All chemicals and reagents were obtained from Sigma-Aldrich (Munich, Germany) unless stated otherwise.

The human SC abdominal adipose tissue collection was approved by the Medical Research Council of Hungary (20571-2/2017/EKU) followed by the EU Member States’ Directive 2004/23/EC on presumed consent practice for tissue collection. All experiments were implemented in accordance with the Helsinki Declaration. All participants were informed about the experiments and written informed consent was obtained from them. hASCs were obtained and isolated from stromal vascular fractions (SVFs) of human SC abdominal adipose tissue of healthy donors undergoing plastic surgery, as previously described (Kristóf et al., 2019; Szatmári-Tóth et al., 2020). Briefly, the tissue samples were immediately transported to the laboratory following plastic surgery. Adipose tissue specimens were dissected from fibrous material and blood vessels, minced into small pieces and digested in phosphate buffered saline (PBS) with 120 U/mL collagenase for 120 min in a 37°C water bath with gentle agitation. The completely disaggregated tissue was filtered (pore size 140 μm) to remove any remaining tissue. The cell suspension was centrifuged for 10 min at 1,300 rpm and the pellet of SVF was suspended and maintained. Floating cells were washed away with PBS after 3 days of isolation and the remaining cells were cultured as described in 2.3. Data of the donors included in the study are listed in Supplementary Table S1.

hASCs were seeded in 6-well plates and cultured in Dulbecco’s Modified Eagle’s Medium/Nutrient F-12 Ham (DMEM-F12) medium containing 17 µM pantothenic acid, 33 µM biotin, 100 U/mL penicillin/streptomycin, and 10% fetal bovine serum (Thermo Fisher Scientific, Waltham, MA, United States) at 37°C in 5% CO₂ until they reach complete confluence. The absence of mycoplasma was verified by polymerase chain reaction (PCR) analysis (PCR Mycoplasma Test Kit I/C, PromoKine, PromoCell, Heidelberg, Germany).

White adipogenic differentiation was induced for 3 days with serum-free DMEM-F12 medium supplemented with 17 µM pantothenic acid, 33 µM biotin, 100 U/mL penicillin/streptomycin, 100 nM cortisol, 10 μg/mL human apo-transferrin, 200 pM triiodothyronine, 20 nM human insulin, 2 µM rosiglitazone (Cayman Chemicals, Ann Arbor, MI, United States), 25 nM dexamethasone, and 500 µM 3-isobutyl-l-methylxantine. After the third day, rosiglitazone, dexamethasone, and 3-isobutyl-l-methylxantine were removed from the medium during the remaining 25 days of the differentiation. The medium was exchanged in every third day.

The active beige differentiation was induced for 3 days with serum-free DMEM-F12 medium supplemented with 17 µM pantothenic acid, 33 µM biotin, 100 U/mL penicillin/streptomycin, 10 μg/mL human apo-transferrin, 200 pM triiodothyronine, 20 nM human insulin, 2 µM rosiglitazone, 25 nM dexamethasone, and 500 µM 3-isobutyl-l-methylxantine. After the third day, dexamethasone and 3-isobutyl-l-methylxantine were removed and 500 nM rosigliazone was added to the medium for the remaining 25 days of differentiation. On the 14th day, a 4 h long 500 µM dibutyryl-cAMP treatment was carried out to mimic the in vivo cold-induced thermogenesis (Szatmári-Tóth et al., 2020). After the treatment, the aforementioned beige differentiation medium was applied until the end of the differentiation.

The inactive beige differentiation was induced in the first 14 days as in the case of the active beige adipocytes, but after the dibutyryl-cAMP treatment, the medium was replaced to the white cocktail without rosiglitazone, dexamethasone, and 3-isobutyl-l-methylxantine for additional 14 days.

Adipocytes were collected in TRIzol reagent (Thermo Fisher Scientific), followed by manual RNA isolation by chloroform extraction and isopropanol precipitation. The RNA quality was evaluated by Nanodrop (Thermo Fisher Scientific). cDNA generation was carried out by TaqMan reverse transcription reagent kit (Thermo Fisher Scientific) and followed by qPCR analysis (Szatmári-Tóth et al., 2020; Shaw et al., 2021). Gene expressions were normalized to the geometric mean of β-actin (ACTB) and GAPDH. All TaqMan assays are listed in Supplementary Table S2.

Total RNA sample quality was checked on Agilent BioAnalyzer using Eukaryotic Total RNA Nano Kit according to the Manufacturer’s protocol. Samples with RNA integrity number (RIN) value >7 were accepted for the library preparation process. RNA-Seq libraries were prepared from total RNA using MGIEasy RNA Library Prep Set V3.0 (MGI, Shenzhen, China) according to the manufacturer’s protocol. Briefly, poly-A RNAs were captured by oligo-dT conjugated magnetic beads then the mRNAs were eluted and fragmented at 94°C. First-strand cDNA was generated by random priming reverse transcription, then in the second strand synthesis step, double-stranded cDNA was generated. After repairing ends, A-tailing and adapter ligation steps, adapter-ligated fragments were amplified in enrichment PCR and finally sequencing libraries were generated. In the next step double-stranded libraries were denatured and single strand circularization was performed, after enzymatic digestion step circularized single-stranded library was generated for DNA nano ball (DNB) generation. After making DNB step, single-end 100 cycles sequencing run was executed on MGI DNBSEQ G400 instrument. After sequencing, the reads were aligned to the GRCh38 reference genome (with EnsEMBL 95 annotation) using STAR aligner (version 2.7.0a). FeatureCounts was used to quantify our reads to genes. Significantly differentially expressed genes (DEGs) were defined based on adjusted p values <0.05 and log2 fold change threshold >0.85. Heatmap was generated by using GraphPad 8.0 Software and interactome map was constructed by using Gephi 0.9 based on interaction from STRING (https://string-db.org/). Pathway analysis was performed by subjecting the list of DEGs to STRING and overrepresented KEGG pathways were selected based on the false discovery rate (FDR) < 0.05. Brown adipocyte content and browning capacity was estimated by BATLAS and ProFAT webtools, respectively, by subjecting the transcriptomic data of both groups of markers (Perdikari et al., 2018; Cheng et al., 2018).

DNA isolation was performed as previously described (Klusóczki et al., 2019). Rs1421085 SNP was genotyped by qPCR using TaqMan SNP Genotyping assay (Thermo Fisher Scientific, 4351379) according to the Manufacturer’s instructions.

The separation of investigated proteins was performed by SDS-PAGE, followed by transfer to a PVDF membrane, which was blocked by 5% skimmed milk solution (Szatmári-Tóth et al., 2020; Shaw et al., 2021). The primary antibodies were utilized in the following dilutions: anti-β-actin (1:5000, A2066), anti-UCP1 (1:750, R&D Systems, Minneapolis, MN, United States, MAB6158), anti-OXPHOS (1:1,000, Abcam, Cambridge, MA, United States, ab110411), anti-SLC7A10 (1:500, Abnova, Taipei City, Taiwan, H00056301-B01P), anti-PGC1α (1:1,000, Novus Biologicals, Centennial, CO, United States, NBP1-04676) anti-GPT2 (1:2000, Thermo Fisher Scientific, PA5-62426), and anti-SHMT1 (1:2000, Thermo Fisher Scientific, PA5-88581). The following species corresponding secondary antibodies were used: HRP-conjugated goat anti-mouse (1:5000, Advansta, San Jose, CA, United States, R-05071-500) or anti-rabbit (1:5000, Advansta, R-05072-500) IgG. The expression of the visualized immunoreactive proteins were quantified by densitometry using the FIJI ImageJ software (National Institutes of Health (NIH), Bethesda, MD, United States) as previously described (Szatmári-Tóth et al., 2020).

hASCs were seeded and differentiated on Ibidi eight-well µ-slides as described in Section 2.3. Cells were washed once with PBS and fixed by 4% paraformaldehyde, followed by permeabilization with 0.1% saponin and blocking with 5% skimmed milk. Primary antibody incubations were kept overnight with anti-TOM20 (1:75, WH0009804M1) and anti-LC3 (1:200, Novus Biologicals, NB100-2220). Secondary antibody incubation was for 3 h with Alexa Fluor 647 goat anti-mouse IgG (1:1,000, Thermo Fisher Scientific, A21236) and Alexa Fluor 488 goat anti-rabbit IgG (1:1,000, Thermo Fisher Scientific, A11034). Propidium iodide (PI, 1.5 μg/mL, 1 h) was used for nuclei labeling. Images were obtained with an Olympus FluoView 1,000 (Olympus Scientific Solutions, Tokyo, Japan) confocal microscope and FluoView10-ASW (Olympus Scientific Solutions) software version 3.0, as previously described (Szatmári-Tóth et al., 2020; Vámos et al., 2022). LC3 and TOM20 immunostaining images were converted to binary form, followed by processing with FIJI. The LC3 punctae count was determined by size (pixel²) 50–infinity AU with circularity 0–1 AU. Fragmented mitochondria were analyzed from the binary TOM20 immunostaining images with size (pixel²) 0–100 AU and circularity 0–1 AU. The optimum size values for the LC3 punctae and fragmented mitochondria were determined based on an analysis of all immunostaining images and manual verification of the counting accuracy by checking the outlines of counts. Both LC3 punctae and fragmented mitochondria content were normalized to per nucleus for individual images. Co-localization of LC3 and TOM20 was evaluated by calculation of the Pearson correlation coefficient (PCC) (Szatmári-Tóth et al., 2020; Vámos et al., 2022). 30 cells per three donors were recorded and analyzed.

Supernatants of the cells were collected at the end of the differentiation process and examined as previously described (Guba et al., 2022; Nokhoijav et al., 2022). Briefly, the media were filtered using a 3 kDa filter (Pall Corporation, Port Washington, NY, United States) and 10 µL of this filtrate was derivatized with AccQ Tag Ultra Derivatization Kit (Waters, Milford, MA, United States). Chromatographic separation was executed on H-class UPLC (Waters) using AccQ Tag Ultra Column (2.1 × 100 mm), AccQ Tag Eluent A and B, and gradient was ensured by the AccQ Tag Ultra Chemistry Kit (Waters). Amino acid (both L and D enantiomers) derivatives were detected at 260 nm in the PDA detector of the UPLC. Detection of the amino acid concentrations were calculated by Empower software (Waters) using a 7-point calibration curve. Flux of amino acids into or from adipocytes was calculated by comparing concentration differences measured in the conditioned media, which was administered to the cells at day 25 and collected at the end point of 28 days differentiation, and the unconditioned medium. The concentration of amino acids was normalized to the number of cells as described previously (Arianti et al., 2021).

Cells were seeded and differentiated on XF96 assay plates (Seahorse Biosciences, North Billerica, MA, United States) to white, active beige, or inactive beige using the protocols as described in Section 2.3. The OCR and ECAR were measured with XF96 oximeter (Seahorse Biosciences). Dibutyryl-cAMP stimulated OCR and ECAR, etomoxir-resistant (ETO-R) OCR, and stimulated proton leak OCR were quantified using previously utilized protocols (Kristóf et al., 2015; Arianti et al., 2021; Nagy et al., 2022). 10 μM antimycin A was used for baseline collection (measuring non-mitochondrial respiration). The OCR was normalized to protein content of each well.

All results are expressed as mean ± SD. The normality of the obtained data was tested by Shapiro-Wilk test. Datasets with a normal distribution was analyzed using one-way ANOVA with a Tukey’s post hoc test. Data was analyzed and visualized by GraphPad Prism 8 Software.

Primarily, we aimed to investigate the global gene expression patterns of the three types of differentiated SC adipocytes, white, active, and inactive beige (see Methods for their differentiation protocol) by performing RNA-seq analysis. We found that general adipocyte markers, such as SLC2A4, FABP4, LPL, ADIPOQ, AGPAT2, PLIN1, LEP, and LEPR were not expressed differentially among the three types of adipocytes (Figure 1A) suggesting that their differentiation rate was similar. The thermogenic markers, such as UCP1, ELOVL3, PGC1a, CIDEA, CITED1, AQP3, GK, CKMT1a/b, and PM20D1 had higher expression in active beige as compared to white or inactive beige adipocytes (Figure 1B). Next, we subjected our transcriptomic data from RNA-seq analysis to open source webtools to estimate brown adipocyte content by BATLAS (Perdikari et al., 2018) and browning capacity by ProFAT (Cheng et al., 2018) based on the expression levels of well-defined marker genes. We did not find significant differences in brown adipocyte content (Figure 1C), however, active beige adipocytes showed higher browning capacity score as compared to white or inactive beige ones (Figure 1D). According to our RNA-seq analysis data, we found that 211 and 147 genes had higher expression in the comparison of active beige vs. white and active beige vs. inactive beige, respectively; out of those, 100 genes were common (Figure 1E, top panel; Supplementary Tables S3, S4). Among the commonly highly expressed 100 genes, thermogenic markers, such as GK, PM20D1, PLIN5, CITED1, and AQP3 were found (Figure 1F). Interestingly, SLC7A10, encoding ASC-1, which was described as an important transporter during thermogenic activation (Arianti et al., 2021; Jersin et al., 2021) was also commonly upregulated in both comparisons (Figure 1F). 248 and 226 genes had lower expression in the comparison of active beige vs. white and active beige vs. inactive beige, respectively, and 164 genes had commonly lower expression (Figure 1E, bottom panel; Supplementary Tables S3, S4). We did not find any DEGs when we compared the gene expression profile of white and inactive beige adipocytes. We also analyzed the mitophagy rate and mitochondrial morphology by co-immunostaining of microtubule-associated protein 1 light chain 3 (LC3) and translocase of outer mitochondrial membrane 20 (TOM20) [Szatmári-Tóth et al., 2020] (Figure 1G). Confocal images were used to quantify the co-localization of LC3 and TOM20 by measuring the correlation between pixel intensities of two detection channels [Szatmári-Tóth et al., 2020]. We observed lower LC3 punctae counts per cell (Figure 1H) and PCC values in active beige as compared to white or inactive beige adipocytes (Figure 1I). We also found that active beige adipocytes had higher amounts of fragmented mitochondria, which were shown to support uncoupled respiration and enhanced energy expenditure (Pisani et al., 2018), as compared to white or inactive beige cells (Figure 1J). Altogether, these data suggests that thermogenesis-related genes were upregulated, the mitophagy rate was lowered, and mitochondria were more fragmented when human abdominal SC adipocytes were activated for thermogenesis.

Next, we intended to investigate whether FTO rs1421085 SNP affected the browning capacity of human abdominal SC adipocytes, which were differentiated into white, active, or inactive beige. Therefore, we genotyped the hASCs for FTO rs1421085 SNP by using SNP genotyping assay and obtained the allelic discrimination plot (Figure 2A). Then, we selected samples from 4 individuals with homozygous TT (risk-free) and from 4 individuals with homozygous CC (obesity-risk) genotypes for further analysis. We found that active beige adipocytes carrying FTO obesity-risk alleles exerted lower brown adipocyte content estimated by BATLAS (Perdikari et al., 2018), but no effect of the allelic distribution was observed in case of white or inactive beige adipocytes (Figure 2B). We also found that the risk-free genotype carrier active beige adipocytes had higher tendency of BATLAS and significantly higher ProFAT (Cheng et al., 2018) scores as compared to white ones that carried the same TT genotype (Figures 2B, C). Interestingly, active beige adipocytes carrying FTO obesity-risk genotype showed similar estimated brown adipocyte content and browning capacity as compared to white adipocytes (Figures 2B, C) suggesting that active beige differentiation could not overcome the browning inhibitory effect of the CC alleles. Active beige adipocytes with FTO risk-free genotype expressed the BATLAS marker genes at the highest level, whereas ones with the obesity-risk allele carriers expressed them at a level similar to those observed in the white or inactive beige adipocytes (Supplementary Figure S1). The expression of ProFAT markers was high in risk-free carrier active beige adipocytes and showed large donor variability in the obesity-risk allele carrier ones (Supplementary Figure S2).

A total of 175 genes including thermogenic markers (UCP2, CKMT2, and CITED1) and 5 BATLAS markers (PPARGC1B, ACO2, ACSF2, NNAT, and DMRT2) were expressed less in active beige adipocytes carrying FTO obesity-risk variant as compared to risk-free carriers (Figure 2D, top panel; Supplementary Table S5). Only 10 genes (7 of them were common in both comparisons) were expressed at a lower extent in CC as compared to TT carrier inactive beige adipocytes (Figure 2D, top panel; Supplementary Table S6). We found only 14 and 46 genes (8 of them were common in both comparisons) which were expressed more in active or inactive beige adipocytes, respectively, carrying the FTO obesity-risk as compared to the risk-free variant (Figure 2D, bottom panel; Supplementary Tables S5, S6). In white adipocytes, we did not find any DEGs which was affected by the FTO rs1421085 SNP. Next, we investigated the gene expression pathways affected by the FTO rs1421085 SNP and found that genes, which were less expressed in active beige adipocytes that carried the obesity-risk genotype, were overrepresented in several pathways, such as metabolic, PPAR signaling, lipolysis, fatty acid metabolism, or TCA cycle (Figures 2E, F). More expressed genes in active beige adipocytes with CC alleles were not significantly overrepresented in any of the pathways. We did not find any overrepresented pathway with respect to the DEGs found in inactive beige adipocytes either.

Because we observed significant effects of FTO rs1421085 SNP on the gene expression pattern of active beige SC adipocytes, we also compared the transcriptomic data of active beige and white or inactive beige adipocytes that carried FTO risk-free or obesity-risk genotypes, respectively. We found a total of 226 genes that were expressed more in active beige as compared to white adipocytes with FTO risk-free genotype, whereas only 54 higher expressed genes were found between the two types of differentiations in obesity-risk carriers; 25 genes were overlapping in the two comparisons (Supplementary Figure S3A; Supplementary Tables S7, S8). There were 93 and 130 genes which were expressed less in active beige as compared to white adipocytes carrying FTO risk-free or obesity-risk alleles, respectively. Out of those, 46 were common (Supplementary Figure S3B; Supplementary Tables S7, S8). The PPAR signaling pathway was the only one in which more expressed genes in active beige as compared to white adipocytes carrying FTO risk-free variant were overrepresented. The less expressed genes in active beige adipocytes as compared to white ones were overrepresented in several pathways, such as axon guidance or longevity regulating only in FTO obesity-risk genotype carriers (Supplementary Figure S3C). When adipocytes carried the risk-free alleles, no overrepresented pathway was found in this comparison.

We found a total of 172 genes that were expressed more in active as compared to inactive beige adipocytes carrying FTO risk-free variant, whereas only 18 genes were found in the same comparison with obesity-risk genotype carrier cells; nine genes were overlapping in the two comparisons (Supplementary Figure S3D; Supplementary Tables S9, S10). There were 116 and 158 genes (61 common), which were expressed less in active as compared to inactive beige adipocytes carrying FTO risk-free or obesity-risk alleles, respectively (Supplementary Figure S3E; Supplementary Tables S9, S10). Genes expressed in a lower extent in active beige adipocytes as compared to white ones were overrepresented in several pathways, such as AMPK signaling and TGF-beta signaling only in FTO obesity-risk allele carriers (Supplementary Figure S3F). When adipocytes carried the risk-free genotype, we did not find any overrepresented pathway in comparison of active with white or inactive beige adipocytes. These results suggest that FTO rs1421085 SNP only affects the gene expression profile, particularly that of the thermogenesis-related genes, in active beige but not in white or inactive beige adipocytes. In addition, the applied differentiation protocols resulted in more pronounced differences in the gene expression patterns of adipocytes with FTO risk-free alleles which suggest their significant browning potential when thermogenic cues are constantly present.

Since we observed that the allelic discrimination at FTO rs1421085 SNP influences the expression of thermogenic and BATLAS markers, we went further to investigate the expression of thermogenic genes at mRNA and protein levels in abdominal SC adipocytes. Our results showed that the mRNA expression of UCP1 was higher in active beige as compared to white or inactive beige adipocytes with FTO risk-free alleles, however, this difference was not observed in obesity-risk carriers (Figure 3A). At the protein level, we observed that active beige adipocytes expressed more UCP1 as compared to white or inactive beige adipocytes regardless the FTO rs1421085 genotypes, however, less UCP1 protein amount was detected in obesity-risk than in risk-free allele carrier active beige adipocytes (Figure 3B). Other thermogenic genes, such as UCP2, PM20D1, CIDEA, CITED1, CKMT1 and 2, CPT2, and PLIN1 were also expressed higher in active beige adipocytes carrying risk-free alleles as compared to white or inactive beige adipocytes with the same TT variant, however, we did not observe these differences in FTO obesity-risk carrier samples (Figures 3D–J). As compared to risk-free carriers, active beige adipocytes with FTO obesity-risk genotype had lower expression of these thermogenic genes and also that of the neurotrophic factor S100b, which was postulated to stimulate sympathetic axon growth and to play an important role in BAT innervation (Zeng et al., 2019) (Figures 3A–K). These results are in accordance with our RNA-seq data suggesting the critical importance of FTO rs1421085 SNP in active beige adipocytes and the compromised effect of active beige differentiation protocol in the presence of obesity-risk alleles.

Having observed that FTO rs1421085 SNP affected the expression of thermogenic genes, our next aim was to investigate whether the expression of mitochondrial complex subunits and cellular respiration were also suppressed in adipocytes with obesity-risk alleles. We found that active beige adipocytes carrying FTO risk-free genotype had higher amounts of mitochondrial complex subunits I, II, and IV as compared to white or inactive beige adipocytes carrying the same TT genotype (Figures 4A–C, E). However, no difference was found between the three types of differentiation programs when adipocytes carried the obesity-risk variant (Figures 4A–C, E). Active beige adipocytes with FTO obesity-risk alleles had lower expression of mitochondrial complex subunits I, II, and IV as compared to the risk-free carriers (Figures 4A–C,E). We observed a similar but statistically not significant trend in the case of mitochondrial complex subunit III (Figure 4D), while the expression of complex V subunit was similar for all types of adipocytes regardless the FTO rs1421085 genotype (Figure 4F). Next, we measured the cellular respiration of the three types of adipocytes carrying FTO risk-free or obesity-risk alleles. In accordance with the mitochondrial complex subunit expression, we found that active beige adipocytes carrying risk-free genotype had higher respiration (at both basal and maximal stimulated conditions), stimulated proton leak respiration, and extracellular acidification as compared to white or inactive beige adipocytes, but this difference was not pronounced when the adipocytes carried the risk variant (Figures 4G, H). Stimulated ATP-dependent respiration was not affected significantly by either the applied differentiation protocols or the FTO rs1421085 genotype (Figure 4G). Active beige adipocytes with FTO obesity-risk variant showed lower cellular respiration, especially stimulated proton leak respiration which reflects UCP1-dependent heat production, and extracellular acidification, which associates with glycolytic activity, as compared to risk-free carriers (Figures 4G, H). Intriguingly, the effect of FTO obesity-risk alleles on cellular respiration was observed in active beige but not in white or inactive beige adipocytes highlighting its exclusive effect in human abdominal SC adipocytes only when they are activated for thermogenesis.

Active thermogenic adipocytes utilize metabolic substrates, such as carbohydrates, fatty acids, or amino acids to generate heat (Onogi and Ussar, 2022). Therefore, we aimed to investigate the fuel utilization by abdominal SC adipocytes with distinct FTO genotypes in three types of differentiation programs. Fatty acids are primarily released from lipid droplets via lipolysis (Cannon and Nedergaard, 2004; Townsend and Tseng, 2014). We found that the regulation of lipolysis in adipocytes and fatty acid metabolism gene expression pathways were downregulated in active beige adipocytes with FTO CC as compared to TT allele carriers (Figure 2E). Although murine brown adipocytes can compensate the lack of lipid droplets-derived fatty acids (Chitraju et al., 2020), our ex vivo model did not include fatty acids in the applied differentiation media. Furthermore, stimulated etomoxir-sensitive oxygen consumption, which correlates with the activity of fatty acid oxidation (Nagy et al., 2022), was not affected significantly by either the differentiation programs or the FTO genotypes (Supplementary Figure S4). ASC-1, which is encoded by SLC7A10, plays an important role in mediating alanine, serine, cysteine, and glycine consumption in human adipocytes derived from abdominal SC and deep neck regions (Jersin et al., 2021; Arianti et al., 2021). To evaluate the preferable energy sources during thermogenic activation, we monitored the oxygen consumption of adipocytes upon etomoxir (inhibitor of carnitine palmitoyltransferase-1) administration. ETO-R respiration, which reflects the activity of carbohydrate and amino acid oxidation (Nagy et al., 2022), was higher in active beige adipocytes with FTO risk-free genotype than in white or inactive beige adipocytes carrying the same TT genotype (Figure 5A). Active beige adipocytes with FTO obesity-risk alleles had lower level of ETO-R oxygen consumption as compared to risk-free carriers (Figure 5A). These observations suggest less pronounced carbohydrate and/or amino acid utilization in active beige adipocytes of CC carriers at FTO rs1421085.

Because we found SLC7A10 (encoding alanine-serine-cysteine transporter, ASC-1) as a DEG among the most abundantly expressed genes in active beige adipocytes with FTO risk-free genotype (Figure 1H; Supplementary Table S7), we decided to investigate the effect of the applied differentiation protocols and FTO rs1421085 SNP on the expression of ASC-1 and the consumption of ASC-1 cargos by the adipocytes. We found that active beige adipocytes with risk-free alleles expressed higher mRNA level of SLC7A10 as compared to white or inactive beige ones that carried the same TT variant (Figure 5B), which could be confirmed at ASC-1 protein level (Figure 5C). The presence of the FTO rs1421085 SNP resulted in lower expression of SLC7A10 in active beige adipocytes; this effect was statistically significant at protein level but not at mRNA level (Figures 5B,C). Next, we measured the consumption of amino acids in the conditioned media obtained from the three types of differentiated adipocytes with CC or TT alleles, respectively. The applied differentiation programs did not affect the release of alanine regardless of the FTO allele status (Figure 5D). In the case of adipocytes with FTO risk-free genotype, we found that active beige ones consumed higher amounts of serine (Figure 5E) and cysteine (Figure 5F) and released more glycine as compared to white or inactive beige adipocytes (Figure 5G). In accordance with the aforementioned gene expression and ETO-R oxygen consumption results, we did not observe any differences in the fluxes of these amino acids between the three types of differentiation protocols in adipocytes with obesity-risk alleles. Active beige adipocytes with obesity-risk genotype consumed lower amount of cysteine as compared to risk-free carriers (Figure 5F), but significant effect of the SNP was not observed on serine consumption (Figure 5E) suggesting that other amino acid transporters might compensate for the reduced expression of ASC-1.

Our RNA-seq data showed that serine hydroxymethyltransferase (SHMT) 1, which catalyzes the conversions of L-serine and tetrahydrofolate (THF) to glycine and 5,10-methylene-THF (5,10-CH₂-THF), was expressed lower in active beige adipocytes with FTO obesity-risk as compared to risk-free carriers (Supplementary Table S5). We validated our RNA-seq data of SHMT1 expression by RT-qPCR (Figure 5H) and immunoblotting (Figure 5I). We also found that active beige adipocytes had higher protein content of SHMT1 as compared to white or inactive beige adipocytes in the presence of the FTO risk-free variant, but this difference was not observed in obesity-risk carrier samples (Figure 5I). The proposed model of ASC-1 mediated amino acid flux and possible interconversion of serine and glycine by SHMT1 are summarized in Figure 5J.

According to the RNA-seq data, active beige adipocytes carrying obesity-risk genotype expressed lower mRNA level of glutamic pyruvic transaminase (GPT) 2 as compared to risk-free allele carriers (Supplementary Table S5). In the case of adipocytes with FTO risk-free alleles, active beige cells expressed more GPT2 both at mRNA (Figure 6A) and protein level (Figure 6B) as compared to white ones. We also observed lower expression of GPT2 in active and inactive beige adipocytes with FTO obesity-risk genotype as compared to risk-free allele carriers (Figures 6A, B). Active beige adipocytes carrying risk-free genotypes also consumed higher amount of glutamine as compared to white or inactive beige adipocytes carrying the same TT genotype, but no difference was observed among the three types of differentiation protocols in the case of the samples with obesity-risk alleles. On the contrary to risk-free carrier cells, adipocytes with FTO obesity-risk genotype did not consume glutamine irrespective to the applied differentiation protocols (Figure 6C). In contrast, we found that active beige adipocytes released higher amount of glutamate as compared to inactive ones, while white adipocytes rather consumed that amino acid. The consumption of glutamate did not depend on the allelic discrimination at FTO rs1421085 locus (Figure 6D). In the case of adipocytes with TT alleles, active beige cells consumed less aspartate and more leucine than white or inactive beige ones, however, the consumption of asparagine, isoleucine, threonine, valine, histidine, lysine, tyrosine, tryptophan, proline, methionine, phenylalanine, and arginine was not significantly affected by the applied protocols. Histidine was consumed by adipocytes with risk-free alleles, however, it was released by obesity-risk allele carrier ones irrespective of the applied differentiation. The same trend was found in the case of lysine without observing statistically significant differences. Adipocytes with CC genotype tended to consume valine, however, it was released by the adipocytes with TT genotype. The FTO alleles did not affect the consumption of aspartate, asparagine, leucine, isoleucine, threonine, tyrosine, tryptophan, proline, methionine, phenylalanine, and arginine at a statistically significant level (Supplementary Figure S5). These results suggest that active beige adipocytes with FTO TT alleles utilize higher amounts of glutamine, which can be converted to glutamate, presumably to generate more of the TCA cycle intermediate, α-ketoglutarate which may contribute to elevated heat generation (Figure 6E).