mTORC2-AKT signaling to ATP-citrate lyase drives brown adipogenesis and de novo lipogenesis

mTORC2 phosphorylates AKT in a hydrophobic motif site that is a biomarker of insulin sensitivity. In brown adipocytes, mTORC2 regulates glucose and lipid metabolism, however the mechanism has been unclear because downstream AKT signaling appears unaffected by mTORC2 loss. Here, by applying immunoblotting, targeted phosphoproteomics and metabolite profiling, we identify ATP-citrate lyase (ACLY) as a distinctly mTORC2-sensitive AKT substrate in brown preadipocytes. mTORC2 appears dispensable for most other AKT actions examined, indicating a previously unappreciated selectivity in mTORC2-AKT signaling. Rescue experiments suggest brown preadipocytes require the mTORC2/AKT/ACLY pathway to induce PPAR-gamma and establish the epigenetic landscape during differentiation. Evidence in mature brown adipocytes also suggests mTORC2 acts through ACLY to increase carbohydrate response element binding protein (ChREBP) activity, histone acetylation, and gluco-lipogenic gene expression. Substrate utilization studies additionally implicate mTORC2 in promoting acetyl-CoA synthesis from acetate through acetyl-CoA synthetase 2 (ACSS2). These data suggest that a principal mTORC2 action is controlling nuclear-cytoplasmic acetyl-CoA synthesis.

A cell's capacity to sense and respond to nutrients is essential for metabolic homeostasis and losing this flexibility is a hallmark of diabetes and cancer. The mammalian target of rapamycin (mTOR) kinase is a master regulator of intracellular metabolism that responds to insulin signaling and nutrient availability. Its biochemical functions are split between two complexes called mTOR complex 1 (mTORC1), which contains the essential subunit Raptor, and mTOR complex 2 (mTORC2), which contains the essential subunits Rictor and Sin1 1,2 . Amino acids and insulin stimulate mTORC1 to promote protein, lipid, and nucleic acid synthesis through well-defined downstream signaling networks. mTORC2 promotes glycolysis and lipid synthesis; however, its regulation and downstream mechanisms are less well defined.
Many models show mTORC2 as an obligatory AKT activator 1,5 ; however, its exact function in downstream AKT signaling in vivo remains elusive because, in many tissue-specific conditional Rictor knockout (KO) models, mTORC2 loss has seemingly minimal-to-no effect on the phosphorylation of several AKT substrates 2,10 . For example, conditionally deleting Rictor in brown adipose tissue (BAT) (e.g., with ucp1-cre or myf5-cre) or in white adipose tissue (WAT) (with adiponectin-cre) ablates AKT HM phosphorylation without obviously impairing AKT activity toward classic substrates, including PRAS40, TSC2, GSK3β, AS160, and FoxO1 [10][11][12][13][14] . Yet, despite seemingly normal AKT signaling, Rictor loss profoundly downregulates transcription of ATP citrate lyase (ACLY), acetyl-CoA carboxylase (ACC), and fatty acid synthase (FASN), which catalyze de novo lipogenesis (DNL) 11,12 . Rictor loss also reduces glucose uptake and inhibits the expression of ChREBPβ 12,13 , which is a constitutively active isoform of the carbohydrate response element binding protein (ChREBP) transcription factor that stimulates carbohydrate and lipid metabolic gene expression 15 . Similarly, inducible deletion of Rictor in brown preadipocytes has seemingly no effect on downstream AKT signaling, yet renders these cells incapable of differentiating in vitro 11 . Nevertheless, expressing recombinant AKT1 containing a phospho-mimetic S473 residue in Rictordeficient preadipocytes rescues differentiation. The latter observation suggests that mTORC2 may be more essential for some AKT actions that are required for brown adipocyte differentiation and de novo lipogenesis, but dispensable for many others.
ACLY cleaves extra-mitochondrial citrate to generate acetyl-CoA, which is the precursor for glucose-dependent de novo lipid and cholesterol biosynthesis. Acetyl-CoA is also used to acetylate lysine residues on histones and metabolic proteins to regulate gene expression and enzyme activity, respectively. ACLY serine 455 lies within a basophilic phosphorylation motif (RxxS) that is similar to the AKT consensus motif (RxRxxS/T). Phosphorylation of this site stimulates ACLY activity; however, the serine 455 kinase remains controversial as AKT 16 , PKA 17,18 , mTOR 19 , or the branched chain ketoacid dehydrogenase kinase 20 can reportedly phosphorylate this site. Nevertheless, because ACLY functions at the interface of glucose-dependent DNL and epigenetic control of gene expression, it is poised to be a key link between hormonal signaling and carbohydrate and lipid metabolism in adipocytes.
Here we test the hypothesis that mTORC2 promotes brown adipocyte differentiation and glucose-driven DNL (gluco-lipogenesis) through a distinctly mTORC2-dependent AKT pathway. To do this, we analyzed mTORC2-regulated AKT phosphosites and metabolites by mass spectrometry. This led us to identify a subset of AKT substrates and metabolites that are regulated by mTORC2, including the enzyme ACLY and its product acetyl-CoA. We provide evidence that ACLY functions downstream of a uniquely mTORC2-dependent AKT pathway required for DNL, differentiation, histone acetylation, and ChREBPβ and glucolipogenic gene expression. Substrate utilization studies suggest an additional role for mTORC2 in promoting acetyl-CoA synthesis from acetate through acyl-CoA synthetase short chain family member 2 (ACSS2). These data uncover a previously unappreciated selectivity in mTORC2-dependent AKT signaling in Rictor-deleted brown adipocytes and reveal mTORC2's central role in controlling nuclear-cytoplasmic acetyl-CoA synthesis.

Results
Stratification of AKT substrates by mTORC2 sensitivity. To test the hypothesis that some AKT substrates are more sensitive to mTORC2 loss than for others and to identify the mechanism by which mTORC2 drives brown preadipocyte differentiation, we conducted a targeted phosphoproteomics assay comparing the insulin response between Rictor-iKO precursor brown adipocytes (Rictor-iKO PBAs ) and isogenic control brown preadipocytes. We designed the assay to quantify phosphopeptides encompassing regulatory phosphorylation sites on AKT, previously reported AKT and SGK substrate sites, and other potential mTORC2 effectors. Rictor-iKO PBAs maintain growth-factor-stimulated AKT T308 phosphorylation, albeit at lower levels, and seemingly normal phosphorylation of AKT substrates [ Supplementary Fig. 1A], as previously observed 11 . Rictor-iKO PBAs and their vehicle-treated isogenic controls were then serum deprived, or serum deprived and then stimulated with insulin for 15 min. Phosphopeptide-enriched samples were analyzed with parallel reaction monitoring targeted mass spectrometry, a highly sensitive quantitative technique. The assay quantified 31 known AKT substrates, of which 17 increase phosphorylation at least 1.5-fold in response to insulin in wild-type (WT) cells at Benjamini-Hochberg 5% false detection rate (FDR), indicating that they are likely targets of insulin-stimulated AKT action. Of the 17 insulin-upregulated phosphopeptides, 9 contain AKT substrate motifs for which the phosphorylated residue can be localized with high confidence. The other eight contain phosphorylated residues with ambiguous localization within at most three residues away from the predicted AKT substrate site.
Based on quantitative differences between conditions, we stratified phosphopeptides into three classes, all with a 5% FDR from analysis of variance (ANOVA; see "Methods"): Class I contains phosphosites that are highly dependent upon mTORC2; Class II contains phosphosites that are insensitive to mTORC2 loss; and Class III contains phosphosites that are partially sensitive to mTORC2 loss [ Fig. 1a, Supplementary Data 1]. Among the Class I peptides, we identified the known direct mTORC2 target sites pAKT1 S473 , pAKT2 S474 , and pAKT3 S472 as insulin stimulated and highly sensitive to mTORC2 loss [ Fig. 1b,  Supplementary Fig. 2A]. We also identified a Class I peptide from NDRG1 with phosphorylation on either S364 or T366 that was not significantly stimulated by insulin but is highly dependent of mTORC2 [ Fig. 1b]. This difference is not explained by changes to NDRG1 protein levels as measured by mass spectrometry. pNDRG1 T366 is known to be phosphorylated by SGK, a kinase from the same family of AKT that requires mTORC2-dependent HM phosphorylation for its activation 6 . These findings confirm our ability to identify phosphorylation sites that occur with or without insulin stimulation for which mTORC2 is essential.
Among the Class II substrates, for which mTORC2 is dispensable, we identified several AKT target sites, including pPRAS40 T246 and pGSK3β S9 [Fig. 1c, Supplementary Fig. 2B], validating our previously observed immunoblot analyses using phospho-specific antibodies [ Supplementary Fig. 1A] 11 . Also included in Class II are pTSC2 S939 and pTSC2 S981 , two wellknown AKT phosphorylation events that stimulate the mTORC1 pathway 21 [Fig. 1c, Supplementary Fig. 2B]. Consistently, ribosomal protein S6 peptides containing phosphorylation at S235 or S236 and at S240 (often used as an mTORC1 activity reporter) are also Class II substrates [ Supplementary Fig. 2B]. We additionally identified several predicted AKT substrates that fall into Class III, showing partial sensitivity to mTORC2 loss [ Fig. 1d, Supplementary Fig. 2C]. One peptide is from a region in  Fig. 1 Identification of mTORC2-sensitive phosphorylation sites, including S455 in ATP citrate lyase (ACLY). a Phosphoproteomics work flow and analysis in Rictor-iKO PBAs . Colored circles correspond with the type of motifs analyzed (n = 6 per group). b, c Motifs were stratified by their sensitivity to mTORC2 loss as being highly sensitive Class I sites (b), insensitive Class II sites (c), or partially sensitive Class III sites (d). e Western blots of protein lysates from control and Rictor-iKO PBA cells using the indicated total and phospho-specific antibodies. Cells growing in DMEM were either serum deprived (−FBS) or serum deprived and then stimulated with fresh serum for 15 min (+FBS) prior to lysis. f Western blots of protein lysates treated as in e with or without the AKT inhibitor MK2206. g Western blots using lysates from HEK293 cells in which Rictor was deleted by CRISPR/Cas9 genome editing. The AKT inhibitor MK2206 (10 μM) was administered 1 h prior to lysis in both f and g.
NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14430-w ARTICLE phosphofructo kinase-2 (PFK2) containing either the annotated AKT substrate site S486 or the nearby S489, and another is from a region in ACLY containing the S455 site [ Fig. 1d]. We also identify mTORC2-sensitive sites on BAD (S155) and vimentin (S39) [ Supplementary Fig. 2C]. These data suggest that there is selective inhibition of AKT signaling pathways following induced Rictor deletion in brown preadipocytes. Immunoblotting with commercially available phosphositespecific antibodies independently uncovered serum and insulinstimulated pACLY S455 as an AKT substrate showing mTORC2 dependency [ Fig. 1e, Supplementary Fig. 1B]. We were unable to detect pPFK2 S486 using a commercially available antibody. Importantly, pACLY S455 and pPRAS40 T246 (an mTORC2insensitive AKT substrate in our assays [ Fig. 1c, f, Supplementary  Fig. 1B]) are both sensitive to the AKT inhibitor MK2206 [Fig. 1f, Supplementary Fig. S1B]. This argues against an AKT compensatory kinase acting in the absence mTORC2. Serum or insulinstimulated pACLY S455 , but not pPRAS40 T246 , is also mTORC2 dependent in HEK293E cells deleted for Rictor using CRISPR/ Cas9, while both sites are sensitive to MK2206 [ Fig. 1g, Supplementary Fig. 1C]. We also expressed recombinant Rictor in the Rictor-deleted HEK293E cells to confirm that this restores ACLY S455 phosphorylation [ Supplementary Fig. 1D]. Thus selectivity may be relevant in other Rictor-deleted cell types.
We also measured the steady-state abundance of 139 additional metabolites representing intermediary metabolism, 113 of which were detected in control and Rictor-iKO PBAs [Supplementary Data 2]. Consistent with an acetyl-CoA synthesis defect, acetylcarnitine (which derives from acetyl-CoA and carnitine in a fast and reversible reaction typically in equilibrium with nuclear-cytoplasmic acetyl-CoA 23 ) decreases by 35.04% in Rictor-iKO BPAs [ Fig. 2b]. Acetylserine and pantothenic acid, both of which contribute to Coenzyme A synthesis, are also reduced by 23.6% and 12.1%, respectively [Supplementary Data 2]; the latter possibly resulting from de-repressed pantothenate kinase activity 24 . While most glycolytic intermediates are unaffected by Rictor loss, F1,6BP decreases by 49.1% [ Fig. 2c], which could be due to the mTORC2-sensitive phosphosite in PFK-2 [ Fig. 1d]. We also observe a 42.5% decrease in citrate, the tricarboxylic acid (TCA) cycle intermediate exported from the mitochondria and used by ACLY to generate acetyl-CoA [ Fig. 2d]. It cannot be determined from these data as to whether this represents the intra-or extramitochondrial citrate fraction; nevertheless, this could reflect a metabolite-driven negative feedback effect or inherent instability. The only metabolite that significantly increases is the TCA cycle intermediate malate (by 29.6%) [ Fig. 2d], possibly reflecting reduced demand for citrate to be exported. Taken together, these data support the notion that a principal mTORC2 function in intermediary metabolism is regulating acetyl-CoA levels.
To directly test whether mTORC2 regulates acetyl-CoA levels, we used stable isotope dilution liquid chromatography-mass spectrometry with 13 C 3 15 N 1 -labeled acyl-CoA internal standards to measure total cellular acetyl-CoA levels 25,26 . This revealed a 43.54% decrease in acetyl-CoA in Rictor-iKO BPAs [Fig. 2e]. We also measured DNL from glucose via acetyl-CoA intermediates using a C 14 -glucose tracer assay. This shows a 48.1% reduction in lipid synthesis in Rictor-iKO BPAs relative to controls [ Fig. 2f]. Quantitative PCR and western blotting indicate that ACLY, ACC, and FASN mRNA and protein levels are similar in control and Rictor-iKO precursors [ Supplementary Fig. 1E, F], consistent with decreased lipogenic flux. Thus integrating results from immunoblotting, phosphoproteomics, and metabolomics points to ACLY as a distinctly mTORC2-sensitive AKT substrate.
mTORC2 drives brown adipocyte differentiation through ACLY. To test whether a defect in ACLY phosphorylation explains why mTORC2-deficient brown preadipocytes cannot differentiate, we first determined whether ACLY itself is required for differentiation. To this end, we generated Acly-iKO PBAs and subjected them to a differentiation assay [ Fig. 3a To determine whether mTORC2 and ACLY might function in the same pathway, we asked whether rescuing ACLY S455 phosphorylation in Rictor-iKO PBAs is sufficient to restore their ability to differentiate. As shown previously 11  These data support a model in which ACLY is a critical mTORC2 effector during brown adipogenesis.
To further test this hypothesis, we generated Akt1-iKO BPAs and Akt2-iKO BPAs for comparison in a differentiation assay [ Fig. 4b].
Like Rictor-iKO PBAs , only the Akt1-iKO BPAs are incapable of synthesizing lipid droplets in a differentiation assay [ Fig. 4c], which is in agreement with previous studies 11,27 . Consistently, Akt1-iKO BPAs Fig. 4g]. This is consistent with total AKT1 deletion having broader effects on glucose-dependent acetyl-CoA synthesis than losing only HM phosphorylation. Further consistent with this notion, glucose uptake is reduced by 37.8% in Akt1-iKO BPAs compared to only a 10.89% in Rictor-iKO PBAs [ Supplementary Fig. 4A]. These data support a model in which mTORC2-dependent AKT1 HM phosphorylation stimulates ACLY to promote brown adipocyte differentiation.
A glucose uptake deficiency is not limiting upon mTORC2 loss. Because adipocyte mTORC2 loss impairs glucose uptake in vivo [12][13][14]28 , we had considered the possibility that a glucose uptake deficiency might be preventing Rictor-iKO PBAs from differentiating. This seemed unlikely as Rictor-iKO PBAs have a relatively minor defect in glucose uptake [Supplementary        4E, F]. Thus neither a glucose uptake nor a trapping defect is a major limiting factor for the differentiation of Rictor-iKO BPAs .
Overexpressing recombinant PPARγ2 in Rictor-iKO brown preadipocytes rescues differentiation; however, supplementing rosiglitizone (a synthetic PPARγ ligand) cannot 11 , suggesting that the mTORC2/AKT1/ACLY pathway likely regulates an early transcriptional initiation event rather than a defect in endogenous ligand synthesis and/or amplification. Given that ACLY also generates acetyl-CoA necessary for histone acetylation and Glut4 expression during white adipocyte differentiation 30,31 , we asked whether mTORC2 might facilitate epigenetic programming during brown adipocyte differentiation. Notably, total H3 acetylation decreases in undifferentiated Rictor-iKO PBAs [ Supplementary  Fig. 5A]. At differentiating day 2, just before Pparγ2 is initially induced, analysis of specific acetylation marks also reveals a marked decrease in H3K27Ac (a key marker of active enhancers during adipogenesis 32  mTORC2 regulates acetyl-CoA synthesis from acetate. While glucose fuels acetyl-CoA synthesis through ACLY, acetyl-CoA can also be generated from acetate through ACSS2 33 . Therefore, we asked whether the ability of Rictor-iKO PBAs to differentiate might also be rescued by supplementing high levels of exogenous acetate into the culture medium. We compared this by adding excess exogenous citrate or pyruvate, which are acetyl-CoA precursors in the glucose pathway. Neither citrate nor pyruvate supplementation rescues differentiation [ Fig. 5a, b]. However, supplementing acetate partially rescues lipid droplet synthesis (to 45.57% of control) and Pparγ2 expression (to 23.47% of control) as well as protein expression without restoring acetyl-CoA level [ Fig. 5a-c, Supplementary Fig. 5G]. Interestingly, performing the acetate supplementation experiment with Acly-iKO PBAs indicates that acetate is nearly twice as effective at restoring lipid droplet synthesis and Pparγ expression (to 85.4% and 44.0% of control levels, respectively) vs Rictor-iKO PBAs [Fig. 5d, e, Supplementary  Fig. 5H]. That Rictor-iKOPBAs are less efficient than Acly-iKO PBAs at using acetate, combined with the fact that acetate supplementation did not restore acetyl-CoA levels, suggested the intriguing possibility that mTORC2 might also regulate acetatederived acetyl-CoA synthesis.
To test whether mTORC2 regulates acetyl-CoA synthesis from acetate, we performed tracer experiments by culturing Rictor-iKO PBAs in the presence of [U-13 C]-glucose (10 mM) and physiological levels of unlabeled acetate (100 μM) or physiological levels of [1,2-13 C]-acetate (100 μM) and unlabeled glucose (10 mM) for 1 h prior to measuring acetyl-CoA. Compared to their isogenic controls, the percentage of labeling of acetyl-CoA from both glucose and acetate is unchanged in Rictor-iKO PBAs [ Fig. 5f]. As total acetyl-CoA levels are halved in these cells [ Fig. 2e], this suggests that Rictor loss may reduce the ability of cells to use either glucose or acetate to label acetyl-CoA by roughly equivalent proportions.
In MEFs, deleting Acly upregulates acss2/ACSS2, thereby enabling cells to engage compensatory acetate metabolism to generate acetyl-CoA 34 . We reasoned that a difference in ACSS2 expression could explain why Rictor-iKO PBAs use acetate less effectively than Acly-iKO PBAs . Like in MEFs, Acly-iKO PBAs induce compensatory ACSS2 expression [ Fig. 5g]. Glucose and acetate tracer analysis also shows that Acly-iKO PBAs switch from using glucose to acetate for acetyl-CoA synthesis [ Fig. 5h], increasing total cellular acetyl-CoA above normal levels (by 118.14%) [ Fig. 5i], also like in MEFs. However, this alternative pool of acetyl-CoA is reportedly not efficiently used for histone acetylation unless high levels of exogenous acetate are supplemented likely explaining why Acly-iKO PBAs cannot differentiate at physiological acetate levels [ Fig. 5d-e] 30 . Consistently, Acly-iKO PBAs have defects in H3K9 and H3K14 acetylation in addition to H3K27 acetylation that are all rescued with high levels of exogenous acetate [ Supplementary Fig. 5D].
In contrast, despite Rictor-iKO PBAs having nearly a 50% drop in total acetyl-CoA levels, they fail to trigger compensatory ACSS2 induction; in fact, ACSS2 levels are reduced in Rictor-iKO PBA cells [ Fig. 5g]. This is consistent with Rictor-deficient cells having reduced capacity to metabolize acetate. Moreover, while Rictor-iKO PBAs and Acly-iKO PBAs overlap in having both H3K14Ac and H3K27Ac deficiencies, only H3K27Ac is rescued with high acetate supplementation in Rictor-deficient cells [ Supplementary Fig. 5B, D]. Finally, we show that compensatory ACSS2 upregulation triggered by ACLY loss is blocked when Rictor is simultaneously knocked out [ Fig. 5j] indicating the glucose-to-acetate metabolic switch triggered upon ACLY deletion also requires mTORC2. Collectively, these data reveal an additional role for mTORC2 in promoting acetyl-CoA synthesis from acetate at least in part by controlling ACSS2 expression.
ACLY is a key mTORC2 effector in mature brown adipocytes. Our prior in vivo observations show that chronic mTORC2 loss (caused by conditionally deleting Rictor in mature brown or white adipocytes with congenitally expressed Cre drivers) attenuates ACLY, ACC, and FASN mRNA and protein expression [11][12][13] . In contrast, more acutely deleting Rictor in precursor brown adipocytes by inducible KO reduces gluco-lipogenic flux but did not impair DNL enzyme expression [ Fig. 2f, Supplementary Fig. 1B, C]. This led us to consider whether the decrease in Acly, Acc, and Fasn gene expression that we observe in mature brown adipocytes may be a secondary effect of impaired ACLY function that occurs with prolonged Rictor loss.
To investigate the temporal effect of acute vs prolonged Rictor loss on gluco-lipogenic gene expression, we performed a timecourse analysis of ACLY phosphorylation and gene expression following tamoxifen-induced Rictor deletion in mature brown adipocytes (hereafter Rictor-iKO MBAs ). Briefly, UBC-Cre;Rictor L/L brown adipocyte precursors were treated with 4-hydroxytamoxifen (4-OHT) at day 8 of differentiation, after Pparγ2 induction, to generate Rictor-iKO MBAs [ Supplementary Fig. 6A, B]. Shortly after 4-OHT treatment (corresponding to differentiation day 12), mature brown adipocytes are depleted of Rictor and have attenuated pAKT S473 and pACLY S455 [Fig. 6a]. This coincides with a 19% reduction in acetyl-CoA levels [ Fig. 6b] and reduced H3K27Ac [ Supplementary Fig. 5I]. At this time point, ACLY, ACC, and FASN protein levels are largely unaffected [ Fig. 6a], while Acly, Acc, and Fasn mRNA show a 28.0%, 28.8% and 48.3% decrease, respectively [ Fig. 6c]. We also observe a 45.3% decrease in Chrebpβ mRNA expression, while Chrebpα and Glut4 expression are normal [ Fig. 6c]. In contrast, more prolonged Rictor loss (differentiation day 14) results in a 62.2% decrease in Glut4 mRNA expression and depletion of ACLY, ACC, and FASN protein [Fig. 6a,   its expression in vitro in mature brown adipocytes [Fig. 6a]. These data are consistent with Rictor loss acutely impairing ACLY activity, histone acetylation, and lipogenic gene expression, which precedes downregulation of DNL proteins and glucose transporter expression.
also moderately rescues gene expression [ Fig. 6d], while in contrast, vector alone has little effect and ACLY-S455A has a mild negative effect [Fig. 6d]. These data are consistent with ACLY being a critical mediator between mTORC2, ChREBP, and glucolipogenic gene expression in mature brown adipocytes. mTORC2 and ACLY loss affect the DNL pathway in vivo. Finally, we explored the in vivo physiological relevance of these findings by comparing the effects of deleting Rictor l/l or Acly l/l alleles in BAT with Ucp1-Cre. As we recently reported 13 , the BAT of Ucp1-Cre;Rictor l/l mice have dramatically attenuated Chrebpβ, Acly, Acaca, and Fasn mRNA [ Fig. 7a] and protein expression [ Fig. 7b] and small lipid droplets [Fig. 7c]. Rictor-deficient BAT also has reduced H3K27 acetylation [ Supplementary Fig. 5J]. However, despite the low Acly expression, which is comparable to deleting Acly itself, Acss2/ACSS2 levels remain largely unchanged [ Fig. 7a, b]. This is consistent with mTORC2 being necessary to invoke the ACSS2-driven glucose-to-acetate metabolic switch. In contrast Ucp1-Cre;Acly l/l mice expresses higher levels of Acaca/ ACC and Fasn/FASN mRNA and protein [ Fig. 7a, b] and have larger lipid droplets [ Fig. 7c]. Increased BAT lipid droplet size is also evident in the BAT of Adiponectin-Cre;Acly l/l mice [Supplementary Fig. 7]. This phenotypic divergence correlates with Acss2/ACSS2 expression, which is induced in Acly but not in Rictor-deficient BAT [ Fig. 7a, b]. Interestingly, Chrebpβ expression decreases in the BAT of both models [ Fig. 7a] indicating that Chrebpβ may not be driving Acss2 expression in Acly-deficient BAT. This is also consistent with ChREBP functioning downstream of both mTORC2 and ACLY 12,35 .
Thus, while the effects of deleting Rictor or Acly during brown adipocyte differentiation are largely overlapping, their loss in mature brown adipocytes has opposite effects on ACSS2 expression, downstream DNL pathway expression, and lipid droplet content. This is despite the fact that deleting Rictor or Acly decreases Acly/ACLY expression to nearly the same extent. These data are consistent with mTORC2 additionally regulating ACSS2 in a physiologically relevant context and supports an overarching model in which brown adipocyte mTORC2 regulates glucose-and acetate-derived acetyl-CoA synthesis [ Fig. 7d].

Discussion
Prior studies implicate adipocyte mTORC2 in controlling insulinstimulated glucose uptake and DNL (gluco-lipogenesis). However, the mechanism has been enigmatic because most AKT substrates examined appear to be phosphorylated normally when Rictor is deleted [11][12][13] . Solving this mystery is important because a similar state of adipose tissue "selective insulin resistance" in which insulin-stimulated glucose uptake, but not AKT activity per se, is defective, occurs in several models of adipocyte insulin resistance 36,37 . Moreover, DNL in human BAT and WAT positively correlates with thermogenesis and insulin sensitivity, respectively 15,[38][39][40] . In this study, we uncover a previously unappreciated selectivity in mTORC2-dependent AKT signaling toward ACLY that provides a plausible mechanistic underpinning for why mTORC2 loss impairs brown adipocyte glucosedependent lipogenesis and differentiation.
One limitation of this study is that it does not address why ACLY S455 is different from other AKT substrates. Although our inducible KO system results in rapid loss of Rictor and AKT HM phosphorylation, it remains possible that a compensatory mechanism stabilizes T308 phosphorylation maintaining AKT activity toward most substrates. A selective mTORC2 inhibitor (i.e., which does not inhibit mTORC1) would be valuable in determining the acute effects of mTORC2; however, such a molecule is not yet available. Nevertheless, why would ACLY phosphorylation differ from other AKT substrates in its mTORC2 dependency? One possibility is that intrinsic differences in the ACLY S455 motif may demand higher AKT activity, or a different biochemical conformation dependent on HM phosphorylation, to be efficiently phosphorylated 41,42 . Aligning mTORC2 dependent and independent phosphosites identified in this study to the AKT consensus motif indicates that more dependent sites are less conserved [ Supplementary Fig. 2D]. Alternatively, the mTORC2 dependency of ACLY phosphorylation could relate to its localization and/or temporal dynamics of phosphorylation. For example, ACLY phosphorylation may occur distal to the plasma membrane and/or membrane PIP3 and therefore may require HM phosphorylation to recruit PDK1, which is how other AGC family kinases lacking PH domains are regulated 4 . Indeed, evidence suggests that AKT is regulated by both PH-domain-dependent and -independent mechanisms in cancer cells 43 . Proteomics and western blotting data also indicate that mTORC2 loss decreases ACLY phosphorylation during serum deprivation suggesting that mTORC2 might regulate acetyl-CoA synthesis in multiple nutritional states. Another possibility is that mTORC2 may suppress an ACLY phosphatase.
Two studies comparing temporal dynamics of the insulinstimulated phosphoproteome and metabolome in 3T3-L1 adipocytes show that total citrate levels acutely decrease, and acetyl-CoA levels increase, roughly when ACLY S455 phosphorylation is stimulated and that this occurs prior to maximum glucose transport 44,45 . This suggests that mTORC2-dependent ACLY stimulation may be an early insulin action that mobilizes metabolites downstream of glucose transport, glycolysis, and the TCA cycle to both relieve citrate's allosteric inhibitory effect on PFK1 and prime gluco-lipogenesis. Data from this study further suggest that mTORC2-dependent ACLY phosphorylation additionally creates an overflow of acetyl-CoA that is siphoned toward histone acetylation, thereby providing an insulin-coupled secondary epigenetic signal that is a direct readout of systemic nutrient (glucose) availability. We suggest a model that, in brown adipocyte precursors, this pathway provides a license for differentiation, while in mature brown adipocytes, it potentiates glucose uptake, glycolysis, and lipid synthesis by upregulating and/or maintaining ChREBP activity and the gluco-lipogenic transcriptional program [ Fig. 7d]. Thus mTORC2-simulated acetyl-CoA synthesis may provide a key missing link between systemic glucose availability and setting the genetic program that determines the glucose-and lipid-handling capacity of brown adipocytes. mTORC2 also promotes Chrebpβ expression, which drives gluco-lipogenic gene expression including that of Acly 12,13 . Interestingly, Chrebpβ expression is also ACLY dependent in white adipocytes 31 consistent with ChREBPα/β functioning downstream of mTORC2/ACLY in a positive feedback loop. Thus ChREBP could also be the key target of mTORC2/ACLY in gluco-lipogenic gene regulation. In fact, hepatocyte ChREBP transcriptional activity is reportedly regulated by acetylation and O-GlcNacylation, which both require acetyl-CoA precursors 46,47 . The most likely scenario is that histone acetylation and ChREBP activity are interconnected, and further resolving how ChREBP is regulated by mTORC2 and ACLY will be important. The mTORC2/ACLY pathway might additionally regulate other posttranslational modifications or allosteric regulatory roles linked to DNL metabolites, such as palmitoylation, which is implicated in Glut4 trafficking 48,49 . The potential role of DNL as a source of signaling metabolites in adipocytes has been underappreciated, but it is tempting to speculate that defects in their second messenger functions could be key steps in the early onset of insulin resistance.
That mTORC2 additionally regulates acetyl-CoA synthesis from acetate indicates a broader role of mTORC2 in controlling  Fig. 7 In vivo, mTORC2 and ACLY both promote ChREBPβ expression, but their loss has opposite effects on ACC and FASN expression that correlates with differential ACSS2 regulation. a qRT-PCR analysis of the indicated genes in brown fat tissue isolated from control (Rictor l/l and Acly l/l ) or UCP1-Cre-Rictor l/l and UCP1-Cre-Acly l/l mice (n = 6 per group). Bar graphs represent mean ± SEM. *p < 0.05, **p < 0.01, and ***p < 0.001. Statistical significance was calculated by using two-way ANOVA with Sidak's test multiple comparisons (a). b Corresponding western blots for a. Black arrows indicate the ACC1 and ACC2 isoforms. c Hematoxylin and eosin stains showing brown fat morphology for the indicated genotypes. Scale bar represents 200 μm. d Our results support the following model of mTORC2 action in brown adipocytes: (1) mTORC2-dependent AKT phosphorylation is distinctly important for ACLY phosphorylation, while for many other AKT substrates, mTORC2 may normally facilitate their phosphorylation, but it is dispensable (indicated by "P" within an unshaded broken circle). This is indicated by the fact that phosphorylation of many AKT substrates is unimpaired by Rictor deletion while ACLY phosphorylation is reduced. (2) mTORC2/AKT-dependent ACLY phosphorylation promotes acetyl-CoA synthesis and primes de novo lipogenesis downstream of glucose uptake and glycolysis. (3) The increased flux to acetyl-CoA additionally stimulates histone acetylation. In brown adipocyte precursors, this coincides with pparγ induction to drive differentiation; in mature brown adipocytes, this coincides with ChREBPβ activity to increase expression of gluco-lipogenic genes. (4) Gluco-lipogenic gene expression then provides a positive feedback effect on glucose transport and de novo lipogenesis. (5) mTORC2 may also stimulate acetate metabolism to acetyl-CoA by regulating the expression and/or activity of ACSS2. Notably, from these data we cannot distinguish between cytoplasmic and nuclear pools of acetyl-CoA.
nuclear-cytoplasmic acetyl-CoA synthesis. The role of acetate as a metabolic substrate in adipocytes is not well understood. However, the differential effects on ACSS2 expression caused by Rictor loss vs Acly loss could explain why Adiponectin-Cre;Rictor l/l mice have a more severe metabolic phenotype compared to Adiponectin-Cre;Acly l/l mice 12,34 . In conclusion, the identification of ACLY and ACSS2 as key targets of mTORC2 action in brown adipocytes fills an important gap in our understanding of this more mysterious mTOR complex. These findings may be relevant to understanding mTORC2 action in other tissues and in the pathogenesis of type-2 diabetes.

Methods
Mice. C57BL/6J mice (JAX stock 000664) were obtained from Jackson Laboratory. Ucp1-Cre (JAX stock 024670) and Adiponectin-Cre (JAX stock 010803) were kindly provided by Dr. Evan Rosen. Akt1 l/l , Akt2 l/l , and Akt3-KO mice were kindly provided by Dr. Birnbaum MJ. Ubc-Cre ERT2 ; Rictor l/l mice and Acly l/l mice were generated by crossing UBC-Cre ERT2 with Rictor l/l or Acly l/l mice 11,34 . UBC-Cre ERT2 ; Akt1 l/l , Akt2 l/l , Akt3-KO were generated by crossing UBC-Cre ERT2 ; Akt1 l/l with UBC-Cre ERT2 ; Akt2 l/l until UBC-Cre ERT2 ; Akt1 l/l , Akt2 l/l was obtained and then once more to Akt3-KO mice until we obtained the final Ubc-Cre ERT2 ; Akt1 l/l , Akt2 l/l , Akt3-KO genotype. Mice were housed in the UMMS Animal Medicine facilities in a room set at 22°C in 45% humidity under daily 12 light/dark cycles. For all in vivo studies, we used 8-week-old Ucp1-Cre-Rictor l/l and Ucp1-Cre-Acly l/l mice. We have complied with all relevant ethical regulations for animal testing and research, and all animal studies have been approved by the Umass Medical School Committee on Animal Care (IACUC).
Cultured cells. Brown preadipocytes were isolated from Ubc-Cre ERT2 ; Rictor l/l , Ubc-Cre ERT2 ; Acly l/l , or Ubc-Cre ERT2 ; Akt1 l/l , Akt2 l/l , Akt3 −/− P1 neonates and immortalized with pBabe-SV40 Large T according to a standard protocol 50  Western blot analysis. Cells were homogenized in a Triton-X lysis buffer containing 150 mM NaCl, 50 mM Tris pH 7.5, 0.5% deoxycholate, 1% Triton X-100, and 0.1% sodium dodecyl sulfate (SDS); a protease inhibitor cocktail (Biotool Cat. Number); and a phosphatase inhibitor cocktail (Sigma Cat. Number P5726). For in vivo analysis, each tissue was collected and frozen down immediately in liquid nitrogen and then stored at −80°C for subsequent lysis. Tissues were homogenized using a TissueLyser (Qiagen) in RIPA lysis buffer containing 150 mM NaCl, 1% NP-40, 0.5% deoxycholate, 0.1% SDS, and 50 mM Tris, pH 7.5 including a protease (Biotool Cat. Number B14011) and phosphatase inhibitor cocktail (Sigma Cat. Number P5726). An equal amount of total protein was loaded into either 10% acrylamide/bis-acrylamide gels for general analysis or 16% acrylamide/bis-acrylamide gels for histone analysis and transferred to polyvinylidene fluoride membranes. Ponceau stain was performed to check for equal transfer and loading. Membranes were incubated in either 5% milk/phosphate-buffered saline with Tween detergent (PBST) or 5% bovine serum albumin (BSA)/PBST for 1 h and with primary antibodies (1:1000 dilution) in 5% milk/PBST or 5% BSA/PBST overnight, washed 3 times with PBST, and then incubated with the correspondent secondary antibody for 2 h. Tubulin was used as an additional protein level control where indicated. HRP-conjugated secondary antibodies (1:10,000 dilution) were given for 1 h. Western blots were developed by enhanced chemiluminescence (PerkinElmer) and detected by X-ray films.
Chromatin extraction. Chromatin extraction was performed prior to the western blot analysis 51 . Briefly, cells were collected at day 2 of the differentiation assay, before Pparγ2 induction, resuspended in lysis buffer (10 mM Hepes pH 7.4, 10 mM KCl, 0.05% NP-40, and a protease (Biotool Cat. Number B14011) and phosphatase inhibitor cocktail (Sigma Cat. Number P5726)) and incubated on ice for 20 min. Then the samples were centrifuged at 16,000 × g for 10 min after which the nuclear pellet was resuspended in HCl 0.2 N and incubated on ice for an additional 20 min. Samples were centrifuged at 16,000 × g for 10 min, and the supernatant containing the chromatin was neutralized in Tris-HCl pH 8 1 M.
Glucose uptake. Cells were preincubated for 3 h in a medium without glucose (KRH medium (Gibco plus 0.5% BSA + 2 mM pyruvate). Deoxy-D-glucose 2-[1,2-3 H(N)] was then added, and incubation was continued for an additional 10 min. The medium was then removed and cells were washed three times with KRH medium to terminate the assay. Cells were then lysed in 1% triton, mixed with scintillation buffer, and the uptake of 3 H glucose was quantified in counts per minute (cpm) using a scintillation counter. The cpm values were normalized to the protein concentration of each sample.
Phosphoproteomics. Control and Rictor-iKO cells were serum deprived for 12 h (unstimulated samples) and then some were treated with 150 nM insulin for 15 min (stimulated). All cells were washed 3 times with cold PBS, snap-frozen in liquid nitrogen, and thawed in a lysis buffer (8 M urea, 50 mM Tris pH 8.2, 75 mM NaCl, 50 mM β-glycerophosphate, 1 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 50 mM sodium fluoride, and 1× EDTA-free protease inhibitor cocktail (Roche)). Each well of a 6-well plate was scraped in 125 μL of ice-cold lysis buffer, and 2 wells were pooled for each replicate. Lysates were probe-sonicated two times, and clarified by centrifugation at 16,000 × g. Protein content was measured by a BCA assay (Pierce). Samples were reduced for 45 min at 55°C using 5 mM dithiothreitol, alkylated with 15 mM iodoacetamide for 30 min in the dark at room temperature, and quenched with an additional 5 mM of dithiothreitol at room temperature for 15 min. Urea was diluted to 1.5 M using 50 mM pH 8.2 Tris, and protein was digested for 16 h using 5 μg/mL trypsin at 37°C. Digested peptides were de-salted using 50 mg Sep-Pak cartridges (Waters) and dried by vacuum centrifugation. Phosphopeptides were enriched using PureCube Fe-NTA magnetic beads (Cube Biotech) in a KingFisher magnetic bead processor (Thermo). Dried phosphopeptide samples were prepared for analysis by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) by dissolution in 4% formic acid and 3% acetonitrile.
To screen for mTORC2-dependent phosphorylation, we designed a targeted, parallel reaction monitoring (PRM) mass spectrometry method that was informed by manual literature curation and prior preliminary experiments. To design the list, we curated a preliminary list of 228 phosphopeptides representing previously characterized AKT substrates, regulatory phosphorylation sites on AGC kinases, phosphorylation sites on proteins with known regulatory function in insulin signaling and/or brown fat, and phosphorylation sites with potential mTORC2 dependence according to preliminary, unpublished data. According to a previously reported method 52 , we referenced an in-house database of mouse phosphoproteome data gathered from several different tissues using datadependent acquisition (DDA) to determine the best phosphopeptide(s) with which to represent each phosphorylation site. Specifically, for each phosphorylation site, we determined which charge state and peptide cleavage form was the most frequently observed in the DDA data. Importantly, this database also enabled accurate prediction of the chromatographic retention time. In some cases, phosphorylation sites were represented by multiple peptides in the preliminary target list because it was not obvious which peptide form to choose. To refine the preliminary target list, three separate PRM assays were used to measure these phosphopeptides in insulin-stimulated WT cells with wide (+/−2.5 min) windows for retention time prediction. Peptides that were not detected or contained complete overlap of phosphorylation sites with another detected peptide were removed from the list, and all retention time predictions were re-calibrated. This final target list was then used for PRM analysis of all the samples with narrow (+/−1.5 min) windows for retention time prediction.
Targeted phosphoproteomic analysis of preadipocyte phosphorylation sites was performed using LC-MS/MS. To normalize for injection amount during LC-MS/MS analysis, we measured all samples using DDA and identified 12 phosphopeptides that correlated strongly with the total MS signal and spanned most of the retention time space. DDA analysis was performed on a Thermo QExactive mass spectrometer connected to a Thermo Easy-nLC II. Chromatographic separation was performed on a 30-cm length column with a 100-μm internal diameter, packed with Reprosil 1.9 μm C18 beads (Dr. Maisch GmbH). The LC method consisted of a 49-min linear gradient from 9% to 25% acetonitrile in 0.15% formic acid. Data were acquired throughout a 70-min total time. Full MS scans were acquired from 300 to 1500 m/z at 70,000 resolution with an AGC target of 3 × 10 6 and a maximum injection time of 240 ms. The top 20 most abundant precursors per MS scan were isolated with a 2-m/z window, an AGC target of 2 × 10 5 , a maximum injection time of 120 ms, and 40 s dynamic exclusion. Isolated precursors were fragmented by high-energy collision-induced dissociation at a normalized collision energy of 27%, and MS/ MS were acquired at 17,500 resolution. Targeted LC-MS/MS using PRM was performed on the same instrument set-up as for DDA. The instrument methods were mostly the same as for the DDA analysis, except that MS/MS spectra were acquired according to the predefined schedule of precursors representing the phosphopeptide target list, as described above. Every 20 scans, a full MS scan was acquired as described above for DDA. All MS and MS/MS spectra for DDA and PRM experiments were collected in centroid mode. PRM data were analyzed using Skyline 53 , and all chromatographic peaks and phosphorylation site localizations were manually inspected. For additional verification of phosphorylation site localization, PRM data were searched using Comet (v2015.02.5) 54 and filtered at 1% FDR using Percolator 55 . Each MS/MS spectrum was scored for localization using Ascore 56 . Ascore values representing confidence of site localization were plotted vs retention time and manually compared to the fragment chromatographic elution peaks in Skyline. Phosphorylation sites were stratified into three categories of mTORC2 sensitivity by calculating ANOVA statistics to determine the significant effects of Rictor loss and correcting to q-values using the Benjamini-Hochberg method. Class I sites had q ≤ 0.05 and at least a 2-fold difference between control and Rictor-iKO PBA . Class III sites had q ≤ 0.05 and at least a 1.33-fold difference between control and Rictor-iKO PBA . All the phosphoproteomic data generated during this study are available and were deposited in Massive.Ucsd.edu [https://massive. ucsd.edu/ProteoSAFe/dataset.jsp?task=2335fcac3c184be69e7efe068387f771].
Metabolite analysis. Cells were incubated in serum-free media for 12 h and then incubated for 3 h with fresh DMEM medium. Samples were prepared by performing an 80% methanol extraction and stored at −80°C until shipping. Polar metabolite profiling methods were developed at the Whitehead Institute for Biomedical Research Metabolomics core facility, where the measurements were performed. Reference standards of each metabolite were used to determine chromatographic retention times and MS multiple reaction monitoring transitions, declustering potentials, and collision energies. All metabolites were normalized to the protein concentration of each sample.
Chromatin immunoprecipitation analysis. Cells were cross-linked with 1% paraformaldehyde for 15 min at room temperature. The reaction was quenched for 5 min at room temperature by adding 0.125 M glycine. After three washes with 1× PBS, cells were lysed with lysis buffer (1% SDS, 20 mM EDTA pH 8, 50 mM Tris-HCl, pH 8) supplemented with protease and deacetylase (TSA) inhibitors. Lysates were sonicated on ice using a Diagenode Bioruptor TM UCD200 sonicator (for 15 min total at high setting in intervals of 45 s on-20 s off and kept on ice during the whole sonication process). Size of fragments obtained (between 200 and 1200 bp) was confirmed by electrophoresis. Soluble chromatin was collected after centrifugation at 16,000 × g at 4°C for 10 min. Soluble chromatin (1%) was kept as input control. Soluble chromatin was precleared with 2.5 μg/mL of protein-A-Sepharose at 50% at 4°C in rotation for 1 h and beads were spin down at 5900 × g for 3 min at 4°C. After centrifugation, supernatants were collected and specific antibodies (for total H3: Cell Signaling #9715; and for K27H3Ac: Abcam ab4729) were added. Mixtures were incubated overnight at 4°C for 8 h in rotation and then incubated for 2 h at 4°C in rotation with protein-A-Sepharose at 50% (Roche). Beads were collected and washed sequentially at 4°C for 10 min with wash buffer A (140 mM NaCl, 0.1% SDS, 0.1% deoxycholate, 1% Triton X-100, 1 mM EDTA, and 50 mM Hepes (pH 7.9)), wash buffer B (500 mM NaCl, 0.1% SDS, 1% Triton X-100, 0.1% deoxycholate, 1 mM EDTA, and 50 mM Hepes (pH 7.9.1)), LiCl buffer (20 mM Tris-HCl (pH 8), 1 mM EDTA, 250 mM LiCl, 0.5% deoxycholate, and 0.5% NP-40), and TE buffer (10 mM Tris-HCL (pH*), 1 mM EDTA). Immunoprecipitates were eluted two times with elution buffer elution buffer (50 mM Tris-HCl (pH 8), 1 mM EDTA, and 1% SDS). Reversion of cross-linking was performed overnight by heating samples and input controls at 65°C. Samples were then treated with 0.2 mg/mL RNAse A (Qiagen) and incubated for 30 min at 37°C followed by addition of 1 μL of (20 mg/mL) Proteinase K (Promega), and samples were incubated at 45°C for 1 h. DNA was then purified using the QIAquick Spin Kit (Qiagen). Real-time reverse transcriptase PCR was performed with primers listed in Supplementary Table 1.
Acetyl-CoA measurements and isotopic tracer analysis. Cells were fasted for 12 h and then incubated in DMEM without glucose plus 5 mM of normal or 3 C 6glucose and 100 μM normal or 13 C 2 acetate at 37°C for 1 h. Samples were harvested in cold PBS, centrifuged 1000 × g, 5 min, and the pellet was quenched in 10% (w/v) ice cold trichloroacetic acid in water, and then frozen at −80°C until extraction and analysis on a QExactive Plus (Thermo) mass spectrometer coupled to an Ultimate 3000 high-performance LC. For quantitation, the abundance of acyl-CoA species was determined by interpolation to a linear standard curve. For isotopic tracer analysis, isotopic enrichment was calculated to compensate for the non-linearity of isotopic enrichment using the FluxFix calculator 57 .
De novo lipogenesis assay. Cells were incubated for 3 days with DMEM in which 0.01% of the total glucose concentration of the medium was comprised of D-[U-14 C]-glucose. Chloroform extraction was performed, and labeled lipids were measured using a scintillation counter. Each sample was normalized to total protein concentration 58 .
Lipid staining. Cells were fixed with 4% paraformaldehyde, washed with distilled water, and treated with 100% propylene glycol twice for 5 min followed by a 15min incubation with Oil Red O solution. Then cells were washed with 85% propylene glycol for 3 min and rinsed twice in distilled water. Oil Red O stain was quantified with the ImageJ software (http://imagej.net/Welcome) and in each case shows the average quantification of four images representing each quadrant of the well.
Gene expression analysis. Cells or tissues were lysed with Qiazol (Invitrogen) and total RNA was isolated with the RNeasy Kit (Invitrogen). Equal amounts of RNA (2 μg) were retro-transcribed to cDNA using a high-capacity cDNA Reverse Transcription Kit (#4368813, Applied Biosystems). Quantitative real-time PCR was performed in 10 μL reactions using a StepOnePlus real-time PCR machine from Applied Biosystems using SYBR Green PCR master mix (#4309156, Applied Biosystems) according to the manufacturer's instructions. Relative mRNA expression was determined by the ΔΔCt method and Tbp expression was used as a normalization gene in all conventional PCR with reverse transcription experiments. Primer information is listed in Supplementary Table 1.