Human bone marrow-derived mesenchymal stromal cells (BMSCs) were purchased from Lonza (Geleen, the Netherlands) and maintained as previously described (23, 27, 28). Two days after seeding, BMSCs were initiated to adipogenic differentiation using 100 nM dexamethasone, 60 μM indomethacin, and 0.5 mM 3-isobutyl-1-methylxanthine. BMSCs at passage 7 were used in the experiments. Culture media was refreshed every 3 or 4 days with various concentrations of compounds during the courses of inductions. These cultures are called adipogenic-BMSCs throughout the remainder of the study.

All tested compounds in this study were either provided by the CMap research group at the Broad Institute or purchased from Sigma (Sigma-Aldrich, St. Louis, MO, USA).

Adipogenic-BMSCs were fixed with 10% formalin, and subsequently incubated with Oil red O solution (Sigma-Aldrich, St. Louis, MO, USA) for 30 min at room temperature (RT) as described (28).To determine cell number, DAPI was used to stain for nuclei. Images (5 per well) were taken using a Zeiss Axiovert 200MOT microscope (Zeiss, Sliedrecht, the Netherlands) and DAPI positive staining was counted using Fiji software. Cell number was calculated from the average of 5 images in each well. Lipid droplets were extracted with isopropanol and absorbance was subsequently measured at 490 nm. Normalized absorbance was calculated as the raw absorbance divided by cell numbers.

Cell viability was determined using Cell Counting Kit-8 (CCK-8) assays. Adipogenic-BMSCs were incubated with CCK-8 reagent (Sigma-Aldrich, St. Louis, MO, USA) for 2 hours in an incubator according to the manufacturer’s manual. The conversion of the tetrazolium salt WST-8 to formazan was measured at 450 nm.

RNA isolation and cDNA synthesis were performed as described (28). Oligonucleotide primer pairs for qPCR were designed to be either on exon boundaries or spanning at least one intron (Table S1). qPCR was performed using the GoTaq qPCR Master Mix system (Promega, Madison, WI, USA). Gene expression was normalized to the expression of 36B4 using the equation 2^^{- (Ct gene of interest – Ct housekeeping gene)}.

Illumina Human HT-12v3 BeadChip arrays were performed as described (29). RNA was isolated from three biological replicates of adipogenic-BMSCs at day 4 and day 13 as described (28). RNA concentration and size distribution profiles were assessed on a 2100 Bioanalyzer (Agilent Technologies B.V., The Netherlands). RNA was amplified using Illumina TotalPrep RNA Amplification Kit according to manufacturer’s instructions (ThermoFisher Scientific, Massachusetts, USA). A total of 750 ng cRNA was hybridized using the standard protocol from Illumina. Data was analyzed after background subtraction using GenomeStudio (V2010.1, Illumina), and processed with R2.10.1 lumi-package in R (30). Differentially expressed (DE) probes (adjusted p-value < 0.001) were identified when comparing to undifferentiated BMSCs using Iimma package from Bioconductor (31).

A connectivity mapping approach was used to prioritize small molecules as previously described (19, 23). The similarities between gene expression profiles from microarray analyses and reference signatures in CMap are computed (21). Top 100 up- and down-regulated probes based on the log₂-fold change were submitted to the query version 1.0 (Dataset S1, https://clue.io/query). The summary score is a perturbagen-centric measure of connectivity that summarizes the results observed in individual cell types, while the connectivity score is a quality control measurement comparing an observed enrichment score from adipogenic gene set to reference gene sets, ranging from -100 to 100. A positive score suggests the facilitated effect on adipogenesis, while a negative score suggests an inhibitory effect on adipogenesis. The magnitude of the score corresponds to the magnitude of similarity or dissimilarity, and the tops and/or bottoms of these connectivity lists can be thought of as hypotheses about the about the relationship of the perturbagen to the query. For further details on these scores, see https://clue.io/connectopedia/.

BMSCs cultured in 96-well plates were treated for 24 h with the compounds following 3 days of adipogenic induction. Compounds used in the screening process were provided by the CMap research group at the Broad Institute in heat-sealed 384- well plate. The concentrations of compounds (3 replicates per concentration per compound) used in adipogenesis cultures were determined based on the data in the L1000 database (21). The vehicle of the compounds, dimethylsulfoxide (DMSO), served as negative control and was added in the same treatment plate. Cell lysate was prepared by using 35 μl of TCL lysis buffer (QIAGEN, Hilden, Germany). The plates were sealed with adhesive film and incubated for 30 minutes at RT prior to storage at -80°C. Compound-induced gene expression profiles were generated using the L1000 platform (21). Raw fluorescent intensity data were detected by Luminex scanners and further deconvoluted to generate expression profiles of 978 landmark genes. Additional values of 12,328 genes not directly measured were inferred based on normalized values of the landmark genes. Compound signatures were represented using robust Z-scores (RZS), which reflect the degree to which a gene was up or down-regulated in a given sample relative to all other samples on the same plate (21). A compound - gene perturbation was classified as DE genes between control and treatment if the absolute difference in median RZS was ≥ 2. Gene set enrichment analysis (GSEA) was performed on the robust Z-score using HALLMARK gene sets in the Molecular Signatures Database (MSigDB) (32).

Data were expressed as means ± SEM of representative experiments. All experiments were performed at least 2 times. Statistical analysis was performed using GraphPad Prism 9 or R package. Significance was calculated using Student’s t-test, one-(two) way ANOVA following post-hoc test. P value less than 0.05 was considered significant.

All of the microarray data and CMap data that are at the basis of this study have been deposited in Gene Expression Omnibus (GEO) with the accession numbers GSE80614 and GSE70138, respectively.

To identify genes that were differentially regulated during adipogenesis, we differentiated BMSCs with adipogenic induction medium for 4 days and 13 days representing the early and late adipocyte differentiation phase, respectively. Cells at day 0, 4 and 13 were collected for microarray-based gene expression analyses. DE probes at day 4 and day 13 were identified by comparing these with the non-induced cells at day 0. Transcriptomic snapshots cannot distinguish causal genes of the adipogenic-BMSCs phenotype from the effect genes, which are in physiological feedback loops and may contribute to a bias over a single representative of such a cluster (33). Therefore, we decided to use a reduced gene set by extracting the top 100 up or down-regulated DE probes, which are more specific to the unique characteristics of adipocytes.

To identify those unique genes, we utilized Gene Ontology (GO) annotations because of its query specificity and ability to distinctively annotate molecular functions, biological processes and cellular components. We began with extracting GO terms related to “lipid” and “fatty_acid”, resulting in 736 and 273 terms respectively (Figure 2). Terms that were not present in homo sapiens (54 in “lipid” and 6 in “fatty_acid”) were excluded. Given our aim to identify compounds that can inhibit molecular pathways or cellular processes during adipogenic differentiation, instead of the formation of certain cellular structures, such as mitochondria or ribosome-enriched in adipocytes, we focused on Molecular Functions and Biological Processes and terms under “cellular component” (234 in “lipid” and 7 in “fatty_acid”) were removed. By further eliminating 26 overlapping GO terms between the “lipid” and “fatty_acid” filters, a total of 682 GO terms specific to molecular functions and cellular events of human adipogenic differentiation were identified. Up- and down-DE probes at day 4 and day 13 annotated by the 682 GO terms were extracted, and the top 100 up- and down-DE probes at these two time points were used for the adipogenic differentiation-specific queries in CMap analyses.

To identify compounds that inhibit adipogenesis, the top 100 up- and down-gene lists at day 4 and day 13 were submitted to CMap as two separate queries. For each query, the entire small molecule library, containing 2,837 chemicals, was ranked based on summary scores of each molecule representing signatures from total 9 cell lines (Figures 3, S1, S2). Summary scores were computed using a perturbagen-centric measure of connectivity that summarizes the results observed in individual cell types (20). The positive scores indicate similarity between input gene sets from the list of DE probes and gene sets associated with each molecule, in this case meaning pro-adipogenic, while negative scores indicate an opposite association implicating it to be anti-adipogenic. Therefore, we reversed the list and only focused on the negative scores. Small molecules of which the summary scores are greater or equal to -99 were removed to select the most anti-adipogenic compounds (Figure 3). At day 4, 30 compounds were found (Figure 4A), whereas 42 compounds were found at day 13 (Figure 4D). Pearson correlation analyses of gene set dissimilarities among all 9 cancer cell lines in the CMap database showed that the list of anti-adipogenic candidates had no clear preferential effect on a particular type of cells in both queries, indicating the summary score was equally influenced by all 9 cell lines (Figures 4B, E). To further enrich the lists, we applied the connectivity score comparing an observed enrichment score from adipogenic gene set to reference gene sets based on L1000 platform, ranging from -100 to 100. A connectivity score less than -99 means only less than 1% of reference gene sets showed stronger dissimilarity than a given molecule in the query lists. Compounds with median connectivity scores greater or equal to -99 were deleted, resulting in 3 compounds and 6 compounds at day 4 and day 13 queries, respectively (Figures 4C, F). Since purmorphamine and DY-131 were present in both queries, we identified 7 potential anti-adipogenic candidates in total. Purmorphamine and chelidonine had been tested for their anti-adipogenic effects in BMSCs (34). Since purmorphamine ranked top 1 in our CMap analyses, demonstrating the validity and strength of our approach, we therefore included purmorphamine as a positive control in the downstream functional analyses. In total, we generated a list of 6 anti-adipogenic candidate from a systematic CMap analysis (Table 1), and 5 of them were eventually used in further experiments. Apart from kinetine-riboside, all the compounds are either in the preclinical phase or phase 2 of clinical trials.

Given that the compound prioritization was performed using cancer-related cell lines, we decided to evaluate their responses in BMSCs as they comprise the stem cells of bone marrow adipocytes. To this end we analyzed the gene expression profiles of adipogenic-BMSC treated with the test compounds or the negative and positive controls. A heat map was constructed with adipogenesis-related genes identified in the HALLMARK_ADIPOGENESIS gene set from the Molecular Signatures Database (MSigDB) (Figure 5A). Our analyses revealed two distinct groups among the 5 compounds, of which DY-131 and SB-225002 were clustered together with the control condition, while emetine and kinetin-riboside showed an opposite expression profile compared to the control group (Figure 5A), which was similar to the previously characterized adipocyte inhibitor purmorphamine. GSEA analyses indicated that adipogenesis genes were either not or positively enriched following treatment with SB-225002 or DY-131 in adipogenic-BMSCs, respectively (Figures 5B, C). These assumptions were further confirmed by Oil Red O assays showing no effect of SB-225002 or DY-131 on adipogenesis (Figures 5D, E). Therefore, we anticipated that emetine and kinetin-riboside are both potential compounds suppressing adipogenesis.

Indeed, GSEA analyses for the emetine, kinetin-riboside and purmorphamine treatments demonstrated enrichments for adipogenesis, cholesterol homeostasis and fatty acid metabolism opposite to the control condition (Figures 5F–H). Importantly, we found that adipogenesis marker genes such as PPARG, LPL, ADIPOQ and LEP were significantly downregulated in the presence of emetine, kinetin-riboside or purmorphamine in adipogenic-BMSCs (absolute difference in RZS ≥ 2, Figure 5I). Taken together, these findings suggest that these 3 compounds induce negative perturbations compared to the control group, suggesting anti-adipogenic effects.

To verify if emetine, kinetin-riboside and purmorphamine have anti-adipogenic effects, in vitro BMSC adipogenesis assays were performed. Adipogenic-BMSCs were cultured for 14 days in the presence of each compound at two different concentrations, or with the vehicle control and next examined by Oil-red O analyses. For each compound, the higher concentration was corresponding to the concentration used in CMap database or published data (35), being 0.1 μM for emetine and 10 μM for kinetin-riboside and purmorphamine, where the lower concentration was one-third of the higher concentration.

The presence of emetine at either 0.03 μM or 0.1 μM significantly reduced Oil red O staining (Figures 6A, B). Total Oil red O absorbances decreased with increasing emetine concentrations in a dose-dependent manner (Figure 6B). In agreement with the reduction in cell number, cell viability indicated by CCK-8 experiment was decreased, suggesting a potential negative effect on adipogenic progenitor proliferation or survival of adipocytes (Figures 6C and S3). A similar dose-dependent response was also shown for normalized Oil red O absorbance (Figure 6D).

Kinetin-riboside also showed a clear Oil red O reduction at 10 μM, but there was no distinguishable difference between the 3 μM and control group (Figures 6E, F). Total cell number as well as the cell viability showed decreases with both concentrations (Figures 6G and S3). Cell number-normalized Oil red O also showed only for 10 μM a strong decrease (Figure 6H).

The positive control purmorphamine (10 μM) also caused a decrease in Oil red O (Figure 6I) both uncorrected and corrected for total cell number (Figures 6J, L). Purmorphamine at this concentration did not change cell number and cell viability (Figures 6K and S3).

To provide more insights into the molecular effect of these compounds, we analyzed the expression of adipogenesis genes (Figure 7). Emetine showed a dose-dependent inhibition of the expression of ADIPOQ but that was not observed in PLIN1, consistent with the transcriptomic analyses. Only high doses of emetine led to FABP4 downregulation (Figure 7A). Since lower doses of kinetin-riboside did not lead to the reduction of Oil red O absorbance, we only evaluated the expression of adipogenic-specific genes using the high dose of kinetin-riboside. ADIPOQ, PLIN1 and FABP4 were significant decreased in the presence of 10 μM kinetin-riboside (Figure 7B). Similar effects were observed following the treatment with the positive control purmorphamine (Figure 7C). Collectively, these data indicate that emetine, kinetine-riboside and purmorphamine all suppress the differentiation of adipogenic BMSCs.

After verifying their bona fide anti-adipogenic effects, we wanted to understand whether such effects were related to different stages of adipogenic differentiation. We therefore divided the 14-day culture time window into three stages, early stage (from day 0 to day 2), mid stage (from day 3 to day 6), and late stage (from day 7 to day 14). Compounds at two concentrations were added only during each of the stages, and cells were collected at day 14 for Oil red O staining and quantification (Figure 8).

Addition of emetine at either concentration during early stage adipogenesis resulted in a mildly reduced Oil red O staining (Figures 8A, B). Significant reductions in Oil red O staining were shown when 0.1 μM emetine was added at mid and late stages, indicating emetine might preferentially target mid and late stages of adipogenic differentiation. Short-term treatment with 0.03 μM emetine was not effective suggesting emetine might need longer incubation times to demonstrate a full anti-adipogenic effect at this concentration (Figures 8A, B, D). Compared to vehicle treatment, mid and late stages incubation of emetine led to 40% and 58% decrease, respectively, which were not as strong as the 14-day incubation (78% decrease vs. vehicle) as shown in Figure 6D. It is worth noting that stage-wise incubation of emetine did not lead to any effect on cell count, indicating its negative effect might be dependent on incubation duration (Figure 8C).

Kinetin-riboside, on the other hand, showed clear Oil red O staining reduction when it was added during any of the three stages with the 10 μM dose (Figures 8E, F). However, the degree of reduction in total and normalized Oil red O absorbance here was less dramatic than that of the 14-day incubation (Figures 8F, H), with 51%,62% and 62% decrease in the early, mid and late stages, respectively, while decreasing by 95% in continues treatment of kinetin-riboside compared with controls. The reduced cell count was still obvious when using the 10 μM dose, but it was also less severe compared to the 14-day incubation, while the negative effect at 3 μM was not observed here (Figure 8G).

Purmorphamine also produced an anti-adipogenic effect as compared with kinetin-riboside in reducing Oil red O staining at all three stages (Figures 8I–J). Similarly, the degree of total and normalized Oil red O absorbance reduction here was smaller than that in the 14-day incubation (Figures 8K, L), showing as 17%, 15%, 33% decrease in the early, mid, later incubation stages, and 68% decrease for continues incubation of purmorphamine compared with controls. Together, these findings suggest that the anti-adipogenic response of BMSCs is primarily driven by the duration of the treatment using any of the 3 compounds.

The use of CMap has already led to the identification of anti-adipogenic molecules for obesity therapy (39, 40). However, to the best of our knowledge, our work represents the first study using this novel method for small-molecule prioritization, and further scale down the screening by analyses of transcriptomic signatures of BMSCs to support the discovery of novel biology and therapeutic targets against adipogenesis. The extraction of genes from GO terms related to ‘lipid’ and ‘fatty acid’ are triggered by the feature of BMAT capable of storing and accumulating lipids as large unilocular lipid droplets (4). This organelle formed during adipogenesis allows for BMAT to be responsible for the storage, mobilization, and metabolism of fatty acids, which are important metabolic substrates for adipose tissue (4, 41). However, in late stages of differentiation, adipocytes also synthesize other adipose tissue-specific products and secreted products (42), which are not included in the GO terms of ‘lipid’ and ‘fatty acid’, leading to the potential bias in search of effective compounds.

One the other hand, we hypothesized that not all of the DE genes directly contribute to a physiological process or a disease. Some of them could represent a “secondary effect” activated by the actual causal genes and do not offer additional information but give a strong noise when used in signature matching. The ideal way is to employ gene signatures, which directly contribute to the phenotypes. However, only using omics data is difficult to identify all of the relevant genes. Therefore, we employed a reduced gene set created by integrating our transcriptomic profiles with gene ontology sources implementing established biological processes and molecular functions related to “lipid” and “fatty acid” to direct target adipogenesis. This method provided us with a unique subset of genes related to important aspects of adipogenesis, and pave the way for exploring compounds targeting adipogenic differentiation.

Dynamic gene expression profiling allows us to capture gene sets characterizing the sequential differentiation stages from BMSCs to fully differentiated adipocyte. DE probes at day 4 were selected for the computational compound screening in CMap on the basis of our previous study revealing that BMSCs adopt to committed adipocyte phenotype following 48–96-hour exposure to adipogenic medium, which was reflected by the stabilization of number of DE probes in the early stage of differentiation (day 0 - day 4) (29). This finding is supported by the observation showing that lipid droplets are visible after 3-5 days of adipogenic differentiation, only to accumulate further from then onwards (43). On the other hand, most DE genes were observed around day 14 following adipogenic induction as opposed to earlier (day 7) or later stages (day 21) (44). These studies provided us with evidence to select day 13 as a second important time point for CMap query analysis.

Emetine is a natural alkaloid originally isolated from the plant Psychotria ipecacuanha (45). This alkaloid has been used as therapeutic target against multiple diseases including amoebiasis (FDA approved), leukemia, Zikavirus, Ebolavirus, cytomegalovirus and SARS-CoV-2 infection (46–50), and was identified as an inhibitor regulating cell autophagy (46). Autophagy is an essential process for adipogenic differentiation (51). Inhibition of autophagy is of significant importance to alter differentiation state and metabolic function of adipocytes. Knockdown of Autophagy-related 7 (Atg7) in 3T3-L1 preadipocytes suppresses lipid accumulation and blocks the expression of adipogenic marker proteins (52), while others have been demonstrated that sustained activation of autophagy suppresses adipocyte differentiation. In our study we found autophagy related genes such as ATG14, ATG7 and ULK1 to be upregulated following treatment with emetine in adipogenic-BMSCs in a dose-dependent manner (Figure S4) indicating the activation of autophagy in cells by emetine. These observations provide mechanistic insight into the involvement of autophagy in emetine-modulated adipogenic differentiation of BMSCs.

The other compound we identified as inhibitory for adipogenic differentiation, kinetin-riboside, was identified as a natural product found in coconut fruits in nanomolar scale (53, 54). It is well established as an anti-tumor molecule able to suppress various kinds of cancer types such as leukemia, colon adenocarcinoma and hepatoma through its anti-proliferative activity and apoptotic effects (55). While its precise mechanism of action is not clear, kinetin-riboside induced apoptosis through classical mitochondria-dependent apoptotic pathways (56). The chemical disrupts mitochondrial membrane potential, induces release of cytochrome C to the cytosol and activates the pro-apoptotic protein caspase 3 (56–59). Mitochondria biosynthesis is well illustrated for its linkage with adipogenesis depending on the type of adipose tissue (59). Studies have shown that proper mitochondrial activity is required for adipogenic differentiation from MSCs and maturation of adipocytes (60, 61). Given the importance of mitochondrial activities in multiple events of adipogenesis, this could explain why kinetin-riboside treatment demonstrated anti-adipogenic effect at all three stages of adipogenesis.

Bone marrow adipocytes have both overlapping and distinct morphology and physiology, along with different transcriptomic profiles compared to adipocytes from white adipose tissue or brown adipose tissue (4, 5). Although our research question related to mesenchymal stromal cell transition to adipocytes does not apply to the other fat depots in our body, we recognize the current limitation of our study of not having included adipocytes from other depots and their response to our newly identified compounds. In fact, future studies should reveal whether the compounds are specifically acting on bone marrow stromal cell-derived adipocytes or whether they also act on adipocytes from other fat depots.

Altogether, we showed that by systemically enriching adipocyte-specific DE probes from transcriptomic data during BMSC-derived adipogenic differentiation, a list of anti-adipogenic compounds was created. All the compounds we tested showed promising anti-adipogenic effects in vitro. We identified two new molecules to have inhibitory actions during adipogenic differentiation or adipocyte maturation, making them potentially interesting drugs to oppose excessive adipocyte accumulation. These findings demonstrate that our analytical and experimental pipeline can be used in the drug repurposing field.