, and the AAF group also weakly displayed the two major patterns of the other groups. B cells showed the largest number of down-regulated genes, whereas satellite cells had the largest number of up-regulated genes in both the AAF and AIF groups. Then, we focused on B cells and performed fluorescence-activated cell sorting (FACS) followed by SMART-seq to validate our scRNA-seq findings. Overall, this work demonstrates transcriptional heterogeneity in frailty and expands the understanding of single-cell molecular mechanisms underlying natural aging and physical frailty. It also uncovers the specific cell types involved in natural aging and physical frailty, the relevance and difference between AAF and AIF and lays a foundation for the clinical prevention and treatment of frailty.

Materials and methods

Animals

The study was approved by the Institutional Animal Care and Use Committee (IACUC) of West China Hospital, Sichuan University (Approval No. 2020251). The study was conducted in accordance with the IACUC of West China Hospital, Sichuan University. The study also reported in accordance with ARRIVE guidelines (https://arriveguidelines.org). Two-month-old (young) and 15-month-old (aged) wild-type C57BL/6J male mice were purchased from Shulaibao Biotechnology Co., Ltd. (Wuhan, China) and housed to the indicated age as cohort 1 and cohort 2.

Measurement of frailty phenotype and frailty index

Consistent with previous preclinical frailty research paradigms31,32, our study utilized aged male C57BL/6J mice as the experimental model, a widely adopted approach in rodent frailty assessment that facilitates direct comparison with prior mechanistic investigations.

A dual assessment approach combining the Valencia Score Frailty Phenotype (FP) developed by Martinez De Toda et al.33 and the Frailty Index (FI) established by Whitehead and colleagues34 was used to evaluate frailty in the mice. The FP assessment involved monitoring weight loss (defined as a > 5% reduction from the previous month) and measuring physical performance using a series of tests, including the treadmill endurance test, the grip strength test, and the tight-rope test (Table S1). A frailty cutoff of 20% was applied to both the young and old cohorts to ensure consistent classification across age groups. The mice were classified as non-frail (0 criteria met), pre-frail (1–2 criteria), or frail (≥ 3 criteria) based on their performance. We also calculated the FI value for each mouse, which quantifies cumulative age-related physiological deficits (Table S2).

ScRNA-seq was performed on mice in the frail old (F-O), non-frail young (NF-O), and non-frail young (NF-Y) groups to delineate age-related and age-independent transcriptomic dynamics in frailty. Pre-frail (PF) mice were excluded from the subsequent analyses as they failed to meet the established frailty criteria at the time of sample collection.

Hematoxylin and eosin staining and quantification of minimum Feret diameters and fiber cross-sectional areas

Paraffin-embedded muscle tissues were sectioned and air-dried at 55℃. The slides were deparaffinized with xylene (Fuchen, Catalog #20200110), rehydrated in an alcohol series, and rinsed with sterilized Milli-Q water. Next, the rehydrated sections were incubated in hematoxylin (Servicebio, Catalog #C191204), rinsed with running water, differentiated in 1% acid alcohol, and rinsed with running water. The sections were counterstained with eosin (Servicebio, Catalog #G1001), dehydrated in an alcohol series and 100% xylene (Fuchen, Catalog #20200110) before being mounted. The minimum Feret diameter and fiber cross-sectional area (CSA) were quantified by applying the SIOX plugin in ImageJ.

Isolation of single cells from mice muscles

Single cells were isolated from mice muscles as described previously35. Mice hind limb muscles were minced into pieces, transferred to dissociation buffer containing Ham’s F10 Medium (Gibco), 10% fetal bovine serum (FBS, Gibco) and collagenase type II (725 U/mL, Worthington) and incubated at 37℃ for 1 h with gentle shaking. After terminating dissociation, the samples were centrifuged at 500 g, and the supernatant was removed, leaving 4 mL. Then, 0.5 mL of collagenase type II stock buffer (1,000 U/mL) and 0.5 mL of dispase stock buffer (11 U/mL, Roche) were added to each sample, followed by resuspension and incubation at 37℃ for 30 min. The samples were slowly passed through an 18-gauge needle 10 times to dissociate the muscle pieces. After centrifuging, the cell pellet was resuspended, passed through a 40 mm filter (BD Falcon), centrifuged, resuspended in 2 mL of red blood cell lysis buffer (Servicebio) for 3 min, washed with cold phosphate-buffered saline (PBS) and resuspended in cold PBS with 0.04% bovine serum albumin.

Primary analysis, quality control, dimension-reduction, and clustering of scRNA-seq data

ScRNA-seq raw reads were aligned against the mus_musculus_ensembl_92 genome (https://ftp.ensembl.org/pub/release-92/gtf/mus_musculus/) and gene expression profiles were generated using CeleScope (v1.5.2, Singleron Biotechnologies) with default parameters. For each sample dataset, the expression matrix was filtered using R package Seurat (v3.1.2) and the following exclusion criteria: (1) cells with gene counts of less than 200 or in the top 2% gene counts, (2) cells in the top 2% UMI counts, (3) cells with mitochondrial contents of \(\:>\) 50%, and (4) genes expressed in fewer than five cells. Gene expression matrices were normalized and scaled using the NormalizeData and ScaleData functions. The top 2000 variable genes were selected for principal component analysis (PCA) using FindVariableFeatures. Cell clusters were visualized using Uniform Manifold Approximation and Projection (UMAP) with the RunUMAP function. Batch effects between samples were removed using Harmony v1.0 and the top 20 principal components from PCA.

Differentially expressed genes analysis and pathway enrichment analysis

Differentially expressed genes (DEGs) were identified using the Seurat FindMarkers function based on the Wilcoxon rank sum test with default parameters. DEGs were defined as genes expressed in more than 10% of the cells in both of the compared groups with an average log(fold-change) value greater than 0.25. A p-value threshold of 0.05 was used to indicate statistical significance. Gene Ontology (GO) analysis was performed using Metascape (https://metascape.org/gp/index.html). Terms with p-values of less than 0.01 were considered to be significantly enriched. Selected significant pathways were plotted as dot plots.

Transcription factor regulatory network analysis

A TF network was constructed using pyscenic (v0.11.0), the scRNA expression matrix, and TFs in AnimalTFDB. First, GRNBoost2 was used to predict a regulatory network based on the co-expression of regulators and targets. CisTarget was then applied to exclude indirect targets and search for TF-binding motifs. Then, AUCell was used to quantify regulon activity in each cell. Cluster-specific TF regulons were identified according to regulon specificity scores (RSSs), and the activity of these TF regulons was visualized using heatmaps.

Cell-cell interaction analysis

Cell-cell interactions between different cell types were predicted based on known ligand-receptor pairs using Cellphone DB (v2.1.0). The permutation number used to calculate the null distribution of average ligand-receptor pair expression in randomized cell identities was set to 1000. Cutoff thresholds for individual ligand or receptor expression were based on the average log gene expression distribution for all genes in each cell type. Predicted interaction pairs with p-values of < 0.05 and an average log expression of > 0.1 were considered significant.

Pseudotime trajectory analysis

The cell differentiation trajectory of monocyte subtypes was reconstructed using Monocle2 (v 2.10.0) and the top 2000 highly variable genes that were selected by Seurat (v3.1.2) FindVairableFeatures. Dimension reduction was performed using DDRTree. The trajectory was visualized using the plot_cell_trajectory function in Monocle2.

Fluorescent-activated cell sorting of B cells

Single cells dissociated from mice muscles were suspended in buffer containing propidium iodide (421301, BioLegend), fluorescein isothiocyanate rat anti-mouse CD45 (103108, BioLegend), PerCP-Cy5.5 rat anti-mouse CD19 (561113, BD), and V500 rat anti-mouse CD45R/B220 (561226, BD). A FACS Aria III instrument (BD) was used for sorting. We used antibodies against the B cell CD19 and B220 marker antigens (specific cell-surface markers) and gated on the CD19+ B220+ cell population.

SMART-seq analysis of B cells

The raw reads were filtered using Seqtk, and then quality control was performed utilizing FastQC (v0.11.9). Adapter sequences were removed using Trim Galore (v0.6.7). Then, hisat2 (v2.0.4) was used to align the trimmed clean reads to GRCm38 (ftp://ftp.ensembl.org/pub/release-83/fasta/mus_musculus/dna/Mus_musculus.GRCm38.dna.primary_assembly.fa.gz). Next, gene count matrices were generated using StringTie (v1.3.0) and DEGs were determined using DESeq2 (v1.26.0). Similar screening criteria were used for SMART-seq and scRNA-seq of differential genes to better compare the results. A p-value threshold of 0.05 was used to evaluate statistical significance. Downstream pathway analysis was performed using Metascape.

Results

Assessing and defining frail status in young and naturally aging mice.

As naturally aging mice might reflect the aging process better than mice with accelerated aging36, all the aging mice used in this study were aging naturally. We included 55 naturally aging male mice (24.5 months old, ~ 80 years old in humans) and nine young male mice (7.5 months old, ~ 30 years old in humans) in cohort 1 (Fig. 1a).

Fig. 1
figure 1

Assembly and construction of scRNA-seq atlas of muscles from young, old and frail mice. (a) Schematic diagram of frailty assessments of young (n = 9) and old (n = 55) mice. NF-Y, 7.5-month-old (young) non-frail mice; NF-O, 24.5-month-old (old) non-frail mice; PF-Y, 7.5-month-old (young) pre-frail mice; PF-O, 24.5-month-old (old) pre-frail mice; F-O, 24.5-month-old (old) frail mice. The figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 4.0 unported license. (b) The prevalence of frailty in young and old mice. Frailty status was based on the cutoff values for each criterion in each respective age. (c) Graphic representation of frailty index values in non-frail, pre-frail, and frail groups between young and old mice. Data are presented as mean ± SD. Statistical differences were tested using the t-test for two groups or one-way ANOVA for three groups. (d-h) Percentage of success in the unintentional weight loss (d), tight-rope test (e), running time (f), running speed (g) and grip strength (h) of NF-Y (n = 6), NF-O (n = 6), and F-O (n = 9). Statistical differences were tested using Pearson’s chi-squared test. (i) Hematoxylin and eosin (H&E) staining was used in each group to assess the tissular integrity of muscle samples. Scale bar, 50 μm. (j) The distribution of myofiber sizes determined using the minimum Feret diameter was quantified for each group. (k) The fiber cross-sectional area (CSA) of each group was assessed. Statistical differences were tested using one-way ANOVA for three groups.

Full size image

Although frailty was closely associated with age, it was not a necessary condition for frailty, and some factors influencing frailty were independent of age9,11,12. We categorized frailty status among individuals of the same age based on the frailty phenotype33 to explore the mechanisms of frailty-dependent/independent aging (Fig. 1a, b, and d-h). The frailty index34 was used as a supplementary measure to further evaluate frailty status (Fig. 1c). Similar to the proportion of frail elderly humans5, the proportion of frail mice among naturally aging mice was ~ 16.4% (Fig. 1b). Similarly, the frailty index values in the F-O group were significantly higher than in the NF-O group (Fig. 1c). Moreover, frailty phenotypes based on all five criteria were stronger in the F-O group compared to the NF-O and NF-Y groups (Fig. 1d-h). Furthermore, hematoxylin and eosin (H&E) staining of muscles showed that NF-Y myofibers assessed using the minimum Feret diameter and fiber cross-sectional area (CSA) were larger than those in the NF-O group, and those in the NF-O group were larger than those in the F-O group (Fig. 1i-k). The myofiber atrophy results in aging and frail mice confirmed the architectural integrity of the muscle tissue and suggested that some factors beyond age might contribute to the development of frailty.

Profiling single-cell transcriptional landscape in relatively healthy aging and physically frail mice using hind limb muscles

We prepared single-cell suspensions of hind limb muscles from NF-Y, NF-O, and F-O mice, followed by scRNA-seq, to investigate the effects of healthy aging and physical frailty on muscle transcriptional profiles (Fig. 2a). After controlling for quality and preprocessing the data, high-quality scRNA-seq data from 70,255 cells (NF-Y = 18,053, NF-O = 26,805, and F-O = 25,397) (Fig. 2b) were further annotated into 15 major cell types, including fibroblasts, fibro/adipogenic progenitors (FAPs), mononuclear phagocytes (MPs), endothelial cells (ECs), smooth muscle cells (SMCs), fast-twitch muscle fibers (FTs), tenocytes, neutrophils, satellite cells, T cells, Schwann cells, B cells, mast cells, erythrocytes, and adipocytes, according to the percent and average expression of canonical cell type-specific markers (Fig. 2c and Fig. S1e-r). For example, SMCs were distinguished based on the high expression level of Tagln and Acta2, as in previous reports37 (Fig. 2c). Pax7 and Myf5-high expressing cells were identified as satellite cells38 (Fig. 2c). B cells were identified by high expression levels of Cd79a and Ms4a116 (Fig. 2c). Altogether, we characterized the intricate organization of muscle tissue down to the single-cell level and obtained a single-cell transcriptional profile of 15 major cell types in the hind limb muscles of relatively healthy aging and frail mice.

Fig. 2
figure 2

Changes in transcriptional profiles of different cell types during relatively healthy aging, age-associated frailty, and age-independent frailty. (a) Flowchart overview of scRNA-seq methodology. (b) Summary of mice and number of cells with scRNA-seq results that passed quality control assessment. (c) Uniform manifold approximation and projection (UMAP) plots showing different cell types in the scRNA-seq samples. The dot plot on the right shows the marker genes for distinct cell types marked with different colors. The color scale indicates the average expression of marker genes in each cell type, and the dot size is proportional to the percentage of cells within each cell cluster expressing the marker genes. (d-f) Bi-directional bar plot showing the number of different genes in 13 cell types in the age-independent frailty (AIF, F-O vs. NF-O) (d), age-associated frailty (AAF, F-O vs. NF-Y) (e) and relatively healthy aging (RHA, NF-O vs. NF-Y) groups (f). Green indicates down-regulated genes, and red indicates up-regulated genes. (g) Heatmaps showing the distribution of DEGs in each cell type in different groups, including AAF, RHA, and AIF. Each row represents one cell type, and each column represents one gene. Red represents up-regulated DEGs, blue DEGs are down-regulated, and gray genes are unchanged. The gene sets were defined as follows: set 1 contained genes not significantly down-regulated in the AIF, RHA, and AAF groups. Set 2 comprised genes not significantly up-regulated in any group, and all remaining genes were assigned to set 3. The genes in set 3 were systematically sorted by cell type (alphabetical ascending) in each comparison group, then by p-value (ascending) and fold-change (descending), revealing three distinct expression patterns: pattern 1 was predominantly observed in AIF-DEGs, pattern 2 was most prominent in RHA-DEGs, and pattern 3 was characteristic of AAF-DEGs.

Full size image

Compositional dynamics of cell types in relatively healthy aging, age-associated frailty, and age-independent frailty

We used split UMAP visualization to depict the cellular composition of the NF-O, NF-Y and F-O groups (Fig. S1a-c). The NF-Y and F-O groups comprised all 15 types of cells, whereas the NF-O group showed no adipocytes (Table S3). This finding agreed with the work of Kedlian et al.39, whose scRNA-seq data did not detect adipocytes in muscle samples from 3-month-old and 18-month-old mice, potentially due to technical limitations of scRNA-seq40,41,42. The cell ratio changed in some specific cell types during the RHA process, including satellite cells, neutrophils, B cells, ECs, and FTs, which were further enhanced or decreased in the AIF stage (Fig. S1d). For example, the ratio of satellite cells was reduced in RHA (NF-O vs. NF-Y) and further reduced in AIF (F-O vs. NF-O), suggesting that aging- and frailty-associated dysfunction might affect the number and functions of satellite cells, as previously reported40. Although these data suggest that the cell-type composition of muscles might be rearranged during relatively healthy aging and physical frailty, additional approaches and further verification are required to validate this conclusion.

Changes in transcriptional profiles of different cell types in relatively healthy aging, age-associated frailty, and age-independent frailty

To assess how factors beyond age promote frailty at the transcriptional level, we defined F-O vs. NF-Y as age-associated frailty (AAF), and F-O vs. NF-O as age-independent frailty (AIF). While the NF-O group represented relatively healthy naturally aging mice, we defined the NF-O vs. NF-Y as relatively healthy aging (RHA). Considering that DEGs reflect the principal molecular events during healthy aging and physical frailty, we identified a total of 5,092 DEGs in 13 cell types (excluding erythrocytes and adipocytes) in a comparison of the three groups (Fig. 2d-f). Of the DEGs in 13 cell types, B cells displayed more down-regulated AIF-DEGs and AAF-DEGs than other cell types, whereas satellite cells showed more up-regulated AIF-DEGs and AAF-DEGs (Fig. 2d, e). The findings suggest that the genes significantly down-regulated in B cells and up-regulated in satellite cells might function in the frailty process dependent or independent of age. As for the RHA-DEGs, mast cells possessed more down-regulated and up-regulated RHA-DEGs than other cell types (Fig. 2f), suggesting that transcriptional changes in mast cells might play a crucial role in the healthy aging process. Then, we surveyed the functional significance of these DEGs in the corresponding cell types and groups. The AAF-DEGs and AIF-DEGs down-regulated in B cells were enriched in the processes of lymphocyte activation, differentiation, and cell-cell adhesion (Fig. S2a, e). The AAF-DEGs and AIF-DEGs up-regulated in satellite cells were associated with mitochondrial translation, muscle structure development, muscle cell development, and muscle system processes (Fig. S2b, f). The RHA-DEGs down-regulated in mast cells were mainly involved in leukocyte differentiation, activation, and the inflammatory response, whereas the RHA-DEGs up-regulated in mast cells were involved in the positive regulation of cell motility, cell migration, and the immune response (Fig. S2c-d). Thus, our analyses showed cell type-specific effects and transcriptional heterogeneities in healthy aging and physical frailty, providing a comprehensive profile for understanding the molecular mechanisms of healthy aging and frailty.

Next, the genes were systematically categorized into three sets (Fig. 2g): set 1 (genes not down-regulated in any of the three groups), set 2 (genes not up-regulated in any group), and set 3 (the remaining genes). Subsequent analysis sorted all genes hierarchically according to (1) AIF, RHA, and AAF, (2) cell type (alphabetical order), (3) p-value (ascending), and (4) fold change (descending). In set 1, Schwann cells showed the most AAF-DEGs and RHA-DEGs (Fig. S3a-b). However, satellite cells in set 1 showed the most AIF-DEGs (Fig. S3c), suggesting that the up-regulated DEGs in satellite cells might affect the frailty process beyond the influence of age. In set 2, B cells consistently displayed the most AAF-DEGs, RHA-DEGs and AIF-DEGs, emphasizing the role of B cells in aging and physical frailty (Fig. S3d-f). We further surveyed the functional significance of these DEGs in the corresponding cell types and groups in sets 1 and 2. The AAF-DEGs and RHA-DEGs up-regulated in Schwann cells were consistently enriched in GO terms related to the positive regulation of cytokine production, the negative regulation of AIM2 inflammasome complex assembly, and leukocyte differentiation (Fig. S3g-h). The AIF-DEGs up-regulated in satellite cells were mainly associated with endoplasmic reticulum organization, nucleotide metabolic processes, and mitochondrion organization (Fig. S3i). The AAF-DEGs down-regulated in RHA and AIF B cells were all mainly enriched in leukocyte differentiation, activation, and the B cell receptor signaling pathway (Fig. S3j-l).

Three distinct transcriptional patterns corresponding to each group in set 3 were seen. Pattern 1 was most obvious in AIF-DEGs, while the most down-regulated and up-regulated AIF-DEGs were seen in B cells and satellite cells, respectively (Figs. 2g and 3a). Pattern 2 was most striking in RHA-DEGs (Fig. 2g), while the down-regulated and up-regulated RHA-DEGs were most associated with mast cells and neutrophils, respectively (Fig. 3b). Pattern 3 was strongest in AAF-DEGs (Fig. 2g), and the down-regulated and up-regulated AAF-DEGs were most concentrated in B cells and satellite cells, respectively (Fig. 3c). These results also revealed the important roles of B cells and satellite cells in physical frailty and the critical function of mast cells in healthy aging. We next explored the biological implications of these DEGs using GO analysis and found that the down-regulated AIF-DEGs in pattern 1 and the AAF-DEGs in pattern 3 in B cells were both involved in lymphocyte activation, differentiation, and immunity (Fig. 3d, e). In contrast, the AIF-DEGs up-regulated in satellite cells in pattern 1 were related to the regulation of apoptosis, cell death, proliferation, and muscle adaptation (Fig. 3f), and the AAF-DEGs up-regulated in satellite cells in pattern 3 modulated mitochondrial gene expression, translation, and organization (Fig. 3g). Additionally, the RHA-DEGs down-regulated in mast cells in pattern 2 were mainly enriched in neuron development and differentiation, leukocyte activation, and the inflammatory response (Fig. 3h), and the RHA-DEGs up-regulated in neutrophils in pattern 2 were mostly related to the negative regulation of leukocyte activation and proliferation, the immune response, and cell-cell adhesion (Fig. 3i).

Fig. 3
figure 3

Changes in transcriptional profiles of different cell types in specific patterns of relatively healthy aging, age-associated frailty, and age-independent frailty.

(a-c) Bi-directional bar plots showing the number of different genes in 13 cell types in the AIF group with pattern 1 (a), the RHA group with pattern 2 (b) and the AAF group with pattern 3 (c). Green indicates down-regulated genes, and red indicates up-regulated genes. (d-e) GO enrichment analysis of genes down-regulated in B cells in the AIF group with pattern 1 (d) and genes down-regulated in B cells in the AAF group with pattern 3 (e). (f-g) GO enrichment analysis of genes up-regulated in satellite cells in the AIF group with pattern 1 (f) and genes up-regulated in satellite cells in the AAF group with pattern 3 (g). (h-i) GO enrichment analysis of genes down-regulated in mast cells (h) and genes up-regulated in neutrophils (i) in the RHA group with pattern 2.

Full size image

Thus, our analyses identified specific DEGs in specific cell types in healthy aging and physical frailty, emphasizing the importance of B cells and satellite cells in frailty, implying the contributions of DEGs up-regulated in satellite cells and down-regulated in B cells in both age-associated frailty and age-independent frailty.

Reconstruction of transcriptional regulatory networks during relatively healthy aging, age-associated frailty, and age-independent frailty

Next, we performed single-cell regulatory network inference and clustering (SCENIC) for all cells in each group to predict the core regulons in each group, and the regulon specificity score (RSS) was calculated to identify group-specific regulons in the F-O, NF-O, and NF-Y groups (Fig. S4a-c). Next, Wilcoxon’s rank-sum test was performed on the activity of all the identified regulons. Of them, 273 regulons with significant differential regulon (DR) activity in at least one group were retained (|L2FC| ≥ 0.25 and a Benjamini–Hochberg corrected P-value of ≤ 0.05) (Table S4). The unsupervised hierarchical clustering of these 273 DRs demonstrated six distinct transcriptional regulon clusters, suggesting the transcriptional heterogeneity between AAF, AIF and RHA with the reconstruction of transcriptional regulatory networks (Fig. S4d). The networks were used to visualize interactions between the top 10 core TFs and their candidate targets in all cells in the NF-Y, NF-O and F-O groups (Fig. S4e-g).

The RSS of each regulon predicted by SCENIC analysis was assessed to identify B cell-specific regulons in the F-O, NF-O, and NF-Y groups (Fig. S5a-c). K-means clustering revealed six distinct transcriptional regulon clusters in B cells, among which cluster 1 and cluster 4 were involved in AIF, cluster 2 and cluster 5 were associated with AAF, and clusters 3 and 6 were specific to RHA (Fig. 4a). The TFs in cluster 1 were enriched in GO terms related to epithelial cell differentiation and skeletal system development, whereas the TFs in cluster 4 were enriched in GO terms related to DNA-templated transcription and muscle tissue development (Fig. S5d-e). The results suggest that these AIF-specific DRs and B cells might mediate muscle function in AIF. The networks between the top 10 core TFs and their targeted genes in B cells in the NF-Y, NF-O and F-O groups were also visualized (Fig. 4b and Fig. S5f-g).

Fig. 4
figure 4

Reconstruction of transcriptional regulatory networks and ligand-receptor interactions during relatively healthy aging, age-associated frailty, and age-independent frailty.

(a) Heatmap visualization of log2 fold-changes based on the average area under the curve (AUC) scores (regulon activities) of B cells in different compared groups. AIF refers to age-independent frailty (F-O vs. NF-O), AAF is age-associated frailty (F-O vs. NF-Y), and RHA is relatively healthy aging (NF-O vs. NF-Y). (b) B cell regulons were visualized as networks. Transcription factors (TFs) and targeted DEGs are shown as square and circular nodes, respectively. The candidate targets of top 10 TFs and DEGs in each group were merged. The intersection was used to plot the regulation networks. (c-e) Network plots showing changes in ligand-receptor interaction events between different cell types in the RHA (c), AAF (d) and AIF (e) groups. Cell-cell communication is indicated by the connected line. The thickness of the lines is correlated with the number of ligand-receptor interaction events. Edge thickness scales with ligand-receptor interaction frequency are shown. Red and blue edges indicate increased and decreased interactions, respectively. (f-h) Network plots showing changes in ligand-receptor interaction events between B cells and other cell types in the RHA (f), AAF (g), and AIF (h) groups. Cell-cell communication is indicated by the connected line. The thickness of the lines is correlated with the number of ligand-receptor interaction events. Edge thickness scales with ligand-receptor interaction frequency are shown. Red and blue edges indicate increased and decreased interactions, respectively.

Full size image

We found satellite cell-specific regulons in the F-O, NF-O, and NF-Y groups (Fig. S6a-c) using SCENIC and the RSSs were estimated. Six distinct transcriptional regulon clusters were identified by K-means clustering, among which clusters 3 and 4 were specific to AIF, clusters 1 and 6 were specific to AAF, and clusters 2 and 5 were specific to RHA (Fig. S6d). Notably, the TFs in cluster 3 were associated with GO items related to DNA-templated transcription and the regulation of the miRNA metabolic process (Fig. S6e), while the TFs in cluster 4 were related to GO terms associated with muscle organ development and stem cell differentiation (Fig. S6f). The transcriptional regulatory networks of the top 10 core regulons in the NF-Y, NF-O, and F-O groups were also similarly visualized (Fig. S6g-i). Satellite cells exhibited distinct top 10 core TFs between the FO/NFO and NFY groups, suggesting the substantial remodeling of transcriptional regulatory networks during aging. Seven TFs (Hoxd8, Lbx1, Mecom, Mef2c, Spi1, Tbx15, and Zfp386) overlapped between the FO and NFO groups, suggesting their collective role in both muscle aging and frailty. In support of this notion, aging associations have been established for six of these factors (excluding Zfp386, also known as kzf1). Hoxd8 is consistently hypermethylated in aged human skeletal muscle tissue and muscle-derived cells, demonstrating conserved age-related epigenetic dysregulation in broader HOX gene family alterations43. Lbx1 exhibits age-related hypermethylation in the rat hippocampus and correlates significantly with spatial memory decline in aged rats44. Mecom is one of the key energy metabolism-related genes down-regulated in Alzheimer’s disease45. Mef2c promoter activity and histone acetylation are suppressed by HDAC11 during myoblast differentiation, indicating its involvement in age-related muscle dysfunction through MyoD-dependent transcriptional regulation46. Spi1 mutation in myeloid cells selectively decreases CD163+/CD206+ M2 macrophage polarization in aged muscle, attenuating age-related fibrosis and preventing sarcopenia47. Tbx15 is significantly up-regulated in senescent bone marrow-derived mesenchymal stem cells (MSCs) from aged donors and associated with malignant transformation potential, suggesting it is a potential biomarker for assessing MSC donor aging and therapeutic safety48. However, their mechanistic roles in satellite cells during aging and frailty require further investigation.

Changes in ligand-receptor interactions between different cell types during relatively healthy aging, age-associated frailty, and age-independent frailty

The transformation of intercellular communication is an integrative hallmark of aging and frailty49. Thus, we compared ligand-receptor pairs in diverse cell types in the NF-Y, NF-O and F-O groups. Most cell-cell communications were enhanced in all three compared pairs, indicating active interactions between various cell types during aging and frailty (Fig. 4c-e and Fig. S7a-c). Cell-cell interactions were enhanced in AAF and AIF compared to RHA (Fig. 4d-e).

In the comparison of the interaction numbers between B cells and other types of cells, B cells exhibited augmented interactions with the most of other cell types in all three groups (Fig. 4f-h). The interactive B cells (sender)-B cell (receiver) pairs were increased in RHA, whereas they were decreased in AAF and AIF. Similarly, both B cell (sender)-fibroblast (receiver) and B cell (sender)-FAP (receiver) pairs were enhanced in AAF and AIF but decreased in RHA. In addition, B cell (sender)-T cell (receiver) pairs were increased in RHA and AAF but decreased in AIF. Thus, our results suggest that communications between B cells and B/T cells were compromised in frailty, whereas communication between B cells and fibroblasts/FAPs was increased (Fig. 4f-h), indicating that the cell-cell interactions between B cells and B cells/T cells/fibroblasts/FAPs might contribute to the frailty process, dependent or independent of aging. The interactions between satellite cells themselves and satellite cells, neutrophils, MPs, mast cells, FAPs, tenocytes, and SMCs were decreased in RHA but increased in AAF and AIF (Fig. S7d-f). These findings suggest the roles of these specific cell types in the communication and function of satellite cells in the frailty process with or without the effects of aging.

Multidimensional dynamic changes and trajectories of B cells during relatively healthy aging, age-associated frailty and age-independent frailty

The scRNA-seq data were validated, and the dynamic changes in B cells during aging and physical frailty were further explored by establishing the naturally aging/ frail mouse model, estimating the frailty phenotype and frailty index, and sorting B cells out from limb muscles using FACS followed by SMART-seq analysis (Fig. 5a-c and Fig. S8a-g). Conjoint analysis showed a significant correlation between the SMART-seq and the scRNA-seq data (Fig. 5d, f). After intersecting the DEGs down-regulated in the SMART-seq and scRNA-seq data to identify the overlapping AAF- and AIF-DEGs, the resulting DEGs were enriched in terms related to cell differentiation and activation (Fig. 5e, g).

Fig. 5
figure 5

Changes in SMART-seq transcriptional profiles of B cells sorted by FACS and trajectories of B cells in relatively healthy aging, age-associated frailty, and age-independent frailty. (a) Schematic diagram of frailty assessment of young (n = 16) and old (n = 50) mice. NF-Y, 6-month-old (young) non-frail mice; NF-O, 24-month-old (old) non-frail mice; PF-Y, 6-month-old pre-frail (young) mice; PF-O, 24-month-old (old) pre-frail mice; F-O 24-month-old (old) frail mice. The figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 4.0 unported license. (b) Graphic representation of frailty index values in the NF-Y, NF-O, and F-O groups. Data are presented as mean ± SD. Statistical differences were tested using the t-test for two groups or one-way ANOVA for three groups. (c) Schematic diagram of FACS sorting of B cells. NF-Y, 6-month-old (young) non-frail mice; NF-O, 24-month-old (old) non-frail mice; F-O, 24-month-old (old) frail mice. The figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 4.0 unported license. (d) Pearson’s correlation analysis of B cell SMART-seq and scRNA-seq data in AAF (F-O vs. NF-Y). (e) The Venn diagram on the left shows the intersecting genes down-regulated in B cells in AAF (F-O vs. NF-Y) analyzed using SMART-seq and scRNA-seq. The dot plot on the right shows the GO enrichment analysis of down-regulated genes intersecting in SMART-seq and scRNA-seq of B cells in AAF (F-O vs. NF-Y). (f) Pearson’s correlation analysis of SMART-seq and scRNA-seq data of B cells in AIF (F-O vs. NF-O). (g) The Venn diagram on the left shows the intersecting genes down-regulated in B cells in AIF (F-O vs. NF-O) analyzed using SMART-seq and scRNA-seq. The dot plot on the right shows the GO enrichment analysis of down-regulated genes intersecting in SMART-seq and scRNA-seq data of B cells in AIF (F-O vs. NF-O). (h-i) Discriminative dimensionality reduction (DDR) tree visualization of B-cell trajectory with age-group (h) and pseudotime (i) mapping. (j) Discriminative dimensionality reduction (DDR) tree visualization of B-cell trajectory with each age-group mapping. (k) Heatmap visualization of the expression levels of genes with correlated and anti-correlated expression profiles to B cell aging pseudotime. (l) Heatmap visualization of expression levels of genes consistent with SMART-seq data in different groups.

Full size image

Next, we performed cell trajectory analysis and mapped B cells from the NF-Y, NF-O and F-O groups on a pseudotime scale to further reveal differentiation changes in B cells. B cells from the NF-Y group were typically clustered at the early end, B cells from the NF-O group were mainly clustered at both the early and late ends, and B cells from the F-O group mostly occupied the late end (Fig. 5h-j). Gene expression patterns mapped on the pseudotime scale were identified and further validated using SMART-seq (Fig. 5k-l). Consistently, genes associated with antigen presentation and the immune response, including Igha, Ighg1, and Jchain, were highly expressed in the late-end B cells from the F-O and NF-O groups (Fig. 5k-l). The genes associated with B cell differentiation, including Ms4a1 (also known as CD20), were highly expressed in early-end B cells from the NF-Y group and decreased in late-end B cells from the F-O (Fig. 5k-l). Additionally, some ribosomal protein-coding genes, including Rpl18a, Rps11 and Fau, and cytoskeleton-related proteins, including Tmsb10 and Coro1a, were expressed at low levels in the F-O group. We further performed subgroup clustering of B cells to investigate whether the differences in cell trajectories were related to different B cell subtypes. The results showed that the proportions of B cell subtypes were different in the three groups (Fig. S8h-j). Consistent with our trajectory analysis, most B cells in the F-O group were plasma cells, and most of those in the NF-Y group were mature B cells, whereas the proportions of plasma cells and mature B cells were approximately equal in the NF-O group (Fig. S8h-j). Overall, B cells exhibited distinct cell trajectories in aging and physical frailty, which might be associated with specific transcriptional patterns and variations in the proportion of cell subtypes.

Discussion

Frailty, representative of pathological aging, is associated with complex demographic, medical, and social problems3,4. The prevalence of frailty is 11–14% in people 60–69 years old and about 26–61% in people 80–90 years old50. However, the precise definition and fundamental mechanism of frailty remain poorly characterized9,28. Frailty is related to age and diverse disease states, neither of which is necessary for frailty occurrence9,11,12. However, few studies have investigated whether frailty has unique and specific changes beyond the influence of age. Thus, the relationship between aging and frailty should be studied further to understand the detailed mechanisms and elaborate processes of AAF and AIF.

Here, we made detailed frailty estimations using the frailty phenotype and frailty index and used them to categorize naturally aging mice into NF-O and F-O groups and robust young mice without frailty as the NF-Y group. Beyond AAF, we also analyzed AIF and RHA transcriptional patterns. scRNA-seq and SMART-seq of the hind limb muscles of mice showed a comprehensive single-cell transcriptional landscape and heterogeneity in RHA, AAF, and AIF, especially in B cells. The transformations in cell composition, gene expression, transcriptional regulation, and cell-cell communication associated with RHA, AAF, and AIF were analyzed to provide insight into the detailed molecular mechanisms underlying aging and physical frailty. Our analysis identified a set of DEGs with distinct expression patterns in these three different conditions. We also found a frailty-related pattern of DEGs, in which down-regulated DEGs were mainly found in B cells while up-regulated DEGs were mostly concentrated in satellite cells, highlighting the importance of muscle-related B cells and satellite cells in frailty. Previous studies reported that satellite cells are involved in aging-related skeletal muscle dysfunction51,52. Our analyses showed that the DEGs up-regulated in satellite cells were principally related to apoptosis and cell death, consistent with previous studies on the exhaustion and depletion of satellite cells in frailty and aging52,53. We also found that the down-regulated DEGs in B cells were mostly enriched in terms associated with lymphocyte activation, differentiation, and the immune response, which mediate B cell dysfunction during aging54,55. In addition, we identified clusters of regulons specific to RHA, AAF, and AIF. Our study supports the hypothesis that the combination of aging-related and aging-independent factors jointly promotes the development of frailty. Thus, all these data might shed light on the investigation, prevention, and intervention of frailty, dependent or independent of aging.

B cell accumulation represents a systemic phenomenon observed in multiple aging organs and is linked to various age-related pathologies56. Single-cell analyses of aged muscle consistently report significant B cell expansion39, consistent with our observations. Our research further revealed increased B cell accumulation during the development of frailty. While single-cell characterizations of muscular B cells in frailty remain scarce, a peripheral blood study reported stable B cell proportions across age groups in human frailty16, reflecting potential tissue-specific differences (blood vs. mice hindlimb muscle in our study). Notably, both studies found impaired B cell functionality. Luo et al. documented the transcriptional dysregulation of antigen presentation in frail individuals16, while our murine model demonstrated compromised antigen presentation in AIF and broader functional impairments in B cell activation and differentiation in the AIF and AAF groups.

Furthermore, we identified significant enrichments in endoplasmic reticulum (ER) unfolded protein response (UPR)-associated TFs (Xbp1、Aft6、Creb3、Aft6b、Creb3l2) among the top 10 core TFs in B cells in the FO-group. The unfolded protein response of the ER (UPRER) serves as a critical mediator of cellular senescence57, where age-associated ER stress characterized by compromised protein-folding capacity and elevated mitochondrial ROS activates all three UPRER pathways (IRE1α, PERK, and ATF6)58. Although B cell activation triggers UPRER to facilitate ER expansion prior to plasma cell differentiation, enabling large-scale antibody synthesis and secretion59,60,61, chronic UPRER activation becomes detrimental by inducing apoptosis of cells and degrading antibody mRNAs62. These UPRER dysregulation may drive immunosenescence via impaired antibody class-switching and autoantibody accumulation, ultimately contributing to frailty through multiple mechanisms13,30,63,64,65, including chronic inflammation-mediated muscle atrophy, pro-inflammatory cytokine disruption of muscle stem cell niches, altered immune homeostasis, and neuroendocrine dysfunction through hypothalamic-pituitary axis suppression. In this model, the interactions between UPRER and the released inflammatory mediators accelerate frailty progression, through both immune-specific mechanisms (increased infection risk) and systemic pathways (chronic inflammation and muscle loss), implying that targeting UPRER components may represent a promising intervention strategy for frailty. However, extensive experimental verification is still needed to demonstrate the generalizability of this model.

Numerous studies have documented the age-related decline satellite cell proportions. As demonstrated by Kedlian et al.39, aged muscle exhibits reduced TNF+ satellite cell numbers, downregulated ribosome biogenesis genes (POLR1D), elevated senescence markers (CDKN2A, TP53), and suppression of pro-growth pathways. The complementary snRNA-seq data confirmed satellite cell depletion in sarcopenia, attributable to proliferative deficits, mitochondrial dysfunction, and inflammatory activation29. Our findings revealed accelerated satellite cell loss during frailty progression, with AIF samples showing activated apoptosis pathways and suppressed proliferation signals and AAF displaying mitochondrial pathway dysregulation, which may collectively explain the decreases in cell numbers.

Notably, the transcriptomic analysis revealed complete divergence in the top 10 core TFs between NF-Y and F-O/NF-O satellite cell, with seven factors (Hoxd8, Lbx1, Mecom, Mef2c, Spi1, Tbx15, and Zfp386) shared between the F-O and NF-O groups, suggesting their potential involvement in both aging and frailty. Regulatory network analysis demonstrated consistent Tbx15-mediated decorin (Dcn) upregulation in the groups, with quantitatively distinct effects (Fig. S6G-I): a 1.98-fold increase was seen in RHA and 2.32-fold in AIF compared to 4.59-fold in AAF (approximately the sum of the RHA + AIF effects). Tbx15 orchestrates mesodermal development and tissue-specific metabolic programming while serving as a multi-disease biomarker and has demonstrated roles in skeletal patterning, glycolytic muscle fiber specification, and associations with aging, malignancy, and Alzheimer’s disease48,66,67. Dcn modulates skeletal muscle plasticity by inhibiting myostatin activity through extracellular binding, regulating fibrillogenesis kinetics, and promoting myoblast proliferation/differentiation68,69. The additive regulatory pattern (AAF was approximately the sum of the RHA + AIF effects) suggests that Tbx15’s promotion of Dcn expression may represent a transcriptional mechanism in which healthy aging-related and frailty-specific factors collectively contribute to gene regulation of AAF. These findings also imply the existence of age-independent pathways in frailty development.

Our literature search identified two single-cell studies related to frailty. Compared to study one16, our muscle-tissue approach captured frailty-specific alterations in the muscle system better than it, which used peripheral blood samples. In addition to the above, both studies revealed consistent patterns in B cell differentiation trajectories during aging and frailty progression. Study 1 demonstrated that cord blood-derived B cells occupied early pseudotime stages while frail group B cells were localized to advanced positions. Our data similarly showed NF-Y group B cells at trajectory origins and F-O group cells at terminal stages. These parallel findings suggest that B cell maturational shifts represent a conserved feature of frailty across different tissue types.

In study two29, the authors used the terms “sarcopenia” and “frail” synonymously and performed bulk RNA-seq comparisons between elderly people with and without sarcopenia. However, due to the limited number of DEGs between these groups, their subsequent single-cell analysis combined all elderly samples into one group and compared them to young controls29. This methodological compromise highlights an important challenge in aging/frailty research; that is, maintaining strict statistical thresholds and capturing biologically relevant but potentially subtle differences. Importantly, our single-cell RNA-seq analysis of NF-O, F-O, and NF-Y murine muscle samples revealed distinct transcriptional regulatory networks in AAF and AIF, providing novel mechanistic insight into both age-dependent and age-independent pathways involved in frailty development.

Our study has several limitations that should be acknowledged: (1) The long-term cost of naturally aging mice, the high mortality and injury rate, and limited funding imposed limitations on the number of mice. Thus, further validations, including incorporating more samples, are necessary. Investigations into the effects of B cells and their interactions with other cell types in the process and mechanism of physical frailty are also needed. (2) This study remains an early-stage exploratory study. To maximize hypothesis generation, we utilized uncorrected p-values for differential gene identification. Subsequent studies should perform multiple testing corrections with larger sample sizes. (3) This study focused on hindlimb muscle tissue in male mice. Thus, whether these findings extend to other sexes or tissues requires further investigation. (4) Growing evidence has shown cell-type capture biases in both scRNA-seq and snRNA-seq of muscle tissue39,41,42. As highlighted in study39, while scRNA-seq provides superior resolution of immune cell subsets, vascular components, and Schwann cell subtypes, snRNA-seq more effectively profiles myonuclei and adipocytes. Thus, the integration of both approaches enables comprehensive single-cell atlas construction. Therefore, future investigations of muscle frailty should combine scRNA-seq and snRNA-seq technologies to generate a more complete cellular map of frailty-associated alterations.