Research Article

Identification of a mesenchymal progenitor cell hierarchy in adipose tissue

See allHide authors and affiliations

Science  26 Apr 2019:
Vol. 364, Issue 6438, eaav2501
DOI: 10.1126/science.aav2501

A singular focus on fat

Fatty tissue can expand in two ways: through increases in the size of individual adipocytes or through increases in the number of adipocytes. The former process promotes metabolic disease, and the latter protects against it. Merrick et al. used single-cell RNA sequencing to define the hierarchy of mesenchymal progenitor cells that give rise to adipose tissue in mice and humans (see the Perspective by Chau and Cawthorn). They found that progenitor cells expressing a protein called DPP4 give rise to two distinct types of preadipocytes in response to different signals. The DPP4 progenitors reside in a fluid-filled network of collagen and elastin fibers surrounding adipose tissue. In principle, therapeutic interventions that increase progenitor cell differentiation into adipocytes could ameliorate metabolic disease.

Science, this issue p. eaav2501; see also p. 328

Structured Abstract

INTRODUCTION

Adipose tissue is a highly plastic organ that plays a central role in regulating whole-body energy metabolism. The capacity for adipocytes to store chemical energy in lipid droplets protects other organs against the toxic effects of ectopic lipid deposition. In the setting of chronic nutritional excess, such as obesity, adipose tissue expands through increases in fat cell size (hypertrophy) and/or fat cell number (hyperplasia). Adipocyte hypertrophy causes fibrosis and inflammation, thereby promoting metabolic disease. Conversely, hyperplastic growth, mediated by the differentiation of progenitor cells into new adipocytes, is critical for preserving adipose tissue function and protecting against metabolic disease. However, the precise cell types, lineage dynamics, and mechanisms governing the development of adipocytes are incompletely understood.

RATIONALE

Adipose mesenchymal progenitors constitute a heterogeneous pool of diverse cell types. Previous attempts to characterize these cells have relied on nonspecific mesenchymal markers and candidate lineage marker genes. We used single-cell RNA sequencing to identify distinct types of progenitor cells in murine and human adipose tissues and to predict lineage relationships in an unbiased manner. Functional assessments of these cell types in vitro and in vivo define a mesenchymal cell hierarchy involved in adipocyte formation.

RESULTS

Single-cell RNA sequencing and cell trajectory analyses identified a lineage hierarchy consisting of several distinct mesenchymal cell types in murine and human adipose. Dipeptidyl peptidase–4–expressing (DPP4+) cells are highly proliferative and multipotent progenitors that are relatively resistant to differentiation into adipocytes. Intercellular adhesion molecule–1–expressing (ICAM1+) cells are committed preadipocytes that express Pparg and are poised to differentiate into mature adipocytes with minimal stimulation. CD142+/Clec11a+ cells represent a distinct adipogenic population in murine subcutaneous adipose that shares many properties with ICAM1+ preadipocytes. In vivo cell transplantation studies showed that DPP4+ progenitors give rise to both ICAM1+ and CD142+ preadipocytes before differentiation into mature adipocytes. DPP4+ cells depend on transforming growth factor–β (TGFβ) signaling to maintain their progenitor identity, whereas ICAM1+ preadipocytes are refractory to the proliferative and anti-adipogenic actions of the TGFβ pathway. Obesity and insulin resistance lead to a depletion of DPP4+ mesenchymal progenitors and a reduction in the adipogenic differentiation competency of preadipocytes, specifically in visceral white adipose tissue. Single-cell analysis of human subcutaneous adipose tissue revealed distinct DPP4+ and ICAM1+ populations that displayed functional properties similar to those of the analogous mouse populations. Histological examination of murine subcutaneous adipose tissue showed that ICAM1+ preadipocytes are intercalated between mature adipocytes, occupying a perivascular niche. The DPP4+ progenitor cells are localized in an anatomically distinct niche surrounding the adipose depot, which we term the reticular interstitium.

CONCLUSION

Our studies define a developmental hierarchy of adipose progenitors consisting of DPP4+ interstitial progenitors that give rise to committed ICAM1+ and CD142+ preadipocytes, which are poised to differentiate into mature adipocytes. Targeting one or more of these cell populations may be beneficial for promoting adaptive hyperplastic adipose growth to ameliorate metabolic disease. A key finding from this work is that adipose progenitor cells reside in the reticular interstitium, a recently appreciated, but not well-studied, fluid-filled network of collagen and elastin fibers that encases many organs, including adipose depots. Our results raise the possibility that DPP4+ cells, in addition to serving as progenitor cells for adipocytes in fat depots, may play important roles in the development and regeneration of other tissues.

DPP4+ progenitors reside in the reticular interstitium and give rise to committed preadipocytes.

DPP4+ multipotent progenitor cells give rise to ICAM1+ and CD142+ committed preadipocytes, which are poised to differentiate into mature adipocytes (top). Committed preadipocytes (green) are intercalated between mature adipocytes, whereas DPP4+ progenitors (red) reside in the reticular interstitium, an anatomically distinct fluid-filled network of collagen and elastin fibers that encases many organs, including adipose depots.

Abstract

Metabolic health depends on the capacity of adipose tissue progenitor cells to undergo de novo adipogenesis. The cellular hierarchy and mechanisms governing adipocyte progenitor differentiation are incompletely understood. Through single-cell RNA sequence analyses, we show that the lineage hierarchy of adipocyte progenitors consists of distinct mesenchymal cell types that are present in both mouse and human adipose tissues. Cells marked by dipeptidyl peptidase–4 (DPP4)/CD26 expression are highly proliferative, multipotent progenitors. During the development of subcutaneous adipose tissue in mice, these progenitor cells give rise to intercellular adhesion molecule–1 (ICAM1)/CD54–expressing (CD54+) committed preadipocytes and a related adipogenic cell population marked by Clec11a and F3/CD142 expression. Transforming growth factor–β maintains DPP4+ cell identity and inhibits adipogenic commitment of DPP4+ and CD142+ cells. Notably, DPP4+ progenitors reside in the reticular interstitium, a recently appreciated fluid-filled space within and between tissues, including adipose depots.

White adipose tissue (WAT) stores excess nutritional energy in the form of triglycerides, which can be released for use in other tissues during periods of energy demand, such as fasting and physical activity. The capacity for adipocytes to store chemical energy in lipid droplets protects other organs against the toxic effects of ectopic lipid deposition. Adipose tissue expands by increases in fat cell size (hypertrophy) and/or number (hyperplasia). Hyperplastic growth, mediated by the differentiation of progenitor cells (adipogenesis), is critical for proper adipose tissue function (13). Defects in this process cause adipose fibrosis and inflammation, contributing to systemic insulin resistance (4). Conversely, enhancing adipogenesis prevents maladaptive adipocyte hypertrophy, reduces ectopic lipid accumulation, and ameliorates metabolic disease (57).

Adipose progenitor cells have been isolated on the basis of their expression of common progenitor or mesenchymal cell surface markers, including platelet-derived growth factor receptor–α (PDGFRα), CD29, CD34, stem cell antigen–1 (SCA1)/LY6A, and CD24 (820). In particular, CD24-expressing (CD24+) progenitor cells, which reside in the vasculature, efficiently form functional adipocytes when transplanted into lipodystrophic mice (21). Preadipocyte factor–1 (PREF1) [also called delta-like homolog–1 (DLK1)] and the adipose lineage marker peroxisome proliferator–activated receptor–γ (PPARγ) are also reported to mark adipogenic cells that are closely associated with blood vessels in adipose tissue (18, 19). Lastly, mural and smooth muscle–related cells that express Acta2 (encoding smooth muscle actin), Myh11, or Pdgfrβ contribute to adipocyte formation under certain conditions (5, 10, 16, 22, 23). However, the overlap, heterogeneity, and developmental interrelationships among these and other described populations are incompletely understood.

Profiling and trajectory analysis of adipose progenitors

We applied single-cell RNA sequencing (RNA-seq) to identify and profile progenitor cells in an unbiased manner from the developing subcutaneous inguinal WAT (iWAT) of 12-day-old (p12) mice. At this stage, adipocytes are rapidly developing from progenitor cells, allowing us to capture the continuum of cell states spanning differentiation. Mature lipid-laden adipocytes, which are incompatible with the downstream analysis, were separated from stromal vascular cells (SVCs) by centrifugation. SVCs were then depleted of CD45+ leukocytes and subjected to single-cell RNA-seq (fig. S1). Unsupervised clustering of the gene expression profiles identified 10 cell types (Fig. 1A).

Fig. 1 Single-cell RNA-seq and cell trajectory analysis delineate the lineage hierarchy of adipocyte progenitors.

(A) Unsupervised clustering of 11,423 cells (mean number of genes per cell = 1977) from the subcutaneous WAT of p12 pooled male and female C57BL/6J mouse pups reveals 10 distinct cell groups represented on a tSNE map (relevant marker genes are listed in parentheses). (B) Individual gene tSNE and violin plots showing the expression levels and distribution of representative marker genes. The y axis is the log-scale normalized read count. (C) Pseudotemporal cell ordering of groups 1 to 4 and adipocytes along differentiation trajectories by using Monocle. Pseudotime (arbitrary units) is depicted from dark to light blue (left). Group identities were overlaid on the pseudotime trajectory map (right).

Canonical mesenchymal progenitor markers Cd34, Pdgfra, Ly6a (Sca1), and Thy1 (Cd90) were expressed predominantly in groups 1 to 3 (figs. S2 and S3). Group 1 cells (blue), which we named “interstitial progenitors,” were marked by the expression of Dpp4 [encoding dipeptidyl peptidase–4 (DPP4)], Wnt2, Bmp7, and Pi16 but did not express adipocyte markers (Fig. 1B and figs. S2 and S3). Group 2 cells expressed Icam1 [encoding intercellular adhesion molecule–1 (ICAM1)] and Dlk1 (Pref1), along with several adipocyte identity genes, including Pparg, Fabp4, and Cd36, suggesting that these cells could be “committed preadipocytes” (Fig. 1B and figs. S2 and S3). Group 3 cells expressed Clec11a and Fmo2, as well as tissue factor (F3/CD142), which is a marker of “adipogenesis-regulatory cells” (Aregs) that were recently reported to inhibit adipogenesis (24). A related cell cluster (group 4) was marked by Wnt6 and Sfrp5 expression but did not show detectable expression of mesenchymal marker genes, such as Pdgfra or Thy1. Group 7 cells (yellow) were classified as “adipocytes” because of their high expression of mature adipocyte–specific genes (e.g., Adipoq, Plin1, and Car3) and lack of expression of progenitor markers. These cells likely represent newly formed adipocytes without substantial lipid stores that copurified with the stromal cell fraction. The SVC compartment of adult adipose tissue displays similar cellular diversity, including prominent populations of mesenchymal cell groups 1 to 3 (fig. S4).

We examined potential precursor-product relationships between the identified mesenchymal populations by using in silico cell trajectory analyses. Pseudotemporal analysis predicted that group 1 interstitial progenitors have two developmental trajectories: group 2 preadipocytes terminating as group 7 adipocytes (cell fate A) and group 3 and 4 cells (cell fate B) (Fig. 1C and fig. S5). Group 3 cells appear to represent an intermediary cell type between group 1 interstitial progenitors and group 4 cells. Altogether, these analyses suggest that group 1 progenitors (Dpp4+) provide a source of both group 2 committed preadipocytes and group 3 Cd142+ cells.

DPP4+ cells are multipotent mesenchymal progenitors

To investigate the function of these cell populations, we developed fluorescence-activated cell sorting (FACS) strategies for the isolation of group 1 [DPP4+, CD142(−)] cells (hereafter referred to as DPP4+ cells), group 2 [ICAM1+, CD142(−)] cells (hereafter referred to as ICAM1+ cells), and group 3 (CD142+) cells from murine subcutaneous adipose tissue (see fig. S14 for the detailed sorting strategy). The iWATs from p12 and adult (8- to 10-week-old) mice contained similar proportions of DPP4+, ICAM1+, and CD142+ cells (fig. S6). RNA-seq profiling of freshly sorted DPP4+, ICAM1+, and CD142+ cells from p12 pups showed that these cells expressed the same marker gene signatures as their respective single-cell groups (fig. S7), validating the cell purification methods. We next assayed the differentiation and proliferative activities of these populations from adult mice, with treatments initiated within 48 hours of cell isolation. All three populations underwent robust adipocyte differentiation and activated adipocyte-specific genes to comparably high levels when treated with the standard complete cocktail of adipogenic inducers (Fig. 2, A and C). However, under minimal adipogenic conditions, where the cocktail contained only a low concentration of insulin, ICAM1+ and CD142+ cells differentiated efficiently into adipocytes (a result similar to that seen with the complete cocktail) whereas DPP4+ cells displayed very low adipogenic capacity (Fig. 2B). ICAM1+ and CD142+ cells expressed adipocyte-specific genes at or near the levels observed in primary adipocytes isolated from adipose tissue, whereas DPP4+ cultures displayed little to no induction of these genes (Fig. 2C).

Fig. 2 DPP4+ progenitors display enhanced proliferative and multilineage differentiation capacity.

Cells were isolated by using the following FACS strategy: Lin− (CD45−, CD31−) cells were stained with anti-DPP4, anti-ICAM1, and anti-CD142. Groups were gated as follows: CD142+ cells were selected first, followed by DPP4+ (CD142−, DPP4+) or ICAM1+ (CD142−, ICAM1+) cells. (A and B) Staining of adipocytes (with Bodipy lipid stain) (green) and quantification of adipogenesis in cell cultures from adult CD1 mice after exposure to the complete adipogenic differentiation cocktail (A) or insulinonly (min) (B) [n = 3 biological replicates (BRs) per condition]. (C) mRNA levels of adipocyte-specific genes in cultures from (A) and (B). Primary adipocytes (adipo) purified directly from adipose tissue were included for reference. (D) Quantification of cellular growth (representative of 3 BRs). (E) mRNA levels of osteocyte-specific genes in cultures exposed to osteogenic differentiation inducers (n = 5 BRs). Statistical testing: not significant, P > 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Dots represent BRs, and error bars indicate SEM. Scale bars, 50 μM.

Classical features of mesenchymal progenitor cells include a capacity for multilineage differentiation and high proliferative activity. We found that DPP4+ cells proliferated at a higher rate than ICAM1+ or CD142+ cells (Fig. 2D). Furthermore, the DPP4+ cell population displayed enhanced competence for differentiation into osteocytes, with higher induction of osteocyte-specific marker genes (Fig. 2E). Together, these data identify DPP4+ cells as highly proliferative multipotent progenitors possessing many properties attributed to mesenchymal stem cells. By contrast, ICAM1+ and CD142+ cells are relatively restricted to the adipocyte lineage.

TGFβ signaling maintains DPP4+ progenitor cell identity

To identify signaling pathways regulating the divergent activities of DPP4+ and ICAM1+ cells, we compared the bulk transcriptomes of freshly sorted DPP4+ cells and ICAM1+ cells by RNA-seq. Gene ontology (GO) analysis identified enrichment of the anti-adipogenic transforming growth factor–β (TGFβ) and WNT signaling pathways in DPP4+ cells (Fig. 3A) (8, 25). To assess the importance of TGFβ signaling for DPP4+ cell activity, we treated freshly isolated DPP4+ cells with either recombinant TGFβ or SB431542, a potent and specific TGFβ receptor inhibitor. TGFβ treatment induced the expression of many group 1 marker genes, including Dact2, Wnt10b, and Ptgs2 (Cox2), while also suppressing group 2 and adipogenic marker genes (Fig. 3B). TGFβ receptor inhibition had the opposite effect, including strongly inducing group 2 and adipocyte genes in DPP4+ progenitor cells (Fig. 3B).

Fig. 3 TGFβ regulates the proliferation and adipogenic differentiation competency of multipotent DPP4+ progenitors.

(A) RNA-seq and GO analysis of signaling pathways enriched in DPP4+ versus ICAM1+ cells from pooled p12 pups (n = 3 BRs). Combined score = log P value multiplied by the z-score of deviation from the expected ranking. (B) mRNA levels of group 1, group 2, and adipocyte (adipo) marker genes in DPP4+ cells treated with vehicle control, TGFβ, or the TGFβ receptor inhibitor SB431542 (n = 4 BRs). (C) Quantification of cell growth in cultures treated with TGFβ or SB431542 (representative of 3 BRs). (D) Bodipy staining of adipocytes (green) differentiated with the complete induction cocktail with or without TGFβ treatment. Relative adipogenesis is the adipogenic index of TGFβ-treated cells relative to that of control cells (right) (n = 3 BRs). (E) Fold changes in mRNA levels of adipocyte-specific genes in cultures from (D). (F) Bodipy staining of adipocytes (green) and quantification of differentiation in the indicated cultures (right) (n = 3 BRs). Relative adipogenesis is the adipogenic index of SB431542-treated cells (SB cpd) relative to that of control cells. Min, minimal cocktail (insulin only). (G) Adipocyte-specific mRNA levels in cultures from (F), displayed as fold change in treated cells relative to control cells. Scale bars, 50 μM. Statistical testing: not significant (ns), P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Dots represent BRs, and error bars indicate SEM.

At a functional level, TGFβ treatment augmented the proliferation of DPP4+ cells, whereas TGFβ receptor antagonism suppressed proliferation down to the levels of ICAM1+ cells (Fig. 3C). TGFβ treatment inhibited adipocyte differentiation and suppressed the induction of adipocyte-specific genes in DPP4+ cells exposed to the complete induction cocktail (Fig. 3, D and E). By contrast, TGFβ only mildly suppressed the adipogenic conversion of ICAM1+ cells and exerted an intermediate inhibitory effect on CD142+ cells (Fig. 3, D and E). Reciprocally, the inhibition of TGFβ signaling enhanced adipocyte differentiation of DPP4+ cells exposed to minimal adipogenic conditions but had no observable effect on ICAM1+ or CD142+ cells (Fig. 3, F and G). Thus, TGBβ signaling regulates the identity and functional character of DPP4+ cells. Moreover, these results show that ICAM1+ cells are especially refractory to the anti-adipogenic effects of TGFβ, reinforcing the notion that, among the mesenchymal cell populations, ICAM1+ cells are the most highly committed to becoming adipocytes.

Adipose tissue is organized into many discrete depots that are distributed throughout the body. Flow cytometry analysis showed that the three major mesenchymal cell populations (DPP4+, ICAM+, and CD142+) identified in iWAT were also present in subcutaneous axillary WAT (axWAT), interscapular brown adipose tissue (iBAT), and visceral epididymal WAT (eWAT) of adult mice (figs. S8 to S10). Of note, the relative proportion of DPP4+ cells was much lower in visceral WAT than in subcutaneous depots (iWAT or axWAT) (figs. S8 to S10). The cell populations isolated from axWAT, eWAT, and iBAT displayed the same pattern of adipogenic activity as those from iWAT (fig. S8 to S10). Regardless of their depot of origin, ICAM1+ cells had greater adipogenic competence and were more resistant to the inhibitory effects of TGFβ than DPP4+ cells (Fig. 3 and figs. S8 to S10). These results suggest that DPP4+, ICAM1+, and CD142+ mesenchymal cells have conserved roles in all the major adipose depots.

Reduced activity of visceral progenitors in obese mice

We next examined the responses of these cell populations to adverse metabolic conditions. DPP4+, ICAM1+, and CD142+ cells were isolated from mice with diet-induced obesity that were glucose intolerant (fig. S11A). The visceral eWAT from obese animals had a lower proportion of DPP4+ cells and a higher proportion of CD142+ cells than that from lean controls (fig. S11B). ICAM1+ and CD142+ cells from the visceral depots of obese mice were less proliferative and had lower adipogenic differentiation capacity than corresponding cells from lean control animals (fig. S11, C and D). By contrast, ICAM1+ and CD142+ cells from the subcutaneous iWAT of obese mice retained high adipocyte differentiation competency (fig. S11, E to G). These results imply that high-fat feeding and/or obesity depletes the DPP4+ mesenchymal progenitor pool and impairs the adipogenic differentiation competency of preadipocytes, specifically in visceral WAT. Thus, progenitor cell exhaustion and reduced precursor differentiation may contribute to the pathologic remodeling of visceral WAT, which is linked to metabolic disease progression.

Human adipose tissue contains analogous stromal populations

An important question is whether analogous populations of adipose stromal cells are present in humans. To address this, we used FACS to identify and purify DPP4+, ICAM1+, and CD142+ cells from fresh subcutaneous adipose tissue of human subjects undergoing cosmetic plastic surgery (Fig. 4A). Human adipose tissue contained proportionally fewer DPP4+ cells than mouse adipose tissue (Fig. 4B). Human DPP4+ cells, like their mouse counterparts, displayed higher proliferative activity (Fig. 4C) and lower potential to differentiate into adipocytes than both ICAM1+ and CD142+ cells, especially under conditions of minimal adipogenic stimulation (Fig. 4, D and E). Activation of TGFβ signaling enhanced proliferation and profoundly inhibited adipocyte differentiation in human DPP4+ cells, while having less effect on ICAM1+ and CD142+ cells (fig. S12).

Fig. 4 Human adipose contains analogous populations of DPP4+ progenitors and ICAM1+ preadipocytes.

(A) FACS isolation of cell populations from human adipose tissue (representative of eight individual donors). Populations were purified as follows: Lin− (CD45−, CD31−) cells were stained with anti-DPP4, anti-ICAM1, and anti-CD142. CD142+ cells were selected first, followed by DPP4+ (CD142−, DPP4+) and ICAM1+ (CD142−, ICAM1+) cells. DP, double positive (DPP4+, ICAM1+); DN, double negative (DPP4−, ICAM1−); FSC, forward scatter; k, thousand. (B) Quantification of relative progenitor abundance from n = 8 human tissue donors. (C) Relative proliferation rates of human cells (n = 5 donors). (D) Bodipy staining of adipocytes (green) in cultures treated with complete (top) or minimal (insulin plus rosiglitazone) (bottom) induction cocktail. (E) Quantification of adipogenic differentiation from cells shown in (D) (n = 6 donors). (F) Unsupervised clustering of 11,338 cells (mean number of genes per cell = 1253) from the abdominal subcutaneous adipose tissue of a 31-year-old female donor (body mass index, 31.6) reveals five distinct cell groups represented on a tSNE map (relevant marker genes are in parentheses). (G) Individual gene tSNE and violin plots showing the expression levels and distribution of representative marker genes. The y axis is the log-scale normalized read count. Statistical testing: ns, P > 0.05; *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001. Dots represent BRs, and error bars indicate SEM. Scale bars, 50 μM.

To investigate the heterogeneity of human adipose stromal cells in an unbiased manner, we performed single-cell expression profiling of CD45-depleted SVCs from human abdominal subcutaneous WAT. Clustering analysis of gene expression profiles identified five cell populations, including two main groups of cells (groups 1 and 2) that expressed mesenchymal markers PDGFRa, PDGFRb, and SCA1. Group 1 cells selectively expressed DPP4, CD55, and WNT2, comparable to murine group 1 cells (Fig. 4, F and G, and fig. S13). Group 2 cells expressed ICAM1, PPARg, and GGT5, comparable to murine group 2 cells. However, murine group 3 markers CD142, CLEC11A, and FMO2 were broadly expressed across human groups 1 and 2 (Fig. 4, F and G, and fig. S13). This suggests that human subcutaneous adipose tissue does not contain a distinct population of CD142+ cells. In support of this, we did not detect any functional differences between sorted human ICAM1+ and CD142+ cells in our assays. We conclude that human and murine WATs contain comparable DPP4+ and ICAM1+ cell populations with similar functional attributes.

DPP4+ progenitors produce ICAM1+ and CD142+ preadipocytes

To determine the in vivo fate of DPP4+, ICAM1+, and CD142+ cells in adipose tissue, we transplanted mTomato-expressing DPP4+, ICAM1+, or CD142+ cells into the developing fat pads of wild-type (WT) animals (Fig. 5 and fig. S14). Flow cytometry analyses of stromal cells isolated from recipient animals showed that DPP4+ cells acquired ICAM1 and CD142 expression as early as day 7 posttransplant and that a subset of the implanted cells down-regulated DPP4 expression by day 14 (Fig. 5A and fig. S15A). Transplanted ICAM1+ cells acquired expression of CD142 but did not produce substantial numbers of DPP4+ cells (Fig. 5B and fig S15B). Transplanted CD142+ cells did not acquire DPP4 expression, but a subset of these cells down-regulated CD142 and retained ICAM1 expression (Fig. 5C). Histologic examination of the transplants after 50 days revealed robust development of mTomato+ mature adipocytes from DPP4+, ICAM1+, and CD142+ cells (Fig. 5, right panels). These results demonstrate that DPP4+ cells act as progenitors for both ICAM1+ and CD142+ cells, with these populations further differentiating into mature adipocytes. Moreover, the results suggest that CD142+ and ICAM1+/CD142− cells interconvert in vivo.

Fig. 5 DPP4+ progenitors give rise to ICAM1+, CD142+ cells and mature adipocytes in vivo.

(A) Sort-purified CD142(−), DPP4(+) and (B) CD142(−), ICAM1(+) cells from mTomato(+) donor mice were analyzed before transplantation (day 0) (left) into the subcutaneous adipose of 10-day-old mTomato(−) recipient mice. Seven and 14 days after transplantation, mTomato(+) cells were recovered and analyzed for the expression of DPP4, ICAM1, and CD142 (shown is one representative transplant from n = 4 BRs). Fifty days after transplantation, mTomato+ cells were visualized by immunofluorescence in recipient iWAT (right) (representative image; n = 3 to 4 BRs). AF, autofluorescence (525/50 nM) showing host adipocytes. (C) Sort-purified CD142(−) cells were analyzed 14 days posttransplant for the expression of DPP4, ICAM1, and CD142 (shown is one representative transplant from n = 4 BRs). The left FACS plot depicts the CD142-positive gate, and the right FACS plot shows analysis of cells from the CD142-negative gate.

DPP4+ progenitors reside in the reticular interstitium

Lastly, given the divergent genetic and functional properties of DPP4+ and ICAM1+ cells, we examined the anatomic relationship of these populations by immunofluorescence staining for DPP4 and PREF1. PREF1 is another marker of group 2 (ICAM1+) cells in developing adipose tissue that was also previously defined as a preadipocyte marker (Fig. 1) (26). Histological examination of transverse sections of iWAT from 2-day-old pups revealed a marked anatomic partitioning of DPP4+ (group 1) and PREF1+ (group 2) cells (Fig. 6A). PREF1+ cells were intercalated between mature adipocytes, distributed throughout the central body of the iWAT. By contrast, DPP4+ cells were localized in the reticular interstitium (RI) (Fig. 6A, inset 2), a recently appreciated fluid-filled network of collagen and elastin fibers that encases many organs, including adipose depots (27). Many cells coexpressing DPP4 and PREF1 were found at the leading edge of p2 iWAT, potentially representing DPP4+ cells in the process of transiting to ICAM1+ cells during adipogenesis (Fig. 6A, inset 1). At even earlier stages of development, doubly positive cells were more prominent throughout the body of the nascent iWAT depot (fig. S16). Staining in adult murine adipose tissue for DPP4 along with the other group 1 markers ANXA3 and CD55 shows a similar localization of these cells in the RI (fig. S17). Previous reports show that preadipocytes and CD142+ cells occupy a perivascular niche, an anatomically distinct structure, characterized by a collagen-rich extracellular matrix, similar to the RI (16, 18, 24). Notably, DPP4+ interstitial progenitors are excluded from the perivascular compartment and lack the expression of classic pericyte markers (fig. S18).

Fig. 6 DPP4+ progenitor cells reside in the RI.

(A) Full-thickness flank tissue from a 2-day-old mouse pup was cross-sectioned. Murine skin and adipose were stained with hematoxylin and eosin (H&E) with labels localizing the dermis (D), dermal fibroblasts (DF), the panniculus carnosus (PC), the RI, and adipocytes (Ad). Anti-DPP4 (red), anti-PREF1 (green), and DAPI (blue) were used for immunofluorescence. (Inset 1) Magnification of the leading edge of the developing iWAT. (Inset 2) Cross-sectional magnification of the body of the iWAT. Arrows show examples of DP (DPP4+, PREF1+) cells. The arrowheads point to DPP4+, PREF1(−) cells. (B) Model depicting the lineage hierarchy relationships of the indicated cell types. (C) Model depicting the anatomical relationships of the indicated cell types at the leading edge of developing adipose tissue.

Discussion

In this study, we identified and functionally assessed three major mesenchymal cell groups in murine adipose: group 1 DPP4+ interstitial progenitor cells, group 2 ICAM1+ preadipocytes, and group 3 CD142+/Clec11a+ cells. Recent reports by other investigators using similar methods identified analogous cell groups (24, 28, 29), demonstrating that these cell populations are present across a wide variety of adipose sources from animals of different ages and metabolic statuses. The high overall concordance between studies highlights the robustness of single-cell RNA-seq technology and strengthens the notion that the identified cell groups represent a general paradigm for adipose progenitor biology.

Burl et al. performed single-cell RNA-seq on SVCs from adult mouse iWAT and eWAT and identified populations similar to those we found in p12 pups (28). Specifically, the group-defining genes for ASC2 (DPP4+/Pi16+) cells overlap extensively with the genes defining our group 1 (DPP4+) population, and the ASC1 (ICAM1+/Col4a2+) group is similar to our group 2 (ICAM1+) cells (fig. S19). Hepler et al. defined two main subtypes of Pdgfrb-expressing (mural) cells in murine eWAT: fibroinflammatory precursors (FIPs), which express many markers overlapping with those of our group 1 DPP4+ cells, and adipocyte precursor cells, which are similar to our group 2 ICAM1+ cells (fig. S20) (29). FIPs from visceral adipose have low adipogenic differentiation capacity. However, the authors noted that the FACS strategy used to isolate FIPs from eWAT (with PDGFRβ+, LY6a+, and CD9+ markers) did not identify an analogous population in subcutaneous adipose. The three markers used to purify FIPs are not specifically enriched in group 1 DPP4+ cells (fig. S20, top three genes), although many other FIP-defining markers are highly concordant between the populations. Therefore, FIPs may represent a subset of DPP4+ cells that are present only in the visceral eWAT depot.

Schwalie et al. identify three cell clusters (24) that largely coincide with the groups that we studied (fig. S21). Schwalie population 1 (P1) expresses a gene signature similar to that of our group 1 (DPP4+) cells, P2 corresponds to our group 2 (ICAM1+) cells, and P3 corresponds to our group 3 (CD142+/Clec11a+) cells. In agreement with Schwalie et al., we found that there was a higher proportion of CD142+ cells in eWAT than in iWAT and in obese versus lean adipose tissue (fig. S11). However, a notable discrepancy concerns the functional properties of CD142+ group 3 and P3 cells. Schwalie et al. show that mouse and human P3 (CD142+) cells, which they named Aregs, have low adipogenic potential and are also capable of inhibiting the differentiation of preadipocytes. By contrast, our study indicates that these cells are highly adipogenic both in vitro and upon transplantation in vivo. This divergence likely stems from key differences in our respective FACS strategies for cell purification. For many of their mouse studies, Schwalie et al. used ABCG1 in addition to CD142 to enrich for the P3 population. Examination of the expression pattern of Abcg1 in our single-cell data shows almost no mRNA detectable in groups 1 to 3 (fig. S21). We performed extensive validation of our flow sorting strategy by bulk RNA-seq analyses, showing that the CD142+ cells we isolated were highly enriched for expression of the corresponding group 3–defining genes (fig. S7). Moreover, our analyses of human adipose reveal that CD142 and ICAM1 are largely coexpressed in the same population of adipogenic cells (Fig. 4). Together, these results demonstrate that group 3 cells are an adipogenic population, related to ICAM1+ group 2 cells, in the adipose lineage. We speculate that the anti-adipogenic activity attributed to P3 CD142+ cells by Schwalie et al. may reside in a different cell type captured in their flow sorting strategy.

In conclusion, we define a developmental hierarchy of adipose progenitors that are active during early murine adipogenesis, and we suggest that this may be a general paradigm for adipogenesis in both mice and humans. In young animals, DPP4+ interstitial progenitors produce CD142+ cells and ICAM1+ preadipocytes that are poised to differentiate into adipocytes (Fig. 6B). We speculate that, in adults, DPP4+ cells undergo adipose lineage commitment in response to various stimuli, providing a renewable source of preadipocytes. It is also conceivable that other cell populations in adult adipose depots, including ICAM1+ or CD142+ cells residing within the perivascular niche, mediate tissue turnover and remodeling without further contribution from DPP4+ cells. Future genetic lineage–tracing studies will ascertain whether DPP4+ cells are the major or exclusive progenitor source of adipocytes during development and in adult animals. A particularly notable finding is that adipose progenitor cells reside in a recently discovered anatomic niche (Fig. 6C). The RI is a large tissue that surrounds many organs but has not been well studied. Our results raise the possibility that in addition to serving as a reservoir for adipocyte progenitor cells in fat depots, the RI plays important roles in the development and regeneration of other tissues. Lastly, the identification of functionally distinct precursor populations could potentially inform the development of more targeted approaches to promote metabolically beneficial adipose growth.

Materials and Methods

Human adipose tissue samples

All human adipose tissue samples were obtained from living donors undergoing cosmetic surgery. Tissue samples consisted of 1- to 10-kg intact blocks of discarded tissue, including skin and subcutaneous adipose, from procedures to remove excess skin and adipose tissue from the abdomen and flanks. Collection of tissue and processing were initiated within 1 to 3 hours of removal from the patient. This work was performed under the approval of Institutional Review Board protocols 812150 and 824825.

Human subject demographics are outlined in fig. S22.

Isolation of SVCs from human subcutaneous adipose

The skin and cauterized edges of human tissue samples were removed to yield an intact block of subcutaneous fat that was unmanipulated by the surgical procedure. Three hundred to five hundred grams of tissue was manually minced and digested with collagenase D (0.75 unit/ml; Roche) and dispase II (1.2 units/ml; Roche) in Dulbecco’s modified Eagle’s medium (DMEM) with Ham’s F12 medium (DMEM/F12) containing 0.4% fatty acid–free bovine serum albumin (Sigma) at 37°C with agitation for 45 min and shaking for 10 s at 15-min intervals. Digestion was performed at a ratio of 1 g of tissue to 1 ml of digestion medium. The digestion was quenched with DMEM containing 10% fetal bovine serum (FBS), and the dissociated cells were filtered twice through a single layer of gauze to remove large undigested particles and then subjected to centrifugation at 400 × g for 5 min at room temperature (RT). The resulting supernatant containing mature adipocytes was aspirated, and the pellet, consisting of SVCs, was resuspended, passed through a 100-μM filter, and subjected to centrifugation at 400 × g for 5 min at RT. Cells were resuspended in red blood cell lysis buffer (Biolegend) for 5 min at RT and then quenched in DMEM containing 10% FBS. Cells were then passed through a 40-μM filter and collected by centrifugation at 400 × g for 5 min. Cells were recovered in FACS buffer [Hanks’ balanced salt solution (HBSS) containing 3% FBS; Fisher] and kept on ice for the duration of processing.

Animals

C57BL/6J WT [stock number (no.) 000664] and C57BL/6J:ROSAmT/mG (stock no. 007676) mouse lines were obtained from Jackson Laboratories. CD1 WT mice (stock no. 022) were obtained from Charles River. 129SVE mice (stock no. 129S6/SvEvTac) were obtained from Taconic. Mice with diet-induced obesity (stock no. 380050) and control animals (stock no. 380056) were obtained from Jackson Laboratories and maintained on their respective diets until harvest. All animal procedures were performed under the guidance of the University of Pennsylvania Institutional Animal Care and Use Committee.

Isolation of SVCs from mouse adipose

Inguinal and axillary subcutaneous white adipose depots were surgically removed from WT mice and processed for SVC enrichment. Briefly, adipose tissues were manually minced and digested with collagenase D (1.5 units/ml; Roche) and dispase II (2.4 units/ml; Roche) in DMEM/F12 containing 0.8% fatty acid–free bovine serum albumin (Sigma) at 37°C with agitation for 45 min and vortexing for 10 s at 15-min intervals. The digestion was quenched with DMEM/F12 containing 10% FBS, and the dissociated cells were passed through a 100-μM filter and then subjected to centrifugation at 400 × g for 5 min. The resulting supernatant containing mature adipocytes was aspirated, and the pellet, consisting of SVCs, was resuspended in red blood cell lysis buffer (Biolegend) for 5 min at RT and then quenched in DMEM/F12 containing 10% FBS. Cells were then passed through a 40-μM filter and collected by centrifugation at 400 × g for 5 min. Cells were recovered in FACS buffer (HBSS containing 3% FBS; Fisher).

FACS

Mouse studies

SVCs from the subcutaneous adipose of ~20 to 30 mice (p10 to p20) or 10 to 20 mice (8 weeks old or older) were pooled and resuspended in FACS buffer for incubation with the following antibodies for 30 min at 4°C: CD26 (DPP4)–fluorescein isothiocyanate (FITC) (Biolegend, San Diego, CA; catalog no. 137806; 1:200), anti–mouse ICAM1–phycoerythrin (PE)/Cy7 (Biolegend catalog no. 116122; 1:100), anti–mouse CD45–allophycocyanin (APC)/Cy7 (Biolegend catalog no. 103116; 1:1000), anti–mouse CD31–APC-Fire (Biolegend catalog no. 102528; 1:1000), and anti–mouse CD142 (Sino Biological catalog no. 50413-R001; 1:100). Antibodies were preconjugated with Biotium Mix-n-Stain CF647 (Sigma catalog no. MX647S100). 4′,6-Diamidino-2-phenylindole (DAPI) (Roche catalog no. 10236276001; 1:10,000) was added for the final 5 min. The cells were washed three times with FACS buffer to remove unbound antibodies. The cells were sorted with a BD FACS Aria cell sorter (BD Biosciences) equipped with a 100-μm nozzle and the following lasers and filters: DAPI, 405 and 450/50 nm; FITC, 488 and 515/20 nm; mTomato, 532 and 610/20 nm; PE/Cy7, 532 and 780/60 nm; CF647, 640 and 660/20 nm; and APC/Cy7 and APC-Fire, 640 and 780/60 nm. All compensation was performed at the time of acquisition in Diva software by using compensation beads (BioLegend catalog no. A10497) for single-color staining and SVCs for negative staining and fluorescence (DAPI and mTomato).

Human studies

SVCs were resuspended in FACS buffer for incubation with the following antibodies for 60 min at 4°C: CD26 (DPP4)–FITC (Biolegend catalog no. 302704; 1:100), anti–mouse ICAM1–PE/Cy7 (Biolegend catalog no. 353116; 1:100), anti–mouse CD45–APC/Cy7 (Biolegend catalog no. 368516; 1:100), anti–mouse CD31–APC/Cy7 (Biolegend catalog no. 303120; 1:100), and anti–mouse CD142–APC (Biolegend catalog no. 365206; 1:100). DAPI (Roche catalog no. 10236276001; 1:10,000) was added for the final 5 min. The cells were washed three times with FACS buffer to remove unbound antibodies. The cells were sorted with a BD FACS Aria cell sorter (BD Biosciences) equipped with a 100-μm nozzle and the following lasers and filters: DAPI, 405 and 450/50 nm; FITC, 488 and 515/20 nm; PE/Cy7, 532 and 780/40 nm; APC, 640 and 660/20 nm; and APC/Cy7, 640 and 780/60 nm. All compensation was performed at the time of acquisition in Diva software by using compensation beads (BioLegend catalog no. A10497) for single-color staining and SVCs for negative staining and fluorescence (DAPI).

Single-cell RNA-seq with 10× Genomics chromium platform

For both mouse and human studies, SVCs were isolated and flow sorted with gating to isolate single cells away from debris, doublets, and dead cells. For the p12 pups (pooled male and female C57BL/6 mice) and the human single-cell study, the cells were further gated against CD45 to exclude leukocytes. For the adult mouse thermoneutral (pooled male 129S6/SvEvTac) sample, CD45 cells were segregated and remixed with the SVCs at a ratio of ~20% of the total cells. The sorted cells were loaded onto a GemCode instrument (10× Genomics, Pleasanton, CA) to generate single-cell barcoded droplets (GEMs) according to the manufacturer’s protocol with the 10× single-cell 3′ v2 chemistry. The resulting libraries were sequenced on an Illumina HiSeq 2500 instrument with the HiSeq rapid sequencing by synthesis (SBS) kit. The resulting reads were aligned, and gene-level unique molecular identifier (UMI) counts were obtained by using Cell Ranger (Pipeline).

The Cell Ranger single-cell software suite v.2.0.1 (mouse studies) or v.3.0.1 (human studies) was used to perform sample demultiplexing, alignment, filtering, and UMI counting (30). Clustering and gene expression were visualized with the Seurat package (version 3.0) (31) on RStudio (32). Cells were first filtered to have >500 detected genes and less than 5% of total UMIs mapping to the mitochondrial genome. Clusters with very few cells were filtered before downstream analysis. Data were scaled to mitigate the effects of the following variables: the number of genes detected per cell, the percentage of mitochondrial reads, and the cell cycle phase. Dimensionality reduction was performed with the t-stochastic neighboring embedding method (tSNE) by using Seurat. Analysis of differential gene expression among clusters was performed by using the Seurat function FindMarkers with the Wilcox test. Violin plots, heatmaps, and individual tSNE plots for the given genes were generated by using the Seurat toolkit VlnPlot, DoHeatmap, and FeaturePlot functions, respectively.

Pseudotime analysis

Pseudotemporal analysis was performed on a filtered subset of clusters (groups 1 to 4 and adipocytes) from the p12 pups’ subcutaneous SVCs by using the Monocle R package (33). Ordering genes were selected by using a cutoff of expression in at least 10 cells and a combination of intercluster differential expression and dispersion with a q value cutoff of <1 × 10−10, which produced a list of 1027 genes. A split heatmap was generated from selected genes showing significant change through pseudotime, high differential expression, or known biologic identity by using the Monocle function plot_genes_branched_heatmap at branch_point 1.

RNA-seq library preparation and analysis

Subcutaneous SVCs from pooled C57BL/6 male and female p10 to p16 pups were sorted as described above into DPP4+ and ICAM1+ populations (Fig. 2) (GO analysis) or DPP4+, ICAM1+, and CD142+ populations (fig. S10) and collected in TRIzol (Invitrogen). Total RNA was isolated by using the RNeasy microkit (Qiagen). Three independent biological replicates (BRs) were collected for each population. RNA concentration and quality were assessed by using Nanodrop2000, Qbit, and Bioanalyzer RNA 2100 (Agilent; Santa Clara, CA). All samples had an RNA integrity number (RIN) greater than 7. Library preparation and cDNA sequencing were performed by Novogene using paired-end 150–base pair reads (20 million reads per sample). FASTQ files were aligned by using STAR, and differential gene expression was analyzed by using the R package DESeq2.

Cell culture and differentiation

Mouse studies

DPP4+, ICAM1+, and CD142+ populations were FACS purified; plated on Corning CellBind 384-, 96-, 48-, or 24-well plates (Sigma-Aldrich); and cultured in DMEM/F12 containing 10% FBS and Primocin (50 ng/ml) (InvivoGen catalog no. ant-pm-1). The cells were incubated for 24 to 48 hours to facilitate attachment before the induction of either adipogenic or osteogenic differentiation. For adipogenic differentiation, cells were plated at near confluence after sorting to ensure the same number of cells per well for each cell type at the start of adipogenic induction. Adipogenic differentiation was carried out in DMEM/F12 containing 10% FBS and Primocin (50 ng/ml) with the addition of either a full adipogenic cocktail [20 nM insulin, 1 nM T3, 1 μM dexamethasone (Sigma catalog no. D4902), 0.5 μM isobutylmethylxanthine (Sigma catalog no. I7018), and 125 nM indomethacin] (complete) or a minimal adipogenic cocktail (20 nM insulin) (minimal). For the full adipogenic cocktail induction, cells were incubated for 2 days and then transferred to adipogenic maintenance medium (20 nM insulin, 1 nM T3). Medium changes were performed every 2 days, and cells were analyzed for all experiments within 4 to 6 days of the induction of differentiation. Osteogenic differentiation was performed by using the MesenCult osteogenic stimulatory kit (Stem Cell Technologies catalog no. 05504). Medium was changed every 2 days, and the cells were analyzed after 8 days of differentiation.

Human studies

DPP4+, ICAM1+, and CD142+ populations were FACS purified; plated on Corning CellBind 384-, 96-, or 48-well plates (Sigma-Aldrich); and first cultured in PM-1 medium (Zenbio catalog no. PM-1) supplemented with 1 nM recombinant fibroblast growth factor 2 (FGF2) (Invitrogen, PHG0261) and Primocin (50 ng/ml). Cells were plated at near confluence and incubated for 48 to 72 hours to facilitate attachment before the induction of adipogenic differentiation. Adipogenic differentiation was carried out in DM-1 medium (Zenbio catalog no. DM-1) containing 3% FBS and Primocin (50 ng/ml) with the addition of either a full adipogenic cocktail [20 nM insulin, 1 nM T3, 1 μM dexamethasone (Sigma catalog no. D4902), 0.5 μM isobutylmethylxanthine (Sigma, I7018), and 1 nM rosiglitazone maleate (Alexis Biochemicals, 350-103-G001)] (complete) or a minimal adipogenic cocktail (20 nM insulin, 1 nM rosiglitazone) (minimal). For the full adipogenic cocktail induction, cells were incubated for 7 days and then transferred to adipogenic maintenance medium (20 nM insulin, 1 nM T3). Medium was changed every 7 days, and the cells were analyzed after 14 to 18 days of differentiation.

Mouse and human studies

In some experiments, cultures were treated with recombinant TGFβ1 (10 ng/ml) (R&D Systems catalog no. 240-B-002) or the TGFβ inhibitor 10 μM SB431552 (R&D Systems catalog no. 1614) throughout differentiation. TGFβ1 was added at the same time that adipogenic induction occurred, whereas SB431552 was added to cells within 24 hours of plating. The cells were incubated in a humidified incubator at 37°C and 5% CO2.

Histology

Tissues were fixed with 2 to 4% paraformaldehyde via transcardiac perfusion, after which dissected adipose pads were fixed overnight. The tissues were subsequently dehydrated through a series of ethanol washes and then embedded in paraffin for thin sectioning. For cross-sectional imaging of the inguinal adipose depots, a full-thickness section of the mouse flank (including the skin and adipose pad) was fixed and embedded in paraffin and then rotated and sectioned in the cross-sectional orientation. Immunohistochemistry analysis was performed by following heat antigen retrieval methods, and samples were stained with the following antibodies: anti–red fluorescent protein (RFP) (rabbit; 1:500; Rockland), anti-DPP4 (goat; 1:250; R&D Systems), anti-Pref1 (rabbit; R&D Systems), anti-Anxa3 (rabbit; Biorbyt), and anti-Pi16 (rabbit; mybiosource).

Transplantation

DPP4+, ICAM1+, and CD142+ populations were purified by FACS from the subcutaneous adipose of pooled male and female p10 to p16 ROSAmT/mG pups (donor cells). mTomato+ donor cells were washed with FACS buffer twice and then concentrated by centrifugation to ~50,000 cells/μl and mixed 1:1 with Matrigel (phenol free and growth factor reduced; Corning) on ice. WT C57BL/6 mouse pups aged 8 to 12 days were anesthetized using an isoflurane nose cone, and abdominal hair was chemically removed before the creation of an ~5-mm midline Y-shaped cutaneous incision to expose the bilateral inguinal adipose pads. Ten microliters of the donor cell–Matrigel slurry was injected along the edge of the inguinal fat pad in several 2-μl depots. The recipient animals were closed with a 6-0 polypropylene suture and placed back with their litter. After the indicated time, donor cells were harvested from the recipient animals by dissection of the entire inguinal adipose pad, followed by the FACS procedures as described above. BRs here mean separate pools of donor cells transplanted into separate individual recipient mice.

Quantification of adipocyte differentiation and cell proliferation

Adipogenesis was assessed by staining with Biodipy 493/503 (Invitrogen catalog no. D3922) for lipid droplet accumulation and by staining with Hoechst 33342 (Thermo Fisher catalog no. 62249) for nucleus number at 4 to 6 days postinduction (mouse cells) and 14 to 18 days postinduction (human cells) in individual wells of a 384-well plate (Sigma-Aldrich catalog no. CLS3770). The cells were imaged on a Keyence inverted fluorescence microscope (BZX-710) by using DAPI (excitation, 360/40 nm; emission, 460/50 nm; Keyence, OP-87762) and green fluorescent protein (excitation, 470/40 nm; emission, 525/50 nm; Keyence, OP-87763) filters. Individual wells were imaged in their entirety at 20× magnification to capture a 7-by-7 tiled/stitched grid (mouse studies) or at 10× magnification to capture a 5-by-5 tiled/stitched grid (human studies).

Image calculations

Tiling and stitching were performed with Keyence BZ-X Viewer software. Image quantification was performed in ImageJ as shown in fig. S23. Images were split into component channels. Nuclei (blue channel) were quantified by applying Gaussian Blur (3 Sigma), thresholding, performing watershed calculation for segmentation, and counting. Lipid accumulation was quantified by applying Gaussian Blur (2 Sigma), thresholding, and quantifying the total area above the threshold. The level of adipogenesis, expressed as the adipogenic index, was assessed by dividing the total lipid area by the total number of nuclei to obtain a number with arbitrary units for comparing absolute levels of adipogenesis. Relative adipogenesis was calculated as the ratio of the adipogenic indices in treatment and control samples; here, the control is the untreated sample from the same cell type and BR (for example, ICAM1–TGFβ BR A/ICAM1 control BR A). This ratio represents induction or suppression compared with the baseline.

Statistics

All image calculations were first performed on a per-well basis with the same thresholds applied to all wells within a single experiment. Once thresholds were determined, calculations were performed on all images in an automated manner. The adipogenic index for BRs was calculated by averaging the results from multiple technical replicates (2 to 8 wells per cell type per treatment per BR). Here, independent wells (i.e., those with cells derived from the same biological and FACS pool but plated into separate wells of the same plate) are considered technical replicates. Statistical testing was performed on BRs.

Cellular proliferation was assessed by plating FACS-isolated DPP4+, ICAM1+, or CD142+ cells into a 96-well plate. For mouse studies, cells were plated at 3600 cells per well (Fig. 2D) or 2000 cells per well (Fig. 3C) in DMEM/F12 containing 10% FBS and Primocin. For human studies, cells were plated at 5000 cells per well in PM-1 medium (Zenbio catalog no. PM-1) supplemented with 1 nM recombinant FGF2 (Invitrogen, PHG0261) and Primocin (50 ng/ml). For both human and mouse studies, the total nuclear content was measured for each well by using the CyQuant direct cell proliferation assay (Thermo Fisher catalog no. C35011). Where indicated, TGFβ1 (10 ng/ml) and SB431542 (10 μM) treatments were initiated after measuring the first time point. Proliferation data are displayed either as a curve showing data from independent wells from a single representative BR (Fig. 2D and 3C and fig. S12C) or as a bar graph from multiple BRs showing relative proliferation rates. Relative proliferation rates were calculated for the human data in order to provide a comparable baseline as follows: The mean slope for DPP4, ICAM1, and CD142 cells was determined and called the patient baseline. Data are displayed as the cell type slope/patient baseline [for example, (DPP4 cell slope for patient 1)/(average slope for DPP4, ICAM1, and CD142 cells from patient 1)].

Quantitative reverse transcription PCR

Total RNA was isolated by using the RNeasy microkit (Qiagen) and quantified by using a NanoDrop spectrophotometer. Five hundred nanograms to 1 μg of RNA was reverse-transcribed by using the high-capacity cDNA synthesis kit (Applied Biosystems) and then subjected to real-time polymerase chain reaction (PCR) with SYBR Green master mix (Applied Biosystems) on a 7900 HT machine (Applied Biosystems). Tata-binding protein (encoded by Tbp) was used as an internal normalization control. Data are presented as the fold change relative to the control. For Fig. 3, E and G, data are calculated as the treatment/control ratio where the control is the untreated sample from the same cell type and BR (for example, treatment sample ICAM1–TGFβ BR A/ICAM1 control BR A). Data were analyzed using analysis of variance (ANOVA) followed by multiple comparisons unless otherwise indicated.

Statistical methods

Statistical methods were not used to predetermine sample size. The experiments were not randomized, and investigators were not blinded in experiments. All P values are reported with adjustment for multiple comparisons. Most statistical tests were performed by comparing DPP4+, ICAM1+, and CD142+ cells and therefore used an ANOVA followed by multiple comparisons. Pairwise comparisons were determined a priori, were two sided, and were performed only if the ANOVA was significant, by using either the Holm-Sidak or Tukey post hoc test. Samples (from each human donor or pool of mice) generated cells from all three populations (DPP4+, ICAM1+, and CD142+). Therefore, two-way ANOVA or repeated-measures ANOVA was used to account for the matched nature of these data. For all parametric tests, quantile-quantile and residual plots were generated to test for deviation from the ANOVA assumptions of normally distributed residuals with equal variance. In cases with unequal variance between samples, the Brown-Forsythe correction (one-way ANOVA) or Geisser-Greenhouse correction (repeated-measures ANOVA) was used. Where ANOVA would be the appropriate statistical test but the data were unbalanced, the mixed effects model for statistical testing was used instead. For fig. S14, multiple t tests followed by Holm correction were performed instead of ANOVA.

Where indicated, BRs represent independent samples (i.e., cells derived from different human donors or from separate pools of mice). For proliferation assays, cells from one human donor or pool of mice were plated into multiple wells, and this was repeated across multiple individuals or pools of mice.

References and Notes

Acknowledgments: We thank M. Lazar, E. Morrisey, S. Shapira, K. Ou, A. Wang, and members of the Seale lab for helpful advice and discussions. We thank L. Cheng and M. Lu for expertise in processing histological samples. We thank the University of Pennsylvania Diabetes Research Center (DRC) for use of the Functional Genomics Core (P30-DK19525). We thank R. Pellegrino da Silva, F. Mafra, and M. Gonzales for assistance with single-cell RNA-seq library preparation and sequencing. Funding: This work was supported by NIH grants RO1DK103008 and RO1DK107589 and ADA grant 1-16-IBS-269 to P.S., NIH grant 5T32DK007314-37 (T32) and the Measey Physician Scientist Fellowship Award to D.M., and NIH grant 5T32GM008216-29 to A.S. Author contributions: P.S., D.M., and A.S. were responsible for conceptualization, investigation, data curation and analysis, methodology, validation, visualization, and writing and review. Z.I. and C.O. performed experiments and assisted with data analysis. M.P.M. performed biostatistical analysis of the single-cell RNA-seq datasets. C.C. and I.P. provided fresh human adipose samples. Competing interests: I.P. is a paid consultant for Galderma. She has no financial interest to declare in relation to the content of this article. Data and materials availability: Single-cell and bulk RNA-seq matrix files are deposited in the Gene Expression Omnibus (GEO) under superseries accession number GSE128889.
View Abstract

Stay Connected to Science

Navigate This Article