Spatial Regulation of Reactive Oxygen Species via G6PD in Brown Adipocytes Supports Thermogenic Function

Adipose tissue plays key roles in the maintenance of systemic energy homeostasis by regulating energy storage and expenditure. With respect to major functions, mammalian adipose tissues are largely divided into white adipose tissue (WAT) and brown adipose tissue (BAT). While WAT functions as a key organ for energy storage and endocrine function, BAT is specialized in transforming stored energy sources into heat and expends energy in response to cold or metabolic stimuli (1–3). Compared with white adipocytes, brown adipocytes exhibit distinct features in terms of subcellular organelle composition and gene expression profiles (4–6). For example, brown adipocytes contain multilocular lipid droplets and abundant mitochondria and express a brown adipocyte–specific marker protein, uncoupling protein 1 (UCP1) (7,8). The thermogenic function of brown adipocytes primarily depends on UCP1, which is localized in the mitochondrial inner membrane. UCP1 mediates proton leakage across the mitochondrial inner membrane and uncouples oxidative phosphorylation from ATP synthesis, resulting in oxygen consumption and heat production (7).

Reactive oxygen species (ROS) play both beneficial and deleterious roles. They participate in various physiological processes, such as proliferation, differentiation, redox regulation, and heat production (9). For example, hydrogen peroxides and mitochondrial ROS facilitate adipocyte differentiation (10,11). Priming of ROS in plants elevates defense mechanisms against abiotic stress (12). However, imbalance in which ROS generation outweighs clearance causes oxidative stress. Excessive ROS accumulation damages intracellular organelles and macromolecules, thus inducing pathological outcomes. Cellular ROS are generated by nonenzymatic and enzymatic reactions in the mitochondria, peroxisomes, and cytosol (13). Mitochondria are one of the major sources of cellular ROS production (13). Since ROS are produced during mitochondrial respiration, mitochondrial oxygen consumption leads to ROS accumulation in the mitochondria (14). Emerging evidence suggests that in brown adipocytes, cold exposure seems to elevate mitochondrial ROS, which could affect thermogenic activity in response to cold and β-adrenergic stimulation (15–17). On the other hand, it has been also reported that cellular ROS accumulation under pathological conditions could impede thermogenic function in BAT (18,19). However, it is not clearly understood whether and how the balance between ROS production and scavenging in brown adipocytes would affect thermogenic execution.

Glucose-6-phosphate dehydrogenase (G6PD), the first and rate-limiting enzyme of the pentose phosphate pathway (PPP), generates cytosolic NADPH. As a cofactor, NADPH exerts opposite functions for prooxidative enzymes, such as NADPH oxidase and inducible nitric oxide synthase, and antioxidative enzymes, such as glutathione peroxidase and thioredoxin reductase (20). Thus, G6PD appears to act as a double-edged sword in the regulation of cellular ROS levels, depending on the cell and tissue types as well as stimuli (20). We and others have demonstrated that G6PD is highly expressed in WAT and that its overexpression in white adipocytes enhances oxidative stress and proinflammatory responses, leading to insulin resistance in obesity (21,22). Although it has been reported that the enzymatic activity of G6PD is higher in BAT than in WAT and is further elevated in BAT upon cold exposure (23,24), the physiological roles of G6PD in BAT are largely unknown.

In this study, we assessed a whole-body X-linked G6PD-deficient mutant (G6PD^mut) mouse model to understand the regulatory mechanisms of ROS homeostasis related to thermogenic activity in brown adipocytes. Here, we found that G6PD in brown adipocytes is required for thermogenesis by scavenging cellular ROS. Upon cold exposure, G6PD^mut mice exhibited cold intolerance and downregulated thermogenic gene expression in BAT. Genetic and pharmacological inhibition of G6PD in brown adipocytes repressed the thermogenic program and oxygen consumption rate (OCR) upon β-adrenergic activation. Mechanistically, G6PD deficiency in brown adipocytes elevated cytosolic ROS level, which suppressed thermogenic gene expression through the activation of extracellular signal–regulated kinase (ERK). Together, these data suggest that G6PD plays a pivotal role in the thermogenic regulation of brown adipocytes by restricting aberrant cytosolic ROS accumulation and ERK activation.

The animal study was approved by the Institutional Animal Care and Use Committee of Seoul National University. As described previously, X-linked G6PD^mut (C3H) mice developed by ethylnitrosourea mutagenesis were backcrossed into C57BL/6 background for >10 generations to minimize the effects other than G6PD mutation (25–27). Theoretically, 99.99% of G6PD^mut mouse background will be from C57BL/6 inbred after 10 generations of backcrossing (28,29). Whole-body G6PD^mut (C57BL/6) mice and their littermates were housed at 22–24°C in a 12-h light/12-h dark cycle and maintained on normal chow diet. For thermoneutral (TN) and cold exposure experiments, 10- to 14-week-old male mice were placed in 30°C or 4–6°C conditions (DBL Co., Eumseong, South Korea), respectively. For in vivo G6PD knockdown via siRNA in BAT, the interscapular region of mice was shaved and then siRNA (20 μg) was injected into BAT using transfection reagent in vivo-jetPEI (Polyplus Transfection, Illkirch, France) according to the manufacturer’s protocol. Three days after transfection, the mice were exposed to 4–6°C conditions. The sequence for siRNA is listed in Supplementary Table 1.

Immortalized murine brown adipocytes (BAC) were kindly provided by Dr. Kai Ge (National Institutes of Health). BAC preadipocytes were grown in DMEM supplemented with 10% FBS to confluence and then incubated in induction medium consisting of DMEM, 10% FBS, 20 nmol/L insulin, 10 nmol/L 3,3,5-triiodo-l-thyronine (T3), 125 μmol/L indomethacin, 0.5 mmol/L isobutylmethylxanthine, and 5 μmol/L dexamethasone. After 2 days, the cells were treated with differentiation medium (DMEM, 10% FBS, 20 nmol/L insulin, and 10 nmol/L T3) for 2 days, and then the medium was changed with DMEM containing 10% FBS. For differentiation of stromal vascular cells (SVCs)-derived brown adipocytes, confluent SVCs were cultured with induction medium supplemented with 1 μmol/L rosiglitazone. After incubation with the above induction medium for 2 days, cells were placed in the differentiation medium supplemented with 1 μmol/L rosiglitazone. 3T3-L1 adipocytes were differentiated as previously described (30).

A protein-protein interactions (PPIs) network of mouse was downloaded from STRING (v11.0) database (31). With use of the mouse PPIs network with “combined score” of no less than 800 as a template network, a template network composed of 13,655 genes and 651,806 interactions was formed. Of the 77 candidate genes, 61 were mapped on the template network and network propagation (NP) was conducted using a random walk with restart algorithm to interpret individual gene-level perturbations at the network-level associations. NP is a graph-based analysis method that propagates information of a node to nearby nodes through the edges at each iteration for a fixed number of steps or until convergence, allowing estimation of protein interactions (32). We denoted the value of each node after NP as the NP score.

Condition-specific subnetworks were obtained using the genes with the top 5% of NP scores. The subnetwork consisted of 686 genes and 28,798 interactions. The centrality of candidate genes in the condition-specific subnetwork from the template mouse STRING PPI network was calculated with the NetworkX package in Python (33). Degree centrality was used as follows. The degree of a vertex ν for a given graph G: = (V, E) with |V| vertices and |E| edges is defined as $deg(v)$⁠. ν* is the node with the highest degree in graph G. The degree centrality of a vertex ν is defined as$CD(v)=deg(v)deg(v*)$⁠.

Cellular OCR was analyzed with a Seahorse XFe96 extracellular flux analyzer (Seahorse Bioscience, North Billerica, MA) according to the manufacturer’s instructions. Prior to analysis, brown adipocytes were incubated in assay medium (25 mmol/L glucose, 1 mmol/L sodium pyruvate, 2 mmol/L l-glutamine, and 2% fatty acid–free BSA in Seahorse XF base medium at pH 7.4). For mitochondrial stress tests, the OCR was measured following treatment with 5 μmol/L oligomycin, 7.5 μmol/L carbonyl cyanide-p-trifluoromethoxyphenylhydrazone, and 6 μmol/L antimycin A with 3 μmol/L rotenone.

G6PD activity was determined by measurement of NADPH production rate. NADPH level was detected with a fluorometric assay kit (Cayman Chemical, Ann Arbor, MI) according to the manufacturer’s guidelines.

Samples were lysed with radioimmunoprecipitation buffer (34). Phosphorylated (p-)ERK (T202/Y204) and p-JNK (T183/Y185) were purchased from Cell Signaling Technology (Danvers, MA); p-p38 (T180/Y182) antibody was purchased from BD Biosciences (San Jose, CA); ERK1 and p38 antibodies were purchased from Santa Cruz Biotechnology (Dallas, TX); and G6PD and UCP1 antibodies were purchased from Abcam (Cambridge, U.K.).

Total RNA was isolated from various tissues and cells (35). Then, cDNA was synthesized with a reverse transcriptase kit (TOYOBO, Osaka, Japan) according to the manufacturer’s instructions. The primers were generated by Bioneer (Daejeon, South Korea), and the sequence information is described in Supplementary Table 2.

Data were analyzed with Student t test or ANOVA in GraphPad Prism software (GraphPad Software, La Jolla, CA). P < 0.05 was considered significant.

The data sets and noncommercially available resources generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Given that BAT and WAT have distinct roles, it is likely that there are intrinsic differences between transcriptomes of BAT and WAT (6,36–38). To explore distinct molecular features of BAT compared with WAT, we analyzed publicly available microarray data of BAT and WAT from mice exposed to TN or cold temperatures using pathway activity inference analysis (36,39) (Fig. 1A and B). The data showed that redox and ROS control–related pathways, including glutathione and selenocompound metabolism, were transcriptionally enhanced in BAT (Fig. 1B). In BAT, metabolic pathways, such as the PPP, were activated by cold stimulation (Fig. 1B). Gene ontology (GO) analysis revealed that genes upregulated in BAT upon cold stimulation were associated with redox regulation pathways (Fig. 1C and Supplementary Fig. 1A). Moreover, the expression levels of antioxidative genes were upregulated, whereas those of NADPH oxidase subunit genes were downregulated, in BAT (Fig. 1D).

To identify the potential key factor(s) involved in the main functions of BAT, we analyzed 77 genes that were commonly upregulated in BAT by cold stimulation, using two microarray data sets (40) (Supplementary Fig. 1B). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis showed that the PPP was significantly upregulated in BAT upon cold exposure (Fig. 1E). In addition, we conducted an in silico NP analysis to prioritize the 77 candidate genes (Supplementary Table 3). G6PD, which is encoded by G6pdx, was one of the crucial genes strongly influencing the other candidate genes (Fig. 1F and Supplementary Fig. 1C). Further, we evaluated gene centrality, which reflects the importance of each gene in the biological network. As shown in Fig. 1G, G6PD was ranked with a high value of degree centrality in a PPI subnetwork. Together, these data imply that G6PD may be important for BAT functions, including thermogenesis and ROS control.

To verify whether G6PD might be involved in BAT functions, we examined its expression levels in BAT of mice exposed to TN, room temperature (RT) (22°C), and cold temperatures. As shown in Fig. 2A and B, G6pd mRNA and G6PD protein levels in BAT were greatly elevated during cold exposure. Compared with other tissues, cold-induced G6PD expression was quite prominent in BAT (Fig. 2C and D). Moreover, the enzymatic activity of G6PD was elevated in BAT, but not WAT, upon cold exposure (Fig. 2E). For further analysis, BAT was fractionated into adipocytes and SVCs. G6pd mRNA levels were elevated in brown adipocytes but not in SVCs isolated from cold-exposed mice (Fig. 2F). In obese mice, G6PD expression is increased in epididymal WAT (eWAT) and mediates proinflammatory and prooxidative responses (21,27). To investigate whether BAT G6PD might have roles similar to those of eWAT G6PD, we examined G6pd mRNA level in BAT from obese mice and the effects of G6PD on inflammatory and oxidative genes in brown adipocytes. While G6pd mRNA levels were increased in eWAT from obese mice, those of G6pd mRNA were somewhat decreased in obese BAT (Supplementary Fig. 2). In addition, G6PD overexpression in brown adipocytes did not stimulate proinflammatory gene expression or prooxidative pathways—unlike in white adipocytes (Supplementary Fig. 3). These results suggest that G6PD is activated in brown adipocytes in response to cold and might have distinct roles.

To investigate the in vivo functions of G6PD in BAT, we examined the effects of cold exposure on thermogenic activity in wild-type (WT) and G6PD^mut mice. Body weight, adipose tissue mass, and serum lipid levels did not significantly differ between the two genotypes after exposure to RT or cold (Supplementary Fig. 4A–C). Interestingly, we found that G6PD^mut mice were cold intolerant (Fig. 3A and B and Supplementary Video 1). Also, mRNA levels of the thermogenic marker genes were downregulated in BAT of G6PD^mut mice upon cold exposure (Fig. 3C). Simultaneously, G6PD-defective BAT comprised larger lipid droplets upon cold exposure than WT BAT (Fig. 3D). Accordingly, the UCP1 level in BAT of G6PD^mut mice was downregulated upon cold exposure (Fig. 3E and F). In contrast, the mRNA levels of thermogenic marker genes in inguinal WAT (iWAT) were somewhat comparable between WT and G6PD^mut mice (Supplementary Fig. 4D). When metabolic activities of WT and G6PD^mut mice were evaluated, G6PD^mut mice displayed decreased oxygen consumption and energy expenditure upon injection of β3-adrenergic agonist CL316,243 (CL) (Fig. 3G and H). The respiratory exchange rate, locomotive activity, and food intake did not significantly differ between the two genotypes (Supplementary Fig. 4E–G). Unlike in WT mice, thermogenic gene expression in BAT was not upregulated in G6PD^mut mice upon CL injection (Fig. 3I).

Next, to address the question of whether BAT-specific G6PD deficiency might affect body temperature during cold exposure, G6PD was selectively suppressed via siRNA in BAT of WT mice (Supplementary Fig. 5A). Similar to results in whole-body G6PD^mut mice, body temperature was downregulated by BAT-specific G6PD suppression (Fig. 3J and Supplementary Video 2). Simultaneously, G6PD knockdown in BAT attenuated small lipid droplet formation and thermogenic gene expression upon cold stimulation (Supplementary Fig. 5C and Fig. 3K). Collectively, these results propose that G6PD in BAT might mediate thermogenic activity and energy expenditure upon cold or β-adrenergic activation.

The finding that G6PD expression was upregulated in brown adipocytes (Fig. 2) and that G6PD deficiency repressed BAT activation under cold or β-adrenergic stimuli (Fig. 3 and Supplementary Fig. 5) prompted us to ask whether G6PD defect in brown adipocytes might suppress the thermogenic program in a cell-autonomous manner. To address this, we evaluated thermogenic gene expression in primary brown adipocytes fractionated from WT and G6PD^mut mice in the absence or presence of isoproterenol (ISO) to activate β-adrenergic signaling. As indicated in Fig. 4A, G6PD^mut brown adipocytes failed to enhance thermogenic gene expression upon ISO treatment. Also, thermogenic marker gene expression was attenuated by G6PD inhibitors such as dehydroepiandrosterone (DHEA) and 6-aminonicotinamide in the presence of ISO (Fig. 4B and Supplementary Fig. 6A). Moreover, G6PD knockdown in brown adipocytes decreased thermogenic marker gene expressions (Supplementary Fig. 6B). In addition, the level of UCP1 was decreased in brown adipocytes by G6PD inhibition or knockdown after ISO treatment (Fig. 4C and Supplementary Fig. 6C). In contrast, G6PD overexpression in brown adipocytes tended to increase thermogenic gene expression in the presence of ISO (Fig. 4D). Then, to examine the effects of G6PD inhibition on mitochondrial biogenesis or activity in brown adipocytes, we analyzed mitochondrial DNA contents and OCR upon G6PD inhibition. The ratio of mitochondrial DNA–to–genomic DNA was not significantly affected by G6PD inhibition (Supplementary Fig. 6D). In brown adipocytes, G6PD inhibitors did not alter OCR in the absence of ISO (Supplementary Fig. 6E). In contrast, G6PD inhibitors did significantly reduce OCR after ISO treatment (Fig. 4E). Similarly, the OCR and thermogenic gene expression in SVC-derived differentiated brown adipocytes were significantly diminished by G6PD deficiency in the presence of ISO, albeit to a lesser extent rather than for differentiated BAC treated with G6PD inhibitors or BAT from G6PD^mut mice (Fig. 4F and Supplementary Fig. 7). It seems that this difference might result from several factors including the differences between cellular heterogeneity in SVCs and homogenous BAC and/or the differences between in vitro cell culture system and in vivo tissue microenvironments. Together, these data suggest that G6PD in brown adipocytes is crucial for the activation of thermogenic gene expression and mitochondrial oxygen consumption upon β-adrenergic activation.

G6PD produces NADPH, which is a key cofactor in the regulation of cellular redox (20). For verification of whether decrease in thermogenic activity in G6PD-defective brown adipocytes might be associated with ROS, cellular ROS level was determined. The basal level of cellular ROS was increased after G6PD inhibition with DHEA, and ISO further augmented the cellular ROS levels in brown adipocytes in the presence of DHEA (Fig. 5A). Cellular ROS accumulation was increased in primary brown adipocytes, but not in eWAT from G6PD^mut mice, under both RT and cold conditions (Fig. 5B and Supplementary Fig. 8). In BAT of G6PD^mut mice, cellular ROS level was further elevated upon CL injection (Fig. 5C). Nonetheless, mitochondrial superoxide levels in BAT and brown adipocytes were not different between the two genotypes (Fig. 5D and E).

The fact that most G6PD is localized in the cytosol led us to investigate whether G6PD in brown adipocytes might affect cytosolic ROS levels, rather than mitochondrial ROS levels, upon β-adrenergic activation. To measure cytosolic ROS levels in brown adipocytes, we used an engineered ascorbate peroxidase (APEX)-generated biotin labeling system. In the presence of hydrogen peroxides, the subcellular compartment–targeting APEX produces biotin-phenoxyl radicals, which in turn biotinylate nearby proteins (41). For investigation of the levels of cytosolic ROS or mitochondrial ROS, brown adipocytes transfected with APEX2-nuclear export signal (NES) (cytosol) or matrix-APEX2 (mitochondrial matrix) were treated with ISO in the absence or presence of DHEA. As shown in Fig. 5F and G, G6PD inhibition accelerated cytosolic ROS accumulation, whereas it did not significantly affect mitochondrial ROS levels upon β-adrenergic activation. Together, these results suggest that G6PD in brown adipocytes modulates cytosolic ROS levels.

For investigation of whether the impaired thermogenesis in G6PD-deficient brown adipocytes might result from oxidative stress, brown adipocytes were treated with antioxidant N-acetyl-l-cysteine (NAC). As shown in Fig. 6A, NAC decreased cellular ROS level in brown adipocytes. In DHEA-treated brown adipocytes, NAC restored the OCR and thermogenic gene expression in the presence of ISO (Fig. 6B and C). For verification of whether accumulated cellular ROS were responsible for the disrupted BAT activation in G6PD^mut mice, NAC was injected into mice, followed by CL administration. Cellular ROS levels in BAT were decreased in NAC-treated G6PD^mut mice (Fig. 6D). In BAT of G6PD^mut mice, NAC administration stimulated thermogenic gene expression and promoted small lipid droplet formation (Fig. 6E and F). Accordingly, UCP1 level in BAT was increased in NAC-treated G6PD^mut mice (Fig. 6G). Moreover, NAC elevated body temperature in the interscapular area in G6PD^mut mice after CL injection (Fig. 6H and Supplementary Video 3). These data indicated that increased cytosolic oxidative stress induced by G6PD defect damaged the thermogenic program in brown adipocytes.

Next, we investigated whether the rescue effects of an antioxidant on thermogenic activity in G6PD-deficient brown adipocytes might depend on the removal of mitochondrial ROS. To address this, we used a mitochondria-targeted ROS scavenger, MitoQ. Mitochondrial ROS levels were reduced after MitoQ treatment (Supplementary Fig. 9A). In DHEA-treated brown adipocytes, MitoQ failed to restore thermogenic gene expression in the presence of ISO (Supplementary Fig. 9B). For further examination of whether the elimination of mitochondrial ROS might modulate the thermogenic function in G6PD^mut mice, mice were injected with MitoQ and exposed to cold. MitoQ treatment decreased the body temperature and mitochondrial superoxide levels in BAT from both WT and G6PD^mut mice (Supplementary Fig. 9C and D). Although there was a clear difference in body temperature between the two genotypes upon MitoQ treatment (Supplementary Fig. 9C), thermogenic gene expression in BAT of WT and G6PD^mut mice was not significantly altered by MitoQ (Supplementary Fig. 9E), implying that impaired thermogenic activity in brown adipocytes with G6PD defect would not be associated with mitochondrial ROS.

Oxidative stress potentiates signaling of mitogen-activated protein kinases (MAPKs) including ERK, p38 MAPK, and c-JUN N-terminal kinase (JNK) to regulate gene expression (42). To unravel the underlying mechanisms by which G6PD-deficient brown adipocytes downregulate thermogenesis, we focused on MAPK signaling under cold conditions. Phosphorylation of ERK, but not of p38 MAPK or JNK, was upregulated in BAT of G6PD^mut mice exposed to cold or injected with CL (Fig. 7A and Supplementary Fig. 10A). In BAT of G6PD^mut mice, the increase in ERK phosphorylation was suppressed by NAC treatment (Fig. 7B). To clarify the involvement of ERK activation in the suppressed thermogenic execution by G6PD defect, G6PD-deficient brown adipocytes were treated with an ERK inhibitor, PD98059 (PD). In DHEA-treated brown adipocytes, ERK inhibition restored thermogenic gene expression after ISO treatment (Fig. 7C and Supplementary Fig. 10B). For investigation of whether ERK activation is responsible for the dampened thermogenic activity in G6PD^mut mice, PD was administrated into WT and G6PD^mut mice, followed by cold exposure. In G6PD^mut mice, PD not only decreased ERK activation in BAT (Supplementary Fig. 10C) but also increased the body temperature during cold exposure (Fig. 7D and E and Supplementary Video 4). In BAT of G6PD^mut mice, ERK inhibition stimulated small lipid droplet formation and potentiated thermogenic marker gene expression (Fig. 7F and G). These data indicated that ERK in brown adipocytes would mediate the deleterious effects of G6PD defect on thermogenic execution.

Emerging evidence suggests that cellular ROS in brown adipocytes are closely associated with thermogenic activity (16–19,43). Elevated mitochondrial ROS in BAT contribute to thermogenesis (16,17), whereas age- or obesity-induced oxidative stress impairs BAT function (18,19,43). However, it remains largely unknown how cellular ROS in brown adipocytes are modulated to avoid oxidative stress. This study provided several lines of evidence that G6PD in brown adipocytes would play a critical role in the thermogenic program by restricting cytosolic oxidative stress during cold exposure. First, G6PD defect in brown adipocytes downregulated the expression of thermogenic genes upon β-adrenergic activation. Second, G6PD^mut mice were cold intolerant and had enhanced cellular ROS levels in BAT. Third, cytosolic ROS level was elevated in G6PD-defective brown adipocytes and was further augmented upon β-adrenergic activation. Fourth, antioxidant NAC treatment in G6PD-inhibited brown adipocytes rescued thermogenic gene expression and OCR. Finally, in BAT of G6PD^mut mice, NAC promoted thermogenic activity, accompanied by small lipid droplet formation and induced thermogenic gene expression.

In WAT, we and others have demonstrated that G6PD acts as a key prooxidative enzyme by generating NADPH, which supports NADPH oxidase–mediated ROS production (21,22,27,44,45). In white adipocytes, G6PD overexpression promotes cellular ROS accumulation and oxidative stress–induced inflammatory responses (21). Moreover, G6PD defect alleviates chronic inflammation and insulin resistance in obese WAT (27). On the other hand, it is well-known that NADPH is a key electron donor for antioxidative enzymes, such as glutathione reductase, which ameliorate oxidative stress (46). Thus, it has been proposed that the role of G6PD in the regulation of ROS homeostasis would be determined by intracellular properties, depending on the cell types (20). Here, we found that physiological roles of G6PD in the regulation of cellular ROS levels would be quite different between white and brown adipocytes. Compared with white adipocytes, brown adipocytes contain more mitochondria, which may render them more susceptible to cellular oxidative stress. Bioinformatics analysis results revealed that redox control system and antioxidative pathways in brown adipocytes seemed to enhance to alleviate excessive ROS accumulation—unlike in white adipocytes. The expression levels of electron donor–producing enzymes and antioxidative enzymes were upregulated, whereas those of NADPH oxidase subunits were downregulated in BAT compared with WAT (Fig. 1). In addition, G6PD overexpression in brown adipocytes reduced cellular ROS levels without altering the expression levels of prooxidative enzymes, which are the targets of G6PD in white adipocytes (Supplementary Fig. 3). In brown adipocytes, G6PD defect significantly augmented the levels of cytosolic ROS, which could provoke oxidative stress (Fig. 5). Therefore, these data suggest that G6PD in brown adipocytes plays an antioxidative role, whereas G6PD in white adipocytes has prooxidative functions, probably due to the distinct intracellular context.

Cellular ROS are produced in several subcellular compartments, including the cytosol and mitochondria (47). In particular, cytosolic ROS are produced by NADPH oxidase and xanthine oxidase (48). On the other hand, mitochondrial ROS are generated in the electron transport chain and can diffuse into the cytosol through channels such as voltage-dependent anion channel or aquaporin (49,50). Recently, it has been reported that subcellular organelle-specific spatial regulation of ROS is crucial for physiological processes such as life span and endothelial function (51,52). Cytosolic ROS and mitochondrial ROS play opposite roles in longevity (51). In brown adipocytes, the level of total cellular ROS, including ROS in the cytosol, mitochondria, and other subcellular organelles, rapidly increased and then decreased in the presence of β-adrenergic activation. In contrast, it seemed that mitochondrial ROS levels were elevated and then maintained after β-adrenergic activation (Supplementary Fig. 11). These observations indicate that the change in total cellular ROS upon β-adrenergic activation would reflect decreased ROS in subcellular compartments other than mitochondrial ROS. In this regard, it is feasible to speculate that brown adipocytes might have distinct spatial regulatory mechanisms of ROS homeostasis in the mitochondria and other subcellular compartments, including the cytosol. Given that G6PD produces cytosolic NADPH, we hypothesized that G6PD in brown adipocytes could be preferentially involved in the spatial regulation of ROS by regulating cytosolic ROS rather than mitochondrial ROS. Although DCF-DA intensity was enhanced in G6PD-deficient brown adipocytes, MitoSOX intensity did not differ between brown adipocytes from WT and G6PD^mut mice. Intriguingly, using subcellular compartment–specific APEX systems, we found that G6PD inhibition in brown adipocytes could alter cytosolic ROS level, but not mitochondrial ROS level, upon β-adrenergic activation (Fig. 5). In addition, NAC restored thermogenic activity in BAT of G6PD^mut mice, whereas MitoQ did not mitigate impaired thermogenic activity in BAT of G6PD^mut mice (Fig. 6 and Supplementary Fig. 9), implying that G6PD would preferentially decrease cytosolic ROS in brown adipocytes, thereby preventing oxidative stress and supporting thermogenesis. Although it remains to be investigated whether increased cytosolic ROS due to G6PD defect might be primarily produced in the cytosol and/or released from other subcellular organelles, it is important to study the spatial regulation of ROS homeostasis at the subcellular level to better understand the roles of ROS in thermogenesis.

MAPK pathways are involved in the regulation of metabolic functions upon various stimuli. For instance, prooxidative stimuli activate ERK, whereas scavenging of cellular ROS with antioxidants diminishes ERK signaling cascades (53,54), indicating that ERK mediates the effects of ROS on physiological or pathological processes. In this regard, accumulating evidence suggests that ERK could regulate thermogenesis in adipocytes (55–58). Also, it has been reported that ERK is a negative regulator of UCP1 expression in brown adipocytes (55). It is of interest to note that G6PD defect would selectively activate ERK in brown adipocytes upon cold or β-adrenergic stimuli. In addition, ERK activation by G6PD defect was attenuated upon NAC administration. Furthermore, ERK inhibition with PD stimulated the thermogenic program in G6PD-inhibited brown adipocytes and restored body temperature of cold-intolerant G6PD^mut mice. Even though we cannot exclude the possibilities that other kinases and/or signaling pathways would mediate G6PD defect in brown adipocytes, current results suggest that activated ERK due to cytosolic ROS in response to G6PD defect appears to be one of the mediators to impair thermogenesis in brown adipocytes.

In conclusion, we demonstrated that G6PD in brown adipocytes would have a distinct role in the regulation of ROS, which is attributed to their unique intracellular context that differs from that of white adipocytes. In brown adipocytes, cytosolic G6PD positively contributes to thermogenic activity by preventing oxidative stress–induced ERK activation upon cold or β-adrenergic stimulation (Supplementary Fig. 12). Taken together, our data suggest that the removal of harmful cytosolic ROS by G6PD in brown adipocytes could be a key process in mediating thermogenic homeostasis under cold challenge.

Acknowledgments. The authors thank Dr. Kai Ge at National Institutes of Health for providing BAC. The authors also appreciate the Korea Mouse Phenotyping Center for analysis of metabolic cages.

Funding. This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT [MSIT], no. NRF-2020R1A3B2078617 and no. NRF-2018R1A5A1024340). In addition, J.H.S. was supported by Basic Science Research Program through NRF funded by the Ministry of Education (no. 2021R1I1A1A01056574). S.M.H. was supported by the BK21 Plus program.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.H.S. and J.B.K. designed the study. J.H.S. conducted most of the experiments and wrote the manuscript. Y.J. contributed to microscopy experiments. H.N. performed in vivo transfection experiments. C-.Y.C., S.L., and S.K. conducted the bioinformatics analysis. Y.G.J., S.M.H., and J.S.H. helped with analysis of data. I.P. and H.-W.R. discussed data. J.B.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in the doctoral dissertation of J.H.S., Roles of Lipid Metabolites and Reactive Oxygen Species in the Regulation of Adipocyte Function, Seoul National University, 2021.