A unique adipocyte progenitor population promotes age-related adiposity

30 31 The average fat mass in adults increases dramatically with age, and older people often suffer 32 from visceral obesity and related adverse metabolic disorders. Unfortunately, how aging leads to 33 fat accumulation is poorly understood. It is known that fat cell (adipocyte) turnover is very low in 34 young mice, similar to that in young humans. Here, we find that mice mimic age-related fat 35 expansion in humans. In vivo lineage tracing shows that massive adipogenesis (the generation of 36 new adipocytes), especially in the visceral fat, is triggered during aging. Thus, in contrast to most 37 types of adult stem cells that exhibit a reduced ability to proliferate and differentiate, the 38 adipogenic potential of adipocyte progenitor cells (APCs) is unlocked by aging. In vivo 39 transplantation and 3D imaging of transplants show that APCs in aged mice cell-autonomously 40 gain high adipogenic capacity. Single-cell RNA sequencing analyses reveal that aging globally 41 remodels APCs. Herein, we identify a novel committed preadipocyte population that is age42 specific (CP-A), existing both in mice and humans, with a global activation of proliferation and 43 adipogenesis pathways. CP-A cells display high proliferation and adipogenesis activity, both in 44 vivo and in vitro. Macrophages may regulate the remodeling of APCs and the generation of CP-A 45 cells during aging. Together, these findings define a new fundamental mechanism involved in fat 46 tissue aging and offer prospects for preventing and treating age-related metabolic disorders. 47

with age (Extended Data Fig. 1a). Bodyweight gain with age was much higher in males than 97 females (Extended Data Fig. 1b, c). Among the two types of fat depots, gonadal WAT (gWAT), 98 a typical type of visceral WAT, had the most significant weight gain with age (4.6-fold) (Fig. 1e), 99 while subcutaneous WAT (sWAT) increased 2.8-fold (Fig. 1f). The aged mice exhibited 100 dramatically reduced oxygen consumption (Fig. 1g), slightly reduced physical activity (Fig. 1h) 101 and total movement (Extended Data Fig. 1d), as well as dramatically reduced energy expenditure 102 (Fig. 1i). Food intake was also reduced in aged mice (Extended Data Fig. 1e). The drop in oxygen 103 consumption and energy expenditure suggested that the aged mice have significantly lower basal 104 metabolism, similar to humans 7 . The decline in basal metabolism causes a positive energy balance 105 (energy intake exceeds expenditure), eventually leads to fat accumulation. Intraperitoneal glucose 106 tolerance tests (GTT) (Fig. 1j, k) and insulin tolerance tests (ITT) (Fig. 1l, m) showed that aged 107 mice have impaired glucose tolerance and are more insulin resistant. Therefore, mice display a 108 pattern of age-related visceral adiposity, accompanied by positive energy balance and insulin 109 resistance, similar to human sarcopenic obesity. 110 111 WAT accumulates with age through massive adipogenesis. 112 To better understand adipocyte dynamics, we previously developed the AdipoChaser mice for 113 tracking adipogenesis 16,17,19,31,32 . This mouse model is a doxycycline (dox)-based, tet-responsive 114 labeling system for the inducible, permanent labeling of Adiponectin (Adipoq) expressing cells as 115 Data Fig. 1f). To determine if aging is accompanied by adipogenesis, we 116 utilized AdipoChaser-LacZ mice to label all adipocytes as LacZ+ cells in 3-month-old male mice 117 and tracked adipogenesis during early aging (Extended Data Fig. 1g). Consistent with our 118 previous studies, in 6-month-old mice, we observed no new adipocytes (i.e., no lacZ-negative cells) 119 in the gWAT or sWAT, indicating that the turnover rate of adipocytes is extremely low in young 120 adults (Fig. 1n). In contrast, in 9-month-old mice, clusters of newly generated adipocytes started 121 to appear in gWAT. New adipocytes also started to appear in sWAT but did not yet form clusters. 122 As mice continue aging, in 18-month-old mice, gWAT showed numerous new adipocytes, which 123 formed large clusters; sWAT also had clusters of new adipocytes, which are relatively smaller than 124 gWAT. Adipocyte size was also increased in sWAT and gWAT, confirming the age-related 125 adipocyte hypertrophy 21 (Fig. 1n). Female mice were found quite different from the males 126 (Extended Data Fig. 1h). In the female gWAT, we observed some adipogenesis in 18-month-old 127 female mice, but the rate is much lower than that in male mice. In the female sWAT, we did not 128 observe any new adipocytes generated during aging. Thus, adipogenesis is the major contributor 129 to age-related gWAT expansion in male mice.  Fig. 2b, c). After dimerization, apoptosis was initiated in every adipocyte, as adipocytes were 139 uniformly dead, with disrupted morphology and negative perilipin staining (Extended Data Fig.   140 2b, c). One week after dimerization, gWAT volume was markedly reduced in both young and aged 141 mice (Extended Data Fig. 2b, c). At two-four weeks after dimerization, the aged mice 142 demonstrated strong adipocyte regeneration with essentially more adipocytes positive for perilipin. 143 gWAT from young mice, while found to have some perilipin-positive adipocytes, still had an 144 abnormal morphology with many lost cells and vacuoles (Extended Data Fig. 2b, c). By four 145 weeks of treatment, the gWAT of aged mice had regained more tissue mass than young mice 146 (Extended Data Fig. 2d). Thus, aged mice exhibit higher adipocyte regeneration capacity, through 147 earlier and faster de novo adipogenesis.

149
APCs from aged mice cell-autonomously have a high adipogenic capacity in vivo 150 Do APCs cell-autonomously undergo adipogenesis during aging? Alternatively, is there a 151 systemic stimulus required for age-related adipogenesis? APCs were reported to be less 152 immunogenic and less subject to rejection than other cell types following transplantation into 153 mice 33 . We then set out to test the in vivo adipogenic ability of APCs from gWAT of young vs. 154 aged male mice through transplantation. An equal number of APCs (Lin-, CD45-CD31-Ter119-) 155 from the stromal vascular fraction (SVF) of gWAT were mixed from 2.5-month-old male CAG-156 EGFP mice (GFP+, "young" APCs) and from 12-month-old male Rosa26-loxp-mTomato-stop-157 loxp-GFP (Rosa26-mT/mG) mice (Tomato+, "aged" APCs) (Fig. 2a). The cell mixture was 158 transplanted into the sWAT of 2.5-month-old male wide type (WT) mice ( Fig. 2b-f, Extended 159 Data Video 1). When the Matrigel transplants were dissected one month later, we observed that 160 more "aged" APCs (Tomato+) had differentiated into adipocytes (Fig. 2b, Extended Data Video 161 1), resulting in a 2.5-fold more adipocytes and 2.2-fold greater adipocyte volume compared to 162 "young" APCs (GFP+) (Fig. 2c, d). These GFP+ and Tomato+ cells (differentiated from GFP+ or 163 Tomato+ APCs) were indeed mature adipocytes, as they have a unilocular morphology (single 164 lipid droplet) (Fig. 2e) and were perilipin positive (Fig. 2f). To further confirm our observation, 165 we also transplanted Tomato+ APCs from the gWAT of 2.5-month-old and 12-month-old male 166 Rosa26-mT/mG mice separately into the different sides of sWATs of the same male WT mice (Fig.   167 2g). Four weeks after transplantation, the "aged" APCs had a significantly higher (3 folds) 168 adipogenesis rate, compared to the "young" APCs ( Fig. 2h-j, Extended Data Video 2, 3). Thus, 169 the APCs from older mice maintain their high adipogenic potential after transplantation into a 170 "young" environment. To test if age-related systemic stimulation is sufficient to trigger APC adipogenesis, we then 173 took the reverse approach and transplanted APCs from young mice into aged mice. Equal numbers 174 of "young" Tomato+ APCs from gWAT of 2.5-month-old male Rosa26-mT/mG were transplanted 175 into sWATs of 12-month-old and 2.5-month-old male WT mice (Extended Data Fig. 3a). Four 176 weeks after the transplantation, no significant adipogenesis of the "young" APCs was observed in Altogether, these transplantation results indicate that the "aged" APCs cell-autonomously 183 exhibit greater adipogenic rate in the in vivo setting, regardless of "young" or "aged" 184 microenvironment.

186
Aging generates a new APC population with a globally activated adipogenic program 187 To explore the molecular underpinning of the adipogenic properties of the APCs from aged 188 mice, we performed scRNA-seq of the CD45-CD31-Ter119-stromal vascular cells from the 189 gWAT of aged and young mice (n=3 mice/group) (Extended Data Fig. 4a). Among the 19,534 190 cells sequenced, 15,194 cells (78%) were identified as progenitor cells (Extended Data Fig. 4b), 191 based on the expression of classic APC markers CD34, CD29, PDGFRα, and Sca-1. While the 192 variation between mice within the same group was minimal (Extended Data Fig. 4c), there were 193 dramatic transcriptomic shifts between young and aged cells (Extended Data Fig. 4d). 194 195 We combined the "young" and "aged" progenitor cells and used unsupervised clustering of 196 gene expression profiles to identify five cell clusters (Fig. 3a). Based on recently published 197 scRNA-seq analysis of WAT APCs 28,34,35 , we named these clusters: 1) Adipocyte stem cell (ASC); 198 2) intermediate adipocyte progenitor (IAP); 3) committed preadipocyte 1 (CP-1); 4) committed 199 preadipocyte 2 (CP-2); 5) committed preadipocyte, age-specific (CP-A). We found that aging 200 significantly remodeled the progenitor cell population, as "aged" cells had "shifted" away from 201 "young" cells, and cells from the two groups had little overlap (Fig. 3b, c). Using Slingshot 202 trajectory analysis, we revealed potential lineage relationships among these APC clusters (Fig. 3d,   203 Extended Data Fig. 4e). Within the "young" progenitor cells, ASC had two developmental 204 trajectories, both through IAP, and terminated as CP-1 and CP-2 ( Fig. 3d, Extended Data Fig.   205 4e). Interestingly, within the "aged' progenitor cells, ASC had one additional, unique 206 developmental trajectory, which terminated as CP-A (Fig. 3d, Extended Data Fig. 4e). 207 Quantifying the percentage of each cell cluster showed that the CP-A population only existed in 208 aged mice, whereas the CP-1 population was dramatically reduced in aged mice (Fig. 3e). These 209 results suggest that the aging process generated a new adipogenic lineage from ASC to CP-A. 210 Next, we used CytoTRACE analysis and sorted the order of these APC clusters based on 212 differentiation status as predicted by transcript abundance (Fig. 3f, Extended Data Fig. 4f, g). 213 From stem cells to committed precursor cells, the order of these clusters was: ASC, CP-2, IAP, 214 CP-1, and CP-A. For each type of cluster, the "Young" APCs were always more stem cell-like 215 compared to the "aged" APCs. And among all clusters, the "aged" CP-A cluster was identified as 216 the most committed APC. The mesenchymal progenitor marker Pdgfra was universally expressed 217 in all clusters (Fig. 3g), but each cluster also showed unique markers (Extended Data Fig. 4h).  Fig. 4h, j). To determine which cluster(s) within the "aged" APCs undergo active 224 adipogenesis during aging, we assessed the biological pathways enriched in the marker genes of 225 each cluster. Interestingly, many genes involved in the adipogenic pathway is highly upregulated 226 in the CP-A cluster, which is unique to the aged group (Fig. 3l). The enrichment of the adipogenic 227 pathway in the CP-A cluster indicates that this population is uniquely generated during aging and 228 may underlie age-related adipogenesis. (ASC1 and ASC2) in gWAT representing different cell states during adipogenesis. In comparison 15 , the ASC cluster we noted highly overlapped with the Dpp4/Pi16/Cd55+ ASC2 235 cluster. Our CP-1 cluster highly overlapped with the Icam1/Igf1/Adam12+ ASC1 cluster. Our IAP 236 and CP-2 clusters partially overlapped with both ASC1 and ASC2 clusters, and the CP-A cluster 237 partially overlapped with the ASC1 cluster (Extended Data Fig. 5a, b). Sárvári et al. utilized 238 single-nucleus RNA-seq to identify four major APC clusters in gWAT 20 . Our ASC cluster 239 overlapped with the FAP2, FAP3, and FAP4 clusters. IAP, CP-1, CP-2, and CP-A clusters partially 240 overlapped with the FAP1 and FAP2 clusters (Extended Data Fig. 5c, d). Schwalie et al.   Fig. 6a). Unsupervised clustering of gene expression profiles divided these APCs 254 into eight cell clusters (Hu0-Hu7) (Fig. 3m). To compare the identified APC populations from 255 mice with the human clusters, we calculated the ASC and CP-A gene module score as a sum of 256 mouse marker genes (Fig. 3n). Clusters Hu0, 2, 3, 6 were enriched with mouse ASC markers, 257 while clusters Hu1, 5, 7 were enriched with mouse CP-A markers (Fig. 3n). Thus, both mouse CD235a-PDGFRα+ cells) (Extended Data Fig. 6j), and found a trend for the LIFR+ population 267 to dramatically increase with age (Fig. 3o). Thus, APCs in human vWAT may undergo similar 268 remodeling during aging, and the human CP-A population may also contribute to sarcopenic 269 obesity.  Table 1). Moreover, the adipogenesis pathway is highly enriched in the age-specific CP-A 275 population (Fig. 4b, Extended Data Table 2). This analysis suggests that the "aged" ASC and  We then determined the capacity of total APC to proliferate and differentiate into adipocytes 280 in 3D culture in vitro (Fig. 4c). The enriched APCs were first mixed with Matrigel and cultured in 281 the growth medium for 72 hours. An adipogenic cocktail was then added to differentiate these cells 282 into adipocytes. Starting with the same number of cells, "aged" APCs proliferated into more cells 283 compared to the "young" APCs (Extended Data Fig. 7a). However, the adipogenesis rate (the 284 percentage of Bodipy+ adipocytes in Tomato+ cells) for both "young" and "aged" APCs were 285 similarly high, close to 100%. The advantage of the 3D culture system is that we do not need the 286 APCs to grow densely and attach prior to adipogenesis. We then added the adipogenic cocktail 16 287 hours later after seeding to the enriched APC populations (Extended Data Fig. 7b). "Aged" APCs 288 had higher (but not significant) adipogenesis rate, and significantly higher proliferation rate, as the 289 total Tomato+ cell number from "aged" APCs was more than 2 folds greater than the "young"   Fig. 7g). This also validates that aged ASCs are more 307 differentiated than young ASC as suggested by the CytoTRACE analysis (Fig. 3f). The "aged" 308 ASC population had an adipogenesis rate similar to the "young" ASCs, but they had a significantly 309 higher proliferation rate (Fig. 4d, e). We also compared the age-specific CP-A population with the 310 "young" CP-1 population, which shares the highest molecular similarity with the CP-A population 311 according to the scRNA-seq results. Importantly, the CP-A population not only had a more than 312 four folds higher proliferation rate compared to the CP-1 population, but had the highest 313 adipogenesis rate among all populations tested (Fig. 4f, g). Thus, the CP-A population, uniquely 314 generated with age, has the most remarkable capacity for proliferation and differentiation in vitro. CP-A populations were collected from freshly isolated SVFs from gWAT of 2.5-month-old and 319 12-month-old male Rosa26-mT/mG mice, and equal numbers of these enriched "young" and "aged" 320 populations were transplanted into the different sides of the sWATs of male WT mice (Fig. 4h). 321 The "aged" DPP4-APCs grew substantially larger in volume compared to the "young" DPP4-  from the "aged" macrophages to "aged" ASC. Signals sent from the "aged" macrophages to the 356 "aged" CP-A were compared with signaling sent from the "young" macrophages to the "young"  Rosa26-mT/mG mice were mixed with macrophages from young and aged male WT mice.

369
Matrigel containing the cell mixture was seeded in the growth medium for 16 hours, then the 370 adipogenic cocktail was added to drive adipocyte differentiation (Extended Data Fig. 10a). 371 Macrophages, regardless of the age of the donor mice, increased the proliferation rate of APCs of 372 both "young" and "aged" mice. However, co-culture with macrophage did not increase the 373 adipogenesis rate of APCs of both "young" and "aged" mice (Extended Data Fig. 10b). We note 374 that the in vitro co-culture experiment may not capture the role of macrophages in the remodeling 375 of APCs during aging, which may take weeks and months. As a highly plastic organ, WAT alters in volume in adaptation to a variety of physiological and 379 pathological metabolic challenges. We and others have shown that without any metabolic 380 challenge, healthy young adult mice or humans have extremely low adipocyte turnover rates. WAT 381 plasticity under many physiological and pathological challenges has been extensively studied, such 382 as cold exposure and high fat diet feeding 9,10 . However, the expansion of WAT during early aging 383 is surprisingly understudied. In this study, we provide the first cellular and molecular evidence that 384 aging triggers massive adipogenesis of unique APC populations (Extended Data Fig. 10c).

386
Among the two major fat depots, vWAT and sWAT, vWAT has the most adipogenesis during 387 aging. As vWAT and sWAT are fundamentally distinct organs with distinct development timelines, 388 molecular signatures, and metabolic functions 9,10,29 , it is not surprising that age-related adiposity 389 happens in the vWAT at a much higher rate. Compared to hypertrophy, which gives WAT only 390 limited potential to expand, adipogenesis is alarming because it grants WAT unlimited potential 391 for growth. Our findings on the age-related adipogenesis, therefore, highlights adipogenesis but 392 not hypertrophy, as the key process to intervene for age-related obesity and other associated 393 diseases. WAT expansion during aging is closely linked to increased healthspan and lifespan. Our study 402 expands on this and suggests that to limit WAT expansion during aging, it is crucial to understand 403 the underlying mechanism of age-related adipogenesis.  In addition to the discovery of the new age-specific CP-A population, our scRNA-seq data 457 showed that the other APC populations also had little overlap between the aged and young groups, 458 indicating that the aging process significantly transformed all APC sub-populations. Indeed, we 459 found that the "aged" ASC population had a higher proliferation rate, which is also essential for   Co-culture with ATMs did not increase the differentiation of APCs, but did promote cell growth, 484 suggesting that ATMs regulate APCs during aging. Ligand-receptor analysis suggests that the cell- In conclusion, we found that massive adipogenesis is triggered during aging, and largely 494 contributes to age-related visceral adiposity. We then discovered a new, age-specific APC sub-

Isolation of mouse adipose SVFs and enrichment of APC subpopulations
gWAT was minced and transferred to a 50 mL Falcon tube containing 10 mL digestion buffer (100mM HEPES pH 7.4; 120mM NaCl; 50mM KCl; 5mM Glucose; 1mM CaCl2, 1.5% BSA, and 1 mg/mL collagenase D). The tissue was incubated in a 37℃ shaking water bath for 30 min.

Isolation of human adipose SVFs
Peri-pancreatic WAT samples from deceased cadaveric donors were dissected from the pancreas (processe time within ~8-12 hours). All of the proposed experiments will be immediately performed using the fresh tissue samples, without freezing or any preservation.

Identification of cell clusters
The Seurat R package version 3.0 4 was used to determine cell clusters by the Louvain algorithm based on similarities in the transcriptome patterns and visualized with t-Distributed Stochastic Neighbor Embedding (t-SNE). The optimal number of Principal Components used for t-SNE/UMAP dimensionality reduction and Louvain clustering was determined using the Jackstraw permutation approach and a grid-search of the parameters. The identities of the cell types were resolved by comparing the cell cluster specific marker genes expressed in each cluster in our own dataset, as identified with a Wilcoxon rank sum test, with known cell-type specific markers curated from literature, single cell atlases, and previous studies in the SVF. To be considered in the cell cluster marker analysis, a gene had to be expressed in at least 10% of the single cells from the cluster of interest and there had to be at least a 0.25 log fold change in the cell cluster of interest than in other cells. A multiple testing was corrected using the Benjamini-Hochberg method to estimate false discovery rate (FDR).

Identification of differentially expressed genes (DEGs) and pathways
To quantitatively determine which genes were affected by aging, we compared the cell transcriptome of each cell type between age groups using a Wilcoxon rank sum test. To be considered in the analysis, a gene had to be expressed in at least 10% of the single cells from at least one of the two groups for that cell type and there had to be at least 1.1-fold change in gene expression between the groups. Multiple testing correction was done using the Benjamini-Hochberg method to estimate FDR.
For ASC, CP-1 and CP-2 populations shared between both the young and aged mice, the DEGs were identified by comparing the two age groups and then subjected to pathway enrichment analysis. Since CP-A is only present in the aged group, the CP-A marker genes in the aged condition were used as DEGs and subject to pathway enrichment analysis. Pathways from HALLMARK 5 were assessed for overlap with the DEGs using Fisher's exact test and corrected for multiple testing using the Benjamini-Hochberg method to estimate FDR.

Single-cell trajectory analysis
To infer cell lineages of complex communities of heterogeneous cells, pseudo-temporal reconstruction was performed by two methods. The Slingshot R package 6 was used first to infer multiple branching lineages based on the minimum spanning tree followed by simultaneous principal curves. The Cytotrace R package 7 was then used to estimate the developmental potential of ten cell sets (five APC subtypes in "aged" and "young" conditions) based on feature selected gene-count measures.

Cell-cell communication analysis
Both the Cellphone DB database 8 and the iTALK database 9 were used to curate ligand-receptor pairs that may mediate intercellular cross-talk between the APC subtypes and immune cell types.
Pathway enrichment of the ligand-receptor pairs between cell types were further analyzed based on the overlap with selected pathways from KEGG 10 , Reactome 11 , BIOCARTA 12 , and All the images and videos were made using Imaris. Quantification of mature adipocyte number and volume were analyzed by the spot function of Imaris. The function parameters were kept the same between different samples.

In vitro 3D differentiation and differentiation of APCs and APC subpopulations
Lin-SVF or enriched ASC, CP-1, and CP-A populations were isolated from gWAT of Rosa26-loxP-STOP-loxP-mT/mG mice that were 12-month-old or 2.5-month-old. The cells were cultured as described previously 13

In vitro 3D macrophage and APC co-culture
Lin-SVF were isolated from gWAT of Rosa26-loxP-STOP-loxP-mT/mG mice that were 12month-old or 2.5-month-old. Macrophages were isolated from gWAT of C57BL/6 mice that were 12-month-old or 2.5 month-old by FACS. "Aged" Lin-SVF and "aged" macrophages, "aged" Lin-      total APCs (all Tomato+ cells that seeded) and GFP+ cells (BODIPY+, adipocytes). The adipogenic cocktail was added 3 days post-seeding. Other experimental conditions are the same as described in Fig. 4c. n=2 per group. All data represent the mean ± s.e.m.