Alterations in Lipid Signaling Underlie Lipodystrophy Secondary to AGPAT2 Mutations

The institutional review boards of the University of Michigan approved the study protocol, and all subjects gave written informed consent. A percutaneous muscle biopsy was obtained from the lateral portion of the vastus lateralis. The biopsy (∼100 mg) was minced and digested in collagenase-dispase (10 and 1 mg/mL, respectively) for 30 min. Nondigested tissue was allowed to sediment, and the supernatant was filtered (70 μm). The supernatant was centrifuged and preplated on type I collagen–coated dishes for 4 h and transferred to collagen-coated dishes (14).

MDMCs were maintained in an undifferentiated state in Ham-F10 media/20% FBS/0.5% chicken embryo with antibiotic and antifungals. 3T3-L1 preadipocytes were propagated and maintained in Dulbecco’s modified Eagle’s medium containing 10% (volume for volume) FBS with antibiotic and antifungals. Differentiation of 3T3-L1 cells was as previously described (15). To induce differentiation of human MDMCs, 2-day postconfluent cells were fed Dulbecco’s modified Eagle’s medium with insulin (I), dexamethasone (D), and 3-isobutyl-1-methylxanthine (M) and 10% FBS. On day 3, cells were incubated in I media for 2 days and then in IDM for 2 days. This process was repeated for three cycles, until day 21. Oil Red O staining was performed as previously described (15).

3T3-L1 cells were transfected with 20 nmol/L AGPAT1 or AGPAT2 small interfering RNA (siRNA) SMARTpools (Dharmacon, Lafayette, CO) or siCONTROL nontargeting siRNA using Dharmafect 3 transfection reagent. For cells undergoing differentiation, transfection was performed on day −2 of differentiation. Relative AGPAT mRNA levels were determined after 48 h. For overexpression experiments, cells were infected with retrovirus expressing either green fluorescent protein (GFP) or GFP-AGPATs and selected with G418 for 1 week. For transient expression, cells were transfected with V5-tagged AGPAT1, AGPAT2, or empty vector.

cDNA was synthesized using random hexamers (Promega Biosciences) from 1 μg of cellular RNA. Expression of the specific genes was assessed by quantitative reverse transcriptase PCR, as previously described (15), using 18s mRNA to normalize expression.

MDSCs and 3T3-L1 cells were differentiated as previously described, harvested at defined intervals, fixed with 90% ethanol, and incubated with a staining solution containing RNase A (30 μg/mL) and propidium iodide (50 μg/mL) in PBS. Cellular DNA content was analyzed by flow cytometry using a Becton Dickinson laser-based flow cytometer. At least 1,000 events were used for each analysis.

Cells were lysed in a solubilizing buffer and processed for Western blot analysis as previously described.

Cells were fixed in 4% formaldehyde in PBS for 15 min at room temperature, permeabilized, and blocked with 0.03% Triton X-100 and 10% normal goat serum for 30 min at room temperature, followed by incubation with specific antibodies in 10% normal goat serum overnight at 4°C and developed with secondary florescent-conjugated antibodies.

For phosphatidylinositol 4-phosphate (PIP) analyses, lipids were extracted from cultured cells (100-mm plate) and eluted as previously described (16,17).

PIP extracts were separated on an XBridge C8 column (Waters, Milford, MA). Mobile phase A was methanol/water/70% ethylamine (50:50:0.13). Mobile phase B was 2-propanol/70% ethylamine (100:0.13). The gradient was 95%/5% (A/B) to 10%/90% (A/B) over 15 min, after which it was held for 1 min, before returning to 95%/5% (A/B) over 1 min, followed by holding for 10 min for reequilibration. The flow rate was 30 μL/min at 25°C using an LTQ Orbitrap mass spectrometer (Thermo-Fisher Scientific, Waltham, MA). Ion spray voltage was set to −4.0 kV in negative ion mode. Phospholipids were measured by a full scan of parent ion (MS1) within charge/mass ratio (m/z) 380−1,500 in negative ion mode using an Orbitrap FTMS (Fourier transform mass spectrometry) analyzer with a resolution of 60,000. Mass accuracy was within 3 ppm. MS/MS (MS2) measurements at a collision energy setting of 35% were performed with an LTQ-IT (ion trap) analyzer when needed. PIP₁- and PIP₂-specific fragments (m/z 321 and 401, respectively) observed in MS2 were used for identification.

Cells were extracted with methanol:chloroform (2:1) containing 1 nmol/L ¹³C₁₆ 16:0 LPA (18). The organic layer was dried under nitrogen and redissolved in 200 μL of 0.1 mol/L ammonium acetate in methanol. Cyclic phosphatidic acid (CPA) and LPA were analyzed by LC-ESI-MS/MS (liquid chromatography coupled to tandem MS with electrospray ionization) using a modified protocol of Shan et al. (19). High-performance liquid chromatography separation was performed using an XBridge C18 column (50 × 2.1 mm, 3 μm; Waters). Mobile phase A consisted of 5 mmol/L ammonium acetate in water, adjusted to pH 9.9 with ammonium hydroxide. Mobile phase B was 60% acetonitrile and 40% isopropanol with a linear increase from 20% B to 99.5% B over 15 min, followed by a 5-min hold at 99.5% B. Online tandem MS analysis was performed using negative ion electrospray. Optimum multiple reaction monitoring (MRM) parameters were determined by flow-injection analysis of authentic standards and are listed in Supplementary Table 1. The parent ion for LPA and CPA species is the deprotonated molecular anion (M-H). Daughter ions for all LPA species were a common fragment at m/z 153.1. CPA species gave daughter ions, which were characterized by a common loss of 136. For lysophosphatidylcholine (LPC) analysis, samples were extracted by a mixture of methanol:chloroform and analyzed as previously described (20) using an Ascentis C18 column 15 cm × 2.1 mmol/L × 3 μm (Supelco) and a TOF-MS (Agilent) operated in negative ion mode. LPC 16:0 was detected as a formic acid adduct with accurate mass of 540.3306 (<1 ppm error). Thin layer chromatography for PA and LPA analysis was performed as previously described (18).

NIH3T3 cells (5 × 10⁶) were electroporated with vectors containing PPARγ1, AGPAT2, RXRα, and FATP-luciferase (1 μg each) and 1 μg of either empty vector, Agpat1, or Agpat2 expression plasmids. After plating, the cells were treated with the vehicle or rosiglitazone for 18 h. Luciferase activity was normalized to β-gal.

Data are expressed as mean ± one SEM. Statistical significance was determined using a noncorrected two-tailed Student t test, unpaired assuming equal variance or ANOVA, as appropriate. A P value of <0.05 was considered significant.

Expanded single-cell MDMC isolates from vastus lateralis muscle biopsies from two control individuals were differentiated with either adipogenic (MDI) or myogenic (2% horse serum) media to determine their lineage potential (Supplementary Fig. 1). About 30% of the cultures showed bipotential differentiation, as evidenced by the formation of multinucleated cells staining for the myogenic marker Myf20 or the accumulation of lipid droplets positive for perilipin. Approximately 20% of the cells underwent adipogenesis only, 40–50% showed only myogenic differentiation, and a small percentage of the single-cell cultures could not be differentiated. Thus, individual human MDMCs undergo both myogenic and adipocytic differentiation.

MDMC cultures were obtained from two CGL subjects harboring defined AGPAT2 mutations (subjects 1 and 2) (Supplementary Table 2) (3). MDMC cultures from both control and CGL subjects were able to undergo myogenesis after exposure to 2% horse serum, demonstrated by the appearance of elongated, multinucleate cells (Fig. 1A) and increased expression of myosin heavy chain mRNA (Fig. 1B). In contrast to controls, CGL MDMC cultures exposed to MDI showed no accumulation of lipid droplets (Fig. 1C). C/EBPβ and C/EBPδ mRNA rose in both control and CGL-derived culture (Fig. 1D) after 1 day of MDI treatment, and by 3 and 21 days, control culture showed induction of C/EBPα and PPARγ whereas CGL cells showed no (threshold cycle > 40) PPARγ or C/EBPα expression (Fig. 1D), indicating an ability to initiate adipogenesis while displaying a block in terminal differentiation.

The adipogenic defect in MDMCs from CGL patients was overcome by retroviral expression of GFP-tagged AGPAT2, as demonstrated by accumulation of lipid droplets in GFP-expressing cells (Fig. 1E) and increases in PPARγ and adiponectin mRNA expression (Fig. 1F) after MDI exposure, consistent with the induction of the adipogenic program after AGPAT2 rescue. Unlike AGPAT2, infection of cells with GFP alone or GFP-tagged AGPAT1 did not result in the formation of lipid droplets (Supplementary Fig. 2).

Due to the limited replicative life of MDMC cultures, we assessed the effect of manipulation of AGPAT2 in 3T3-L1 preadipocytes. Using a pool of siRNA for AGPAT1 or AGPAT2 or a nontargeting siRNA we were able to decrease mRNA expression and protein levels of AGPATs (Fig. 2A). Knockdown of AGPAT1 had almost no effect on accumulation of lipid in 3T3-L1 cells, whereas AGPAT2 knockdown significantly impaired adipogenesis, as assessed by Oil Red O staining (Fig. 2B) and decreased expression of C/EBPα, PPARγ, and adiponectin mRNA levels (Fig. 2C). Consistent with a role in adipogenesis, AGPAT2 mRNA and protein levels increase markedly during differentiation of both 3T3-L1 cells and MDMCs (Supplementary Fig. 3). When 3T3-L1 preadipocytes were transfected with either V5-tagged AGPAT2 or empty vector, we found that overexpression of AGPAT2 increased protein levels of PPARγ when compared with cells treated with an empty vector (Fig. 2D). Thus, modulation of AGPAT2 levels in 3T3-L1 preadipocytes parallels the phenotype of CGL-derived MDMCs.

We noted that during differentiation of MDMCs from CGL patients and in 3T3-L1 cells after AGPAT2 knockdown, there was an increase in cells lifting from the plates, suggesting cell death. AGPAT2, but not AGPAT1, knockdown resulted in a statistically significant reduction in the number of cells in the first 24 h of differentiation (Fig. 3A). This correlated with an increase in cell death, as demonstrated by propidium iodide staining (Fig. 3B). A similar effect was found in MDMCs from CGL patients assessed by flow cytometry. Prior to addition of the adipogenic media, no significant difference in annexin staining was seen between the control and CGL subjects (Fig. 3C). After 3 days of differentiation, the control MDMCs showed between 9.4 and 10.5% cell death, and the CGL cells from subjects 1 and 2 had 24 and 18.3%, respectively, annexin-positive cells (Fig. 3C), indicating significant cell compromise in the absence of AGPAT2.

A key survival signal with a central role in adipogenesis is the PI3K/Akt pathway. Previous studies with small-molecule AGPAT2 inhibitors showed a reduction in Akt and Erk signaling in cancer cells (21–24). In adipocytes, insulin stimulated Akt phosphorylation in MDMC cultures isolated from control individuals but was reduced in CGL-derived MDMC cultures, whereas Erk1/2 phosphorylation was preserved (Fig. 4A). Similarly, insulin-stimulated 3T3-L1 preadipocytes transfected with AGPAT2 siRNA showed significant attenuation of Akt phosphorylation as well as decreased phosphorylation of its downstream target p70-S6K1 (Fig. 4B). We corroborated our findings using two different siRNA sequences and both demonstrated a decrease in pSer473-Akt with good correlation between degree of AGPAT2 knockdown and downregulation of Akt phosphorylation (Supplementary Fig. 4). Unlike the Akt pathway, and consistent with the findings in human MDMCs, Erk1/2 phosphorylation was modestly upregulated in preadipocytes after AGPAT2 siRNA treatment (Fig. 4B).

To investigate gain of function by AGPAT2 overexpression, 3T3-L1 preadipocyte cells were transiently transfected with empty vector, V5-tagged AGPAT1, or V5-tagged AGPAT2. After 36 h, the cells were starved overnight and stimulated with insulin. AGPAT2 overexpression increased p473-Akt and p308-Akt phosphorylation, even in the absence of insulin stimulation, which correlated with an increase in the phosphorylation of p70-S6K1 (Fig. 4C). In contrast, AGPAT1 did not upregulate Akt phosphorylation but did increase the phosphorylation of p70-S6K1 (Fig. 4C). The efficacy of AGPAT2 to increase Akt phosphorylation appeared to be augmented by the addition of a mixture of free fatty acids, with increases in Akt phosphorylation in the presence of low insulin concentrations (50 nmol/L) and increased levels with higher concentrations (100 nmol/L) (Supplementary Fig. 5), suggesting an interactive role of AGPAT2 and fatty acids in enhancing preadipocyte signaling.

Consistent with its ability to increase Akt signaling, treatment of 3T3-L1 cells with insulin caused the nuclear exclusion of FoxO1 (Fig. 4D, a and b) (25). Cells overexpressing AGPAT2 demonstrated nuclear exclusion of FoxO1, even in the absence of insulin (Fig. 4D, c–e). Akt has been reported to migrate into the nucleus in response to a variety of proliferative stimuli, including insulin (26,27). Immunostaining 3T3-L1 cells for Akt in the presence or absence of insulin stimulation in control cells demonstrated the characteristic nuclear pattern of Akt (Supplementary Fig. 6A and B). Immunostaining of Akt in cells overexpressing GFP-AGPAT2 after starvation in serum-free media demonstrated translocation of Akt, even in the absence of insulin stimulation (Supplementary Figure 6C–E). Finally, we expressed constitutively active Akt (myr-Akt) after knockdown of AGPAT2 and then treated cells with MDI to induce differentiation. myr-Akt increased the expression of adiponectin, but not PPARγ mRNA (Fig. 4E), and led to an increase in Oil Red O staining (Fig. 4F). These data demonstrate that disruption of AGPAT2 results in a significant reduction in Akt activation and downstream Akt signaling events, and bypassing this defect can rescue adipogenesis.

The generation of phosphatidylinositols is important for the activation of Akt signaling (28–32). Using LC/MS, we analyzed lipid extracts from 3T3-L1 preadipocytes expressing either GFP or AGPAT2-GFP. We found that overexpression of AGPAT2 results in a significant increase in both PIP₂ 36:1 and PIP₂ 36:2 levels without changes in PIP₁ levels (Fig. 5A). PIP₃ could not be assayed by LC/MS; however, immunofluorescence detected with anti-PIP₃ antibody was increased after AGPAT2 overexpression and was confirmed by quantification using direct detection on nitrocellulose membranes (Fig. 5B and C). Consistent with a role for AGPAT2 regulating the upstream signaling to Akt, we found that in serum-starved 3T3-L1 preadipocytes, AGPAT2 overexpression resulted in increased Akt phosphorylation, which was completely abrogated by the PI3K inhibitor LY294002 (Fig. 5D). The findings support a role for AGPAT2 as regulator of phosphatidylinositol levels in preadipocytes, which in turn modulates Akt signaling.

LPA has been described as an endogenous PPARγ agonist, whereas CPA was recently identified as a PPARγ antagonist with nanomolar affinity (33,34). CPA is a naturally occurring analog of LPA, and although both share a similar structure, their biological effects differ significantly (35). We assessed the effect of AGPAT1 and AGPAT2 on PPARγ transactivation and endogenous levels of these bioactive lipids. When NIH3T3 cells were transfected with AGPAT1 or AGPAT2, only AGPAT2 increased expression of a PPARγ luciferase promoter driven by a canonical PPAR-responsive element (PPRE) (36), both in the absence and presence of the PPARγ ligand rosiglitazone (Fig. 6A). This was not associated with a significant increase in PPARγ protein levels or migration, suggesting that AGPAT2 overexpression modulates the transactivation of PPARγ (Fig. 6B). It was possible that the effect of AGPAT2 overexpression on PPARγ activity was dependent on the activation of the PI3K/Akt pathway. Cells treated with the PI3K inhibitor LY294002 demonstrated ∼50% downregulation of luciferase activity (Fig. 6B). However, AGPAT2 overexpression increased the luciferase expression in both control and LY294002-treated cells, suggesting that the effect of AGPAT2 was independent of the direct activation of Akt.

Knockdown of AGPAT2 resulted in an increase in different LPA species (Fig. 7A), and, surprisingly, overexpression of AGPAT1 and AGPAT2 also resulted in an increase in the levels of LPA (Fig. 7B) while simultaneously decreasing PA. Conversely, knockdown of AGPAT2 resulted in increased levels of the PPARγ inhibitor CPA, in particular the most highly abundant subspecies, CPA18:0 and CPA16:0 (Fig. 7C). Overexpression of AGPAT2 decreased CPA levels of the same subspecies, whereas overexpression of AGPAT1 had no effect or a trend toward increasing CPA species (Fig. 7D). LPC, the precursor of CPA, has been implicated as a negative regulator of insulin signaling and is also reduced (∼20%) after overexpression of AGPAT2 (Supplementary Fig. 7) (37). The ability of pioglitazone and AGPAT2 to enhance transactivation of the PPARγ reporter was inhibited by treatment with exogenous CPA, suggesting that AGPAT2 modulates the production of CPA rather than increasing its turnover (Fig. 7E). Finally, to determine the role of PPARγ transactivation on the adipogenic defect of CGL-derived MDMCs, we treated cells derived from subject 2 with the PPARγ agonist pioglitazone along with MDI. Pioglitazone increased the expression of terminal differentiation markers C/EBPα and PPARγ, and this associated with the accumulation of lipid droplets, as shown by Oil Red O staining (Fig. 7F), though this effect was still significantly reduced compared with the differentiation observed in cells from control individuals.

AGPAT2 mutations have been found in a subset of patients with CGL, and it was presumed that these subjects were unable to generate triglycerides, resulting in “empty adipocytes” (3). However, earlier studies suggested that AGPAT2 is involved in adipogenesis. Our work in MDSCs from humans with AGPAT2 mutations as well as in 3T3-L1 cells shows that AGPAT2 plays a central role in adipogenesis by generating a proadipogenic lipome, leading to the activation of Akt signaling while simultaneously regulating the production of CPA, a PPARγ inhibitor.

AGPAT2 is upregulated in a number of tumors, including ovarian cancer (38,39) and small-molecule inhibitors have been developed that specifically inhibit AGPAT2. These agents result in the attenuation of a number of signaling pathways, including both the Ras/Raf/Erk and PI3K/Akt pathways (9). Our work demonstrates that gain and loss of AGPAT2, but not AGPAT1, has concordant effects on the phosphorylation of Akt and its downstream targets. Further, we find that constitutively active Akt can partially rescue the adipogenic defect, demonstrating that loss of AGPAT2 is working proximal to Akt activation. In the absence of AGPAT2, differentiating CGL-derived progenitor cells had increased cell death, similar to AGPAT2-expressing tumor cells after treatment with an AGPAT2-specific small-molecule inhibitor, suggesting that apoptosis of differentiating adipocytes may play a role in the severe lipodystrophic phenotype seen in CGL patients. Both Akt and PPARγ are involved in adipocyte survival (40). A provocative observation is that addition of fatty acids to cells overexpressing AGPAT2 further increases Akt phosphorylation, suggesting a physiological mechanism whereby increased flux of fatty acids may participate in the stimulation of the adipogenic program. Further work on the specificity of fatty acids and concentration and duration of exposure will be important to define these relationships.

We continue to try to understand the apparently paradoxical effects of AGPAT2 on PA and LPA levels in cells. Ovarian cancer cells in which AGPAT2 is highly expressed show an increase in LPA levels, and knockdown of AGPAT2 in OP9 cells results in increases in PA levels, identical to our findings (41). As with AGPAT2, disruption of the Lipin-1 gene results in lipodystrophy in mice (42, 43). PA results in Lipin-1 exclusion from the nucleus and accumulation in the endoplasmic reticulum (44,45). One potential explanation for the paradoxical PA changes is that the PA pool affected by AGPAT2 is not reflected in total cells extracts. In preliminary studies, we have found that AGPAT2 overexpression causes the nuclear exclusion of Lipin-1 (A.R.S., unpublished observation), which is consistent with a localized increase of PA in the endoplasmic reticulum. Further studies will be needed to sort out the complex changes in lipid metabolism that result in unexpected changes in bulk PA and LPA levels in cells after alterations in AGPAT2 expression.

Although AGPAT2 mRNA is detected in 3T3-L1 preadipocytes prior to differentiation, there is a significant increase after induction of adipogenesis, paralleling that of PPARγ, and it is also increased in vivo after treatment with the PPARγ agonist pioglitazone (46), placing AGPAT2 as a PPARγ-responsive gene. Sequence analysis shows a PPAR direct repeat at −723 from the transcriptional start site. The finding that AGPAT2 reduces CPA levels in preadipocytes could suggest that the progressive increase in PPARγ with differentiation is accompanied by reduction in CPA levels, allowing activation of PPARγ. These findings support a model in which AGPAT2 joins PPARγ and c/EBPα as part of a positive feedback loop promoting the mature adipocyte phenotype.

Treatment of CGL-derived MDMCs with pioglitazone was able to increase the expression of adipocyte terminal differentiation markers and generate lipid droplets, although this effect was markedly attenuated compared with MDMCs derived from control individuals. Despite the almost complete absence of adipose tissue, subject 3 (Supplementary Table 2) responded to therapy with PPARγ agonists. Her HbA_1c was reduced from 14.6 to 9.7% and her triglycerides decreased from 727 mg/dL to 213 mg/dL after 6 months of treatment (47). Considering the in vitro response of MDMCs to pioglitazone, it is tempting to suggest that generation of adipocytes from precursor cells helped alleviate some of the metabolic derangements in this patient, although it is possible that thiazolidinedione exerted its metabolic effect independent of adipose tissue (48–50). AGPAT2 and PPARγ are both expressed in other insulin-responsive tissues, including liver and skeletal muscle. This would raise the possibility that AGPAT2 is important for the regulation of PPARγ and PI3K/Akt in tissues, which contributes to the severe insulin resistance in CGL patients. Although AGPAT2 is expressed in liver, the hepatic steatosis seen in CGL subjects appears to be secondary to the insulin resistance and not to a direct effect of AGPAT2 on hepatic lipogenesis (51).

In summary, this study provides a further understanding of the pathogenesis of lipodystrophy in patients with AGPAT2 mutations. Our findings demonstrate that CGL secondary to AGPAT2 mutation is not a disease of empty adipocytes, rather a disease of defective adipose tissue development due to a disruption in the modulation of the lipome during adipogenesis (Fig. 8). Normally, upon activation of the differentiation program, AGPAT2 expression increases and reduces cyclic PA levels, possibly by directing LPA generated from LPC to PA. This allows endogenously derived PPARγ activators to increase PPARγ-mediated activation of the adipogenic program. The increased flux to PA could be the basis for increasing the generation of PIP₃, which is important for activation of Akt. Because of the paradoxical change in LPA in cells with AGPAT2 manipulation, we believe that the PA generated by AGPAT2 is compartmentalized from that generated by AGPAT1 (Fig. 8A, dashed circle). In patients with CGL due to AGPAT2 mutations, cyclic PA increases and inhibits PPARγ, and the reduced generation of PA could be leading to reduced PIP₃ levels, impaired activation of Akt, and an increase in apoptosis (Fig. 8B). Understanding AGPAT2 actions in adipogenesis in relation to the modulation of the lipome provides the potential for understanding the intimate relationship between lipids, adipogenesis, lipid metabolism, and insulin resistance in common obesity.

This work was supported by the Michigan Nutrition Obesity Research Center (DK-089503) Metabolomics Core and Pilot Grant Program and by the Robert C. and Veronica Atkins Research Foundation.

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

A.R.S. designed and performed the experiments and wrote the manuscript. A.K.D., X.L., B.E., C.E., M.E.A., and M.T. performed the experiments. E.O. and T.L. contributed to the experiments and manuscript preparation. C.F.B. contributed to experimental design and writing and editing the manuscript. C.F.B. and A.R.S. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.