FoxO maintains a genuine muscle stem-cell quiescent state until geriatric age

Tissue regeneration declines with ageing but little is known about whether this arises from changes in stem-cell heterogeneity. Here, in homeostatic skeletal muscle, we identify two quiescent stem-cell states distinguished by relative CD34 expression: CD34High, with stemness properties (genuine state), and CD34Low, committed to myogenic differentiation (primed state). The genuine-quiescent state is unexpectedly preserved into later life, succumbing only in extreme old age due to the acquisition of primed-state traits. Niche-derived IGF1-dependent Akt activation debilitates the genuine stem-cell state by imposing primed-state features via FoxO inhibition. Interventions to neutralize Akt and promote FoxO activity drive a primed-to-genuine state conversion, whereas FoxO inactivation deteriorates the genuine state at a young age, causing regenerative failure of muscle, as occurs in geriatric mice. These findings reveal transcriptional determinants of stem-cell heterogeneity that resist ageing more than previously anticipated and are only lost in extreme old age, with implications for the repair of geriatric muscle. García-Prat, Perdiguero, Alonso-Martín et al. show that skeletal muscle contains a subpopulation of quiescent stem cells, maintained by FoxO signalling, that is preserved into late life but declines in advanced geriatric age.

S tem-cell ageing is a nodal cause of the decline in regenerative capacity of most tissues over time and a vivid example is provided by skeletal muscle 1,2 . Its stem cells (also known as satellite cells) are quiescent for most of their life. In response to injury, these quiescent cells activate, expand and acquire distinct fates, with some differentiating and forming new myofibres and others self-renewing to replenish the homeostatic quiescent stem-cell pool 3,4 . Mouse transgenesis, in vivo labelling and recent single-cell transcriptomic analyses suggest heterogeneity among quiescent satellite cells (QSCs) [5][6][7][8][9][10][11][12] ; however, whether this heterogeneity is affected by age and whether it influences muscle stem-cell progeny, fate and regenerative function with ageing is unknown.
The present study addresses the dynamics of the diversity in muscle stem-cell quiescence throughout the entire postnatal life of an animal. We show that quiescent stem cells expressing high levels of CD34 emerge at a young age (or in juveniles) and are set to self-renew following injury, whereas those expressing low levels of CD34 are primed and restricted to the myogenic fate. Based on this classification, we term these states 'genuine' and 'primed' . We further show that the genuine stem-cell state is preserved into old age through active repression of gene programmes characteristic of the primed state. However, in extreme old (geriatric) age, inappropriate expression of primed-state genes causes regenerative failure.   The cells were cultured for 72-96 h in differentiation medium to obtain 'reserve QSCs' (self-renewal; Pax7 + Ki67 − cells from total DAPI + cells; n = 8 from four mice; two replicates per mouse). Paired t-test. j, Representative image (left) and quantification (right) of CD34 H and CD34 L QSCs from ROSA26-tdTomato and β-actin-GFP mice (n = 9 transplanted muscles from three mice) transplanted into the pre-injured tibialis anterior (TA) of the recipient immunodeficient mice for 3 d. Equal cell numbers were used. Scale bar, 75 μm. k, CD34 H and CD34 L QSCs (ten mice) were transduced with a lentivirus expressing GFP (LV-GFP) and transplanted for 21 d. Representative images (left) and quantification (middle and right) of GFP + fibres (middle, n = 16 CD34 L and 20 CD34 H transplanted muscles; right, self-renewal (GFP + Pax7 + cells), n = 7 CD34 L and 9 CD34 H transplanted muscles). The mice were 2-4 months old. Scale bar, 100 μm. Data are the mean ± s.e.m.; a two-tailed unpaired t-test with Welch's correction was used unless otherwise indicated. SCs, satellite cells.
(including Myog (Myogenin), Tmem8c (Mymk; Myomaker) and Actc1). Single-cell RNA-seq of QSCs confirmed an association of Myod1 and Myog transcripts with cells expressing low Cd34 levels (Extended Data Fig. 1c). Through RNAscope, we validated the existence of two QSC fractions based on Cd34 expression in intact muscle from Pax7 cre-ER ; ROSA2 tdTomato mice (where Pax7-expressing satellite cells are permanently labelled): cells expressing high Cd34 levels also expressed high Notch3 levels, whereas those expressing low Cd34 levels expressed high Myog levels (Fig. 1e). Compared with activated cells, both the CD34 H and CD34 L satellite-cell fractions were reduced in size and showed no proliferative traits (Extended Data Fig. 1d), consistent with full quiescence. Thus, QSCs exist in distinct transcriptional states in vivo and can be distinguished by the relative expression of CD34. Compared with freshly isolated CD34 L QSCs, CD34 H QSCs had delayed activation kinetics and completion of the first division cycle, increased dye retention (PKH26) and displayed greater clonogenic and self-renewal capacities (Fig. 1f-i). Competitive transplantation experiments with CD34 H and CD34 L QSCs from green fluorescent protein (GFP)-or tdTomato-expressing mice revealed that CD34 H cells expanded more efficiently and had a repopulation advantage over CD34 L cells (Fig. 1j). CD34 H QSCs also gave rise to more regenerated myofibres and self-renewing satellite cells than CD34 H cells (Fig. 1k). To validate these findings, we used Pax7 cre-ER ; ROSA26 YFP mice, in which satellite cells are permanently labelled with yellow fluorescent protein (YFP; Fig. 2a). Immunostaining of snap-frozen resting muscle confirmed differential CD34-protein fluorescence intensity in YFP + QSCs (Extended Data Fig. 1e), whereas proteins corresponding to genes that were upregulated in CD34 H cells, such as Ryr3, were expressed at higher levels in satellite cells with higher CD34 expression (Extended Data Fig. 1f), thus excluding influences of the isolation/sorting protocol on the CD34-based stem-cell classification. Pax7 cre-ER ; ROSA26 YFP mice also contained a small YFP + CD34 − satellite-cell fraction (2%; Fig. 2a). Bioinformatic analyses of the RNA-seq of these three YFP-labelled subsets (YFP + CD34 H , YFP + CD34 L and YFP + CD34 − ) revealed similar transcriptomic profiles in YFP + CD34 H and α7 + CD34 H as well as in YFP + CD34 L and α7 + CD34 L cells (Extended Data Fig. 1g-j). Ex vivo and in vivo experiments confirmed that the α7 + CD34 H and YFP + CD34 H QSC fractions were functionally similar, both displaying higher stemness potential than α7 + CD34 L , YFP + CD34 L and YFP + CD34 − cells (Fig.  2b,c). Finally, nGFP + CD34 H satellite cells in Pax7-nGFP mice (where Pax7 satellite cells expressed nuclear GFP (nGFP)) showed higher stemness potential than nGFP + CD34 L and nGFP + CD34 − cells, which was retained after muscle re-injury (Extended Data Fig. 2a-d).
We next compiled the transcriptomic data from all young mice (n = 15, 2-4 months old; see below) and interrogated them for 19,074 gene sets 21 . Compared with CD34 L QSCs, the CD34 H QSC transcriptome showed lower cellular activities, including at the myogenic, metabolic, transcriptional and translational levels (Extended Data Fig. 2e), reminiscent of genuine (or more naive) long-term stem cells 22 , with high engrafting and self-renewing capacities. In contrast, CD34 L QSCs were in a primed state, more prone to enter the cell cycle and committed to myogenesis. Based on this functional separation, we termed the CD34 H QSCs as genuine (QSC genuine ) and CD34 L cells as primed (QSC primed ). Transcriptomics of all CD34 H and CD34 L QSC datasets determined that the QSC genuine state core signature was composed of 219 genes and included Thy1 (CD90), Tek (Tie-2), Ryr3 and Notch3 (refs. 23,24 ), whereas the QSC primed core signature was composed of 168 genes, including Myog, Tmem8c, Tnnt2 and Cdk2 (Fig. 2d,e and Supplementary Table 2).
The genuine stem-cell state persists into advanced old age, only collapsing at geriatric age. We next investigated the origin and development of the quiescent stem-cell populations from birth to advanced age. In neonatal mice (8 d post birth), the satellite cells were mostly CD34 − (Fig. 3a), reflecting their highly proliferative status [25][26][27] . CD34 expression was first detected in juvenile mice (21 d post birth), coinciding with the entry of satellite cells into quiescence 27 and witnessing of the emergence of the QSC genuine state (Fig.  3a). The levels of CD34 expression increased further in young mice (2-4 months old), coinciding with satellite cells in the QSC genuine state reaching full quiescence 25,26 . Surprisingly, the QSC genuine state persisted in old mice (18-22 months old) despite the decline in satellite-cell numbers 5,17,18,28,29 (Fig. 3a and Extended Data Fig. 3a). Moreover, based on CD34 expression, the QSC genuine state was preserved even in extreme old age (geriatric; 28-30 months old; Fig.  3a). Neonatal satellite cells were transcriptionally close to juvenile cells and both were distant from young cells (Fig. 3b). Importantly, neonatal and juvenile satellite cells were transcriptionally closer to the QSC primed state, whereas young cells were closer to the QSC genuine state (Fig. 3b). To our surprise, old satellite cells were transcriptionally closer to the young QSC genuine than the young QSC primed state (Fig. 3b).
We next performed RNA-seq of CD34 H and CD34 L cells from young, old and geriatric mice. Hierarchical clustering indicated that the QSC genuine state in old and geriatric mice is transcriptionally distant from the QSC primed state in this age range and closer to the young QSC genuine state (Fig. 3c,d). Principal component analysis (PCA) confirmed that the transcriptome distance between the QSC genuine and QSC primed states was maintained in old mice (Fig. 3c). However, the QSC genuine and QSC primed transcriptomes of geriatric mice drifted from the corresponding states in young cells, and the transcriptional distance between them was reduced (Fig. 3d). In transplantation assays, QSCs genuine from young and old mice showed a similar capacity to regenerate myofibres, which was impaired in geriatric QSCs genuine (Fig. 3e). The age-dependent functional difference between young/old and geriatric QSCs genuine was reproduced at the level of the whole stem-cell population (Extended Data Fig.  3b). Thus, the QSCs that remain in old muscles are age resistant, preserving a high regenerative potential, but this potential is no longer maintained in extreme old age. In contrast, the endogenous muscle regenerative capacity (in the absence of transplanted cells) was already attenuated in old mice (Extended Data Fig. 3c) but was accentuated in geriatric mice, despite a similar reduction in the total satellite-cell number in old and geriatric mice (Extended Data Fig.  3a), as reported previously 18 . Notably, this endogenous regenerative decline in old mice was largely due to a numerical decline in the QSC primed state, whereas geriatric mice showed a similar numerical decline in both QSC states (Extended Data Fig. 3d).
Pathway enrichment analysis of the RNA-seq data revealed that two molecular networks identified as hallmarks of the young QSC primed state (protein translation, and tricarboxylic acid cycle and oxidative phosphorylation (OxPhos)-both indicative of higher cellular activities; Extended Data Fig. 2e) were enriched in the geriatric (but not the old) QSC genuine state (Extended Data Fig. 3e). Thus, the clear demarcation between the genuine and primed stem-cell states is lost in extreme old age due to a blurring of the transcriptome features that earlier define the young and old QSC genuine state clusters.

The genuine stem-cell state is maintained by FoxO signalling.
To investigate the molecular specification of the QSC genuine state, we searched for the most enriched transcription-factor (TF) binding motifs in the genes belonging to the QSC genuine and QSC primed core signatures. This search revealed enrichment for the occupancy of the Forkhead box O (FoxO) family DNA recognition motif ( Fig. 4a and Extended Data Fig. 4a). We next performed transposase-accessible chromatin high-throughput sequencing (ATAC-seq) of freshly sorted CD34 H and CD34 L cells (and in the minor CD34 − cell fraction; Fig. 2a) and analysed the accessibility of genes defining the QSC genuine and QSC primed core signatures. Despite comparable chromatin accessibility in all quiescent populations, the QSC genuine core genes were more accessible in CD34 H cells, whereas the QSC primed core genes were more accessible in CD34 L and CD34 − cells (Fig. 4b).
Open-chromatin peaks in the three cell subsets contained FoxO consensus motifs (Extended Data Fig. 4b). The number of FoxO motifs per peak was elevated in the promoters of the QSC genuine core genes in all three QSC fractions but was the highest in CD34 H cells (Extended Data Fig. 4b), thus highlighting the potential control of the genuine satellite-cell state by FoxO. Intracellular FACS analysis revealed higher expression of FoxO TFs in the QSC genuine state than QSC primed (Fig. 4c). FoxO3a (FoxO3) showed strong nuclear localization, FoxO1 was observed in both the nucleus and cytoplasm, and FoxO4 was cytoplasmic ( Fig. 4d and Extended Data Fig. 4c). We used the Pax7 cre-ER driver to specifically ablate FoxO3a in satellite cells after tamoxifen administration to FoxO3a ΔPax7ER mice (Extended Data Fig. 4d,e). FoxO3a ΔPax7ER mice showed no differences in satellite-cell numbers compared with FoxO3a WT mice (in agreement with a previous study 30 ) or in the relative cell abundance of the QSC genuine and QSC primed states (based on CD34 expression; Extended Data Fig. 4f), which is consistent with a lack of the muscle regenerative phenotype after one round of   , young (n = 4) and old (n = 4) WT mice as well as QSCs genuine and QSCs primed freshly isolated from young WT mice (n = 3; bottom). Heatmap illustrating the expression of genes that are differentially expressed between QSCs genuine and QSCs primed freshly isolated from young WT mice of different ages (top). c, PCA of the full transcriptome of QSCs genuine and QSCs primed freshly isolated from young (n = 4 for both genuine and primed) and old (n = 3 genuine and 4 primed) WT mice (bottom). Heatmap showing the expression of genes that are differentially expressed between QSCs genuine and QSCs primed freshly isolated from WT mice of different ages (top). d, As in c but from young (n = 4 for both genuine and primed) and geriatric (n = 3 for both genuine and primed) WT mice. e, QSCs genuine and QSCs primed isolated from young (n = 12 CD34 H and CD34 L transplanted muscles from six mice), old (n = 6 CD34 H and CD34 L transplanted muscles from eight mice) and geriatric (n = 4 CD34 H and CD34 L transplanted muscles from eight mice) WT mice, which were transduced with LV-GFP and transplanted for 21 d into pre-injured TA muscle of recipient immunodeficient mice. Equal numbers of freshly isolated cells were used. Representative images (top) and the quantification of the GFP + fibres are shown (bottom; 4-12 transplanted muscles QSC genuine state-as indicated by a loss of CD34 H quiescent cells after 30 d (Fig. 4e)-and spontaneous satellite-cell activation and differentiation ( Fig. 4f and Extended Data Fig. 4i). These results were validated by lineage tracing in FoxO1,3a,4 ΔPax7ER ; ROSA26 YFP mice ( Fig.  5a and Extended Data Fig. 4j,k), as abrogation of all three FoxO genes (FoxOs) triggered the loss of CD34 H cells and impacted the total YFP + QSC population (30% reduction; Fig. 5a). Bromodeoxyuridine (BrdU) labelling demonstrated that these FoxO1,3a,4-deficient cells were able to incorporate BrdU and spontaneously fuse to uninjured myofibres (Fig. 5b). Furthermore, FoxO1,3a,4 ΔPax7ER satellite cells showed reduced clonogenic capacity ex vivo (Fig. 5c) and severely impaired capacity to form new myofibres and self-renew following transplantation (Fig. 5d), consistent with the loss of QSCs genuine . The numerical and functional satellite-cell deficit in FoxO1,3a,4 ΔPax7ER mice was evident in impaired muscle regeneration even 15 d post injury (Fig. 5e).   and QSC primed (right) gene signatures using position weight matrices from the entire UCSC genome browser (through Enrichr). FoxO binding-site enrichment is highlighted (red). b, ATAC-seq libraries were generated from YFP + CD34 H , CD34 L and CD34 − QSCs isolated from young Pax7 cre-ER ; ROSA26 YFP mice (2-3 months old; n = 6 mice). Tag density plots around the TSS of genes involved in the QSC genuine and QSC primed signatures are shown. One-sample t-test P values for the QSC genuine genes: CD34 H versus CD34 L , 1.87 × 10 −51 ; CD34 H versus CD34 − , 3.87 × 10 −77 ; and CD34 L versus CD34 − , 4.25 × 10 −42 . One-sample t-test P value for QSC primed genes: CD34 H versus CD34 L , 1.38 × 10 −9 ; CD34 H versus CD34 − , 2.02 × 10 −26 ; and CD34 L versus CD34 − , 1.37 × 10 −17 . c, FoxO1, FoxO3a and FoxO4 MFI of freshly isolated QSCs genuine and QSCs primed (n = 5 mice for all three groups; 2-4 months old), determined using flow cytometry. d, Representative immunofluorescence images of FoxO1, FoxO3a and FoxO4 in QSCs freshly isolated from WT mice (2-4 months old). Scale bar, 10 μm. This experiment was repeated three times. e, Representative flow cytometry analysis (left) and quantification of the percentage of α7-integrin + CD34 + satellite cells within the live cell population of FoxO1,3a,4 WT and FoxO1,3a,4 ΔPax7ER mice (n = 3 mice; 2-4 months old) 30 d after the last tamoxifen injection (right). f, Representative immunofluorescence images (left) and quantification of MyoD + and Myogenin + satellite cells in muscle sections from FoxO1,3a,4 WT and FoxO1,3a,4 ΔPax7ER mice (n = 4 mice; 2-4 months old) 7 d after the last tamoxifen injection. Scale bars, 50 μm (main image, left) and 10 μm (magnified view, right). Data are the mean ± s.e.m.; a two-tailed unpaired t-test with Welch's correction was used unless otherwise indicated. Tmx, tamoxifen. closer to QSCs primed and distant from QSCs genuine (Fig. 6b). Many upregulated genes in the FoxO1,3a,4 ΔPax7ER satellite cells were also upregulated in QSCs primed . Moreover, this shared gene list was enriched in skeletal-muscle-differentiation genes, including Myog and Tnnt2 among 43 QSC primed core signature genes, and mitochondrial and OxPhos genes such as Cox8b or Mtx2 (Extended Data Fig. 5a). In contrast, 49 QSC genuine core signature genes were downregulated in the absence of FoxOs, including Cd34 and Igfbp4 (Extended Data Fig. 5b). FoxO signalling thus maintains the expression of important QSC genuine -state molecules, and FoxO silencing correspondingly causes the disappearance of major QSC genuine core genes and inappropriate expression of genes characteristic of the QSC primed state.

The genetic loss of
To identify early, and thus probably direct, FoxO-target genes, we performed RNA-seq of satellite cells isolated only 4 d after the first tamoxifen administration to FoxO1,3a,4 WT ; ROSA26 YFP and FoxO1,3a,4 ΔPax7ER ; ROSA26 YFP mice. The FoxO1, -3a and -4 genes were deleted at this early post-tamoxifen stage in YFP + cells (Extended Data Fig. 5c). Compared with the analysis at 30 d post tamoxifen treatment, substantially fewer genes showed dysregulated expression at 4 d after deletion of the FoxOs, forming a set of candidate direct FoxO-target genes ( Fig. 4a and Extended Data Fig.  5d,e). Notably, downregulation of QSC genuine and upregulation of QSC primed core gene signatures, including altered expression of Myog and Cd34, was already evident early after deletion of the FoxOs ( Fig.  6c and Extended Data Fig. 5d,e). Furthermore, gene set enrichment analysis (GSEA) revealed alterations to muscle-differentiation, but not mitochondria or OxPhos, pathways (Extended Data Fig. 5f).
Data Fig. 5g). Similar accessibility results were obtained for the promoters of Igfbp4 and Tnnt2 (Extended Data Fig. 5g), matching the reduced and increased expression, respectively, in FoxO1,3a,4-null satellite cells ( Fig. 6a and Extended Data Fig. 5g).
Analysis of the Myog promoter identified a putative FoxO DNA-binding site (TGTTTAC) at −62 to −49 base pairs (bp) from the transcription start site (TSS), and chromatin immunoprecipitation (ChIP) confirmed promoter association of FoxO3a (Extended Data Fig. 6a). In addition, ChIP confirmed that Cd34 is a direct FoxO3a target (Extended Data Fig. 6a), in agreement with the loss of CD34 expression in the absence of FoxOs. These results are consistent with increased and decreased acetylation of the Myog and Cd34 gene promoters, respectively, in FoxO1,3a,4 ΔPax7ER cells, as shown by histone H4 acetylation in ChIP (Fig. 6e). Ablation of the FoxOs thus provokes de-repression of Myog and premature activation of the myogenic-differentiation gene programme, with a concomitant loss of Cd34. Furthermore, ChIP also demonstrated the association of FoxO3a with the Igfbp4 and Tnnt2 promoters (Extended Data Fig. 6a), matching their reduced and increased expression, respectively, in cells lacking FoxOs (Fig. 6a). Finally, the delivery of short interfering RNA (siRNA) targeting Myog messenger RNA (siMyog) into YFP + satellite cells from FoxO1,3a,4 ΔPax7ER ; ROSA26 YFP mice (to silence Myog) restored their capacity to engraft and form new regenerating myofibres after in vivo transplantation ( Fig. 6f and Extended Data Fig. 6b). Transcription from the Myog promoter was increased in the absence of FoxO, as shown by the induction of  Myog-promoter-driven luciferase activity in FoxOs-silenced satellite cells, whereas FoxO3a overexpression repressed Myog transcription in differentiating cells ( Fig. 6g and Extended Data Fig. 6c,d).
Thus, FoxO signalling maintains the QSC genuine state, whereas its loss imposes the QSC primed gene expression programme and depletes the genuine state, with deleterious consequences on regeneration.
Establishment and life persistence of the genuine stem-cell state requires FoxO activity. We generated a new mouse line, FoxO1,3a,4 ΔPax7 , with constitutive deletion of the three FoxOs in the Pax7 satellite-cell lineage to investigate the relevance of FoxO signalling to the initial establishment and evolution of the QSC genuine state (Extended Data Fig. 6e). The gradual acquisition of CD34 expression and emergence of QSCs genuine in WT juvenile mice (Fig. 3a) did not occur in FoxO1,3a,4 ΔPax7 mice (Fig. 7a); FoxO1,3a,4 ΔPax7 satellite cells were CD34 − (Fig. 7a) and continued to express cell-cycle-and myogenic-differentiation-related genes (Extended Data Fig. 6f), thus resembling QSCs primed . In vivo BrdU labelling from postnatal days 21-30 detected BrdU + satellite-cell nuclei in myofibres of FoxO1,3a,4-deficient, but not WT, mice (Extended Data Fig. 6g), thereby demonstrating that satellite cells without FoxOs fail to enter quiescence and establish the genuine state, and this failure persisted throughout life (Fig. 7a) and was exacerbated in old age (Fig. 7b). Thus, in absence of FoxO signalling, the QSC population is never established. Transcriptomic analysis showed that juvenile and old FoxO1,3a,4-deficient satellite cells failed to induce the QSC genuine core signature while showing persistent activation of the QSC primed core signature (Fig. 7c) and enrichment in QSC primed -related pathways-such as, cell-cycle, OxPhos and myogenic differentiation (Extended Data Fig. 7a). We next tested for changes in the expression of FoxOs in the QSC genuine and QSC primed states in the two most distant ages. In young     QSCs, FoxO3a and FoxO4 levels were higher in the genuine state than the primed state ( Fig. 7d and Extended Data Fig. 7b); a similar trend was observed for FoxO1 (Extended Data Fig. 7b). However, the FoxO3a levels in geriatric QSCs genuine were lower than in their young counterparts (Fig. 7d), suggesting that reduced FoxO signalling in geriatric QSCs genuine may account for the stem-cell regenerative decline in extreme old age ( Fig. 3e and Extended Data Fig. 3c,d). Furthermore, acute genetic impairment of FoxOs in FoxO1,3a,4 ΔPax7ER ; ROSA26 YFP geriatric mice did not exacerbate the already pronounced functional deficits of geriatric satellite cells ( Fig. 7e and Extended Data Fig. 7c) despite affecting the CD34 levels ( Fig. 7f and Extended Data Fig. 7d). In agreement with these results, no significant differential enrichment of the genuine and primed signatures was detected between WT and FoxO1,3a,4-null geriatric satellite cells in contrast to those from juvenile, young and old mice (Extended Data Fig. 7e and Figs. 6c, 7c). Thus, young QSCs, equipped with higher FoxO levels, are more sensitive than geriatric cells to FoxO1,3a,4 genetic depletion.
Computational prediction of Igf-Akt pathway inhibition for reinforcement of the genuine stem-cell state. We next aimed to identify signalling molecules emanating from the niche that might maintain the two quiescent stem-cell-state phenotypes throughout life and which might impact FoxOs. For this, we implemented a computational method, NicheHotSpotter, that is based on a probabilistic model of sustained signal transduction, which was previously developed to predict hotspots (key molecules) in signalling pathways that are constantly activated/inhibited by the niche to robustly maintain distinct cellular states 31,32 . For this predictive modelling, we used RNA-seq data for QSCs genuine and QSCs primed from young and geriatric mice. The signalling molecules predicted to be active or inactive in young genuine or primed states are shown in Fig. 8a. Akt signalling (Akt1 and Akt2) was predicted to be inactive (or inhibited) in young QSCs genuine (Fig. 8a and Extended Data Fig. 8a). Akt is activated through phosphorylation 33 and, consistent with Akt inactivation, the Akt phosphatase PPP2CA (known to dephosphorylate Akt in myogenic cells 34 ) was predicted to be active in young QSCs genuine (Fig. 8a). In contrast, Akt signalling might be active in young QSCs primed (Fig. 8a). Higher levels of phosphorylated Akt (p-Akt) were found in young QSCs primed than QSCs genuine (Extended Data Fig. 8b). Consistent with Akt being inhibitory for the transcriptional activity of FoxO 35 , NicheHotSpotter-derived networks revealed that FoxOs target several differentially expressed TFs, which are directly inhibited by Akt1 in young QSCs genuine (Fig. 8a and Extended Data Fig. 8a). This prediction is consistent with the higher FoxO3a protein levels in young QSCs genuine than QSCs primed (Fig. 7d). Furthermore, NicheHotSpotter predicted that the signalling network centred around Akt is controlled upstream by several growth factors/receptors, including insulin (Ins)/insulin-growth-factor (Igf) and receptor (Igf1r; Fig. 8a and Extended Data Fig. 8a), suggesting that niche-derived Ins/ Igf-driven Akt signalling may inhibit FoxO in young QSCs primed . To test this prediction, we administered the Akt inhibitor wortmannin to young mice for 2 weeks, which resulted in reduced p-Akt and increased nuclear FoxO3a in QSCs primed (Extended Data Fig. 8b,c). The QSCs genuine and QSCs primed from young mice treated with wortmannin both showed increased engraftment capacity in vivo and clonogenic capacity in vitro ( Fig. 8b and Extended Data Fig. 9a). Furthermore, we observed enrichment of the QSCs genuine signature in CD34 L QSCs from the wortmannin-treated mice (Fig. 8c). Thus, Akt inhibition reinforces the genuine stem-cell state by activating FoxO and this benefits the QSCs primed population, which has reduced function.
NicheHotSpotter predicted the Igf-Igfr-Akt pathway to be more active in geriatric than young QSCs genuine (Fig. 8d and Extended Data   Fig. 9b), consistent with the stronger FoxO inhibition in geriatric satellite cells (Fig. 7d). The method also predicted that activation of the Igf and Ins receptors by extrinsic ligands would have negative consequences on young satellite cells related to Akt and FoxO ( Fig. 8a and Extended Data Fig. 8a). To test this prediction, we used transgenic mice that overexpress a locally secreted form of IGF1 (Tg-IGF1) in skeletal muscle 36 , and found decreased nuclear FoxO3a (Extended Data Fig. 9c), correlating with increased p-Akt, which was reflected in altered satellite-cell clonogenic potential (Extended Data Fig. 10a,b), thus partially phenocopying FoxO1,3a,4-deficient satellite cells. Consistent with this, CD34 H QSCs from the Tg-IGF1 mice showed partial enrichment of the QSC primed signature (Extended Data Fig. 10c). Thus, Akt-signalling inhibition and FoxO-nuclear relocation enhance the stemness traits of young QSCs primed and may constitute a promising strategy to rejuvenate geriatric QSCs genuine that have acquired primed-state traits.

Discussion
Despite studies showing quiescent muscle-stem-cell heterogeneity 5,7,8,37 , the demonstration of this heterogeneity in unperturbed skeletal muscle, the dynamics of heterogeneity throughout life and its requirement during tissue repair remain unclear. Through orthogonal approaches, including RNAScope, we demonstrate the existence of two stem-cell states in undisturbed muscle tissue. We further provide a high-resolution reconstruction of the molecular events that maintain functional muscle stem-cell heterogeneity in quiescence and that will determine the generation of diverse progeny following stress and how these properties change over time.
Stem-cell quiescence is progressively lost with ageing due to alterations affecting the intrinsic and niche factors 5,17,18 . Based on differences in the expression levels of CD34, we show that QSCs encompass a genuine state (QSC genuine ), with inferred stemness features and low metabolism, and a primed state (QSC primed ), which is more prone to myogenic differentiation and metabolic activity. In contrast to the concept of a general malfunctioning of aged stem cells, our study surprisingly reveals that the genuine stem-cell state is age resistant, given that old QSCs genuine perform as well as their young counterparts in vitro and during muscle regeneration. Notably, acquisition of terminal cell fate in yeast also seems to be determined quite early in life 38 . Nonetheless, in extreme old age, the genuine stem-cell state undergoes a steep functional decline. Molecular features of the QSC primed gene signature emerge in the geriatric, but not the old, QSC genuine state, indicating that the identity of the young QSC genuine state prevails in old age but becomes blurred in extreme old age. Although geriatric QSCs genuine maintain CD34 expression, they concomitantly upregulate genes involved in OxPhos metabolism. The acquisition of these inappropriate traits, specifically in geriatric satellite cells, provides a probable explanation for the previously described sharp decline in muscle regeneration in extreme old age 17,18 .
Our results show that FoxO TFs act as a sentinel of the genuine-stem-cell state (Extended Data Fig. 10d). Thus, the separation between the QSC genuine and QSC primed states is disrupted by the deletion of FoxOs. Consistent with this, FoxOs-deficient satellite cells largely phenocopied QSCs in the primed state, exemplified by the shared high expression of the master regulator Myog and myogenesis-related genes, and downregulation of classical stemness-and quiescence-related genes such as Notch3 and Calcr 14,[39][40][41] . We therefore propose that long-lived quiescent stem cells may rely on FoxOs to preserve diversity and stemness, thus extending FoxO functions beyond cellular differentiation 42,43 .
It is unclear why QSCs genuine are more responsive to FoxO signalling. The newly implemented computational method NicheHotSpotter predicted that niche-produced Igf and Akt signalling regulate the primed state by inhibiting FoxO activity. This was confirmed by the conversion of QSCs primed into QSCs genuine following Akt blockade, whereas IGF1 overexpression in the muscle niche causes Akt activation and FoxO inhibition, and deteriorates the genuine-stem-cell state. The enhanced activity of the Akt inhibitory pathway could therefore underly the reduced FoxO activity of geriatric QSCs genuine and provoke acquisition of primed features by QSCs genuine in this extreme old age.

Fig. 8 | A Markov-chain-based computational method predicts reinforcement and rejuvenation of the genuine state by igf-Akt pathway inhibition.
a, Summary of the steps involved in NicheHotSpotter for the prediction of niche-induced signalling hotspots (left). A simplified representation of the NicheHotSpotter-derived Akt-centric network that controls FoxO and other TFs (right; see Extended Data Fig. 9a for the full network). Predictions of active (compatibility score > 0.5) and inactive (compatibility score < 0.5) signalling hotspots for young QSCs genuine and QSCs primed . b, Equal numbers of QSCs genuine and QSCs primed isolated from young mice (n = 3 transplanted muscles from four mice; 4 months old) previously treated with wortmannin or vehicle for 2 weeks (see Methods for details) were stained with Dil and transplanted into pre-injured TA muscle of the recipient immunodeficient mice or 5 d. The levels of Dil + cells (right) and representative immunostaining images are shown (left; 12 transplanted muscles). Scale bars, 50 μm (main image), 10 μm (insets). c, GSEA showing significant (false-detection-rate q-value < 0.05) enrichment of the QSC primed signature in CD34 L (QSC primed ) satellite cells isolated from WT mice treated with vehicle (CTRL; n = 5 mice; 4 months old) compared with CD34 L satellite cells isolated from WT mice treated with wortmannin (+Wort; n = 4 mice; 4 months old). d, Predictions of signalling hotspots for young QSCs genuine in comparison to geriatric QSCs genuine as in a. Data are the mean ± s.e.m.; a two-tailed unpaired t-test with Welch's correction was used; NS, not significant. and technical standpoints, despite reports suggesting the nonequivalence of muscle stem-cell populations [5][6][7][8][9]44 , it is worth mentioning that no previous study has achieved the separation of distinct quiescent states in the absence of transgenesis or labelling, probably due to a lack of cell-surface markers. Our study shows that CD34 reports stemness in quiescence.
In conclusion, our findings set forth the idea that quiescent stem cells in low-turnover tissues are already pre-programmed to adopt divergent future fates in response to regenerative demands. This molecular partitioning becomes less demarcated in geriatric age, with detrimental functional consequences. We anticipate that understanding this molecular diversity and how it changes throughout life will be critical to harnessing the potential of stem cells for regenerative medicine in sarcopenia.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41556-020-00593-7.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply 2020 45 , β-actin-GFP, LC3-GFP, Pax7-nGFP 46 , Pax7 cre-ER ; ROSA26 YFP and Pax7 cre-ER ; ROSA26 Tomato mice, and the offspring from intercrossing FoxO3a fl/fl or FoxO1,3a,4 fl/fl mice with the Pax7 Cre , Pax7 cre-ER , ROSA26 YFP or Tg-IGF1 (MLC-IGF1) 46 lines were used at the indicated ages. Mice with the FoxO3a or FoxO1,3a,4 genes deleted in their satellite cells, inducible or constitutive, were generated by breeding the previously described 47 FoxO3a fl/fl or FoxO1,3a,4 fl/fl mice with the Pax7 cre or Pax7 cre-ER lines (provided by C. Keller and M. Capecchi). FoxO3a ΔPax7ER and FoxO1,3a,4 ΔPax7ER mice were intercrossed with ROSA26 YFP mice to generate the FoxO3a ΔPax7ER ; ROSA26 YFP mice. Fox Chase SCID CB17/Icr-Prkdc scid /IcrIcoCrl (strain code, 236) mice were from Charles River Laboratories. The mouse experiments were not randomized or blinded. No statistical methods were used to predetermine the sample size. Cre activity was induced by intraperitoneal injection (one injection per day for 4 d) with 2 mg per 30 g bodyweight tamoxifen (Sigma; 20 mg ml −1 in corn oil). Every animal procedure involved in this work-using sex-, age-and weight-matched littermate animals-was supervised by the Ethical Committee of Animal Experimentation of the PRBB (CEEA-PRBB) and previously authorized by the corresponding Catalan committee, the Section for the Domestic Animal Protection, General Direction of Environmental and Nature Politics, Department of Territory and Sustainability.

Animals. Male C57BL/6J (WT), ROSA26-tdTomato
In vivo treatments. For the BrdU treatment, BrdU (Sigma; 1 mg ml −1 in 2.5% sucrose) was provided to the mice in drinking water every 3 d for 1 month. For the wortmannin (Penicillium fumiculosum) treatment, the mice were injected intraperitoneally (three injections per week for 2 weeks) with 0.03 mg per 30 g body weight (Sigma; 1 mg ml −1 in dimethylsulfoxide), and dimethylsulfoxide (10% in PBS) in control animals.

Muscle regeneration.
Mice were anaesthetized with ketamine-xylazine (80 and 10 mg kg −1 , respectively; intraperitoneally) or isoflurane. Muscle regeneration was induced by injection with CTX (Latoxan; 10 µM) in the TA as described 48 . The mice were killed at the indicated times post injury, and muscles were dissected, frozen in liquid-nitrogen-cooled isopentane and stored at −80 °C. For GFP immunostaining, muscles were prefixed 2 h in 2% paraformaldehyde (PFA) at 4 °C, embedded in 15% sucrose overnight at 4 °C, and frozen in liquid-nitrogen-cooled isopentane.
Single myofibre isolation. The extensor digitorum longus muscles were dissected and digested in collagenase type I (Sigma-Aldrich) solution for 1.5 h as described 49 , manually rocked before individual fibres were harvested with a fire-polished Pasteur pipet, pre-flushed with 0.5% horse serum and fixed for immunofluorescence.
Satellite-cell transplantation. FACS-isolated satellite cells were collected, resuspended in GM medium, plated in six-well plates coated with collagen (3 mg ml −1 ; 40-50×10 3 cells per well) overnight as described previously 18,50 and transferred to a six-well plate coated with RetroNectin (Takara); lentiviral particles were added and the plate was centrifuged at 50g for 5 min and left overnight. The cells were recovered the following day, centrifuged at 900g for 15 min, resuspended (100,000 cells per 100 µl) and 10,000 cells were injected with a Hamilton syringe into the recipient mice (10 µl per muscle).
In the siRNA experiments, satellite cells were transfected with siRNA for 3 h on ice and transplanted as indicated for GFP labelling. For transfections, a Lipofectamine RNAi max kit (Invitrogen) was used with 20 nM of non-targeting siRNA control pool or SMART mouse siMyog pool oligomers (ThermoScientific Dharmacon).
For short-time engraftments, to avoid the culture effect on freshly isolated satellite cells, the cells were stained with 1/200 Vybrant Dil cell-labelling in FACS buffer (Invitrogen). When using fewer than 3,000 satellite cells, 20,000 C2C12 myoblasts (Supplementary Table 3) were added. The TA muscles of recipient CB17/Icr-Prkdc scid /IcrIcoCrl mice were injured by freeze-crush 1 d before transplantation. For the second-injury experiments, we used Pax7-nGFP mice. Different CD34-sorted subpopulations were grafted into pre-injured TAs; 21 d later we induced a second injury and muscle was recovered after 21 d. The muscles were processed for histology 3 (satellite-cell expansion) or 21 d (muscle regeneration) after satellite-cell transplantations. The results were expressed as the number of labelled fibres per muscle section or number of satellite cells per mm 2 of the damaged area.
Clonogenic assays. For clonal analysis, satellite cells were plated at low density on collagen-coated dishes (50-75 cells per well in 96-well plates) and cultured for 3 d in proliferation medium: DMEM GlutaMAX containing 20% FBS, 10% horse serum, 1% penicillin-streptomycin, 1% glutamine, 1% HEPES, 1% sodium pyruvate and 0.005 µg ml −1 bFGF. The cells were transfected with siRNA for 3 h and plated for the transplantation experiments. The cells were finally fixed for 10 min in 4% PFA and washed with PBS. The cells in each well (4-8 wells per sample) were quantified. A colony was defined as ≥3 cells together.

Reserve cells.
Satellite cells were plated on collagen-coated dishes at a density of 5-8 × 10 3 cells cm −2 , cultured in GM for 3 d and then: (1) in experiments with CD34 H , CD34 L and CD34 − QSCs, the medium was switched to differentiation medium (5% horse serum) for an extra 4-5 d; or (2) in experiments with WT and FoxO1,3a,4 ΔPax7ER QSCs, the cells were plated at 12-13 × 10 3 cells cm −2 , cultured for 24 h and then switched to differentiation medium for an extra 4-5 d. The siRNA transfection of the sorted satellite cells was allowed to proceed for 3 h. PAX7 and KI67 immunostaining was performed and the PAX7 + KI67 − cells in five random fields per well were counted.
RT-qPCR: RNA extraction, cDNA synthesis and PCR. Total RNA from FACS-isolated satellite cells was obtained using an RNeasy micro kit (Qiagen) and analysed by quantitative PCR with reverse transcription. Digestion with DNase (Qiagen) was performed. Complementary DNA was synthesized from 100 ng total RNA using a First-strand cDNA synthesis kit (Roche). Real-time PCR was performed on a LightCycler 480 system using the LightCycler 480 SYBR Green I master reaction mix (Roche Diagnostics) and specific primers (Supplementary Table 3 Table 3) or 10 μg rabbit IgG (Supplementary Table 3). Magna ChIP protein A + G magnetic beads (20 μl; Millipore) were added and incubated for 1 h at 4 °C. The beads were washed with 1 ml of different buffers: low-salt immune complex wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.1 and 150 mM NaCl), high-salt immune complex wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8.1 and 500 mM NaCl), LiCl immune complex wash buffer (0.25 M LiCl, 1% NP-40, 1% deoxycholate, 1 mM EDTA and 10 mM Tris-HCl, pH 8.1) and TE (1 mM EDTA and 10 mM Tris-HCl, pH 8.0). To elute immunocomplexes, the beads were incubated twice in elution buffer (1% SDS and 100 mM NaHCO 3 ) for 15 min at room temperature. The crosslinking was reverted by incubating the DNA overnight at 65 °C. The DNA was phenol-chloroform extracted, ethanol precipitated and resuspended in 80 μl H 2 O for analysis. The bound fraction and input were analysed by quantitative PCR using specific primers.
Satellite cells were transduced with LV-GFP or LV-FoxO3a (1:500 from a titre of 3-10 × 10 10 virus particles ml −1 ). Doxycycline was added the next day (1:1,000; to a final concentration of 2 µg ml −1 ) for 48 h. The cells were then transfected as described earlier. Following transfection, the cells were cultured in GM for 24 h or 24 h in GM and 48 h in differentiation medium; the cells were harvested, and the luciferase activity was determined using a luciferase assay kit (Promega) and normalized to the β-galactosidase activity. For each experimental group, a minimum of four independent transfections (in triplicate) were performed.
Western blotting. Satellite-cell lysates and western blotting were performed as described previously 52 . The antibodies used are listed in Supplementary Table 3.  Table 3) after blocking with a high protein-containing PBS solution (Vector Laboratories) for 1 h at room temperature. Subsequently, the slides were washed with PBS and incubated with the appropriate secondary antibodies and labelling dyes. For immunofluorescence, the secondary antibodies were coupled (Supplementary Table 3) and nuclei were stained using DAPI (Invitrogen). Tissue sections were mounted with Fluoromount G (SouthernBiotech).
Immunofluorescence. Immunofluorescence in isolated satellite cells was performed in glass slides (Thermo Scientific), whereas it was performed in suspension in single myofibres as described 53 (Supplementary Table 3).
RNAscope multiplex fluorescence in situ hybridization. In situ hybridization was performed using a RNAscope multiplex fluorescent reagent kit v2 (Advanced Cell Diagnostics) following the manufacturer's protocol with some modifications. The TA muscle from Pax7 cre-ER ; ROSA26 tdTomato mice was dissected and fixed in 2% PFA for 2 h at 4 °C, frozen and processed into 10-μm longitudinal sections. The slices were digested with Protease Plus. Probe hybridization was performed by incubation of the mRNA target probes for 2 h at 40 °C. The slices were hybridized with two probe sets of dual probes using Cd34-C3 as the common probe in each set and Myog-C2 or Notch3-C2 as the companion probes. The Cd34-C3 probe was conjugated to the Opal 690 fluorophore, whereas Myog-C2 and Notch3-C2 were conjugated to the Opal 570 fluorophore (Akoya Biosystems). For satellite-cell immunofluorescence, the sections were blocked in 3% BSA for 1 h at room temperature before an overnight incubation with anti-RFP primary antibody (Rockland, 600-401-379; 1:200) at 4 °C. Anti-rabbit Alexa Fluor 488 secondary antibody (A11008, Life Technologies; 1:500) and DAPI (Invitrogen) were added for 1 h at room temperature. After washing, the sections were mounted with ProLong gold antifade mountant (Thermo Fisher). Confocal images were taken using a Zeiss LSM-700 confocal system with a Plan-Apochromat ×40/1.4 numerical aperture oil objective.
Videomicroscopy. Freshly sorted satellite cells (5,000-10,000) were plated on eight-well glass slides (Lab-Tek) coated with collagen and recorded for 96 h. A cell was considered to have divided when cytokinesis was completed. Using the same settings, the cells were labelled with the red fluorescent membrane dye PKH26 (Sigma).
Digital image processing and automated counting. Image processing and quantification was carried out using Fiji 54 Life-Line (2017 version) using macro language to create these pipelines: (1) for the myofibre size, individual fibres were automatically segmented using a machine-learning approach (trainable weka segmentation) complemented by a manual-correction macro; the cross-sectional area was measured on the resulting mask; and (2) transplanted cells labelled with GFP or Dil were automatically segmented and the fluorescence intensity of the selected proteins or PKH26 dye were quantified for each cell; the average of the relative fluorescence was expressed as the mean or sum of the fluorescence intensities. The macros and their detailed explanation are available at https:// github.com/MolecularImagingPlatformIBMB. Sample preparation for RNA-seq and ATAC-seq. Total RNA was isolated from FACS-isolated satellite cells using a RNeasy micro kit (Qiagen) for regular or low-input RNA-seq. For low-cell numbers (100-400), the samples were collected directly into either 96-conical-well plates or 200-µl conical Eppendorf PCR tubes containing 10 µl lysis buffer (SMART-Seq v4 ultra low input RNA kit for Sequencing), spun down and frozen at −80 °C before processing.
The RNA-seq of CD34 H , CD34 L satellite cells from young WT mice; or CD34 H , CD34 L and CD34 − satellite cells from young Pax7 cre-ER ; ROSA26 YFP mice or from FoxO1,3a,4 WT and FoxO1,3a,4 ∆Pax7ER mice was completed on a HiSeq2000 instrument (Illumina) using cDNA libraries from poly A+ purified mRNA and sequenced using 50 bp single-end reads.
For RNA-seq of CD34 H , CD34 L and CD34 − satellite cells, 0.2-0.4 ng total RNA (or 400 cells) were used to amplify the cDNA with a SMART-Seq v4 ultra low input RNA kit (Clontech-Takara). The amplified cDNA (1 ng) was used to generate barcoded libraries using a Nextera XT DNA-library preparation kit (Illumina). The cDNA was fragmented, adaptors were added in a single reaction, amplified and cleaned up. The size of the libraries was checked using an Agilent 2100 Bioanalyzer high-sensitivity DNA chip and the concentration determined using a Qubit fluorometer (Thermo Fisher Scientific). The libraries were sequenced on a HiSeq2500 (Illumina) to generate 61-base single-end reads. Single-cell RNA-seq of mononucleated cells from young muscle was performed as described 12 . For ATAC-seq, freshly sorted satellite cells were processed as described 55 .

Statistical analysis of RNA-seq.
1. For all samples (except those in point 2), the following steps were sequentially performed to generate matrices of read counts from the demultiplexed, single-end read fastq files. (1) Adaptor sequences were trimmed using TrimGalore.
(2) The trimmed reads were mapped to the mm10 Mus musculus reference genome using TopHat v2.0.9 (TopHat calls the mapper Bowtie2 v2.1.0). The following options were selected: -library-type fr-firststrand for stranded libraries (dUTP protocol) and -library-type fr-unstranded for unstranded libraries (low input RNA-seq). (3) Reads with an mapping quality < 3 were filtered out. (4) Data-quality check: all samples were initially checked for the raw number of sequenced reads, duplication levels and number of final mapped reads. Duplicate reads were not removed for downstream analysis. (5) Read counts were performed using HTSeq v0.7.2 . (6) A matrix of counts was generated for each experiment, with one row for every feature (genes) and one column for every sample. This data structure was used as input for the subsequent data analysis in R. The R pipeline was used for statistical analysis. Each experiment was analysed using the R package Deseq2 1.8.2 (ref. 56 ). This package assumes a negative binomial distribution to generate lists of differentially expressed genes. A matrix containing all of the relevant experiment information (variables, potential batches and so on) was constructed for each experiment. A model was defined for each experiment based on the experimental design to accept or reject the alternative hypothesis. Before running the statistical tests, all experimental data were subjected to an exploratory analysis. To this end, variance across all median values within samples was stabilized using a regularized log transformation (rlog; returns log 2 values normalized to the library size and in a way that minimizes variation in genes with a small/large number of counts). The rlog-transformed data were the subject to a PCA and/or clustering to visualize the sample-to-sample distances. The samples were also visualized before/after the removal of known batch effects through the application of the Limma's removeBatchEffect function and PCA/clustering visualization. In all cases, the experiments were subject to surrogate-variable analysis to avoid the unwanted effects of potentially unknown batch effects. If surrogate variables were identified, these were added to the deseq2 model and data were once again visualized through PCA/ clustering analysis before/after surrogate-variable removal. If the addition of the surrogate variable into the model resulted in better sample clustering/ separation, the surrogate variables were kept in the model for statistical testing. Lists of differentially expressed genes were generated through the results function in Deseq2. Differentially expressed genes were corrected for multiple testing (an adjusted P value was generated using the Benjamini-Hochberg correction). Gene counts for the differentially expressed genes were visualized using the pheatmap v1.0.12 function in R. 2. For satellite cells isolated from young WT and Tg-IGF mice as well as young WT mice treated with dimethylsulfoxide (control) or wortmannin for 2 weeks, mRNA expression profiling was performed at the NIAMS Genome Core Facility at NIH. Illumina Hiseq 3000 runs were demultiplexed and converted to FastQ format using bcl2fastq2 (Illumina). FastQ reads were mapped to mm10 using TopHat 2.1.

TF binding analysis.
Comprehensive GSEA for transcriptional machinery was carried out using the Enrichr 58 library of published position weight matrices from the UCSC genome browser. Z-score was used to represent the TF lists, defined as the likelihood that the number of transcription-factor-binding-site nucleotides detected for the included target genes/sequences is significant compared with the number of transcription-factor-binding-site nucleotides detected for the background set. The Z-score was expressed in units of magnitude of the standard deviation. Alternatively, TF targets were interrogated in GSEA 57 using the TFT library included in MSigDB database v6.2. (https://www.gsea-msigdb.org/gsea/ msigdb/index.jsp). The Enrichr libraries can be found at https://amp.pharm.mssm. edu/Enrichr/#stats.

ATAC-seq analysis.
Libraries were sequenced on a HiSeq2500 system (Illumina) to generate paired-end reads of 50 bases. A published ATAC-seq data analysis method 59 was followed to generate ATAC-seq peaks from fastq files. Briefly, redundant paired-end reads were removed using fastquniq 60 . Paired-end reads of 50 bases were aligned to the mouse genome build mm10 using Bowtie 1.1.1 following the guidelines presented previously 58 . Customized Python scripts were used to calculate the fragment length of each pair of uniquely mapped paired-end reads and identify reads from fragments of fewer than 175 bases, which were then used for peak calling with MACS 1.4.2 (ref. 61 ). HOMER v4.8 (ref. 62 ) and BEDtools 2.27 (ref. 63 ) were used to generate raw and peak bigWig files. The read density profile around the TSS was made using ngs.plot 64 . The ATAC-seq data have been deposited in the GEO under the accession code GSE155642.

Computational analysis by NicheHotSpotter.
NicheHotSpotter is based on a method that was previously developed to identify hotspots in signalling pathways from single-cell RNA-seq data 31,32 . Here we have extended the approach to be applicable to bulk RNA-seq by considering that steady-state protein levels are more likely to have a high correlation with the respective steady-state mRNA levels. Importantly, we focused on the effect of signalling molecules via TFs that are not differentially expressed but controlled by post-translational modifications to exert their regulatory effect on downstream differentially expressed TFs. A detailed description of the method is available elsewhere 32 . Briefly, the method involves two-main steps: (1) Markov-chain model-based identification of the signalling molecules that exhibit the high probability of signal flow through them and (2) topological characterization-based assessment of the compatibility with the differential expression status of the downstream TFs to infer the activity status of such molecules. The differentially expressed TFs were based on the analyses that defined the signatures of CD34 cells. The molecules that exhibit high signalling probability as well as high (or low) compatibility were identified as active (or inactive) signalling hotspots. To build the signalling networks around the predicted hotspots, we traced all of the shortest paths from the Niche node to the predicted hotspot (Akt1) and from the hotspot to the downstream differentially expressed TFs. This integrated network (Extended Data Figs. 8 and 9) of signalling and TF molecules serves as a model to understand the potential mechanisms of how niche cues could regulate downstream TFs to stably maintain the cellular phenotype.

Statistics and reproducibility.
No specific blinding method was used for the mouse experiments, but the mice for each sample group were selected randomly. The sample size (n) of each experimental group is described in each corresponding figure caption, and all of the experiments were conducted with at least three biological replicates, unless otherwise indicated. For the majority of the experiments, young and old mice were processed in an alternating manner rather than in two large groups to minimize the group effect. The GraphPad Prism software was used for all statistical analyses except for sequencing-data analysis. Quantitative data displayed as histograms are expressed as the mean ± s.e.m.
(represented as error bars). The results from each group were averaged and used to calculate the descriptive statistics. A two-tailed unpaired t-test with Welch's correction was used for pairwise comparisons between groups unless otherwise indicated. Statistical significance was set at P < 0.05.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The RNA-seq and ATAC-seq data that support the findings of this study have been deposited in GEO under the accession code GSE155642. All other data supporting the findings of this study are available from the corresponding author on reasonable request. Source data are provided with this paper.

Code availability
Codes for RNA-seq and ATAC-seq analysis (GSE155642) are available from the corresponding author on request. NicheHotSpotter is available at https://gitlab.com/srikanth.ravichandran/signalingfactorscd34positive. Automated counting macros for Fiji are available at https://github.com/ MolecularImagingPlatformIBMB. Fig. 1 | CD34 stratifies quiescent stem-cell states in satellite-cell reporter mice. a, Heatmap generated by ImaGEO algorithm using the indicated data sets (from gene expression omnibus, GEO) and which defined the "Quiescence Core" list of genes (see Supplementary Table 1    box limits indicate the 25th and 75th percentiles. b, QSCs isolated from young (2-3 mo, n=10 transplanted muscles from 6 mice), old (22-24 mo, n=7 transplanted muscles from 3 mice) and geriatric (>28 mo, n=7 transplanted muscles from 8 mice) WT mice were transduced with GFP-expressing lentivirus and transplanted for 21 days into pre-injured TA muscles of recipient immunodeficient mice. Equal number of freshly-isolated cells were used. Representative images and quantification of GFP + fibres are shown. Scale bar, 50 μm. c, Representative images and frequency distribution analysis of positive embryonic myosin heavy chain (eMHC) fibre size in cryosections of regenerating TA muscles from young (2-3 mo, n=3 mice), old (22-24 mo, n=3 mice) and geriatric mice (>28 mo, n=3 mice), one week after cardiotoxin (CTX)-induced injury. Scale bar, 50 μm. For the frequency distribution representation. 500 fibres per muscle were measured. d, Percentage of QSCs genuine and QSCs primed in young (2-3 mo, n=18 mice), old (22-24 mo, n=18 mice) and geriatric (>28 mo, n=17 mice) WT mice. Percentages were set at 15% for each population in one young WT mice from each FACS experiment. e, Enrichment map of gene sets enriched in QSCs genuine from geriatric vs young WT mice (FDR q value ≤ 0.05). Node size is proportional to the number of genes identified in each gene set (minimum 10 genes/gene set). Grey edges indicate gene overlap. Clusters were automatically annotated using Autoannotate app in Cytoscape.         Extended Data Fig. 8 | NicheHotSpotter predicts niche-derived ins/igf-driven Akt signalling to inhibit FoxO in the young primed stem-cell state. a, Complete predicted signalling network around Akt1 for young QSCs genuine in comparison to QSCs primed . Akt1 and FoxO TFs are shown as orange and purple coloured nodes, the green and red triangles are upregulated and downregulated downstream TFs respectively. All other grey nodes are intermediate molecules that are expressed. Green and red edges indicate activation and inhibition interactions, respectively. The network depicts potential flow of external signals via the predicted signalling hotspots to the differentially expressed TFs. b, Quantification of p-Akt signal in freshly-isolated QSC genuine and QSC primed cells from young mice (2-3 mo, n=1 mice, 35 total cells). Representative immunofluorescence images and quantification of p-Akt signal of freshly-isolated QSC primed cells from young mice 2-weeks after treatment with vehicle (control) or wortmannin (2-3 mo, n=3, 135 total cells). Scale bar, 5 μm. c, Quantification of FoxO3a nuclear signal in freshly-isolated QSC genuine and QSC primed cells from young mice (2-3 mo, n=3, 232 total cells). Representative immunofluorescence images and quantification of FoxO3a nuclear signal of freshly isolated QSC primed cells from young mice 2-weeks after treatment with vehicle (control) or wortmannin (2-3 mo, n=1, 63 total cells). Scale bar, 5 μm. Means ± s.e.m.; two-tailed unpaired t-test with Welch's correction unless otherwise indicated.

Extended Data
Extended Data Fig. 9 | NicheHotSpotter predicts the igf-igfr-Akt pathway to be more active in geriatric than in young genuine satellite cells. a, Clonogenic assay of freshly-sorted QSCs genuine and QSCs primed from young mice 2-weeks after treatment with vehicle (control) (n=20 wells from 4 mice, 5 replicates/mouse) or wortmannin (n=18 wells from 4 mice, 4-5 replicates/mouse). Number of colonies and cells were quantified. b, Complete predicted signalling network around Akt1 for young QSCs genuine in comparison to geriatric QSCs genuine . The network colours representation is the same as in Extended Data Fig. 8a. c, Representative immunofluorescence images and quantification of FoxO3a on QSCs freshly isolated from young WT and MLC-IGF1 (Tg-IGF1) mice (2-3 mo, n=88 cells from 2 WT mice and n=135 cells from 4 MLC-IGF1 mice. Scale bar, 5 μm. Means ± s.e.m.; two-tailed unpaired t-test with Welch's correction unless otherwise indicated.