β-Catenin Stabilization in Skin Fibroblasts Causes Fibrotic Lesions by Preventing Adipocyte Differentiation of the Reticular Dermis

The Wnt/β-catenin pathway plays a central role in epidermal homeostasis and regeneration, but how it affects fibroblast fate decisions is unknown. We investigated the effect of targeted β-catenin stabilization in dermal fibroblasts. Comparative gene expression profiling of stem cell antigen 1- (Sca1-) and Sca1+ neonatal fibroblasts from upper and lower dermis, respectively, confirmed that Sca1+ cells had a preadipocyte signature and showed differential expression of Wnt/β-catenin–associated genes. By targeting all fibroblasts or selectively targeting Dlk1+ lower dermal fibroblasts, we found that β-catenin stabilization between developmental stages E16.5 and P2 resulted in a reduction in the dermal adipocyte layer with a corresponding increase in dermal fibrosis and an altered hair cycle. The fibrotic phenotype correlated with a reduction in the potential of Sca1+ fibroblasts to undergo adipogenic differentiation ex vivo. Our findings indicate that Wnt/β-catenin signaling controls adipogenic cell fate within the lower dermis, which potentially contributes to the pathogenesis of fibrotic skin diseases.


INTRODUCTION
The epidermis is maintained by distinct subpopulations of stem cells whose fate is regulated by intrinsic mechanisms and external signals from the niche (Kretzschmar and Watt, 2014;Lim and Nusse, 2013). Signaling between epidermal stem cells and cells in the underlying dermis is reciprocal (Millar, 2002;Sennett and Rendl, 2012). Such interactions can occur at close range-for example, via extracellular matrix (ECM) deposition (Fujiwara et al., 2011)-and by long-range signaling mediated by epidermal secreted factors (Collins et al., 2011;Donati et al., 2014). Furthermore, factors secreted by proliferating adipocyte progenitors (Festa et al., 2011) and differentiated adipocytes (Plikus and Chuong, 2014;Plikus et al., 2008) regulate the hair growth cycle.
The dermis contains a variety of subpopulations of mesenchymal cells with different locations and functions (Driskell and Watt, 2015). At the E12.5 stage of embryonic development, mouse dermal cells are capable of differentiating into all the different fibroblast types present in postnatal skin. However, at about E16.5, the dermal mesenchyme undergoes commitment to two different lineages. In late embryonic and neonatal skin, the upper dermal lineage gives rise to papillary fibroblasts, the cells of the arrector pili muscle (responsible for piloerection), the dermal sheath, and the dermal papilla (DP). The lower dermal lineage gives rise to the reticular fibroblasts, which deposit most of the skin fibrillar collagen, preadipocytes and mature adipocytes (Driskell et al., 2013). The different fibroblast lineages are functionally significant, because the upper papillary lineage is required for new hair follicle (HF) formation, and the lower (or reticular) lineage is responsible for the first wave of dermal repair after wounding (Driskell et al., 2013). However, the molecular mechanisms that determine fibroblast fate decisions are largely uncharacterized (Driskell et al., 2009).
Epidermal Wnt/b-catenin signaling causes profound changes in the underlying dermis, leading to expansion of both the upper and lower dermal lineages, de novo formation of DPs, and an increase in adipocyte differentiation (Collins et al., 2011;Donati et al., 2014;Lichtenberger et al., 2016). In addition, there is compelling evidence that Wnt/b-catenin signaling in fibroblasts regulates the composition of the dermis. b-Catenin is necessary and sufficient to specify dermal fate in different body regions during mouse embryonic development (Atit et al., 2006;Ohtola et al., 2008). Wnt/b-catenin signaling has a well-characterized inhibitory effect on adipogenic differentiation (Gesta et al., 2007;Kennell and MacDougald, 2005;Longo et al., 2004) and is required in the DP to promote HF formation and control DP activity and size (Enshell-Seijffers et al., 2010; Kaushal et al., 2015;Tsai et al., 2014).
Furthermore, it was shown recently that expression of stabilized b-catenin in the fibroblasts of mouse ventral dermis at E16.5 results in progressive skin fibrosis, with thickened collagen fibers and altered collagen fibril morphology (Hamburg-Shields et al., 2015).
In this study, we performed comparative gene expression profiling of distinct populations of neonatal fibroblasts, which showed differential expression of Wnt pathway genes. Using a conditional targeting approach, we examined the effects of b-catenin stabilization in all fibroblasts or selectively in reticular fibroblasts. Our findings indicate a key role of Wnt/b-catenin signaling in regulating the differentiation of reticular fibroblasts into adipocytes.

Differential expression of adipocyte genes in upper and lower dermal fibroblasts
As previously reported (Collins et al., 2011), in PdgfraEGFP mice (Hamilton et al., 2003) enhanced green fluorescent protein (EGFP) expressed under the control of the plateletderived growth factor receptor a (PDGFRa) promoter is detected in nuclei of all dorsal skin fibroblasts ( Figure 1aed). In neonatal (P2) skin, stem cell antigen 1 (Sca1) is absent from the upper (papillary) dermis but is expressed by reticular fibroblasts, preadipocytes, and skin adipocytes in the lower dermis (Donati et al., 2014;Driskell et al., 2013;Festa et al., 2011). Immunolabeling confirmed that most reticular fibroblasts express Sca1 at P2 (Figure 1c and d). We have previously shown that fibroblasts that express the protein Delta homolog 1 (Dlk1) at E16.5 give rise to the Sca1 þ fibroblasts and mature adipocytes present in P2 and adult skin, although Dlk1 is no longer expressed in adult skin (Driskell et al., 2013;Lichtenberger et al., 2016). To compare fibroblasts from the upper and lower dermis, we sorted EGFP þ cells from P2 dorsal dermis and fractionated the cells further on the basis of Sca1 surface marker expression (Figure 1e). We also sorted EGFP -/Sca1cells as a control. expression. (fem) Quantitative real-time PCR analysis of mRNA levels in sorted cell populations: PdgfraEGFP -/Sca1 -, PdgfraEGFP þ /Sca1and PdgfraEGFP þ / Sca1 þ . Genes were normalized to the Gapdh gene. Error bars represent standard error of the mean of replicates from four mice. *P 0.05, **P 0.005, ***P 0.0005 compared with GFPcells; # p 0.05 compared with EGFP þ /Sca1cells. ad, adipocyte; epi, epidermis; der, dermis; EGFP, enhanced green fluorescent protein; GFP, green fluorescent protein; HF, hair follicle; pap, papillary; PDGFRa, platelet-derived growth factor receptor a; ret, reticular. M Mastrogiannaki et al. b-Catenin Regulates Fibroblast Lineages Quantitative real-time PCR (QPCR) analysis of mRNA isolated from the different dermal subpopulations showed that mRNAs corresponding to Pdgfra and the fibroblast marker genes Col1a2 and Vimentin were highly enriched in both Pdgfra þ subpopulations relative to Pdgfracells (Figure 1feh), whereas mRNA corresponding to Ly6a (Sca1) was highly enriched in the Pdgfra þ /Sca1 þ fraction (Figure 1i), confirming the relative purity of the sorted cell populations. The adipocyte/preadipocyte marker genes Fabp4, Perilipin, Pparg, and Dlk1 were also enriched in Pdgfra þ /Sca1 þ cells, consistent with previous reports (Driskell et al., 2013;Festa et al., 2011) (Figure 1jem). The QPCR results were confirmed by immunofluorescence labeling of P2 dorsal skin with antibodies to Fabp4 and Perilipin ( Figure 1neq).

Differential expression of Wnt pathway genes in upper and lower dermal fibroblasts
To explore the differences between Pdgfra þ /Sca1 þ and Pdgfra þ /Sca1fibroblasts, we carried out gene expression profiling using RNA from flow-sorted cells. We found that 1,457 entities were regulated by more than 2-fold (t test, P < 0.05) (Figure 2a; see Supplementary Table S1 online), showing that global differences in gene expression distinguish the two fibroblast subpopulations. In addition to differential expression of adipogenic genes, there was differential expression of genes encoding zinc finger proteins (Gupta et al., 2012) and regulators of the Wnt, BMP, Notch, and PDGF signaling pathways (Figure 2b and c).
Because Wnt/b-catenin signaling is known to regulate dermal development, the differential expression of genes associated with this pathway was of particular interest ( Figure 2c). Several Wnt/b-catenin pathway genes were differentially regulated in Pdgfra þ /Sca1 þ and Pdgfra þ /Sca1fibroblasts, which we confirmed by QPCR in independent biological samples ( Figure 2dej). Pdgfra þ /Sca1 þ fibroblasts expressed significantly lower levels of Wnt5a ligand, the Wnt receptor Frzb, and the Wnt effector Lef1, as well as several other Wnt pathway genes including Axin2 and Dkk1 (Figure 2deh; see also Driskell et al., 2013). However, Sca1 þ cells expressed significantly higher levels of the Wnt receptor Fzd4 and the Wnt effector Tcf7l2 (Figure 2i and j). Tcf7l2, commonly known as Tcf4, is expressed in human adipose tissue, and gene variants are associated with susceptibility to Type 2 diabetes and inability to lose weight after lifestyle interventions (Cauchi et al., 2006;Haupt et al., 2010). There was no significant difference in b-catenin mRNA levels in Pdgfra þ /Sca1and Pdgfra þ /Sca1 þ fibroblasts at P2 (Figure 2k). However, immunostaining showed differential protein expression of b-catenin in the upper and lower dermis of neonatal skin, with high levels of nuclear b-catenin in papillary fibroblasts and only few nuclear b-cateninepositive cells within the adipose tissue ( Figure 2lep).
We conclude that neonatal dermis is compartmentalized such that Wnt/b-catenin signaling pathway components are differentially expressed in Sca1 þ and Sca1fibroblasts.
Constitutive b-catenin stabilization in postnatal skin fibroblasts reduces the adipocyte layer and disturbs the hair growth cycle Given the inhibitory effect of Wnt/b-catenin signaling on adipogenic differentiation (Gesta et al., 2007;Kennell and MacDougald, 2005;Longo et al., 2004), we speculated that activating the pathway in postnatal skin fibroblasts would change the composition of the dermis by altering neonatal fibroblast lineages or differentiation. To determine the effect of active Wnt/b-catenin signaling in all neonatal fibroblasts, we produced crosses between PdgfraCreER T2 (Rivers et al., 2008) and Ctnnb1 Exon3 Flox/þ (Harada et al., 1999) mouse strains. Recombination of Ctnnb1 Exon3 Flox/þ (referred as cDex3) produces a variant of b-catenin that is resistant to phosphorylation by glycogen synthase kinase 3b and degradation by the adenomatous polyposis complex. Littermates were treated topically with 4-hydroxy-tamoxifen (4-OHT) on the day of birth and analyzed at different time points thereafter ( Figure 3a).
The recombination efficiency was similar in different fibroblast subpopulations ( Figure 3b). Lrig1 þ /Sca1cells are resident in the papillary dermis; Dlk1 þ /Sca1cells are a subpopulation of P2 reticular fibroblasts; Sca1 þ cells, which can be Dlk1 þ or Dlk1 -, are found in the lower dermis and subdermal fat layer (Driskell et al., 2013). Analysis of skin sections confirmed that tdTomato þ cells were present in all dermal layers (Figure 3c and d).
A high abundance of nuclear b-catenin was detected in tdTomato þ fibroblasts of cDex3 skin at P4 (see Supplementary Figure S1cef) and was even more pronounced at P56 ( Figure 3eeh). Furthermore, QPCR showed increased (albeit not significant) expression levels of known Wnt/b-catenin target genes such as Nov, Sp5, and Gpr165 (Hamburg-Shields et al., 2015) in flow-sorted tdTomato þ fibroblasts isolated from mutant skin compared with control skin (see Supplementary Figure S1g) at P4. Interestingly, at P56 expression levels of Nov were similar in tdTomato þ /Sca1fibroblasts in wild-type and mutant dermis but significantly up-regulated in tdTomato þ /Sca1 þ fibroblasts, thereby reaching similar levels between papillary and reticular fibroblasts (see Supplementary Figure S1h).
No differences in skin morphology were observed between mutant and control mice at P4 (Figure 3i and j). However, at we observed fibroblast-rich regions within the adipocyte layer of cDex3 mice (Figure 3k and l). At P35 these fibrotic regions were prominent (Figure 3m and n), and at P56 the adipocyte layer had been largely replaced by fibrotic dermis (Figure 3o and p).
To assess whether Wnt/b-catenin signaling in fibroblasts affects the hair cycle, we classified different stages of HFs according to their morphology (Muller-Rover et al., 2001;Paus et al., 1999) and also examined their length. For each time point we examined at least 8 HFs per skin sample and skin biopsies from 2e8 different mice of the same genotype. We examined skin at P4 and P11 (n ¼ 3e4), during the growth phase of HF morphogenesis; at P18 (n ¼ 3e5), P24 (n ¼ 2e4), and P28 (n ¼ 2), during the first postnatal telogen; at P35 (n ¼ 6), during anagen; and at P42 (n ¼ 4e8) and P56 (n ¼ 3e6), when the follicles are entering catagen and the second telogen, respectively. Regardless of genotype and age, all HFs were synchronized, with the exception of P35 cDex3 mice, when two of eight mice were still in telogen and the others were in anagen.
HF morphogenesis occurred normally in cDex3 mice (day 4 and 10; Figure 3q and r), as did the first telogen (day 18) and first anagen (day 35; Figure 3q and r). However, from day 42 the hair cycle was disturbed in both male ( Figure 3q) and female ( Figure 3r) cDex3 mice, and cDex3 follicles remained in anagen, as assessed both by morphology and length, when wild-type follicles were entering telogen (Figure 3oer). Based on the observation that in two of eight mutant mice the HFs were still in telogen at P35, which were not scored in (Figure 3m and n), we speculate that there may be an extended growth phase before the second catagen. This is supported by the observation that there was an increased number of proliferating phosphohistone H3 þ cells in the outer root sheath of HFs in mutant mice (Figure 3s). Nevertheless, the HFs in mutant skin eventually entered catagen, and all HFs were in telogen at the age of 3 months (n ¼ 4; see Supplementary Figure S1i and j). Thus, our findings indicate that fibroblast-specific b-catenin stabilization perturbs the hair cycle and extends anagen.

Constitutive b-catenin activation results in dermal fibrosis
To characterize the effect of activated Wnt/b-catenin signaling in fibroblasts, we focused on P35 skin (Figure 4). The fibrotic areas within the adipose layer of cDex3 dermis labeled pink with the histochemical Herovici's stain, indicative of the mature fibrillar collagen (Collins et al., 2011) (Figure 4a and b). QPCR showed reduced expression of differentiation markers such as Fabp4, Perilipin, and Cebpa ( Figure 4c) in mutant tdTomato þ /Sca1 þ fibroblasts compared with controls. Immunostaining confirmed reduced protein expression of Perilipin and Fabp4 but showed higher levels of the preadipocyte marker Cd24 (Figure 4dei). The reduction in the number of differentiated adipocytes was confirmed by LipidTox (Invitrogen/Thermo Fisher Scientific, Waltham, MA) staining (Figure 4j and k) and was statistically significant (Figure 4l). Fibroblast proliferative activity, measured by 5-Ethynyl-2'-deoxyuridine (EdU) labeling, was increased in both the adipocyte layer and in the rest of the dermis, including papillary fibroblasts of the dermal sheath and reticular fibroblasts, the effect being most pronounced in the adipocyte layer ( Figure 4meo). However, we did not detect significant numbers of EdU þ DP cells, consistent with our previous finding that the number of proliferating cells in adult DP is very low, even when b-catenin is stabilized and the DP increases in size (Kaushal et al., 2015).
Selective Wnt/b-catenin stabilization in Dlk1 D fibroblasts results in fibrosis within the adipose layer Given the differential expression of adipogenic genes in Sca1 þ and Sca1fibroblasts at P2 (Figure 2) and the fibrosis within the adipose layer on fibroblast-specific stabilization of b-catenin (Figure 3), we hypothesized that the fibrotic effect of activating Wnt/b-catenin signaling in all Pdgfra þ fibroblasts would be reproduced by selectively targeting the lower dermal lineage during skin development. Because fibroblasts that express Dlk1 at E16.5 give rise to the Sca1 þ fibroblasts and mature adipocytes present at P2 and in adult skin (Driskell et al., 2013), we crossed Ctnnb1 Exon3 Flox/þ , Dlk1CreER T2 , and tdTomato-LSL mice and treated them with tamoxifen at E16.5 to target the lower dermal lineage (Figure 5a).
In contrast to the effect of activating Wnt/b-catenin signaling in all fibroblasts, selective activation in Dlk1 þ cells did not disturb the hair cycle, and at P56 both control and Dlk1LTDex3 skin was in telogen (Figure 5f and g). Nevertheless, there was an accumulation of fibroblasts within the adipose layer in with activated Wnt/b-catenin signaling. Note that the PDGFRaCreER T2 transgene is also active in cells of the dermal papilla (white arrowheads in g).   Dlk1LTDex3 skin (Figure 5f and g). These regions stained positively for fibrillar ECM using Herovici's dye (Figure 5h and i) and expressed high levels of b-catenin (Figure 5j and k). The reduction in adipocyte numbers was confirmed by labeling for caveolin-1, a marker for preadipocytes and adipocytes (Figure 5l and m) and was statistically significant, in accordance with the appearance of fibrotic lesions (Figure 4n and o).
Stabilization of b-catenin via Dlk1CreER T2 also resulted in an increase in fibroblast proliferation both in the adipocyte layer and throughout the dermis (Fig. 5per).

Postnatal Wnt/b-catenin stabilization inhibits adipogenesis of Sca1 D fibroblasts in culture
To establish whether b-catenin stabilization within preadipocytes resulted in an inhibition of differentiation, we analyzed the behavior of single cells captured in ECM-functionalized hydrogels, as described previously (Driskell et al., 2012;Driskell et al., 2013). PdgfraCreER T2 , tdTomato-LSL, and Ctnnb1 Exon3 Flox/þ mice were crossed and treated with 4-OHT on the day of birth (Figure 6a). Sca1 þ tdTomato þ fibroblasts were isolated by flow cytometry 2 days later (Figure 6b and c). Activation of Wnt/b-catenin signaling did not affect the proportion of Sca1 þ cells that expressed the preadipocyte marker Cd24 (Festa et al., 2011) (Figure 6d). Sca1 þ cells were encapsulated at clonal density in 3-dimensional hydrogels and cultured in control or adipogenic medium. After 10 days in culture, individual clones were scored for cell number and the total intensity of LipidTox staining (Figure 6e). Whereas b-catenin stabilization in vivo stimulated proliferation within the adipose layer (Figure 4m), there was no effect on clone size when Sca1 þ cells were cultured in adipogenic medium, and in standard medium the only effect was to increase the percentage of two cell clones at the expense of clones with higher cell numbers (Figure 6f and g). Wnt/b-catenin signaling activation did, however, decrease adipocyte differentiation, as evaluated by decreased LipidTox staining, both in standard medium ( Figure 6h) and in medium supplemented with adipogenic factors, regardless of clone size (Figure 6i). These findings suggest that the effect of b-catenin stabilization in neonatal fibroblasts is to prevent Sca1 þ fibroblasts from undergoing adipocyte differentiation in a cell-autonomous manner, rather than changing the proportion of Cd24 þ /Sca1 þ preadipocytes.

DISCUSSION
Here we have examined the effect of dermal Wnt/b-catenin signaling in the context of fibroblast heterogeneity. We show that expression of Wnt pathway genes such as Wnt5a, Lef1, Tcf4, or Dkk1 differs between upper (Sca1 -) and lower (Sca1 þ ) dermis. Interestingly, immunostaining showed that b-catenin and Lef1 are more abundant in Sca1cells. b-Catenin stabilization in all fibroblast populations in neonatal skin resulted in a decrease in mature adipocytes and the appearance of fibrotic regions in the adipose layer, accompanied by stimulation of fibroblast proliferation throughout the dermis. The replacement of adipocytes by ectopic reticular fibroblasts also occurred when the lower dermal lineage was selectively targeted at E16.5, before adipocyte differentiation.
Our in vitro studies showed that b-catenin stabilization did not drive fibroblast proliferation cell-autonomously. The effect of b-catenin stabilization was to inhibit terminal differentiation of Sca1 þ cells rather than to promote expansion of preadipocytes or selective proliferation of reticular fibroblasts. This is similar to the effect of b-catenin stabilization on cultured human keratinocytes, which is to expand the stem cell compartment without stimulating proliferation (Zhu and Watt, 1999). However, although we did not detect an expansion of preadipocytes within 4 days of b-catenin stabilization in vivo, Cd24 expression was increased in P35 skin. This could either be caused by an expansion of preadipocytes at a later stage, which is supported by increased fibroblast proliferation in vivo, or by the differentiation defect.
Our conclusion that targeting the lower dermal lineage accounts for the appearance of fibrotic regions is consistent with other studies highlighting the contributions of different dermal cell subpopulations to fibrosis. For example, fibroblasts that express engrailed 1 during development are responsible for the bulk of ECM deposition in dorsal dermis and mediate dermal fibrosis in response to irradiation (Rinkevich et al., 2015). Other dermal subpopulations that contribute to fibrosis include Sox2 þ cells (Liu et al., 2014) and adiponectin-positive adipocyte precursors (Marangoni et al., 2015). In addition, Adam12 þ perivascular mesenchymal cells expressing Pdgfra and Sca1 are profibrotic in response to injury (Dulauroy et al., 2012).
The inhibition of adipocyte differentiation was not the only effect of dermal b-catenin activation that we observed. There was an increase in proliferation throughout the dermis, consistent with an earlier report (Cheon et al., 2002), and a disruption of the HF cycle. The effects on the hair cycle cannot be attributed solely to a decrease in mature adipocytes (Donati et al., 2014;Festa et al., 2011), because they were observed when we targeted all fibroblast subpopulations via PdgfraCreER T2 and not when we selectively targeted the lower dermal lineage. This raises an interesting question as to whether increased proliferation of the fibroblast subpopulations that are known to regulate hair growth, in particular the DP, dermal sheath, and other papillary fibroblasts (Driskell et al., 2013;Enshell-Seijffers et al., 2010;Kaushal et al., 2015), is responsible. We have previously shown that b-catenin stabilization in Prominin-1eexpressing DP cells results in an increase in DP size, but because the number of EdU þ cells is very low, the increase could be attributable to a number of factors, such as increased migration from the dermal sheath (Kaushal et al., 2015). Because the PdgfraCreER T2 transgene is active in DP cells (Figure 3g and h), it is possible that the perturbed hair cycle reflects, at least in part, Wnt/b-catenin activation in DP cells. and causes a down-regulation of adipogenic effector pathways, such as cEBP/PPARg, in culture (Gesta et al., 2007). In addition, expression of the Wnt inhibitor Dkk1 interferes with profibrotic signaling . Together, these studies suggest that pharmacological modulation of Wnt signaling could be beneficial in the treatment of fibrotic skin diseases, including scleroderma (Ohgo et al., 2013). Similar mechanisms could also be responsible for the regulation of mesenchymal progenitors in other organs, including Pdgfra þ /Sca1 þ muscle fibroblasts with proliferative and myofibroblast/adipogenic potential in response to muscle regeneration and ectopic fat formation (Uezumi et al., 2010).

In vivo experiments
The following mouse strains were maintained on a C57 Bl6/CBA background: Dlk1CreER T2 Institut Clinique de la Souris (ICS, Illkirch-Graffenstaden Cedex Alsace, France), PdgfraEGFP (PdgfraH2B-eGFP) (Jackson Laboratories, Bar Harbor, ME), Ctnnb1lox(ex3)/ þ (Harada et al., 1999). PdgfraCreER T2 (Rivers et al., 2008) and Rosa-CAG-LSL-tdTomato (Jackson Laboratories; 007905) strains were maintained as homozygotes. Cre-mediated recombination was induced by topical application of 2 mg of 4-hydroxy-tamoxifen (Sigma, St. Louis, MO; diluted in acetone þ 10% DMSO) on the back skin of neonatal pups at P0eP1. For lineage tracing experiments, plugged females received a single intraperitoneal injection of dissolved in corn oil (10e20 mg/ml) by intermittent sonication at 37 C for 20e30 minutes. Pups were harvested and fostered after surgical removal between E18.5 and E21. For in vivo proliferation assays, mice received a dose of 500 mg 5-ethynyl-2 0 -deoxyuridine (EdU; 2.5mg/ml in phosphate buffered saline) by intraperitoneal injection 2e4 hours before they were killed. Cohorts included male and/or female littermates, and results are representative of at least three biological samples. All experimental procedures were carried out under the terms of a UK Home Office project license after ethical review at Cambridge University or King's College London.

Histology
Skin samples were harvested and processed to generate paraffin (5 mm) or thick cryopreserved whole-mount tissue (50e100 mm) sections as previously described (Collins et al., 2011;Driskell et al., 2013). Immunostaining was performed using the following antibody combinations: rabbit anti-Rfp ( Invitrogen) or glycerol was used for mounting slides or wholemount sections, counterstained with 4 0 , 6-diamidino-2phenylindole (DAPI). The EdU click-it imaging kit (Invitrogen) was used for detecting incorporated EdU nucleoside analogs in proliferating cells. Herovici's staining was performed as previously described (Collins et al., 2011).

Image acquisition and quantification
Immunostained tissue sections were imaged using a Nikon A1R confocal microscope. Adobe Photoshop CS6 (Adobe, San Jose, CA) was used to adjust images and correct background. Bright field images were collected using either a Hamamatsu NanoZoomer slide scanner (Hamamatsu, Japan) or a Zeiss Axiophot microscope (Carl Zeiss, Oberkochen, Germany) with a Â10 objective and an Axio-Cam HRc camera. Image measurements were performed on at least six microscopic fields per biological sample.

Hydrogel culture and high-content imaging
Sorted fibroblasts were collected and encapsulated in Extracel (Glycosan Biosystems) as previously described (Driskell et al., 2012), at a density of 5 Â 10 5 /ml in mClear 96-well plates (Corning). Cultures were maintained for 10 days at 37 C/5% CO 2 in standard medium (Dulbecco's modified Eagle's medium þ 10% fetal bovine serum þ 1% penicillin/streptomycin) or Adipogenic medium (StemXVivo Osteogenic/Adipogenic Base Media with Adipogenic Supplement; R&D Systems), and the medium was changed every 2e3 days. Cultures were fixed with 4% paraformaldehyde for 10 minutes, washed twice in phosphate buffered saline, permeabilized with 0.1% Triton X-100 for 5 minutes and stained with LipidTox, DAPI, and rabbit anti-Rfp (1:300 dilution; Abcam). Multiple image stacks of each colony were recorded using an Operetta High Content Imaging system (Perkin Elmer, Waltham, MA). Spheres were identified by staining for tdTomato, and the number of nuclei per sphere was determined using quantification algorithms in the Columbus analysis software (Perkin Elmer, Waltham, MA). The fluorescence intensity of LipidTox staining was measured for each sphere.

Quantification and Statistics
GraphPad Prism 6 software (GraphPad Software Inc., La Jolla, CA) was used for all analyses of numerical data and generation of graphs and statistical tests, including one-way analysis of variance or Student t test. Error bars represent standard error of the mean.

Microarray
Genome-wide expression profiling was carried out by the Paterson Institute Microarray Core Facility, as described previously (Collins et al., 2011). The data are deposited in the National Institutes of Health GEO repository under accession number GSE76751. Complementary DNA was hybridized to Affymetrix MG430.2A arrays (Affimetrix, Santa Clara, CA). Array images produced by the Affymetrix PICR 3000 scanner were imported as CEL files into Genespring GX11 (Agilent Technilogies, Santa Clara, CA) for analysis. Robust Multi-array Average normalization (baseline to median of all samples) was used. Analyses were performed on genes selected for expression above the bottom 20th percentile in all three samples within at least one of two experimental groups. To identify differentially expressed genes, we compared the two groups using the Student t test (P-value cutoff of 0.05). Entities regulated by greater than 2-fold were selected for further analysis.