Aging alters the epigenetic asymmetry of HSC division

1Hematopoietic homeostasis depends on the ability of hematopoietic stem cells (HSCs) to balance symmetric and asymmetric divisions over an organism’s lifespan. Asymmetric divisions allow one daughter cell to differentiate while the other retains its stem cell potential. In contrast, symmetric divisions result in daughter cells that adopt equivalent fates. The asymmetric or symmetric distribution of cellular components to daughter cells during mitosis is thought to determine their fate. This concept has been proposed as a way of controlling the size of the HSC pool. However, the mechanisms that control the mode and outcome of HSC divisions with respect to the potential and ultimately function of daughter cells remain incompletely understood.

2While the number of phenotypic HSCs increases upon aging, their regenerative potential decreases, both in mice and in humans. Aged HSCs differentiate preferentially into myeloid cells and less into cells of the lymphoid lineage. Aging of HSCs is caused by intrinsic changes in HSCs and also by extrinsic factors from the niche. Epigenetic modifications and altered gene expression profiles (e.g., high expression of myeloid genes), together with loss of the polar distribution of tubulin and cell division control protein 42 (Cdc42) in the cytoplasm and loss of the polarity of the acetylated form of the epigenetic marker H4K16 (H4K16ac) in the nucleus, are hallmarks of intrinsic HSC aging. Cdc42 is a small RhoGTPase that cycles between an active guanosine triphosphate (GTP)-bound state and an inactive guanosine diphosphate (GDP)-bound state. A central, HSC-intrinsic molecular mechanism that causes aging of HSCs and that results in the above-mentioned phenotypic and functional impairments of aged HSCs is an increase in the activity of Cdc42 in aged stem cells. Consequently, treatment of aged HSCs ex vivo with a Cdc42 activity specific inhibitor (CASIN), which reduces the age-associated increase in Cdc42 activity, increased polarity in aged HSCs and rejuvenated their function in vivo. The increase in Cdc42 activity in aged HSCs is caused by an intrinsic increase in the level of Wnt5a in HSCs. Treatment of young HSCs with Wnt5a induced elevated activity of Cdc42 and a loss of cell polarity and premature aging of HSCs.

3We hypothesized here that HSC polarity determines the mode (symmetry/asymmetry) of HSC divisions and therefore that there is a difference in the symmetry/asymmetry of division upon changes in the activity level of Cdc42 in HSCs. We thus investigated whether the targeted alteration of Cdc42 activity by either CASIN (to decrease Cdc42-GTP levels in aged HSCs) or Wnt5a treatment (to increase Cdc42-GTP levels in young HSCs), which translates into polarity changes, influences the balance of asymmetric or symmetric HSC divisions (mode of division) and therefore drives functional differences among daughter cells (output of division) upon aging and rejuvenation (S1A Fig). By comparing the mode of HSC division in vitro and the output of division in vivo, we demonstrate that young and aged CASIN-treated HSCs divide asymmetrically while aged and young Wnt5a-treated HSCs undergo symmetric divisions. We conclude that, by modulating Cdc42 activity and consequently cell polarity, it is possible to adjust the mode and the epigenetic outcome of HSC divisions.

4We first microscopically tracked the kinetics of single HSC doubling in vitro (Fig 1A).

5There was no difference in the kinetics of first and second divisions between young, aged, aged treated with CASIN, and young treated with Wnt5a HSCs (Fig 1B) nor in the relative distribution of mitotic cells to metaphase, anaphase, or telophase (S1B Fig). Next, 3D confocal immunofluorescence (IF) was used to quantify Cdc42 and H4K16ac allocation to daughter cells during mitosis. Young HSCs were found to segregate Cdc42 and H4K16ac asymmetrically into daughter cells. Aged HSCs displayed a symmetric distribution of both Cdc42 and H4K16ac in daughter cells, and inhibition of Cdc42 activity in aged HSCs increased the frequency of asymmetric divisions among aged HSCs. Consistently, in the presence of Wnt5a, young HSCs presented a shift towards a symmetric allocation (Fig 1C and 1D and Fig 2A and 2B; S1C–S1J Fig; S1–S8 Movies; S1 Table; S2 Fig).

6To note, polarity was not affected by cell culture conditions (S1K Fig), and Cdc42 protein was observed asymmetrically or symmetrically already during metaphase (S1L Fig). H4K16ac levels were stably detected throughout mitosis (S1M Fig), and the difference in allocation was quantified at telophase. When setting the threshold to a 75% difference in Cdc42 amount between daughter cell pairs (which translates into at ratio of 1 over 4, see also cartoon scheme in Fig 1D), 80% of young HSCs divided asymmetrically (Fig 2C). The frequency of asymmetric division was significantly reduced upon aging (S1 Table and Fig 2D). Consistent with polarity/apolarity before division, aged CASIN-treated HSCs divided more frequently asymmetrically, while Wnt5a treatment of young HSCs increased the frequency of symmetric divisions (Fig 2E and 2F).

7Because it is not yet possible to image in living cells Cdc42 and H4K16ac allocation to nascent daughter cells, we used mathematical association analyses to investigate whether there is a link between polarity in mother HSCs and the mode of HSC division. The probability that a polar stem cell will undergo an asymmetric division was termed da (with probability for symmetric division defined as 1 − da). Similarly, ds represented the probability that an apolar stem cell divides symmetrically (with 1 − ds for asymmetric) (Fig 3A).

8By applying a maximum likelihood estimation to the previously published observed frequencies of polar versus apolar HSCs in young, aged, young Wnt5a-treated, and aged CASIN-treated cell populations, we found that the probabilities da ≈ 0.96 and ds ≈ 0.82 best explained the corresponding modes of division (Fig 3A). Modelling thus indicates that there is a 96% chance that a polar cell will divide asymmetrically, and apolar cells will divide with an 82% chance symmetrically.

9Encouraged by this strong association between HSC polarity/apolarity and asymmetric/symmetric mode of division, we further designed a mechanistic mathematical model to investigate the causative link to the fates of the daughter cells (see S1 Text and S3 Fig). To mimic the experimental observations, we assumed that Cdc42 is either distributed uniformly (apolar) or peaked (polar) along a circular, abstracted cell shape. We directly coupled the active proportion of intracellular Cdc42 (Cdc42-GTP) to the shape of this distribution, thereby functionally connecting the observed increase in Cdc42 activity with the loss of polarity during aging. The mathematical approach suggested that a switch-like construct could determine daughter cell fate, in which different fates manifest based on the intracellular protein concentration immediately after cell division. We also implemented a transcriptional autoregulative feedback for total Cdc42 concentration, thereby establishing bistability (Fig 3B). For certain parameter choices, such a system yielded 2 characteristic states: either Cdc42 was highly expressed due to self-activation (e.g., indicative of HSCs) or Cdc42 was lowly expressed (e.g., indicative of the progenitor state) (Fig 3C and 3D). The choice of either the HSC or the progenitor state was determined by Cdc42 concentration right after cell division. Distinct concentrations between daughter cells due to an asymmetric mode of division resulted also in distinct states (Fig 3C and 3D). Because Cdc42 activity, and thereby HSC polarity, is altered during aging, the model predicted that asymmetric outcomes of division are more likely for young HSCs, while a higher incidence of symmetric divisions in aged HSCs will preferentially promote self-renewal, predicting a progressive expansion of the stem cell pool over time (Fig 3E). Indeed, an increase in the number of HSCs in aged mice and humans has been verified by several groups.

10To experimentally verify the predictions of the mathematical models and to further determine the function of daughter cells, we performed single daughter cell transplants using Rag2−/−γc−/−KitW/Wv recipient mice. After the first division in vitro, we injected each single daughter cell into an individual recipient mouse (Fig 4A) and assessed its contribution to peripheral blood (PB) every 4 to 8 weeks up to 24 weeks after transplantation.

11A daughter cell that was able to contribute to B cells, T cells, and myeloid cells 24 weeks after transplantation with an overall engraftment of more than 0.1% qualified as a stem cell. When a contribution at 24 weeks was not detected in all cell lineages tested or the overall contribution of all donor-derived cells was below 0.1%, the initially transplanted cell was scored as a progenitor cell. Based on the scores of the outcomes of the daughter cell transplants, the initial HSC division was then defined as symmetric (two daughter stem cells or two daughter progenitor cells) or asymmetric (one daughter stem cell and one daughter progenitor cell) (Fig 4B; S4A–S4D Fig; S5A Fig). Young HSCs divided primarily asymmetrically, while aged HSCs showed mainly symmetric division outcomes (Fig 4C; S2 and S3 Tables; S5A Fig). Treating aged HSCs with CASIN significantly increased the frequency of asymmetric divisions of aged HSCs; in contrast, treating young HSCs with Wnt5a (Cdc42 inducer) increased the frequency of symmetric divisions (Fig 4C; S2 and S3 Tables; S5A Fig). The highest overall contribution to PB was detected in mice transplanted with young daughter stem cells at both early and late time points after transplantation (Fig 4D and 4E; S5A Fig). Young Wnt5a-treated daughter stem cells presented with a decrease in engraftment only at early time points (4 weeks), while there was no difference in engraftment between young and young Wnt5a-treated daughter stem cells at the end point (24 weeks) (Fig 4D and 4E). Aged daughter stem cells from CASIN-treated mother HSCs showed an about 2-fold increase in overall engraftment compared to daughter stem cells from aged untreated HSCs (Fig 4E; S5A Fig). Young daughter stem cells presented with the highest level of contribution to B cells, while aged and young Wnt5a-treated cells displayed a significantly reduced B cell output (Fig 5A and 5B; S5A Fig).

12Daughter stem cells from aged CASIN-treated HSCs differentiated significantly more than the other experimental samples into T cells (Fig 5A and 5B; S5A Fig). As for daughter progenitor cells, no difference was detected among samples, besides the observation that young and aged CASIN-treated sample sets presented with a trend in higher reconstitution at 24 weeks after transplant (S5B–S5D Fig). Our data further demonstrate that the initial population of mother HSCs the experiments were initiated with contained, on average, 77.5% of cells with stem cell function (based on their ability to give rise to at least one daughter stem cell), and there was no difference between the groups of young and aged HSCs and treated HSCs (S3 Table and S6 Fig).

13Frequent symmetric divisions of aged HSCs might result in a local accumulation of HSCs in bone marrow. We set up experiments to test this hypothesis and stained whole-mount preparations of the long bones of young and aged mice (S7A Fig). Indeed, in aged animals, HSCs often were found in clusters of 2 to 3 stem cells, while HSCs in young animals were almost always found as solitary cells (Fig 6A and 6B).

14As a consequence, the distance of HSCs to their nearest neighbor HSC was significantly decreased in aged bones, and around 30% of all HSCs were found less than 20 μm (proximity) from another HSC (Fig 6C and 6D and S7B and S7C Fig). This translated into 34% of all aged HSCs being in clusters compared to less than 2% for young HSCs (Fig 6E and S7D Fig). We next investigated whether HSC clustering was intrinsic to aged HSCs. To this end, we transplanted 1,000 young and aged HSCs from AcRFP mice (that ubiquitously and constitutively express the red fluorescent protein [RFP]) into young Rag2−/−γc−/−KitW/Wv recipient mice and imaged the bone marrow 6 weeks after transplant. We observed a higher frequency of HSC clusters (58% of all HSCs) in the bone marrow of recipients of aged HSCs compared to recipients of young HSCs (11% of all HSCs) (Fig 6F and 6G and 6J). Aged HSCs were also closer to each other and more frequently located in proximity of another HSC when compared to the distribution of young HSCs (Fig 6H and 6I). The frequency of clusters, compared to non-transplanted controls, increased for both aged and young donor HSCs (Fig 6J and S7E Fig), suggesting a contribution of the transplantation to aging of HSCs. In summary, clusters of HSCs in bone marrow are intrinsic to HSCs, are primarily formed by aged HSCs, and are consistent with an elevated frequency of symmetric divisions of aged HSCs.

15To investigate the extent of the correlation between the divisional asymmetry/symmetry of HSCs and the transcriptome of daughter cells, we performed single-cell RNA sequencing (scRNA-seq) on paired daughter cells (Fig 7A).

16A global supervised “between-group analysis” (BGA) showed distinct clustering of daughter cells derived from the different experimental arms (Fig 7B; S4 Table; S8 Fig; S9A–S9C Fig). Daughter cells from aged and aged CASIN-treated HSCs showed a higher transcriptional heterogeneity compared to daughter cells from young and young treated-with-Wnt5a HSCs. When daughter pairs were separated based on chronological age, the aged CASIN-treated arm was closer to the young than to the aged control arm. However, the daughter cells from young Wnt5a-treated HSCs were also closer to the young control than to the older control arm (Fig 7B).

17To evaluate asymmetry/symmetry based on the difference of the transcriptome between the daughter cells, we first performed Self-Organizing Maps (SOM) analyses followed by three different analytical approaches (S10 Fig). We selected statistically significant genes (p-value and q-value < 0.05) that fall into distinct clusters of highly correlated genes (metagenes) that, summarily, presented a portrait that captures the global transcriptional landscape of a given cell. Cells with similar portraits and significant sets of genes will implicitly have high levels of transcriptional similarity. In the first analysis, we performed a hypergeometric-model–based gene set enrichment analysis using 13 previously published HSC and polarity signatures by simultaneously interrogating all the gene sets (Fig 7C; S11 Fig; S5 Table). When we then compared cells belonging to a pair, the majority of the pairs showed a high degree of similarity (concordance) across all arms (Fig 7D). In the second approach, we quantified the degree of similarity between SOMs of daughter pairs by comparing the genes that showed statistically significant log fold-change compared to the mean global expression profile (Fig 7E). Each pair was classified as concordant or discordant based on the magnitude of genes showing the same direction of change. Again, we observed a high level of concordance for all daughter pairs (Fig 7F).

18Finally, taking into consideration the results of the transplants (Fig 4C) and the mathematical modeling (Fig 3E), we sought to confirm higher asymmetric outcomes (discordant transcriptome profiles) in young and aged CASIN-treated HSCs and increased symmetric division outcomes (concordant transcriptome profiles) in aged and young Wnt5a-treated HSCs. To this end, we analyzed differentially expressed genes among daughter cells from young and aged CASIN-treated HSCs to highlight discordant divisions among the different pairs (Fig 7G; S6 Table). Results showed a percentage of asymmetric divisions slightly higher for all arms compared to the previous gene set enrichment and fold change analysis approaches; however, again results were not in agreement with the transplants and the mathematical prediction, and no significant difference was observed among the four arms (Fig 7H).

19In the end, we analysed enrichment for biological processes in the 134 up-regulated genes in young and aged CASIN-treated daughter pairs. We observed a significant enrichment in epigenetic regulation of gene expression and chromatin organization (Fig 7I). In the down-regulated gene sets, structural constituents of ribosomes were significantly enriched (Fig 7I; S9D Fig). No significant enrichment was found for the genes up-regulated in aged and young Wnt5a-treated cells (S7 Table). Therefore, the transcriptome in daughter cells to a large extent does not correlate with the function of a daughter cell as determined in the transplantation experiments.

20It is well established that genes specific for certain cell types or that encode crucial lineage determining regulators might be activated in a stepwise fashion at the levels of chromatin accessibility first, followed by transcriptional activation. Therefore, to investigate whether chromatin accessibility might be distinct in daughter, cells we performed single-cell transposase-accessible chromatin sequencing (scATAC-seq) on daughter cells from young and aged HSCs (Fig 8A; S12 Fig; S13 Fig).

21ATAC-seq was selected because it reflects alterations in chromatin accessibility that are possibly also linked to changes in the level of H4K16ac (see Fig 2A and). The ATAC-seq profile of single daughter cells showed the expected distribution of insert sizes and an enrichment at transcription start sites (TSSs) of nucleosome-free fragments and the expected profile with respect to CCCTC-binding factor (CTCF) binding sites (S13 Fig). ATAC-seq peaks were annotated predominantly to TSSs, introns, and intergenic regions up to 1 kb apart from TSSs (S14A Fig). The overall scATAC-seq profile of daughter pairs of young HSCs showed a marked increase in the value of peak ratio compared to daughter pairs of aged HSCs. This resulted in a difference in ATAC-seq peak amounts between daughters from young HSCs, while daughter pairs of aged HSCs presented more often with having equal amounts of ATAC-seq peaks and a ratio quite uniform around 1 (Fig 8B). The difference in the ATAC-seq peak amount between paired daughters allowed for scoring of divisions as symmetric or asymmetric. Briefly, for those pairs for which we had a sufficient number of peaks for the reshuffling simulation test of the null hypothesis that cells in a pair show the same peak counts in the same genomic positions, asymmetry assignment was based on the significance of no overlap and the ratio of the peak counts (as a measure of global chromatin accessibility) greater than 2 (S8 Table and S15 Fig). Based on these statistical parameters, young HSCs presented with 57% of asymmetric divisions, while aged HSCs showed 40% of asymmetric divisions (Fig 8C; S8 Table; S15 Fig). The symmetry/asymmetry difference between young and aged cells determined by 3D-IF for Cdc42 and H4K16ac, in transplantation assays and as predicted by the mathematical model, is thus similar to the difference in the ATAC-seq peak profile of paired daughters of young and aged HSCs.

22We next annotated scATAC-seq peaks to genes. Similar to the analyses performed for the scRNA-seq data, we used gene set enrichment analysis to align daughter pairs to known stem cell signatures (Fig 8D and S16 Fig) and presented them again in radar plots. The data revealed also for these analyses a higher frequency of asymmetric divisions for young HSCs (64%) compared to asymmetric division for aged HSCs (40%) (Fig 8E and S16 Fig). Individual daughter cells from young and aged pairs were subsequently grouped according to the mode of division as established based on the ATAC-seq peak ratio (S8 Table and Fig 8B and 8C): both daughter cells of all symmetric divisions were assorted as stem cells because the majority of daughter cells stemming from symmetric divisions were, according to the transplantation assay, stem cells. As for asymmetric divisions, the daughter cells with the highest ATAC-seq peak number were assigned to the daughter progenitor cells group because daughter cells of symmetric divisions, which result in daughter stem cells, presented with a quite uniform and low peak count (Fig 8F). Next, transcription factor (TF) footprint analyses on ATAC-seq peaks identified significantly enriched TF binding sites common to daughter stem cells of both young and aged HSCs (S14B and S14C Fig and S17 Fig and S9 Table). Among them were TFs like NFY, SpiB, Sp5, Sp1, Runx, ERG, PU.1, and Maz (S14B–S14D Fig). Of these, the frequencies of target sequences with a Sp5, Sp1, Maz, and ERG motif were even further increased in daughter stem cells of aged HSCs (S14D Fig). In daughter progenitor cells from both young and aged HSCs, the enriched TFs set was very distinct from the one enriched in stem cells (S14E and S14F Fig), and factors like c-Myc, Usf2, and Stat3 further showed a decline in enrichment in aged daughter progenitors, while PU.1:IRF8, CEBP, CEBP:AP1, and Nanog showed the opposite trend (S14E Fig). However, only the frequency of target sequences for the E2F6 motif was significantly more enriched in aged daughter progenitors (S14G Fig).

23Finally, gene ontology (GO) analysis revealed 603 and 490 significantly enriched pathways in young and aged daughter stem cells and 474 and 360 pathways in young and aged daughter progenitor cells (S18 Fig and S10 Table). More than 75% of young daughter progenitor cells shared enrichment for only 1.47% of all significantly enriched GOs. These daughter-progenitor–specific enriched GOs for young HSCs were platelet homeostasis, mitosis, interferonγ and interferon signalling, IL-2 and IGF1 signalling, and pyruvate and lipid metabolism (Fig 8G). In daughter progenitors of aged HSCs, 100% of the cells were enriched for only 2.77% of the GOs, which included platelet homeostasis, TNF signalling, propionyl-CoA catabolism, and cholesterol biosynthesis (Fig 8H). To further dissect differences between young and aged daughter progenitor cells, we performed gene set enrichment analyses for a previously reported multipotent hematopoietic progenitor cell signature. Despite only 2 aged and 3 young daughter progenitor cells showing significant enrichment, both young and aged daughter progenitor cells showed a similar set of genes overlapping across cells in a given age group but also across the two age groups, indicating a similar pattern of chromatin accessibility in and around these genes in both young and aged daughter progenitor cells (S14H–S14I Fig).

24Most of the daughter stem cells from both young and aged HSCs were enriched for GO pathways linked to glycolysis and gluconeogenesis and also for signaling linked to RhoGTPases, implying an important role of regulation of the glycolytic metabolism and of small RhoGTPase signaling (like Cdc42) for the regulation of stem-ness of daughter stem cells (Fig 8I and 8J), as also predicted by our mathematical model (Fig 3). Similarly informative might be the finding that major differences in GO enrichment among daughter stem cells from young and aged HSCs were linked to signalling pathways involved in the interaction of HSCs with the stem cell niche: while daughter stem cells from young HSCs are enriched for GOs associated with TGFβ, VEGFR2, and EGFR signalling, daughter stem cells from aged HSCs are enriched for Wnt signalling (Fig 8I–8J). In summary, open chromatin configuration and thus the epigenetic make-up of a daughter cell strongly correlates with the potential of the daughter cell.

25The asymmetric or symmetric sorting of epigenetic information upon HSC division has been proposed as a way of controlling the potential of nascent daughter cells. Here, we demonstrate, using a comprehensive set of paired daughter analyses that include single-cell 3D confocal imaging, single-cell transplants, scRNA-seq, and scATAC-seq, that the mode of HSC division is strongly linked to the polarity status before mitosis, which is in turn determined by the level of the activity Cdc42 in stem cells. These results consequently imply that the level of activity of Cdc42 in mother HSCs regulates the mode (symmetric/asymmetric) of HSC divisions, which results in a distinct allocation of epigenetic regulators like H4K16ac as well as of Cdc42 itself to daughter cells. Our data, combined with our mathematical model on the mode of divisions and the role of Cdc42 in daughter cells, further support that polar HSCs (that are the majority in young animals) undergo asymmetric divisions while apolar HSCs (that are the majority in aged animals) undergo primarily symmetric divisions. Aging and rejuvenation of HSCs, as a consequence of changes in the activity of Cdc42 in HSCs, also imply a change in the mode of HSC divisions and thus link the mode of division of HSCs to changes in the function of daughter stem cells. In addition, our data (both experimental and mathematical modeling) strongly support that symmetric divisions, which are the primary mode of division of aged HSCs, might be directly linked to the increase in the number of HSCs in bone marrow upon aging. A striking new feature, clustering of HSCs within the bone marrow of aged animals (Fig 6), is most likely a consequence of the elevated frequency of symmetric divisions of aged HSCs. An interesting novel concept derived from HSC clustering is that, in such clusters, HSCs might serve as niche cells for other HSCs.

26Recently published data support a dominant role of hematopoietic progenitor cells but not HSCs in maintaining long-term hematopoiesis in an unperturbed setting in young animals. Our findings are consistent with this novel concept because young HSCs will divide primarily asymmetrically and provide the progenitor cells that support hematopoiesis. Our data might predict that, upon aging, this dominant role of progenitor cells in driving hematopoiesis will shift towards a more relevant contribution of stem cells to hematopoiesis due to the reduced number of progenitors generated from aged HSCs.

27The final fate of daughter cells after HSC divisions might be defined by signals intrinsic to stem cells that are allocated during division to daughter cells but might also be driven by extrinsic signals from the environment acting on daughter cells after mitosis is completed. The stem cell intrinsic signals need to be, by definition, epigenetic in nature (not linked to the DNA sequence), as daughter cells present with a genome that is identical to their mother cell. Our data and our mathematical model support that the decision on the potential of daughter cells is driven, to a large extent, by the mode of the division of the mother HSC, which is decided upon in the mother HSCs. We show distinct allocation of a polarity protein (Cdc42) and an epigenetic mark (H4K16ac) to daughter cells, which correlates with the potential of the daughter cells themselves. Our data also demonstrate that the overall transcriptome in daughter cells does not correlate with their potential, while the amount and the location of open chromatin regions is tightly linked to it. Furthermore, mathematical modelling and our ATAC-seq-linked GO signatures in daughter stem cells predict an important role for RhoGTPase signaling as well as metabolic signaling for the maintenance of the stem cell potential in daughter stem cells upon stem cell divisions. While a role of glycolysis in stem cell maintenance has been already described by multiple laboratories, an important role for RhoGTPase signaling for stem cell maintenance has only been recently suggested. The daughter progenitor cell signature is distinct from the stem cell signature, while sharing for both young and aged daughter-cell gene loci linked to platelet homeostasis. It has been described that HSCs can rapidly differentiate into platelets. In addition, progenitor cells display enrichment for signatures linked to interferon signaling, which has been, in the past, primarily associated with stem cell quiescence/activation.

28Recently, it has been reported that H4K16ac can directly control chromatin accessibility, so it is likely that a symmetric or asymmetric distribution of H4K16ac to daughter cells (due to polarity/apolarity of the mark in mother stem cells) might contribute to distinct levels of chromatin accessibility and thus affect the fate of daughter cells. These findings are consistent with recent data supporting that epigenetic memory in HSCs is persistent and guides cellular function. The data presented here also imply that an open chromatin configuration precedes changes in gene expression in daughter cells. Our ATAC-seq analyses imply that young HSCs might also establish a specific chromatin configuration that allow stem-cell–extrinsic signalling (TGFβ, VEGF, IGF1, and EGF signalling) to act on nascent daughter stem cells, while aged daughter stem cells present with open chromatin regions linked to Wnt signalling. A role for all of these factors for the biology of young and aged HSCs has already been described, further supporting these conclusions.

29In summary, our data imply an important role for Cdc42 activity/polarity in HSCs for driving the symmetric/asymmetric mode of division as well as a role of epigenetic mechanisms for determining the potential of daughter cells. The frequency of polar HSCs decreases upon aging, which results in more symmetric divisions but daughter stem cells with impaired potential.

30This study involves animal research (mice, C57BL/6 strain). All mice were housed in the animal barrier facility under pathogen-free conditions either at the University of Ulm or at CCHMC. All mouse experiments were performed in compliance with the German Law for Welfare of Laboratory Animals and were approved by the Institutional Review Board of the University of Ulm or by the IACUC of CCHMC (approval number: TVA 0.1172 and IACUC2013-0154). This study does not involve human participants and/or tissues, and it does not involve collection of plant, animal, or other materials from a natural setting.

31C57BL/6 mice (10- to 12-week-old) were obtained from Janvier. Aged C57BL/6 mice (20- to 26-month-old) were obtained from the internal divisional stock (derived from mice obtained from both The Jackson Laboratory and Janvier) as well as from NIA/Charles River. Congenic young and aged C57BL/6.SJL-Ptprca/Boy (BoyJ) mice were obtained from Charles River Laboratories or from the internal divisional stock (derived from mice obtained from Charles River Laboratories). Pan-RFP mice carrying constitutively active ROSA26-tdRFP alleles (indicated in the manuscript as Ac-RFP mice) were obtained from Professor Hans Joerg Fehling (Institute of Immunology, Ulm University) and were previously generated by intercrossing C57BL/6-Gt(ROSA)26Sortm1Hjf/Ieg mice with animals from a germline Cre-deleter strain. Offspring in which the ROSA26-driven fluorescent tdRFP reporter had been activated irreversibly as the result of loxP/Cre-mediated recombination in the germline were backcrossed for >10 generations onto C57BL/6, thereby eliminating the Cre recombinase transgene. Ac-RFP mice were used as homozygotes. Rag2−/−γc−/−KitW/Wv mice were obtained from the internal divisional stock (derived from mice obtained from Hans-Reimer Rodewald). All mice were housed in the animal barrier facility under pathogen-free conditions either at the University of Ulm or at CCHMC. All mouse experiments were performed in compliance with the German Law for Welfare of Laboratory Animals and were approved by the Institutional Review Board of the University of Ulm or by the IACUC of CCHMC.

32SMART-Seq version 4 Ultra Low Input RNA kit from Clontech (catalog number 634892) was used. The kit generates Illumina compatible RNA-Seq libraries. Sorted single cells were cultured with and without treatment in the presence of cytokines until first cell division (40–44 hours). The daughter cells were manually separated, washed with phosphate-buffered saline (PBS), and collected for RNA sequencing. The cDNA synthesis and amplification was done as recommended by Clontech. The amplified cDNA generated from single cells were used to make libraries using Illumina’s Nextera XT DNA Library Preparation kit (catalog number FC-131-1096) as per Illumina’s instructions. The generated libraries were quantified using an agilent bio-analyzer, pooled, and subjected to next-generation sequencing in a Hi-Seq 2500 for pair-end 75 bp sequencing condition.

33Bone marrow mononuclear cells were flushed out from long bones (tibiae and femurs) of young and aged mice and were isolated by low-density centrifugation (Histopaque 1083, Sigma). Low-density bone marrow cells were stained with a cocktail of biotinylated lineage antibodies. Biotinylated antibodies used for lineage staining were all rat anti-mouse antibodies: anti-CD11b (clone M1/70), anti-B220 (clone RA3-6B2), anti-CD5 (clone 53–7.3), anti-Gr-1 (clone RB6-8C5), anti-Ter119, and anti-CD8a (clone 53–6.7) (all from eBioscience). After lineage depletion by magnetic separation (Dynalbeads, Invitrogen), cells were stained with anti-Sca-1 (clone D7; eBioscience), anti-c-kit (clone 2B8; eBioscience), anti-CD34 (clone RAM34; eBioscience), anti-CD127 (clone A7R34; eBioscience), anti-Flk-2 (clone A2F10; eBioscience), and Streptavidin (eBioscience). Single long-term HSCs (gated as LSK CD34−Flk2−) were sorted using a BD FACS Aria III (BD Bioscience) into Terasaki plates. HSCs were cultured in IMDM plus 10% FBS plus P/S plus 100 ng/mL TPO, G-CSF, and SCF (PeproTech) at 37°C, 5% CO2, 3% O2. CASIN was used at a dose of 5 μM. Wnt5a (R&D) was used at a dose of 100 ng/mL. Each well of the Terasaki plate was microscopically tracked, and within 1 hour from HSC division, daughter cells were separated by mechanical pipetting, deposited in distinct Terasaki wells, and controlled again by microscopy. Eventually, the daughter pair was injected into recipient mice (1 daughter in 1 mouse, the second daughter into a littermate).

34For daughter cell pair transplantation, young (2- to 4-month-old) and aged (24-month-old) C57BL/6.SJL-Ptprca/Boy (BoyJ) (Ly5.1+) mice were used as donors. Single daughter cell HSCs were separated by pipetting under microscopy control and were directly injected into the retro-orbital vein of recipient mice. PB chimerism was determined by FACS analysis at week 4, 12, 16, and 24 post transplant. The transplantation experiment was performed 8 times with a cohort of 10 to 16 recipient mice per group each transplant. In total, 110 recipients were injected, and 94 mice showed detectable chimerism (contribution to PB >0.1%) at a minimum of 1 time point of the analyses.

35PB immunostaining was performed according to standard procedures, and samples were analyzed on a LSRII flow cytometer (BD Biosciences). Monoclonal antibodies to Ly5.2 (clone 104, eBioscience) and Ly5.1 (clone A20, eBioscience) were used to distinguish recipient from donor cells. For PB lineage analysis, the antibodies used were all from eBioscience: anti-CD3ε (clone 145-2C11), anti-B220 (clone RA3-6B2), anti-Mac-1 (clone M1/70), and anti-Gr-1 (clone RC57BL/6-8C5). The engraftment is plotted as percentage of donor-derived cells among total white blood cells. B cells, T cells, and myeloid cell lineage data are plotted as the percentage of B220+, CD3+, and Myeloid (Gr-1+, Mac-1+, and Gr-1+Mac-1+) cells among donor-derived cells. Please note that data are plotted in a logarithmic scale to accommodate sample heterogeneity.

36Freshly sorted HSCs were seeded on fibronectin-coated glass coverslips. At 32 to 34 hours after culturing in IMDM plus 10% FBS plus P/S plus 100 ng/mL TPO, G-CSF, and SCF (PeproTech) ± CASIN 5 μM ± Wnt5a (R&D) 100 ng/mL at 37°C, 5% CO2, 3% O2, cells were fixed with BD Cytofix Fixation Buffer (BD Biosciences). After fixation, cells were gently washed with PBS, permeabilized with 0.2% Triton X-100 (Sigma) in PBS for 20 minutes, and blocked with 10% Donkey Serum (Sigma) for 30 minutes. Primary and secondary antibody incubations were performed for 1 hour at room temperature. The primary antibodies were anti-alpha tubulin antibody (Abcam, rat monoclonal ab6160), anti-Cdc42, and anti-H4K16ac obtained from Millipore and Abcam (we tested 2 different antibodies for each target, and results were consistent; all 4 antibodies were rabbit polyclonal; anti-Cdc42 from Millipore was previously validated). The secondary antibodies for IF were anti-rat DyLight488-conjugated antibody, anti-rat DyLight647-conjugated antibody, and anti-rabbit DyLight549-conjugated antibody (all obtained from Jackson ImmunoResearch). Coverslips were mounted with ProLong Gold Antifade Reagent with or without DAPI (Invitrogen, Molecular Probes). Samples were imaged with an AxioObserver Z1 microscope (Zeiss) equipped with a 63× PH objective. Images were analyzed with Zen software. Alternatively, samples were analyzed with an LSM710 confocal microscope (Zeiss) equipped with a 63X objective. Primary raw data were imported into the Volocity Software package (version 6.2, Perkin Elmer) for further processing and conversion into 3D images.

37After intravenous injection of APC-anti-CD31 (clone MEC13.3, BioLegend) and Alexa Fluor 647-anti-CD144 (clone BV13, BioLegend) antibodies, long bones were harvested after postmortem heart perfusion with 4% paraformaldehyde (PFA) in PBS and were post-fixed in 4% PFA/PBS solution for 24 hours at 4°C. Subsequently, bones were embedded without bisecting in optimum cutting temperature (OCT) compound (Tissue-Tek) and were snap frozen in liquid nitrogen and stored at −80°C. Bones were shaved along the longitudinal axis on a cryostat until the BM cavity was exposed (S5A Fig). The bones were purified from melting OCT. Specimens were fixed again in 4% PFA/PBS at RT for 30 minutes. Tissues were blocked and permeabilized with buffer containing 20% donkey serum and 0.5% Triton X-100, were incubated with a fluorescent-labeled antibody PE-anti-CD150 or AlexaFluor488-anti-CD150 (clone TC15-12F12.2, BioLegend) as well as Biotin-labeled primary antibodies anti-CD41 (clone MWReg30), anti-CD48 (clone HM48-1), anti-CD11b (clone M1/70), anti-B220 (clone RA3-6B2), anti-CD5 (clone 53–7.3), anti-Gr-1 (clone RB6-8C5), anti-Ter119, and anti-CD8a (Clone 53–6.7) (all from eBioscience) 1 to 3 days at 4°C and stained with Streptavidin-eFluor450 (eBioscience) or Streptavidin- FITC (eBioscience) for 2 hours at RT. The fluorescently labeled bone tissues were placed cut-face down onto a 4-well-μ-Slide and were covered in antifade to prevent tissue desiccation. The preparations were immediately examined under Zeiss LSM 710 or Leica TCS SP8 confocal microscopes and analyzed with the image analysis software Volocity (version 6.2; Perkin Elmer).

38Raw Fastq files were adapter trimmed using Trimm Galore (Babraham Institute), a wrapper tool around Cutadapt and FastQC. Reads with Phred score of 20 or higher were kept for further analysis. Reads were aligned to the mouse reference genome version 10 (GRCm38/mm10) using tophat. Transcript abundance estimation, fragments per kilobase of exon per million fragments mapped (FPKM) calculations, and normalization was done using Cufflinks, a package of tools for transcriptome assembly and differential expression analysis of RNA-Seq data. Correspondence-based BGA was done using R and bioconductor package made4. SOM and metagene analysis was performed using oposSOM. Calculation of concordant and discordant daughter pairs was done using quadrant count ratio (QCR) by counting genes that fall in the 4 quadrants. A chi-squared test was used to test the significance of difference of concordant (quadrants I and III) and discordant genes (quadrants II and IV). Gene set enrichment analysis was carried out using stem-ness and cell-polarity–related gene sets. The analysis approach simultaneously interrogated 13 previously published HSC and polarity signatures, and each daughter cell pair was depicted in a radar plot such that vertices in the plot correspond to each of the considered gene sets: Mm_HSC_Runx1_Wu, Mm_HSC_Tcf7_Wu, Mm_LT_HSC_Venezia, Mm_Proliferation_Venezia, Mm_Quiescence_Venezia, Polarity_factors_Ting, Novel_HSC_regul_polar_Ting, Grover et al, 2016 and Kowalczyk et al, 2015 (S6 Table). Cells were first individually analyzed, and significance of gene set enrichment was tested using hypergeometric test. Concordance between daughter pair was then visualized using radar plots and tested using goodness-of-fit test. Knowledge-driven comparison between young (young LT-HSC/CASIN-treated aged LT-HSC cells) versus aged (aged LT-HSC/WNT5A-treated young LT-HSC) arms was made by populating genes that show same direction of significant change (p-value cutoff 0.05) in expression in a given arm but not in the other. Genes fulfilling this criterion were used to generate the Venn diagram shown in Fig 4G. Daughter pair concordance and discordance in each treatment group was determined based on the level of significance of correlation that each pair shows. GO analysis and analysis of enrichment for biological processes was done according to standard GO Consortium dataset by using PANTHER version 11.

39The protocol was adjusted from previously published procedures. Briefly, single HSCs from young and aged C57BL/6 mice were sorted into Terasaki plates with IMDM supplemented with 10% FBS, P/S, and 100 ng/mL of SCF, G-CSF, and TPO at 3% O2, 5% CO2. Cells were tracked microscopically, and after the first division, the daughter cells were separated by pipetting and singularly subjected to fragmentation of open chromatin regions using Tn5 transposase (Illumina), followed by a pre-amplification step, library preparation, and subsequent paired-end sequencing. For the pre-amplification, NEBNext Ultra II Q5 Master Mix was used with Primer 1: 5’GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG3’ and Primer 2: 5’TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG3’. For dual-indexing, 10 μL of the pre-amplified ATAC reaction was used as input for Nextera index kit (Illumina). The generated libraries were quantified using an agilent bio-analyzer and a qPCR kit (New England Biolabs), pooled, and subjected to next-generation sequencing in a Hi-Seq 3000 for paired-end 75 bp sequencing condition.

40We used the ENCODE guidelines and current standards of ATAC-seq data analysis. This was achieved by using a combination of ENCODE prototype scripts and in-house algorithms. Briefly, sequences were aligned to the UCSC mouse reference genome mm10 using Bowtie 2. Peak calling was performed using MACS. Additional analysis, visualization, and DNA footprint analyses were done using various in-house R scripts and published Bioconductor packages (ATACSeqQC). Overlap analysis between peak files was done using bedtools. Random chromosome-controlled reshuffling of peaks was performed using bedtools followed by Monte Carlo Simulation based on in-house script. TF motif analysis and annotation were done using HOMER. Gene set enrichment analysis was carried out using stemness and cell-polarity–related gene sets. The analysis approach simultaneously interrogated 13 previously published HSC and polarity signatures, and each daughter cell pair was depicted in a radar plot in which vertices in the plot correspond to each of the considered gene sets: Mm_HSC_Runx1_Wu, Mm_HSC_Tcf7_Wu, Mm_LT_HSC_Venezia, Mm_Proliferation_Venezia, Mm_Quiescence_Venezia, Polarity_factors_Ting, Novel_HSC_regul_polar_Ting, Grover et al, 2016, and Kowalczyk et al, 2015 (S6 Table). Cells were first individually analyzed, and significance of gene set enrichment was tested using hypergeometric test. Concordance between daughter pair was then visualized using radar plots and tested using goodness-of-fit test.

41Data were tested to meet normal distribution. The variance was similar between groups that were statistically compared. All data are plotted as mean plus 1 SEM unless otherwise stated. The SEM is used to indicate the precision of an estimated mean. Such a data representation does not affect the statistical analyses as variance information is utilized in the test statistics. A paired Student t test was used to determine the significance of the difference between means of 2 groups. One-way ANOVA or two-way ANOVA were used to compare means among 3 or more independent groups. Bonferroni post-test to compare all pairs of dataset was determined when overall p-value was < 0.05. All statistical analyses were determined with Prism 4.0 c version. In order to choose sample size, we used GraphPad StatMate Software version 2.0b, estimating an SD between 2 and 8 (depending on the experiment and the possibility of increasing sample size). For transplantation experiments, mice showing signs of sickness and with clear alterations of blood parameter and/or showing signs of major disease involving also nonhematopoietic tissues were excluded from analysis. As for in vitro experiments, samples were excluded from analysis in the case of clear technical problems (error in immune-blotting or staining procedures or technical problems with reagents). All criteria for exclusions of samples from in vivo or in vitro experiments were preestablished. Each figure legend contains the number (n) of biological repeats (samples obtained from experiments repeated on different days and starting from different mice) of the statistical analysis. Mice for experiments were randomly chosen from our in-house colonies or from suppliers. All mice were C57BL/s6 females unless otherwise stated. The investigator was not blinded to the mouse group allocation nor when assessing the outcome (Rag2−/−γc−/−KitW/Wv mice, aged mice, or young mice transplanted with aged BM stem cells require particular care and follow-up).