Learn more: PMC Disclaimer | PMC Copyright Notice
Single Cell Approaches to Address Adipose Tissue Stromal Cell Heterogeneity
Associated Data
Abstract
A central function of adipose tissue is in the management of systemic energy homeostasis that is achieved through the coordinated regulation of energy storage and mobilization, adipokine release, and immune functions. With the dramatic increase in the prevalence of obesity and obesity-related metabolic disease over the past 30 years, there has been extensive interest in targeting adipose tissue for therapeutic benefit. However, in order for this goal to be achieved it is essential to establish a comprehensive atlas of adipose tissue cellular composition and define mechanisms of intercellular communication that mediate pathologic and therapeutic responses. While traditional methods, such as fluorescence-activated cell sorting (FACS) and genetic lineage tracing, have greatly advanced the field, these approaches are inherently limited by the choice of markers and the ability to comprehensively identify and characterize dynamic interactions among stromal cells within the tissue microenvironment. Single cell RNA sequencing (scRNAseq) has emerged as a powerful tool for deconvolving cellular heterogeneity and holds promise for understanding the development and plasticity of adipose tissue under normal and pathological conditions. scRNAseq has recently been used to characterize adipose stem cell (ASC) populations and has provided new insights into subpopulations of macrophages that arise during anabolic and catabolic remodeling in white adipose tissue. The current review summarizes recent findings that use this technology to explore adipose tissue heterogeneity and plasticity.
INTRODUCTION
We are in the midst of an epidemic of obesity-related diseases including diabetes, cardiovascular disease, and cancer (1–4). It is now clear that obesity itself, i.e., the excessive accumulation of fat tissue, is not the specific cause of obesity-related disease. Rather, obesity-related diseases occur when the environmental demands to store excess energy disrupts the integrative metabolic, endocrine and immune functions of adipose tissue (reviewed in (5–9)). Viewed from this perspective, an important therapeutic goal is to improve functioning of adipose tissue by augmenting its energy buffering capacity, improving the mixture of released adipokines, and/or reducing levels of persistent inflammation (7, 10–17). Thus, central to the ability to therapeutically target adipose tissue is a more detailed understanding of the cellular composition (i.e., adipocytes, stromal cells and immune cells) among tissue depots and how this composition changes in response to physiological, nutritional, and environmental challenges.
Adipocyte cell number is thought to be determined by early adulthood, with homeostatic turnover estimated at ~8% per year in humans (18) and ~18% per month in mice (19). However, white adipose tissue (WAT) is highly plastic and can undergo extensive remodeling in response to metabolic, nutritional, and pharmacological challenges (reviewed in (20–23)). During periods of excess energy intake, WAT expands by both hypertrophy (increase in cell size) and/or hyperplasia (increase in cell number) (24–26), and differences in the anatomical localization and type of adipose tissue expansion has profound effects on metabolic health (27–31). In general, hypertrophy in visceral adipose tissue (VAT) is linked more strongly to inflammation and insulin resistance than that in subcutaneous depots (32–36). In rodent models, high fat feeding leads to adipocyte hypertrophy in both depots, whereas adipogenesis is thought to occur primarily in VAT (24–26, 37). The ability of adipose tissue to expand by hyperplasia confers a protective advantage on insulin sensitivity and metabolic disease risk (33–36, 38). During chronic overnutrition, adipocyte death coupled with accumulation of pro-inflammatory macrophages and fibrotic changes eventually leads to ectopic fat accumulation in muscle and liver with systemic insulin resistance (reviewed in (7, 20, 39, 40)). In addition to remodeling during overnutrition, exposure to cold temperatures, β-adrenergic receptor agonists, and a number of hormones and metabolites stimulate the appearance of thermogenic brown adipocytes (“beige” or brown in white, “brite”) within WAT (41–54). Brown adipocytes are thought to exert a protective role on metabolism through increased fatty acid and glucose utilization (55). Brown adipocytes can arise from either existing white adipocytes in which the brown adipocyte program is reinstated (43, 44, 47, 48, 54), or via differentiation of new brown/beige adipocytes from progenitors, depending on the location, duration, and type of inductive stimuli (26, 46, 49, 54, 56).
Given its central role in metabolism, there is substantial interest in better defining cellular subtypes involved in adipose tissue homeostasis and the mechanisms that regulate in vivo adipogenesis, plasticity, and inflammation for therapeutic strategies to treat metabolic disease (22, 57). While adipocytes are the major functional cell type in adipose tissue and occupy the greatest mass and volume of the tissue (owing to stored triglyceride), these cells comprise fewer than ~20% of total cells (58). The remaining ~80% of cells is a complex mixture of poorly defined stromal cells (fibroblasts, vascular cells, adipocyte stem cells) and immune cells (58–61). Importantly, these stromal and immune cell types play critical roles in the establishment and maintenance of parenchymal cell function. The composition of stromal cells varies across fat depots (24, 59), likely reflecting tissue specialization and differences in energy storage, vascularization, innervation, and metabolism (62). Traditional methods of addressing heterogeneity such as antibody-based cell sorting approaches (e.g., FACS) have provided useful information for some cellular populations (46, 63); however, this approach is limited by the number of antibodies available to characterize cell populations. Additionally, surface marker expression can be altered during distinct biological processes, as is the case with Pdgfra, which decreases in cells undergoing active adipogenesis (64). Similar challenges exist for immune cells that are dynamic, and whose marker expression can change depending on differentiation and activation states (65, 66). Other methods, such as bulk RNA-seq analysis of whole tissue and/or sorted cells provides a greater depth in gene expression profiles. However, this technology does not completely discriminate differences in gene expression due to shifts in cellular composition versus changes in cellular states (67, 68). Additionally, relatively minor or undefined cell types can be underrepresented or missed entirely (65, 67, 69). Therefore, an important barrier to understanding adipose tissue biology is the ability to define and characterize cellular subtypes and expression profiles in a comprehensive and unbiased fashion. In this regard, single cell RNA sequencing (scRNAseq) technology has recently emerged as a powerful tool to address tissue heterogeneity (65). This review highlights recent findings on the use of scRNAseq to address adipose tissue stromal and immune cell complexity in WAT, with a focus on studies in mouse models.
OVERVIEW OF scRNASEQ
High throughput single cell RNA technology has evolved as a powerful tool to deconvolve tissue heterogeneity (65, 68, 70–74). The concept of scRNAseq is to characterize individual cells based on the expression profile of thousands of genes compared to only a handful of surface markers commonly used in FACS-based sorting (68, 75). One of the benefits of this technique is that no prior knowledge of tissue composition is required. Although subject to some technical noise, scRNAseq provides crucial information about cellular populations, gene expression profiles, as well as changes in cellular states that may arise under different physiological conditions (65, 75). Available scRNAseq platforms differ slightly in design (droplet vs plate-based), throughput, and information acquired (e.g., full length mRNA vs 5′ or 3′ digital end sequencing); for recent reviews see (65, 70, 75–77). Droplet-based microfluidic technology, such as inDrop (71, 72), Drop-seq (73), and 10X Genomics (74) are among the most widely used scRNAseq platforms due to increased throughput and more automated protocols (77, 78). The technical aspects of droplet generation are similar among methods (78). Briefly, cells and barcoded beads are partitioned into separate channels of a microfluidic device, and each bead/cell complex is then mixed with oil to generate an emulsion containing nanoliter-sized single cell droplets (Fig. 1) (71–74, 78). Each bead contains primers, cellular barcodes, and a unique molecular identifier (UMI) to provide transcript counts per cell (71–74, 78–80). Reverse transcription either takes place within the droplets (inDrop and 10X Genomics) or following demulsification and cell lysis (Drop-seq) (71–74, 78). cDNA libraries are then prepared for sequencing through preamplification, tagmentation/fragmentation, and PCR indexing (Drop-seq, 10X Genomics) or by linear amplification (in vitro transcription, RNA fragmentation, and RT-PCR) for inDrop (71–74, 78). A comparison among the different droplet-based technologies has recently been presented (78).
Cells and DNA-barcoded beads are partitioned into a microfluidic device and mixed with oil to generate nanoliter-sized single cell droplets. Each bead contains primers, cellular barcodes, and a unique molecular identifier (UMI). Cells are lysed within droplets, and reverse transcription either takes place within the droplets or following primer annealing and demulsification. Downstream steps differ slightly between platforms (not shown) to generate libraries for sequencing. Following sequencing, reads are demultiplexed, aligned, and then UMI counts per cell are quantified. Modified from (78).
Numerous bioinformatic tools are available and continuously emerging for analyzing scRNAseq output (77, 79, 81). Standardized pipelines involve initial preprocessing of reads (demultiplexing, alignment, and quantification), normalization as well as filtering to remove technical noise, such as cells containing high mitochondrial counts and doublets (75, 77, 79, 82). This is followed by dimensionality reduction, visualization, and identification of cell clusters based on differential gene expression (75, 77, 79, 82). In addition to identifying major and minor cell types within complex tissues, advanced bioinformatic programs are available for analysis of cellular transitions/hierarchies (83, 84), inference of gene regulatory networks (85, 86), predictions of cell-cell interactions (87), and integrative analyses of diverse datasets (88, 89).
ADIPOSE STEM CELL POPULATIONS IN MAJOR WAT DEPOTS AS IDENTIFIED BY scRNASEQ
There is a general consensus that adipocyte progenitors within adult WAT express the mesenchymal surface markers Pdgfrα, CD34, and Sca-1 (Ly6a) (46, 63, 90–92). Based on previous work, these progenitors are located in the interstitial and perivascular space and have a dendritic morphology with long processes that extend and contact several cells (46, 92). Lineage tracing experiments with constitutive and inducible Pdgfrα-Cre drivers indicates that most adipocytes arising in adult tissues are derived from Pdgfrα progenitors (90) and contribute to de novo adipogenesis both following high fat diets (HFD) and during catabolic tissue remodeling (46, 92). However, additional populations of adipocyte progenitors have also been described (38, 49, 93–97). For example, adipogenesis is recognized to be tightly linked to vasculature development (93, 98–101) and some, but not all, studies indicate adipocytes can be derived from endothelial or mural cells that also express Pparγ or Pdgfrβ (38, 49, 90, 93–95). Hematopoietic-derived progenitor cells have also been shown to contribute to neogenesis as well as the heterogeneity of adipocytes within adipose tissue (96, 97).
More recent studies have used scRNAseq to deconvolve progenitor heterogeneity in adipose tissue during development and in adults, under basal conditions (64, 102–104) and during catabolic remodeling induced by β3 adrenergic receptor (ADRB3) activation (64). A summary of the basic designs, including mouse strain, sex, and depots examined are presented in Table 1. In order to characterize the major cell populations present within non-immune stromal cells across studies, we integrated these diverse datasets using canonical correlation analysis (CCA) (89, 105). CCA is designed to reduce variability that can arise under different experimental conditions (e.g., cell isolation, mice strains, sex, and treatment) and clusters cell populations based on common or shared characteristics (89, 105). Approximately 11, 000 cells from eWAT and 28, 000 from iWAT were included in the integration (64, 102–104). Where appropriate, selection markers that were used to further characterize ASC subtypes with cell sorting along with the original cluster identity from each study are also presented (Table 1). A proposed nomenclature for adipocyte stem cells (ASC) based on marker gene expression is also included to facilitate comparisons between studies (Table 1). Additionally, individual findings from each study are summarized and discussed separately below.
Table 1.
Summary of scRNAseq Studies Performed on Non-Immune Stromal Cells from White Adipose Tissue.
Reference | WAT Depot | Sex | Age | Strain | Stimulus | scRNA Cell Sorting Technique | ASC Populations (identified by authors) | Proposed ASC Nomenclature | Cell Sorting Strategy (Functional Assays)** |
---|---|---|---|---|---|---|---|---|---|
Burl et al. (64) | eWAT | M | 8–10 Wk | C57BL6/J | SHAM or 3 days CL-316,243 (0.75 nmol/hr) | Lineage Negative (CD45-, CD45R-, CD11b-, Gr-1 (Ly-6G/C)-, 7–4-, and Ter-119-) | ASC1 ASC2 | ASC1a ASC2 | ND |
Hepler et al. (104) | eWAT | M | 8 Wk | PdgfrbrtTA (C57BL6/J) | None | “Muralchaser” tdTomato- mGFP+ SVC | APC FIPs | ASC1a ASC2 | Pdgfrβ+Ly6C-CD9-Pdgfrβ+Ly6C+ |
Burl et al. (64) | iWAT | M | 8–10 Wk | C57BL6/J | SHAM or 3 days CL-316,243 (0.75 nmol/hr) | Lineage Negative (CD45-, CD45R-, CD11b-, Gr-1 (Ly-6G/C)-, 7–4-, and Ter-119-) | ASC1 ASC1 ASC2 | ASC1a ASC1b ASC2 | ND |
Merrick et al. (102) | iWAT | M/F | PND12 | C57BL6/J | None | Lineage Negative (CD45-) | Group2 Group3 Group1 | ASC1a ASC1b ASC2 | Icam+CD142- CD142+ DPP4+CD142- |
Merrick et al. (102) | iWAT | M | 10 Wk | 129S6/SvEv | None | Lineage Negative (CD45-) | Group2 Group3 Group1 | ASC1a ASC1b ASC2 | Icam+CD142- CD142+ DPP4+CD142- |
Schwalie et al. (103) | iWAT | M/F | 8 Wk | Tg(Pref1-CreER)426 Biat (Pref1-CreER) | None | Lineage Negative (CD31−CD45−Ter119−) | G2 (P2*) G3 (P3*) G1/G4 (P1*) | ASC1a ASC1b ASC2 | Vap1+Adam12+ CD142+Abcg1+ CD55+IL13ra+ |
Abbr: Adipose stem cells (ASC), adipose progenitor cells (APC), fibro-inflammatory progenitors (FIP), stromal vascular cells (SVC).
Across studies, a majority of the non-immune stromal cell population detected by scRNAseq from each depot (>70%) were identified as ASC cells and express genes for previously identified surface markers including Pdgfrα (Pdgfra), CD29 (Itgb1), CD34 (Cd34), and Sca-1 (Ly6a) (Figs. 2B, ,3B)3B) (64, 102–104). Two main populations of ASC (ASC1a and 2) are further distinguishable by scRNAseq, with an additional subpopulation (ASC1b) present in iWAT (64, 102–104). Some of the ASC2 genes commonly expressed within both adipose depots include those associated with inflammation/prostaglandin synthesis (Ptgs2), angiogenesis/endothelial cell migration (Anxa3), proteolysis (Pi16, Dpp4), as well as immune signaling (Cd55) (Supplemental Figs. 2,3). Genes commonly expressed within the ASC1a population include those involved in lipid uptake/metabolism (Plpp1, G0S2, and Apoe) and the metalloreductase Steap4 (Supplemental Figs. 2,3). Several genes encoding extracellular matrix (ECM) components are also strongly expressed in both ASC2 and ACS1, however the type of ECM differs between populations (Supplemental Figs. 2,3). For example, Fn1 and Col14a1 are more enriched in ASC2 whereas Col15a1 and Col4a2 are enriched in ASC1 progenitors (Supplemental Figs. 2, 3) (102–104). The relative proportion of total ACS1 to ASC2 in adult animals is also higher in eWAT than that observed in iWAT (Figs. 2A, ,3A),3A), and functional assays suggest that the ASC1a population is more poised for adipogenesis (102–104).
Publicly available scRNAseq datasets (64, 104) from immune-depleted stromal cells isolated from eWAT were integrated using Seurat (105). (A) tSNE plots displaying the overlay of integrated datasets (left) and cell cluster identification (right). (B) tSNE plots displaying the log2 expression levels for select genes commonly used to purify adipose stem cells by FACS and/or used in lineage tracing experiments. (C) Violin plots displaying relative log2 expression levels of select marker genes in individual clusters. Abbr: Ly6a (stem cell antigen-1 (Sca-1)), Itgb1 (CD29).
Publicly available scRNAseq datasets (64, 102, 103) from immune-depleted stromal cells isolated from iWAT were integrated using Seurat (105). (A) tSNE plots displaying the overlay of integrated datasets (left) and cell cluster identification (right). (B) tSNE plots displaying the log2 expression levels for select genes commonly used to purify adipose stem cells by FACS and/or used in lineage tracing experiments. (C) Violin plots displaying relative log2 expression levels of select marker genes in individual clusters. Abbr: Ly6a (stem cell antigen-1 (Sca-1)), Itgb1 (CD29).
As expected, very few actively differentiating preadipocytes were detected in either depot (Figs. 2, ,3).3). In the study by Burl et al. (64), the proportion of proliferating (Birc5, Cdca8, Top2a) and differentiating preadipocytes (Cebpa, Adig, Plin1, Scd1) was drastically increased in response to ADRB3 stimulation in eWAT, and readily resolved as a separate cluster (Figs. 2A, ,2C)2C) (64). A population of mesothelial-like cells present only in eWAT was also distinguished by scRNAseq (Fig. 2A) (64, 104). The mesothelium is a thin membrane containing a layer of epithelial-like cells that surrounds some internal organs, including visceral adipose depots, and has been proposed to be a source of adipocyte progenitors during development (106–109). This cluster shares some similarity with marker genes for ASC2 (Dpp4, Ackr3, Smpd3), expresses low levels of Pdgfra/Pdgfrb, and highly express genes characteristic of mesothelial/epithelial cells (Msln, Krt8, and Krt14) (104). Wt1, a developmental marker (109), is also strongly expressed in these cells, as well as the other ASC clusters in eWAT, but not in iWAT (Fig. 2B). Other subpopulations present to varying degrees in each depot were identified based on marker gene expression, and included endothelial cells (Plvap, Flt1, Tie1), smooth muscle cells (SMC; Tagln, Acta2, Myh11), and subpopulations of contaminating immune cells tentatively identified as macrophages, NKT cells, B lymphocytes, and dendritic cells (DC) that were somewhat variable across studies (Figs. 2A, ,3A3A).
A number of different genetic tracers have been used to determine progenitor fate both during development and in adult tissue (46, 49, 90–92, 94, 109–112). During development, adipocytes that arise in eWAT have a history of Wt1 and Pax3 (in males) expression, whereas subcutaneous depots are thought to be derived from cells that once expressed Prx1 (91, 109, 111, 112). Among these, only Wt1 was detectable in ASC from adults and expression was restricted to eWAT (Fig. 2B). Dlk1 (Pref-1) also marks mesenchymal ASC progenitors (110) and was largely absent in ASC from adults, but present earlier in development and more strongly in ASC1a (unpublished observations). In addition to Pdgfra, which was broadly expressed in both ASC populations, other markers that have been used to trace the adipocyte lineage in adult tissue including Pdgfrb and Pparg are more strongly expressed in ASC1a. However, these genes are also co-expressed in endothelial/SMC clusters that are positive for Cdh5 and Acta2, but generally negative for Pdgfra (Figs. 2B, ,3B)3B) (64).
In Vivo and In Vitro Functional Characteristics of eWAT Progenitors
Many of the known transcriptional and morphological changes that occur during adipogenesis have been derived from in vitro studies, either using preadipocyte cell lines or progenitors isolated from adipose tissue (reviewed in (113–116)). One of the benefits of scRNAseq, aside from detection of major cell populations, is the ability to detect changes in cellular states that occur in vivo (67). The ADRB3 agonist CL-316,243 (CL) is a potent inducer of brown adipogenesis in white adipose tissue; however, the mechanisms through which CL induces brown adipocyte formation is depot specific (46, 92, 117). In a series of studies, Lee et al. (45, 46, 92, 117) demonstrated that de novo brown adipogenesis in epididymal WAT (eWAT) is dependent on recruitment of alternatively activated (M2) macrophages that proliferate to form “crown like structures” (CLS) involved in the clearance of dead adipocytes (117, 118). The local release of lipid metabolites (e.g., 9-hydroxyoctadecadienoic acid) and other chemoattractants (e.g., OPN) by macrophages that comprise the CLS then potentiates the proliferation, migration, and differentiation of Pdgfrα/CD44+ progenitors into brown adipocytes (45, 46, 92, 117). In iWAT, CLS are rarely induced by CL treatment (92), and most brown adipocytes arise from existing white adipocytes that transform into a brown/beige phenotype (46, 54, 64, 92).
Burl et al. (64) profiled non-immune stromal cells in eWAT with scRNAseq 3 days following CL treatment to further deconvolve the recruited Pdgfrα+ progenitors and identify temporal changes in ASC that occur during brown adipogenesis in vivo. Two major clusters of quiescent ASC (ASC1, ASC2) were identified from both depots (Figs. 2A, ,3A)3A) (64). An additional cluster of proliferating and differentiating preadipocytes was further resolved in the CL-treated group, comprising ~10% of the total ASC population in eWAT (Fig. 2A) (64). Pseudotime analysis mapped a differentiation trajectory of cells undergoing brown adipogenesis into distinct cellular states, spanning proliferating to late differentiating preadipocytes (64). In general, differentiating preadipocytes lost expression of proliferation markers (Birc5, Ccnb2) and progressively gained expression in transcriptional regulators of adipogenesis as well as metabolic markers of maturing adipocytes (e.g., Dgat1/2, Pnpla2, Lipe, Acot5, Agpat2) (64). Several ECM remodeling proteins were also downregulated (Mmp23) whereas Ucp1, a marker of brown adipocytes, was upregulated in late differentiating preadipocytes (64). Moreover, scRNAseq provided better resolution of these “cellular state” changes compared to bulk RNA sequencing of FACS-isolated Pdgfrα cells (64). A number of marker genes defining ASC1a (Agt, Col15a1) were also co-expressed within the proliferating and differentiating cluster, indicating that ASC1a (Pdgfra+) cells are a more committed adipocyte progenitor (64). Similar to previous findings (46, 92), progenitors in iWAT were largely refractory to the effects of ADRB3 agonism (Fig. 3A), with few transcriptional changes detected amongst the ASC clusters (64).
Perivascular cells expressing the surface marker Pdgfrβ are enriched in the adipogenic transcription factor Pparγ and have been suggested to be a source of progenitors in white adipocyte tissue (38, 49, 94). Disrupting Pparg within this subtype impairs, while overexpression promotes, hyperplastic VAT expansion and improves insulin sensitivity during diet-induced obesity (38). Hepler et al. (104) performed scRNAseq on FACS-sorted cells isolated from eWAT of “Muralchaser” mice to further characterize the Pdgfrβ progenitor populations (Table 1). In the Muralchaser mouse model, treatment with doxycycline constitutively activates a membrane green fluorescent protein (mGFP) reporter in a Pdgfrb- and Cre-dependent manner, allowing purification of cells with a history of Pdgfrb expression at the time of pulsing (described in (49, 104)). Using this model, mGFP+ cells were isolated from the total stromal population and used as input for scRNAseq (104). Despite a more selective cell enrichment strategy, distinct clusters of adipocyte progenitor cells (ASC1a, FIPs/ASC2, preadipocytes) and a small population of mesothelial cells (Fig. 2A) were resolved that exhibited an expression profile similar to those found by Burl et al. (64) (Figs. 2B, ,C;C; Supplemental Fig. 1). Hepler et al. (104) further devised a cell sorting strategy to purify and functionally evaluate these populations (Table 1, Fig. 2B). In cell culture, ASC1a cells (Pdgfrβ+CD9-Ly6c-) were found to be more adipogenic than FIPs/ASC2 (Pdgfrβ+Ly6c+) or mesothelial cells (Pdgfrβ+CD9+Ly6c-) (104). Further, ASC1a, but not FIP/ASC2 cells differentiated into adipocytes when transplanted into the subcutaneous fat pad of lypodystrophic mice, providing further evidence that ASC1a is a direct adipogenic precursor (64, 104). In comparison, cell sorted FIPs/ASC2 inhibited the differentiation of ASC1a and displayed a higher pro-inflammatory/pro-fibrotic phenotype in vitro (104). Interestingly, FIPs/ASC2 were also more proliferative than ASC1a in response to a high fat diet in vivo (104). These findings are consistent with those of Marcelin et al. (119) who described a Pdgfrα+CD9high population in eWAT with a similar profile to ASC2 (Fig. 2B) that were also pro-fibrotic and more resistant to adipogenesis in vitro.
In Vivo and In Vitro Functional Characteristics of iWAT Progenitors
Adipose progenitors in the subcutaneous iWAT are generally less sensitive to de novo adipogenesis following ADRB3 stimulation or high fat feeding than progenitors in eWAT (24–26, 46, 64, 92). Two additional studies conducted scRNAseq on non-immune stromal cells isolated from iWAT under basal conditions (103) and during the period of perinatal tissue expansion (102) (Table 1). Schwalie et al. (103) performed scRNAseq on stromal cells from adult mice and (in addition to ASC1 and ASC2) identified a subset of ASC1 cells they named “Aregs” (adipogenesis-regulatory cells; ASC1b). This subpopulation was distinguished by the expression of several gene markers including Fmo2, F3 (CD142), and Gdf10 (Fig. 3), and co-localized to perivascular regions within the parenchyma body (103). In cell culture, FACS-sorted ASC1b cells (Lin-Sca1+CD142+Abcg1+) were more refractory to adipogenesis when compared to total ASC (Lin-Sca1+) or total ASC depleted of ASC1b (Lin-Sca1+CD142-Abcg1-) (103). Further, FACS-isolated ASC1b cells inhibited the differentiation capacity of ASC (Lin-Sca1+CD142-Abcg1-) in vitro, potentially through a paracrine effect (103). FACS-isolated ASC1b cells were also detected in higher proportions within the subcutaneous and visceral depots of ob/ob mice, a genetically obese model, although a similar cluster based on scRNAseq was not readily identifiable in eWAT of mice (104) or in subcutaneous abdominal fat from humans (102). In Matrigel transplantation experiments, depletion of ASC1b from the total ASC population also led to a greater number of mature adipocytes following high fat feeding, suggesting that this subtype maintains regulatory function in vivo (103).
Merrick et al. (102) evaluated immune-depleted stromal cells from iWAT during a period of perinatal adipose tissue expansion (postnatal day (PND) 12) (Table 1). Similar to findings by Schwalie et al. (103), they identified an ASC2 (DPP4+CD142-) and two ASC1 populations (Icam+CD142-(ASC1a) and CD142+ (ASC1b); Table 1), with a proportionally higher number of total ASC1 cells during the period of perinatal adipose tissue growth (102). However, in contrast to the reported inhibitory role of ASC1b, all 3 cell sorted ASC subpopulations were adipogenic both in vitro and in vivo (102). Additional functional assays indicate that ASC2 are more resistant to in vitro adipogenesis when insulin concentrations are reduced, are capable of multilineage differentiation, and are more sensitive to the anti-adipogenic effects of transforming growth factor beta (TGFβ) than ASC1a or ASC1b (102). In Matrigel transplantation studies of isolated populations in vivo, ASC2 cells were found to transition into either ASC1a or ASC1b prior to becoming adipocytes, suggesting ASC2 are a less committed progenitor (102). It is noteworthy that although all ASC progenitors from iWAT were adipogenic in this study, ASC2 cells isolated from eWAT and transplanted into iWAT of lipodystrophic mice failed to develop into adipocytes and were more resistant to in vitro adipogenesis (104, 119). Therefore, despite similarities in gene expression profiles (Figs. 2, ,3),3), at least some characteristics of ASC2 appear to be depot specific.
Immunohistochemical Localization of ASC Subpopulations in iWAT and eWAT
Cluster analysis indicates the major cell populations in eWAT or iWAT exhibit similar marker gene profiles, as well as other functional characteristics. For example, a pro-adipogenic phenotype was observed for ASC1a from both WAT depots (102–104), and an inflammatory, fibrotic, or mesenchymal phenotype for ASC2 (102, 104, 119). Merrick et al. (102) evaluated the distribution of ASC progenitors during perinatal adipose tissue development and found ASC subtypes to be compartmentally localized in iWAT. ASC2 (Pref-1- DPP4+) cells were mainly confined to the collagen-rich fascia (“reticular interstitium”), whereas ASC1a progenitors (Pref-1+DPP4-) were located within the parenchyma body, closely associated with adipocytes (102). To determine whether these subtypes share a similar distribution in eWAT, we evaluated total ASC (Pdgfrα) and ASC2 (DPP4, Pi16) markers in both depots from adult mice by immunohistochemistry (Fig. 4). Similar to findings from Merrick et al. (102), ASC2 (Pdgfrα+DPP4+/Pi16+) cells were restricted mainly to the fascia (extracellular connective tissue) in iWAT, which encapsulates the adipose depot and is several cell layers thick (Fig. 4B). Double positive cells also could be found in a thin connective layer lining individual fat lobules (Fig. 4B). By comparison, ASC1 cells (Pdgfrα+DPP4-) were found throughout the parenchyma body and in areas surrounding vessels, as previously observed (46).
(A, B) Immunofluorescence for Pdgfrα (green) and either DPP4 (red, top panel) or Pi16 (red, bottom panel) was evaluated in formaldehyde-fixed tissue sections (10 μM) from (A) eWAT and (B) iWAT. (A) ASC2 cells (Pdgfrα+DPP4+ or PI16+) cells are localized in areas underlying the mesothelium in eWAT. Note the strong immunoreactivity of DPP4 surrounding eWAT, which is also expressed in mesothelial cells from scRNAseq (Fig. 2C). (B) ASC2 progenitors are localized primarily within the fascia that encapsulates the inguinal adipose depot and is also present in a thin layer of connective tissue lining individual fat lobules. ASC1 cells (Pdgfrα+DPP4-) are localized within the parenchyma body in association with mature adipocytes and surrounding vessels. DAPI was used for nuclear staining as indicated in blue. Orange arrows denote some areas immunoreactive for both Pdgfrα and either DPP4 or Pi16. Abbr: Mesothelium (Mesoth.).
In eWAT, a layer of cells that stained strongly for DPP4 surrounded the entire adipose depot (Fig. 4A). This distribution is consistent with the visceral mesothelium previously described (Fig. 4A) (106, 109). Some double-labeled cells (ASC2; Pdgfrα+DPP4+) were observed underlying the mesothelial layer but were mostly absent within fat lobules (Fig. 4A). To better distinguish cells, we additionally evaluated distribution of Pi16, a marker more specific for this ASC cell type in eWAT (Fig. 2C). Doubly labeled cells (ASC2; Pdgfrα+Pi16+) were localized to areas directly underlying the mesothelium but were generally fewer in number compared to ASC1a cells (Pdgfrα+Pi16-) (Fig. 4A). Pi16+ cells were also found in areas surrounding large vessels in eWAT (not shown). In comparison, ASC1a cells were located near the mesothelium and dispersed throughout the fat lobules. Mesothelial cells are thought to serve as a protective barrier, with additional roles including immunomodulatory functions as well as tissue repair (107, 108, 120–123). It is interesting that ASC2 cells were closely associated with this cell type in eWAT and may be related to the higher pro-inflammatory and fibrotic phenotypes observed in vitro and in vivo (104, 119). Whether ASC2 can transition into ASC1a as suggested for iWAT (102) or if there are distinct progenitor lineages for each population is currently under investigation.
ADIPOSE STEM CELL POPULATIONS IN DERMAL WAT (dWAT) AND THE LACTATING MAMMARY GLAND AS IDENTIFIED BY scRNASEQ
In addition to ASC identified within the major WAT depots, Scherer and colleagues (124) recently used scRNAseq to address plasticity of dermal white adipose tissue (dWAT) during hair follicle cycling and in the mammary gland during pregnancy and lactation (125). The dermal adipocyte layer expands by both hypertrophy and hyperplasia during active periods of hair follicle growth (anagen) and then diminishes in thickness during the regression of follicles and rest (catagen/telogen) (124, 126). Using the “AdipoChaser-mT/mG” model, whereby mGFP is expressed on the membrane of cells with a history of adiponectin expression at the time of pulsing (second anagen), dermal adipocytes (GFP+) were found to dedifferentiate into Pdgfrα/GFP+ progenitor cells following hair follicle regression/rest (second telogen) (124). FACS-isolated Pdgfrα (CD45-/CD31-) progenitors were then profiled with scRNAseq during the second telogen to evaluate transcriptomic differences between dedifferentiated (GFP+) and unlabeled (GFP-) progenitors (124). Using this model, two main populations of ASC were identified that share some similarity in marker gene profiles with those observed in the other major WAT depots (ASC2, Ptgs2; ASC1a, Apoe) (64, 102–104, 124). Although few transcriptional changes were observed between dedifferentiated (GFP+) and GFP- ASC, dedifferentiated cells were slightly more adipogenic in vitro (124). Similar to the above, the fate of adipocytes during lactation and recurrence at involution were evaluated in the “Adipochaser-mT/mG” model (125). Adipocytes (GFP+) labeled prior to pregnancy were found to dedifferentiate into Pdgfrα+/GFP+ SVC during lactation and re-differentiate into adipocytes upon weaning of pups (125). Using scRNAseq, the Pdgfrα+/GFP+ (CD31-/CD45-) progenitors were found to have a similar expression profile to Pdgfrα progenitors isolated from virgin females (125). However, due to limitations in sample size, it is unknown if the dedifferentiated cells represent more than one ASC subtype (125). Nonetheless, these findings highlight the cellular plasticity of adipose tissue and further suggest that adipocytes can fully dedifferentiate into ASC under certain physiological conditions (124, 125).
IMMUNE CELL POPULATIONS IN eWAT DETECTED WITH scRNASEQ
Visceral adipose tissue is known to have differential susceptibility to inflammation and neogenesis compared to subcutaneous fat (24–26, 37, 46, 92, 127). Infiltration and local proliferation of adipose tissue macrophages (ATMs) is frequently observed in obesity and contributes to insulin resistance and other metabolic derangements through local secretion of pro-inflammatory mediators (16, 37, 128–131). In obese states, a large proportion of ATMs in the visceral depot surround dead adipocytes, forming “crown-like structures” (CLS) (37, 118, 127), that are thought to be an important source of these mediators (37). The recruitment and proliferation of macrophages is also tightly linked with de novo adipogenesis and is important for initiating WAT remodeling in response to energy excess, β-adrenergic stimulation, and following tissue injury (45, 92, 117). Recruited macrophages also exhibit heterogeneity in phenotypes depending on the type and duration of the stimulus (37, 45, 92, 130), although the distinct populations involved and activation states have not been thoroughly characterized. Several studies have used scRNAseq to characterize immune cell populations during WAT remodeling, either following ADRB3 stimulation or during diet-induced obesity (64, 132–134). However, due to the extent of data evaluated, results are only briefly summarized below.
Immune Cell Populations Associated with ADRB3-Induced Remodeling
ADRB3-induced remodeling in eWAT is associated with the rapid accumulation of adipose tissue macrophages that are involved in phagocytosis and lipid clearance (45, 46, 92, 117). Compared to other types of WAT remodeling (92), ADRB3-induced remodeling involves an alternatively activated (M2-polarized, non-inflammatory) macrophage response that is dependent on the chemokine osteopontin (OPN; Spp1) (45, 92, 117, 135). Burl et al. (64) used scRNAseq to profile immune cell populations following CL treatment. Adipose tissue macrophages (ATM) were found to be the most prominent immune cell type present (~58%), and the relative proportion of macrophages expanded by ~2.5 fold following CL treatment (64). Clustering of ATM populations based on transcriptional profiles identified at least 6 different subgroups (64). Among these, CL modestly expanded a subset that was enriched for proliferation markers and whose gene expression profile was associated with processes including lipid uptake and metabolism and ECM remodeling (64). Osteopontin (OPN, Spp1), a chemoattractant that is highly upregulated in CLS macrophages (92, 117), was also enriched within this subpopulation (64). OPN is a chemoattractant for ASC1a progenitors and would likely explain their recruitment to CLS and formation of an “adipogenic niche” previously described during catabolic tissue remodeling (45, 46, 64, 92, 117).
Immune Cell Populations Associated with HFD-Induced Remodeling
Jaitin et al. (133) performed scRNAseq on isolated lineage positive (CD45+) cells to identify immune cell populations driving WAT remodeling during diet-induced obesity. Cells were isolated from eWAT at 6, 12, and 18 weeks following HFD and cell populations (“metacells”) were classified into 15 different immune cell types (133). Between 6 and 12 weeks following high fat feeding, there was a dramatic shift in immune cell populations, including a substantial increase in the proportion of macrophages and concomitant decrease in monocytes (133). Aside from resident/perivascular macrophages, they identified two populations of macrophages (“Mac2” and “Mac3”), characterized by differential expression of Cd9 and Nceh1, that were prominent only in obese animals, derived from recruited monocytes, and by histochemical analyses, accumulated in CLS (133). Consistent with localization to the CLS, the Mac3 population was enriched for genes involved in lipid metabolism and phagocytosis, and the receptor Trem2 was found to be exclusive for this immune cell population (133). In loss of function studies, Trem2-deficiecy prevented the appearance of the Mac3 subset during obesity, without affecting monocyte/macrophage infiltration (133). Further histochemical and metabolic analyses revealed an important role for Trem2 signaling in preventing adipose tissue dysfunction by promoting the formation of CLS to initiate remodeling and limit/regulate adipocyte size (133).
Two other studies evaluated immune cell heterogeneity in eWAT (132, 134). Hill et al. (132) profiled CD11b+Ly6c- ATMs and identified 2 distinct populations with scRNAseq. One subset was enriched for Cd9 and found to be the main population of Ly6c- ATMs that accumulate during diet-induced obesity (132). Similar to findings from Jaitin et al. (133), both Cd9+Ly6c- and Cd9-Ly6c+ cells were recruited from bone marrow-derived monocytes, and Cd9+Ly6c- cells localized to CLS in obese states (132). The transcriptional profile and chromatin landscape of Cd9+Ly6c- macrophages further suggested a proinflammatory role compared to that of the Cd9-Ly6c+ subset, which were implicated more in tissue maintenance/homeostasis (132). Interestingly, the subset of CLS macrophages identified by Burl et al. (64) was also enriched for the genes Trem2 and Cd9. Therefore, this transcriptional “signature” appears to be a conserved pathway for CLS-associated macrophages during WAT remodeling.
Weinstock et al. (134) profiled immune cell changes that occur in eWAT during high fat feeding and following 2 weeks of caloric restriction (CR; 70% kcal) and compared these profiles to that of lean mice (64). They identified 15 distinct immune cell populations in eWAT with scRNAseq (134). Seven of the clusters were categorized as macrophages and collectively represented more than one-half of the total immune cell types present (134). HFD was found to expand the “major macrophage” population by ~50% and also enriched for a subset of “phagocytic macrophages” (7%) that were not present in lean control mice (64, 134). Following a two week period of CR, this phagocytic population further increased to 30% of the total immune population, whereas the major macrophage population decreased to proportions observed in lean mice (134). The increase in phagocytic macrophages following CR was suggested to enhance clearance of lipid and adipocyte remnants to restore ATM homeostasis (134).
SUMMARY OF scRNASEQ STUDIES IN WAT
Collectively, studies exploring adipose tissue composition using scRNAseq have provided relatively consistent data regarding WAT progenitor populations, expression profiles, and insight into immune cell phenotypes during catabolic and anabolic tissue remodeling. The adipocyte progenitors identified from subcutaneous and visceral depots express several of the previously identified genes for stem cells including Pdgfra, Itgb1, Cd34, and Ly6a; however, 2–3 distinct subpopulations could be further resolved by scRNAseq. Based on both in vitro characterization (102–104) and in vivo trajectories (64), ASC1a appears to represent an immediate adipocyte progenitor. Further, immunochemical evidence indicates this cell type is localized in the parenchyma body in association with adipocytes (102).
Some inconsistencies exist regarding ASC1b in iWAT, as both pro-adipogenic and anti-adipogenic roles have been reported (102, 103). The reason for this discrepancy is currently unknown but may be due to the different selection markers used to purify cell populations, as suggested (102). ASC2 appears to be a relatively new defined progenitor subset, and available data suggest a pro-inflammatory role (104, 119), at least in eWAT. This subpopulation is detected by immunohistochemistry in the collagen-rich fascia encapsulating iWAT (102) and in close contact with mesothelial cells in eWAT. ASC2 are further characterized by a higher expression of Wnt genes generally associated with a less committed, multi-lineage mesenchymal stem cell, and have been hypothesized to be a direct precursor for both ASC1a and b in iWAT (102). Lineage tracing using cell-selective markers identified through these studies will help further delineate a potential hierarchal role for this subtype in vivo (102).
In visceral depots, the most prominent immune cell population is monocyte/macrophages. In separate models of tissue remodeling, scRNAseq was used to detect population and phenotype changes in response to either HFD or ARDB3 stimuli (64, 132–134). A shift in macrophage populations as well as gene expression patterns were resolved using scRNAseq. Further, a subset of macrophages associated with “crown-like structures” were distinguishable based on gene profiling and further confirmed with histochemical analyses (64, 132, 133). Cd9, Spp1, and Trem2 were identified as genes enriched within this subset. Additionally, Trem2 signaling appears to drive the transcriptional and functional responses of this subgroup in obesity, which is suggested to promote formation of CLS to control adipocyte size (133). These findings highlight an important role for CLS-associated macrophages in initiating adipose tissue remodeling and provide new insights into the complexity of immune activation states in adipose tissue.
Although the current review aimed to summarize findings available from preclinical mouse models, a few comparable studies addressing heterogeneity of stromal cells with scRNAseq in humans have also been reported (102, 132, 133). For example, Merrick et al. (102) found similar populations of the two major ASC (ASC1a, ASC2), in human abdominal subcutaneous adipose tissue as those identified in mice, with a proportionally higher number of ASC1a cells. Conversely, ASC1b markers were more diffusely expressed (102). FACS-sorted ASC populations also exhibited similar functional properties in adipogenic assays, with ASC1a being more poised for adipogenesis than the ASC2 subtype (102). The presence of an analogous, lipid-laden CD9/TREM2+ macrophage population associated with CLS in obese human visceral fat was also identified using scRNAseq (133) and by histological and FACS-based analyses (132). Importantly, a positive association between the accumulation of CD9+ macrophages in VAT and obesity was reported by both Jaiten et al. (133) and Hill et al. (132). These studies suggest at least some conserved cell types and cellular responses between species. Nonetheless, further integration and interrogation of single cell datasets will lead to a better understanding of species-dependent differences among adipose tissue depots and potentially aid in more directed treatments for obesity-related metabolic dysfunction.
FUTURE DIRECTIONS
It is evident that scRNAseq differentiates stromal cell populations and cellular states with high resolution, providing a powerful tool for addressing adipose tissue heterogeneity. One limitation from the available studies is that all have enriched for lineage selection, therefore the relative composition of stromal cells within the different depots and proportional changes in cell types under different conditions are not well defined. Cellular input and capture efficiencies differ to some degree among different scRNAseq platforms (65, 78). As scRNAseq becomes increasingly streamlined, the capacity to detect wider range of cells is feasible and should be considered in future studies.
Although the focus of this review was on stromal cell heterogeneity, adipocytes arise from distinct developmental lineages and depot specific differences in adipokine release, lipolysis, and inflammation are well established (reviewed in (60, 136, 137)). Heterogeneity among adipocytes within the same depot have also been reported by both genetic and histological criteria (138, 139). For example, adjacent brown adipocytes within the interscapular brown adipose tissue (iBAT) of mice often express substantially different levels of uncoupling protein 1 (UCP1), yielding a ‘harlequin’ type appearance upon histological staining (138). Due to their fragility and natural buoyancy in suspension, and for brown adipocytes, high mitochondrial content, examining isolated adipocytes by scRNAseq is technically challenging (64). Methods for single nucleus RNA-seq (snRNAseq) are available (140, 141) and results from other tissues suggest good correlation to profiles generated with scRNAseq that are less affected by cell isolation methods (142, 143). Two recent studies have begun to address adipocyte heterogeneity, either in isolated brown adipocytes or from inguinal white adipocyte nuclei using sc/snRNAseq, respectively (139, 144). These studies reported some evidence of heterogeneity within adipocyte gene-expressing clusters (139, 144). Nonetheless, both studies also identified additional clusters of putative adipocytes that co-express markers prominent in other cell types, including endothelial cells (e.g., Flt1, Pecam1) (139, 144) as well as ASC subtypes (e.g., Pi16, Pdgfra, Gdf10) (144). Indeed, the profile of putative brown adipocytes expressing low levels of UCP1 reported by Song et al. (139) is more similar to capillary endothelial cells than brown adipocytes when integrated with multiple independent scRNAseq data sets. Whether these cells represent non-classical adipocytes or contaminating cells present in the floating adipocyte fraction is worthy of future validation.
While scRNAseq provides deep information regarding cell types and states, it does not provide information on localization of cellular populations within adipose tissue. Nonetheless, scRNAseq data provide additive and informative markers that can be used to localize cells within the tissue microenvironment and confirm potential cell-cell interactions. In this way, ASC2 has been localized to the fascia or “reticular interstitium” (102) in iWAT, and Spp1/Trem2 macrophages and activated ASC progenitors to the CLS of visceral depots (46, 64, 92, 117, 132, 133). We anticipate that additional exploration of markers genes identified through current scRNAseq studies as well as future developments in spatial transcriptomics (145) will provide a more complete picture of adipose tissue heterogeneity, cellular interactions, and help dissect mechanisms of tissue plasticity.
Acknowledgements
The authors thank members of CIMER for useful suggestions and Dr. Roger Pique-Regi for guidance in scRNAseq data analysis.
Funding
This work was supported by the National Institute of Diabetes and Digestive and Kidney Disease (NIDDK) [JGG; R01DK062292] and the National Institutes of Health National Institute of Environmental Health Sciences [Center Grant P30 ES020957].
Abbreviations:
(Aregs) | Adipogenesis-regulatory cells |
(ASC) | adipose stem cell |
(ATMs) | adipose tissue macrophages |
(ADRB3) | β3 adrenergic receptor |
(BAT) | brown adipose tissue |
(CCA) | canonical correlation analysis |
(CL) | CL316,243 |
(cDNA) | complementary DNA |
(CLS) | crown-like structures |
(eWAT) | epidydimal white adipose tissue |
(ECM) | extracellular matrix |
(DC) | dendritic cells |
(dWAT) | dermal white adipose tissue |
(DPP4) | dipeptidyl peptidase 4 |
(ECM) | extracellular matrix |
(FIP) | fibro-inflammatory progenitor |
(FACS) | fluorescence-activated cell sorting |
(HFD) | high fat diet |
(iBAT) | interscapular brown adipose tissue |
(iWAT) | inguinal white adipose tissue |
(mGFP) | membrane green fluorescent protein |
(NKT) | natural killer T cells |
(OPN) | osteopontin |
(Prx1) | paired related homeobox transcription factor 1 |
(Pi16) | peptidase inhibitor 16 |
(PPARγ) | peroxisome proliferator-activated receptor gamma |
(Pdgfrα) | platelet-derived growth factor alpha |
(Pdgfrβ) | platelet-derived growth factor beta |
(PCR) | polymerase chain reaction |
(PND) | postnatal day |
(Pref-1) | preadipocyte factor-1 |
(RT-PCR) | reverse transcription-polymerase chain reaction |
(scRNAseq) | single cell RNA-sequencing |
(snRNAseq) | single nucleus RNA-sequencing |
(Sca-1) | stem cell antigen-1 |
(t-SNE) | t-distributed stochastic neighbor embedding |
(UCP1) | uncoupling protein 1 |
(VAT) | visceral adipose tissue |
(WAT) | white adipose tissue |
(UMI) | unique molecule identifier |
(Wt1) | Wilms tumor 1 |
Footnotes
Competing Interests
The authors declare that there are no competing interests.