A subset of UCP1-inducers modulate adipocyte lipid droplet morphology

White adipocytes have a single or a few large lipid droplets, whereas the brown adipocytes contain multiple small lipid droplets. We exploited this morphological difference to evaluate the browning effects of newly identified UCP1-inducer compounds. Following treatment of PSC-WA with selected compounds (Fig. 1b), we used fluorescent dyes to mark nuclei and lipid droplets, and performed high content imaging to quantify the number and size of lipid droplets per cell. After 7 days of treatment with selected JAK3 and SYK inhibitors, we observed changes in lipid droplet morphology that are typical of brown-like adipocytes (Fig. 2a). Contrastingly, a THRB agonist was inert in this assay despite elevation of UCP1 expression (Fig. 2a and ​and1d).1d). We determined the lipid area of small (<1070μm²) and large (>1070μm²) droplets per well (Fig. 2b, curve graphs) and used the ratio small/large droplet area as a ‘brown-like lipid index’ for each compound, as exemplified in Figure 2c for DMSO, JAK3 inhibitor, SYK inhibitor and THRB agonist. A total of 39 compounds that induced UCP1 or displayed visible effects on lipid morphology during the browning screen were thereby quantified and ascribed a brown-like lipid index (Fig. 2d, Y-axis). When the two browning indices were taken into account, i.e. UCP1/FABP4 and lipid droplet morphology, JAK3 and SYK inhibitors were the most potent browning compounds from our screen (Fig. 2d). In addition, this analysis revealed that brown-like lipid morphology was poorly correlated with elevated UCP1 expression, indicating that quantification oflipid droplet size is not a suitable phenotypic readout for adipocyte browning, at least in our system.

Open in a separate window

Figure 2

Selected compounds modulate lipid morphology

a) PSC-WA were differentiated and treated as described in Fig1B. At day 14, cells were fixed, stained and imaged by confocal microscopy. Green: lipids, Red: nuclei. Scale bars, 50μm. Images are representative of n =3 biological replicates.

b) Quantification of changes in lipid morphology is depicted as fraction of total lipid area (Y axis) per lipid droplet size (X axis). JAK3 inhibitor, SYK inhibitor and THRB agonist-treated cells are shown in red and DMSO in gray. Values represent the mean of two biological replicates.

c) Bar graph illustrating the quantification of a brown-like lipid index, determined by calculating the ratio of total area for small (<1070μm²) versus large (>1070μm²) lipid droplets for the graphs in b), normalized to the DMSO control sample. Values represent the mean of two biological replicates.

d) Scatter plot showing the relation between UCP1/FABP4 mRNA (X axis) and brown-like lipid morphology (Y axis) upon treatment with 39 selected compounds. Brown-like lipid index refers to the ratio of small/large lipid droplets normalized on DMSO as determined in b). Values represent the mean of two biological replicates.

Validation of tofacitinib and R406 browning compounds in primary adipocytes

As model browning compounds, we selected the JAK3 inhibitor tofacitinib and the SYK inhibitor R406 for their potency in our browning assay and for their clinical relevance (Supplementary Figure 3). We examined the dose-response of tofacitinib and R406 in PSC-WA and found that the UCP1/FABP4 ratio reaches a plateau at 2 μM for tofacitinib and 1 μM for R406 (Fig. 3a). Graphical representation of UCP1 and FABP4 normalized on housekeeping gene revealed a UCP1-specific effect of tofacitinib, while R406 induced both UCP1 and FABP4 expression when dosed above 1 μM (Fig. 3a). We confirmed that tofacitinib and R406-mediated increases in UCP1 mRNA levels translated into accumulation of UCP1 protein (Fig. 3b) and observed concomitant elevation in PRDM16 protein (Fig. 3b).

Open in a separate window

Figure 3

Validation of tofacitinib and R406 browning compounds in primary adipocytes

a) bDNA analysis of dose response with tofacitinib and R406. At high doses, R406 increases both UCP1 and FABP4 expression but UCP1/FABP4 remains above 2. Values represent the mean of two biological replicates.

b) Western blot analysis showing that up-regulation of UCP1 and PRDM16 protein levels correlates with up-regulation of UCP1 mRNA by tofacitinib and R406.

c) bDNA analysis showing that tofacitinib (tofa.) and R406 increase UCP1 expression in human primary adipocytes. ADSC: Adipose tissue-derived stromal cells. Values are mean ± s.d. of n = three biological replicates and differences from DMSO are significant for * P < 0.05. P values were calculated using the two-tailed paired Student's t-test.

d) Bright field images showing that tofacitinib (tofa.) and R406 induce brown-like lipid morphology (arrows) in human primary adipocytes more prominently than BMP7. ADSC: Adipose tissue-derived stromal cells. Scale bars, 20μm. Data representative of 3 independent experiments. For uncropped images see Supplementary Figure 3.

e) RT-PCR analysis of UCP1 gene expression in mouse subcutaneous white adipose tissue (WAT) explants following 7 days of treatment with the indicated compound. Values are mean ± S.E.M. of n = three biological replicates of pooled tissue from 5 mice and differences from DMSO are significant for * P < 0.05 and ** P < 0.005. P values were calculated using the two-tailed paired Student's t-test.

Importantly, we validated the effects of tofacitinib and R406 on UCP1 and FABP4 expression in lentiviral-free, human primary adipose tissue-derived stem cells (ADSCs) differentiated according to standard procedures (Fig. 3c). Tofacitinib and R406 also caused the emergence of brown-like lipid droplet morphologies in ADSC-derived adipocyte cultures and appeared superior to BMP7 in this assay (Fig. 3d). Finally, we subjected mouse white adipose tissue explants to increasing doses of tofacitinib and R406 and observed up-regulation of UCP1 expression in explants of sub-cutaneous, but not visceral, origin (Fig. 3e and Supplementary Figure 4).

Tofacitinib and R406 block the JAK-STAT1/3 pathway during adipocyte browning

Tofacitinib is a potent inhibitor of signaling through JAK1 and JAK3 with 5-100 fold selectivity over JAK2 in cell-based assays^{23, 24}. R406 is described as a potent and specific SYK inhibitor, but also targets JAKs, c-kit, Lck, and FLT3 when profiled at 2 μM²⁵. We sought to determine the pharmacological signature of tofacitinib and R406 in human adipocytes where target abundance and signaling activity might differ from immune cells. First we used RNA sequencing analysis of PSC-WA to evaluate the abundance of known targets of tofacitinib and R406. JAK1 was the most abundantly expressed member of the Janus kinase family, followed by Tyk2, JAK2 and JAK3 (Figure 4a). The abundance of SYK, c-kit, Lck and FLT3 was below statistical significance suggesting that R406 may exert its browning effect via JAK inhibition.

Open in a separate window

Figure 4

Inhibition of STAT phosphorylation downstream of tofacitinib and R406

a) Transcript abundance in RPKM (reads per kilobase transcript per million reads) of known targets of tofacitinib and R406 indicating that the JAK kinases are predominantly represented in PSC-WA. Values are mean ± s.d. of n = three biological replicates.

b) Western blot analyses of STATs, AKT and MAPKs in PSC-WA previously treated with DMSO, tofacitinib and R406 for 20 minutes or 7 days showing a pronounced inhibition of STAT phosphorylation by tofacitinib and R406 at both time points. R406 also significantly decreased phosphorylation levels of AKT and ERK1/2. The data shown is representative of two independent experiments.

c) Western blot analysis of a dose-response with tofacitinib and R406 showing the correlation between UCP1 accumulation and inhibition of STAT3 phosphorylation. The data shown is representative of two independent experiments.

e) bDNA analysis of PSC-WA differentiated as in figure 1b) and treated with TNFα at day 12, indicating that the negative effect of TNFα on UCP1 expression is rescued by tofacitinib and R406. Values represent the mean of two biological replicates.

f) Tofacitinib (tofa.) and R406 do not synergize during adipocyte browning. PSC-WA were treated with tofacitinib, R406 or a combination of both and analyzed for UCP1 expression by bDNA. Values are mean ± s.d. of n = three biological replicates and differences from DMSO are significant for * P < 0.05. P values were calculated using the two-tailed paired Student's t-test.

d) The JAK 1/2 inhibitor Ruxolitinib positively modulates UCP1 expression (graph) and inhibits STAT1/3 phosphorylation (right panels) in PSC-WA. Graph values of bDNA analysis represent the mean of two biological replicates. Images of western blot analysis are representative of two independent experiments.

g) bDNA analysis showing that the JAK1/2 inhibitors CYT387, AZD1480 and Baricitinib positively regulate UCP1 expression in PSC-WA. Values are mean ± s.d. of n = three biological replicates and differences from DMSO are significant for * P < 0.005. P values were calculated using the two-tailed paired Student's t-test.

h) Tofacitinib (tofa.) synergizes with THRB agonist (right panel) but not with Ruxolitinib (Ruxo., left panel) during adipocyte browning. Values are mean ± s.d. of n = three biological replicates and differences from DMSO are significant for * P < 0.005. P values were calculated using the two-tailed paired Student's t-test.

To profile possible downstream targets, PSC-WA were treated with tofacitinib and R406 at the determined ‘browning dose’ (2 and 1 μM respectively) for 20 minutes and screened by reverse phase protein array (RPPA) covering 51 different phospho-isoforms. RPPA analysis revealed STAT3 as the main downstream target for both tofacitinib and R406 (Supplementary Figure 5). While the effect of tofacitinib was restricted to STAT3 in the RPPA panel, R406 had additional downstream targets including AKT, EGFR and ERK1/2 (Supplementary Figure 5). We further investigated the effect of tofacitinib and R406 on JAK/STAT, AKT and ERK1/2 pathways by immunoblot analysis. As shown in Figure 4b, tofacitinib and R406 blocked tyrosine phosphorylation of STAT1 and STAT3 after 20 minutes, an effect that is exacerbated upon 7 days of treatment. In a dose response experiment, the gradual increase in UCP1 levels negatively correlated with STAT3 phosphorylation (Fig. 4c). Noticeably, STAT1 protein level decreased after treatment with tofacitinib and R406. Tofacitinib had no effect on AKT and MAPKs, whereas R406 significantly decreased the phosphorylation level of AKT, ERK1 and ERK2 (Fig. 4b). AKT phosphorylation was restored at day 7, suggesting an adaptive response to maintain adipocyte homeostasis and insulin signaling. Phospho-STAT5 was undetectable in our assay and p38 and JNKs were unaltered at both times. Thus, JAK inhibitors likely change UCP1 expression via a different route than the cAMP/p38 pathway downstream of β3-adrenergic stimulation^{26, 27}.

Tumor necrosis factor alpha (TNF-α) represses UCP1 expression in brown adipocytes²⁸, an effect also observed in PSC-WA (Fig.4d). Since TNF-α can signal through JAK/STAT and ERK pathways, we tested CP-690550 and R406 ability to antagonize TNF-α repression of UCP1 expression. CP-690550 and R406 partially and fully restored UCP1 levels respectively, in agreement with the JAK/STAT-specific and JAK/STAT-ERK1/2 effects of each compound (Fig.4b). In addition, the combination of CP-690550 and R406 was not superior to R406 alone in increasing UCP1 expression (Fig.4e), supporting evidence that CP-690550 and R406 functionally overlap.

Finally, we tested Ruxolitinib, a JAK1/2 inhibitor, and observed up-regulation of UCP1 expression without significant regulation of FABP4 with concomitant inhibition of STAT1/3 phosphorylation (Fig. 4f). Three other JAK1/2 inhibitors, AZD1480, CYT387 and Baricitinib, also showed up-regulation of UCP1 expression (Fig. 4g). A combination of tofacitinib and Ruxolitinib did not further increase UCP1 expression (Fig. 4h, left panel), unlike a combination of tofacitinib with a THRB agonist (Fig. 4h, right panel). Altogether, the data highlight a central role for the JAK1/2-STAT1/3 pathway in UCP1 expression in adipocytes.

Progressive and stable conversion of adipocytes by JAK inhibition

Molecules that increase UCP1 expression by directly activating the UCP1 promoter²⁶ may not be suitable for in vivo induction of browning or therapeutically useful to combat obesity. Alternative mechanisms of browning such as adipocyte metabolic remodeling are needed, and our screening platform was designed for this purpose. We examined the kinetics of tofacitinib and R406-mediated browning in comparison to THRB agonist, which also ranked in the top-10 best UCP1 inducers but failed to modify lipid droplet morphology (Fig. 1d and ​and2).2). The UCP1 promoter contains a thyroid hormone responsive element (TRE)²⁰ and as expected, a THRB agonist rapidly increased UCP1 levels (Fig. 5a). Contrastingly, UCP1 mRNA levels increased more slowly and progressively in tofacitinib and R406-treated adipocytes (Fig. 5a), consistent with remodeling of cellular metabolism rather than direct modulation of UCP1 promoter activity.

Open in a separate window

Figure 5

JAK inhibition stably induces a brown-like profile in adipocytes

a) bDNA quantification of UCP1 mRNA levels over time, showing that the progressive accumulation of UCP1 induced by tofacitinib and R406 contrasts with the acute effect of THRB agonists. Values represent the mean of two biological replicates.

b) Schematic illustration of experimental design for b), c), and d): PSC-WA were treated with compounds for 7 days, washed 3 times with compound-free medium, and maintained in compound-free medium for an additional 14 day-period. Images were captured at day 28 prior to addition of lysis buffer and bDNA analysis. Only JAK3i and SYKi-pre-treated cells displayed high levels of UCP1 mRNA relative to DMSO. Values represent the mean of two biological replicates.

c) Bright field images showing a reduced lipid vacuole size at day 28 for tofacitinib (tofa.) and R406-pre-treated adipocytes. Scale bars, 50μm. Images are representative of two independent experiments.

d) DMSO, tofacitinib (tofa.) and R406-pre-treated adipocytes were exposed to TNFα from day 26 to 28. bDNA analysis shows that pre-treatment with tofacitinib and R406 protects adipocytes from TNFα-mediated down-regulation of UCP1 expression. Values represent the mean of two biological replicates.

If a stable white to brown-like conversion is achieved via JAK inhibition, the brown-like phenotype should persist upon removal of CP-690550 and R406. To test this, we selected 9 inducers of UCP1 mRNA levels from the original screen and treated PSC-WA with each compound for 7 days, washed with compound-free medium, and maintained in compound-free medium for an additional 14 day-period (Fig. 5b). Remarkably, all cells pre-treated with JAK inhibitors displayed high levels of UCP1 mRNA and reduced lipid droplet size at day 28 compared to DMSO-treated cells while all other compounds lost their browning effect after washout (Fig. 5b-c). Finally, we repeated the washout experiment with tofacitinib and R406 and added TNF-α at day 26, 48 hours prior to cells collection. Even in the absence of compounds, adipocytes pre-treated with tofacitinib and R406 maintained higher level of UCP1 when challenged with TNF-α (Fig. 5d). Altogether, these results indicate that JAK inhibition leads to the stable acquisition of brown-like metabolic properties in human adipocytes.

JAK-inactivated adipocytes acquire a brown-like metabolic program

Brown adipocytes are characterized by high mitochondrial content as well as high metabolic activity¹. Mitochondrial content was assessed by determining the ratio of subunit I of Complex IV (COX-I), which is mitochondrial DNA-encoded, on the 70kDa subunit of Complex II (SDH-A), which is nuclear DNA-encoded. Quantification of COX-1 and SDH-A immunoblots showed a significant up-regulation of mitochondrial content in R406 and tofacitinib-treated PSC-WA and ADSC adipocytes (Fig. 6a). To determine the impact of these changes on metabolic activity, oxygen consumption rate (OCR) was measured. We found that basal, uncoupled and maximal respiration were increased when PSC-WA and ADSC adipocytes were pre-treated for 7 days with tofacitinib or R406 as compare to a DMSO control (Fig. 6b). Thermogenesis in brown adipocytes is fuelled by lipid catabolism, which can be measured by release of glycerol from fatty acid. We quantified the level of free glycerol from adipocyte culture medium and observed a ~2.5 fold increase of basal lipolysis in tofacitinib and R406-treated adipocytes (Fig. 6c), but no significant increase in forskolin (FSK) stimulated lipolysis (Fig. 6c). Altogether, these data suggest the engagement of a brown-like metabolic program upon inhibition of JAK/STAT in human adipocytes.

Open in a separate window

Figure 6

JAK inhibition induces brown-like metabolic properties in adipocytes

a) Mitochondrial content was assessed by determining the ratio of COX-I on SDH-A protein levels by immunoblots as shown in upper panel for PSC-WA. Quantification of immunoblots revealed up-regulation of mitochondrial content in tofacitinib and R406-treated PSC-WA and ADSC adipocytes. Values are mean ± s.e.m. of n = three biological replicates and differences from DMSO are significant for * P < 0.05 and ** P < 0.01. P values were calculated using the two-tailed paired Student's t-test.

b) Tofacitinib and R406-treated PSC-WA and ADSC adipocytes have higher oxygen consumption rate (OCR) compare to DMSO-treated adipocytes. Values are mean ± s.e.m. of n = four (DMSO and tofacitinib) and n = three (R406) biological replicates.

c) The degree of lipolytic activity was assessed by quantification of glycerol release. tofacitinib (tofa.) and R406-treated ADSC adipocytes showed increased lipolysis in the basal state (upper graph), but no in Forskolin (FSK)-stimulated cells (lower graph. Values are mean ± s.e.m. of n = three biological replicates and differences from DMSO are significant for * P < 0.05. P values were calculated using the two-tailed paired Student's t-test.

JAK-inactivated adipocytes maintain a white adipocyte transcriptional identity

To determine whether tofacitinib and R406 treatment had converted the transcriptional identity of white adipocytes, we performed RNA sequencing and analyzed the whole transcriptome of MPC, PSC-WA, PSC-BA and compound-treated PSC-WA. We used multidimensional scaling to project expression data, where distances between samples reflect the differences between global gene expression profiles. As expected, PSC-WA and PSC-BA moved away from undifferentiated PSC-MPC, and away from each other (Fig. 7a). Despite the acquisition of a brown-like metabolic profile, tofacitinib and R406-treated PSC-WA global expression profiles clustered with untreated PSC-WA samples and not with PSC-BA (Fig. 7a). Consistent with our findings, RNAseq analyses revealed a strong induction of UCP1 expression by tofacitinib and R406 (Fig. 7b). Among known positive regulators of brown adipogenesis, R406 promoted mRNA expression of PGC1α, PGC1β and PPARG (Supplementary Figure 6). Intriguingly, PRDM16 mRNA expression was decreased in treated-PSC-WA despite an increase at the protein level (Supplementary Figure 6 and Fig. 3). These data indicate that adipocyte browning by JAK inhibition occurs by functionally remodelling white adipocytes and through the acquisition of brown-like metabolic activities rather than via cell fate conversion.

Open in a separate window

Figure 7

Gene signature and cellular identity of JAK-inactivated adipocytes

a-f) Adipocytes were differentiated according to scheme 1B, treated with tofacitinib and R406 at day 7, and collected at day 8 (24h time point) or day 14 (d7 time point). N= 3 biological replicates. Each independent biological replicate was pooled from two individual wells.

a) Multi-dimensional scaling of RNA sequencing data revealing the white lineage identity of tofacitinib (tofa.) and R406 treated PSC-WA.

b) Levels of UCP1 transcripts served as experimental control for adipocyte browning. UCP1 transcripts are higher in BA versus WA and higher in tofacitinib (tofa.) and R406-treated PSC-WA compare to DMSO-treated PSC-WA.

c) Interferon targets and pro-inflammatory pathways are significantly down-regulated by both compounds at 7d in PSC-WA. Enrichment scores of 9116 gene sets are compared between two time points (24h and 7d) for both compounds. Each circle represents one gene set that is coherently regulated by an upstream pathway. Black lines indicate the change of average scores, and blue lines the change of individual pathways that are significantly reduced (|ΔES|>=2).

d) Differential expression profiles of interferon targets induced by tofacitinib and R406. A substantial subset of target genes are negatively regulated in both cases, making the density curves of logFC shifts toward left and thereby forming a “red shoulder”. Compared with 24h, the expression of interferon pathway targets are repressed by both compounds at 7d (P =2.94E-6 and 2.06E-5, respectively; one-sided Kolmogorov-Smirnov test).

e) Differential expression profiles of selected interferon target genes in heatmap.

f) Whole transcriptome analysis revealed that the sonic hedgehog responsive genes GLI1, SFRP5, KLHL31 and SHH were up-regulated in tofacitinib (tofa.) and R406-treated adipocytes at day 7 compare to DMSO control. Values are mean ± s.d. of n = three biological replicates and differences from DMSO are significant for * P < 0.05 and ** P < 0.01. P values were calculated using the two-tailed paired Student's t-test.

Compound library for adipocyte browning screen

A library of 867 small molecules of known mode of action was assembled by combining Roche internal reference compounds and commercially available screening libraries (Tocriscreen Collections #3514 Kinase inhibitor toolbox, Selleckchem Stem cell #L2100, Selleckchem Epigenetic #L1900 and Selleckchem GPCR #L2200). All compounds were pre-dissolved in DMSO, diluted to 5mM stock solutions and stored at -80°C. A list of compounds ID, mode of action and CAS numbers is available upon request.

Browning screen, statistical analysis and hit validation

PPARG2-MPCs were seeded in 96-well plates, induced and differentiated into PSC-WA as described above. A library of small molecules was applied at a 5 μM final concentration in biological duplicates at day 7, 10 and 12 in fresh adipogenic medium. DMSO and rosiglitazone (0.5 μM final concentration) were included on each plate as negative and positive controls respectively. Adipocytes were collected at day 14 and processed for UCP1 and FABP4 mRNA analysis by bDNA.

For statistical analysis, the raw mRNA signal was log-transformed and normalized to fold-change of the per plate median of the negative controls for all calculations, x_norm = log(s_raw)/M_neg, where M_neg = median_plage{log(s_neg)}. The data quality was assessed by calculating the Z’ factor based on positive and negative controls for each plate (Supplementary Figure 1b). For UCP1, <Z’> = 0.50, and Z’ > 0.50 for 11 out of 24 plates; for FABP4, <Z’> = 0.64, and Z’ > 0.50 for 18 out of 24 plates. This is characteristic for a good quality HCS³⁴. To evaluate the replicates performance, biological duplicates were aggregated to their mean; 5% of the compounds were excluded from further analysis since their UCP1 difference was larger than 0.07; Supplementary Figure 1c shows that the replicate difference distribution is constant over the entire range, suggesting a constant replicate variance.

The remaining 832 compounds were used for hit selection as described before³⁵. Knowing the constant variance from the controls, a p-value could be assigned to each compound following the test statistics T = (x_norm − 1)/σ_neg, where σ_neg = MAD_plate{log(s_neg)}, x_norm(M_neg) = 1, which is normal distributed for inactive compounds (i.e., under the null hypothesis H₀). We next estimated the false discovery rate (FDR) by p-value distribution analysis³⁶ and applied a selection threshold of <0.01. Supplementary Figure 1d, left panel, illustrates the selected primary actives among the non-selected compounds in a quantile-quantile plot and the p-value distribution (inset). Next, a second hit selection process for FABP4 was included, again at a FDR cutoff <0.01 (Supplementary Figure 1d, right panel). We thereby found 135 UCP1 hits, of which 52 were assigned as “Rosiglitazone-like” compounds that induced both UCP1 and FABP4. In the validation experiment, we selected browning hits that induced UCP1 by at least two fold and UCP1/FABP4 by at least 1.5 fold. PPAR agonists were also included for a total of 33 re-tested compounds, of which 26 were confirmed (Supplementary Figure 1d, left panel). 15 validated hits and PPAR agonists are displayed in figure 1d and chemical names are provided in Supplementary Table 1.

High content imaging

At day 14 of browning assay, cells were fixed with 4% Paraformaldehyde in PBS for 15 minutes at room temperature, incubated with blocking buffer (5% Donkey Serum, 0.5% triton X-100 and 0.02% Azid NaN3 in PBS) for 30 minutes at RT and stained with 10 μM Hoechst [B2261, Sigma] and 20ng/ml of Bodipy ® 493/503 (D-3922, Life technologies) in PBS for 30 minutes at room temperature. Samples were kept in PBS. Fluorescence Images where acquired using the Opera QEHS reader (Perkin-Elmer, Hamburg, Germany) equipped with a Nipkow Spinning disc for confocality using a 20× objective. Hoechst stain was imaged with laser excitation at 405nm laser and a 455/70nm band path filter for emission. The lipid stain was imaged using laser excitation at 488nm, and a 535/60nm / BP filter for emission. The image segmentation was done with the instrument integrated software Acapella. A custom made algorithm was used to identify the morphologically heterologous nuclei in the mixed cell population: Larger, round and relatively dark (brown adipocytes, non-differentiated cells); smaller, bright, more irregular shaped nuclei (white). The lipid droplets were segmented independently as either larger round bright objects or darker spots and assigned to individual nuclei based on an equidistant ring around the nuclei. The ratio in the total area of cell related droplets above and below 1070 μm² was quantified as shift between a more white and brown phenotype in the whole cell population. Cell classification; Adipocyte: total droplet area above 750 μm² and/or >10 spots <50 μm². White subtype: Minimum one droplet above 830 μm², brown subtype: No larger droplets, minimum 10 spots.

Browning assay with tofacitinib, R406, JAK inhibitors and BMP7

Following screening and hit validation, browning assays were performed with commercially available JAK3 inhibitor (tofacitinib, Selleckchem S5001) and SYK inhibitor (R406, Selleckchem S1533). PSC-WA and ADSCs were differentiated and treated following the assay design of the browning screen. Tofacitinib was added at 2 μM and R406 at 1 μM final concentration unless otherwise indicated on figures and legends. THRB agonist (CAS 219691-94-8) was added at 5 μM final concentration. Ruxolitinib (Selleckchem S1378), Baricitinib (Selleckchem S2851), CYT387 (Selleckchem S2219) and AZD1480 (Selleckchem S2162) were added at 0.1-5 μM final concentration as indicated on figures.

Recombinant human BMP7 (R&D systems, 354-BP) was added at 10 ng/ml final concentration in adipogenic medium supplemented with 250 μM IBMX (3-Isobutyl-1-methylxanthine, Sigma-Aldrich I7018) at day 0, 3 and 5. In ADSCs, BMP7 was added at 10 ng/ml at day 7, 10 and 12. Tofacitinib and R406 were added following the same conditions as BMP7 for proper comparison.

Browning of mouse adipose tissue explants

Subcutaneous and visceral adipose depots were dissected from 8 week old male C57/Bl6 mice (n=5) and minced in standard media (DMEM containing 10 % FBS and 1% penicillin/streptomycin). Minced pieces were pooled, placed into a 100 μM nylon strainer and rinsed 3 times with standard media to remove red blood cells and excess lipid. Explants were carefully removed from the top of the strainer and plated into 24-well plates. Following 48 hours of acclimatization in adipogenic medium (DMEM, 10% FBS, 1% penicillin/streptomycin, 17 nM insulin, 1 μM dexamethasone) cells were treated with adipogenic medium containing DMSO, tofacitinib or R406 at 1, 2 and 5 μM final concentrations. Each treatment was performed in triplicate. Media with fresh compound was replaced every other day. Following 7 days of treatments, RNA was extracted in Trizol (Life Technologies) following standard protocol. RT-PCR was performed using Superscript III (Life Technologies) followed by analysis on a ViiA7 Real-Time PCR System (Life Technologies). Expression was analyzed relative to 18s RNA levels. Taqman probes were purchased from Life Technologies (UCP1: catalog number Mm01244861_m1, 18S: catalog number Mm03928990_g1).

Treatment with TNFα, INFγ and cyclopamine

TNFα (human recombinant Tumor necrosis factor alpha, Sigma-Aldrich T0157) was added directly to cell medium 48 hours or 24 hours before end of assay, at a final concentration of 0.25 ng/ml or 1 ng/ml respectively. IFNγ (human recombinant Interferon-gamma, Sigma-Aldrich I3265) and cyclopamine (Tocris ,1623) were added to fresh cell medium containing DMSO, tofacitinib or R406, at day 7, 10 and 12. For all treatments, corresponding volume of solvent (PBS, DMSO or ethanol, according to supplier's recommendation) was added to control wells and used for normalization.

Gene expression analysis by branched DNA (bDNA)

Adipocytes grown in 96 well plates were collected by replacing adipocyte medium with 100 μl/well of 1× RNA lysis mixture (Panomics QuantiGene2.0 Assay kit). Lysates were transferred to mRNA capture plates containing a specific capture probe (Supplementary Table 2), incubated overnight, processed for branched DNA amplification and analyzed according to the manufacturer's instructions (Panomics QuantiGene2.0 Assay). ACTB and PPIB probes were used as housekeeping genes for normalization. Data are presented as average and standard deviation of 2 or 3 biological replicates and normalized on DMSO control.

Western blot analysis

For protein analysis, browning assays were performed in 6 well plates initially seeded with 5×10⁵ PPARG2-MPCs. At differentiation day 7, PSC-WA were treated with DMSO, tofacitinib (2 μΜ) or R406 (1 μM) for 20 minutes and lysed in 75 ul/well of cold RIPA buffer (60 mM Tris–HCl at pH 7.4, 150 mM NaCl, 0.25% SDS, 1% NP40) containing Complete protease inhibitor (Roche) and phosphoSTOP (Roche). For browning assays over 7 days, cells were lysed in cold RIPA buffer at day 14. Western blot analysis was carried out using the following antibodies: STAT1 (Cell Signaling 9172, 1:1000), phospho-STAT1-S727 (Cell Signaling 9177, 1:1000), phospho-STAT1-Y701 (Cell Signaling 9167, 1:1000), STAT3 (Cell Signaling 4904, 1:1000), phospho-STAT3-S727 (Cell Signaling 9134, 1:1000), phospho-STAT3-Y705 (Cell Signaling 9131, 1:1000), STAT5 (Cell Signaling 9363, 1:1000), phospho-STAT5-Y694 (Cell Signaling 9351, 1:1000), Akt (Cell Signaling 4691, 1:1000), phospho-Akt-Ser473 (Cell Signaling 4060, 1:1000), Erk1/2 (Cell Signaling 9102, 1:1000), phospho-Erk1/2-Thr202/Tyr204 (Cell Signaling 4370, 1:2000), p38 MAPK (Cell Signaling 8690, 1:1000), phospho-p38-Thr180/Tyr182 (Cell Signaling 4511, 1:1000), JNK (R&D AF1387, 1:1000), phospho-JNK-T183/Y185 (R&D AF1205, 1:400), UCP1 (R&D MAB6158, 1:2000), PRDM16 (R&D AF6295, 1:200), GAPDH (SIGMA, G8795, 1:2000), COX-I (abcam 14705, 1:1000) and SDH-A (abcam 14715, 1:1000). The CHEMIDOC MP Imaging System and ImageLab Analysis Tool (BIO-RAD) or the Licor odyssey system was used to scan & quantitate the band intensity.

Reverse phase protein array (RPPA) analysis

For RPPA analysis, browning assays were performed in 10 cm dishes initially seeded with 1×10⁶ PPARG2-MPCs. At differentiation day 7, adipocytes were treated with DMSO, tofacitinib (2 μM) or R406 (1 μM) for 20 minutes and lysed in 100ul of CLB1 lysis buffer. Lysates were snap-frozen in liquid nitrogen, stored at −80°C and shipped to NMI Technology Transfer GmbH (Reutlingen, Germany) for RPP analysis as follow: Lysates were printed at 4 dilutions in duplicates onto ZEPTOMARK chips (Bayer Technology Services GmbH, Leverkusen, Germany) and analyzed for 51 selected phospho-analytes. Array images were analyzed using the ZeptoVIEW 3.1 software.

Measurement of cellular Oxygen Consumption Rate (OCR)

PPARG2-MPCs and ADSCs were plated at 7×10⁴ cells/well in gelatin-coated XF 24-well cell culture microplates (Seahorse Bioscience), differentiated into adipocytes at confluence and treated with DMSO, tofacitinib and R406 as described above (browning assay). Day 14 adipocytes were incubated in pre-warmed unbuffered DMEM medium (DMEM containing 2 mM GlutaMAX, 1 mM sodium pyruvate, 1.85 g l⁻¹ NaCl and 25 mM glucose) for 1 h. The oxygen consumption rate was measured by the XF24 extracellular flux analyser (Seahorse Biosciences). Mitochondrial biogenesis was profiled by injecting perturbation drugs, 2 μM oligomycin, 0.5 μM CCCP (carbonyl cyanide-p-trifluoromethoxyphenylhydrazone) and 5 μM antimycin A, in succession. OCR were determined by plotting the oxygen tension and acidification of the medium in the chamber as a function of time and normalized by protein concentration (picomoles per minute per milligram), respectively.

Lipolysis assay

To measure lipolysis activity, DMSO, tofacitinib and R406-treated ADSC adipocytes were starved in DMEM containing 1% FBS for 1 h. Cells were then incubated in HBSS (Hank's balanced salt solution) with 2% fatty-acid-free BSA alone or with 10 μM forskolin. The culture medium was collected for glycerol measurement using the free glycerol reagent (Sigma#F6428). Protein concentrations used to normalize glycerol content were measured using the Bradford protein assay (BioRad). Glycerol release was expressed in micrograms of glycerol per milligram of total protein or the lowest value is set to 1.

RNA sequencing of adipocyte browning

A total of six, 6-well plates were prepared as follow: In each plate, three wells were seeded with PPARG2-MPC, one well with PPARG2/CEBPB/PRDM16-MPC and one well with rtTA-MPC (5×10⁵ cells/well). All cells were induced and differentiated as described above. At day 7, PSC-WA received either DMSO, tofacitinib (2 μM) or R406 (1 μM). PSC-BA and MPC received DMSO only. At day 8, three cell plates were collected by adding 350 ul/well of RLT lysis buffer (Qiagen), constituting the 24h-triplicates. In the remaining three plates, treatment with DMSO, tofacitinib and R406 was repeated at day 10 and 12. At day 14, cells were collected by adding 350 ul/well of RLT lysis buffer (Qiagen), constituting the 7days-triplicates. Total RNA from adipocyte lysate was extracted and purified using the RNeasy mini kit (Qiagen) according to the manufacturer's instructions. Genomic DNA was removed from the extracted RNA using the RNase Free DNase Set (Qiagen) and purified with RNeasy Minelute Clean up Columns (Qiagen). For all samples high quality RNA was obtained (RIN between 8.1 and 9.8). RNA Sequencing libraries were prepared from 1μg of total RNA using the Illumina TruSeq RNA Sample preparation Kit v2 (Illumina) following the manufacturer's instructions. Sequencing libraries were quantified using the Kapa Library Quantification kit (Kapa Biosystems) and quality controlled by capillary electrophoresis on a Bioanalyzer using High Sensitivity chips (Agilent Technologies). Libraries were sequenced in randomized pools of three or four libraries per lane on a HiSeq2000 sequencer (Illumina) for 2× 50 cycles using version 3 cluster generation kits and version 3 sequencing reagents (Illumina). One % of PhiX control library (Illumina) was spiked into each lane as a sequencing control.

Bioinformatics and statistical analysis of RNAseq data

RNA-seq reads were first aligned to human transcriptome, and unmapped reads were then aligned to human genome. Reads that map to annotated genes were modelled by negative binomial distribution, and differential expression was quantified by generalized linear models (GLMs) implemented in the edgeR package. For gene-level analysis, the filter for significantly differentially expressed genes was set as |logFC|>=1 (logFC=log fold-change) and FDR<0.05 (Likelihood-ratio test, adjusted with the Benjamini-Hochberg method for multiple testing). The filtering was not necessary for gene-set-level analysis. Gene set analysis was performed with both Gene Set Enrichment Analysis (GSEA) and the CAMERA method. Gene set enrichment score (ES) was defined as ES=sign(NES_GSEA)log10(FDR_GSEA), namely the sign of normalised enrichment score reported by GSEA (+1 if NES_GSEA>0, and -1 otherwise), multiplied by log10-transformed false-discovery rate (the q-value). The enrichment score ranges between -4 and 4, corresponding to negatively or positively enriched gene sets that are statistically highly significant (FDR q<=1E-4). Gene sets in use included both pathway information from public repositories, including Reactome, StringDB and NCI-Nature pathway database, and manually curated sets of genes whose expression are regulated by upstream pathways.

The authors thank Inga Clausen, Martine Kapps, Roland Schmucki and Angelika Schuler for technical support, Klaus Christensen and Martin Graf for stem cell support, Laura Badi for preliminary data analysis, Corinne Solier, Annette Schell-Steven and Tobias Bergauer for experimental planning and Michael Pawlak (Natural and Medical Sciences Institute at the University of Tübingen) for RPPA analyses. A.M. was supported by the Roche Postdoctoral Fellowship (RPF) program (2011-2013). This research was supported in part by F.Hoffmann-La Roche; grant R01DK095384 (C.A.C. and Y.K.L.) and R01DK097768 (C.A.C.) from the United States Institutes of Health (NIH); and Harvard University.