Rab18 promotes lipid droplet (LD) growth by tethering the ER to LDs through SNARE and NRZ interactions

Lipid droplets (LDs), highly dynamic subcellular organelles primarily responsible for energy storage, have been linked to multiple cellular processes, including virus packing, protein storage and modification, and host defense (Herker et al., 2010; Klemm et al., 2011; Anand et al., 2012; Li et al., 2012; Suzuki et al., 2012). LDs contain a monolayer of phospholipids and their specific associated proteins, and undergo dynamic changes including biogenesis, fusion/growth, and degradation (Martin and Parton, 2006; Farese and Walther, 2009; Walther and Farese, 2012; Yang et al., 2012; Thiam et al., 2013; Pol et al., 2014). The dynamics of LDs reflect the lipid metabolic status, and uncontrolled growth of LDs has been linked to the development of multiple diseases including obesity, diabetes, fatty liver diseases, cardiovascular diseases, cancer, and neurodegenerative diseases (Gong et al., 2009; Greenberg et al., 2011; Suzuki et al., 2011; Xu et al., 2012a; Krahmer et al., 2013; Gross and Silver, 2014; Liu et al., 2015).

LD biogenesis is initiated and nascent LDs are formed from ER (Murphy and Vance, 1999; Khandelia et al., 2010; Zanghellini et al., 2010; Gross et al., 2011; Pol et al., 2014; Wilfling et al., 2014; Choudhary et al., 2015). The sizes of nascent LDs in mammalian cells are believed to be <100 nm, whereas most mature cytosolic LDs have diameters ranging from 0.25 to 100 µm depending on cell types (Pol et al., 2014). Several distinct mechanisms by which LDs grow and expand have been discovered. First, nascent LDs may grow to mature ones by acquiring neutral lipids from ER through continuous association with ER (Ohsaki et al., 2008; Jacquier et al., 2011), or by incorporation of ER-synthesized lipids that is dependent on DGAT1 activity through an unknown mechanism (Szymanski et al., 2007; Gross et al., 2011; Cartwright and Goodman, 2012; Xu et al., 2012b; Wilfling et al., 2013). Seipin, a protein originally identified in human general lipodystrophy (Magré et al., 2001; Payne et al., 2008), has shown to play an important role in promoting LD growth (Szymanski et al., 2007; Fei et al., 2008, 2011; Pagac et al., 2016; Salo et al., 2016; Wang et al., 2016) by localizing on a potential ER–LD contact site (Szymanski et al., 2007; Binns et al., 2010; Grippa et al., 2015; Han et al., 2015; Salo et al., 2016; Wang et al., 2016). Second, LD-associated enzymes such as GPAT4 and DGAT2 can promote LD growth by incorporating locally synthesized TAG into LDs (Fujimoto et al., 2007; Kuerschner et al., 2008; Krahmer et al., 2011; Wilfling et al., 2013). Finally, CIDE protein can promote LD growth via atypical lipid transfer and LD fusion in the white adipose tissue, in the liver of high-fat diet treated or obese mice, and in skin sebocytes and lactating mammary epithelia cells (Gong et al., 2011; Wang et al., 2012; Zhou et al., 2012; Wu et al., 2014b; Zhang et al., 2014; Xu et al., 2016). Several factors including Perilipin, Rab8a, As160, and Mss4 that modulate Cidec-mediated LD fusion have been identified (Sun et al., 2013a; Wu et al., 2014a).

The activity of RabGTPases, crucial regulators of vesicle trafficking and membrane dynamics, is regulated by their specific GEFs, GAPs, and downstream effectors (Zerial and McBride, 2001; Grosshans et al., 2006; Stenmark, 2009). Rab18 is shown to be an LD-associated protein in several cell types including 3T3-L1 preadipocytes and differentiated adipocytes, and its expression levels and LD localization are controlled by specific nutritional or hormonal signals (Martin et al., 2005; Ozeki et al., 2005; Martin and Parton, 2008; Salloum et al., 2013). Rab18 has been shown to play a potential role in β-adrenergic–stimulated lipolysis and in insulin-induced lipogenesis in adipocytes (Pulido et al., 2011) and can facilitate hepatitis C virus replication by interacting with its nonstructural protein NS5A (Salloum et al., 2013). Interestingly, point mutations (L24Q or T95R, etc.) that results in defective Rab18 were discovered in autosomal recessive disorder Warburg Micro syndrome (WARBM; Bem et al., 2011). Mutations in Rab3GAP1 (T18P and E24V) and Rab3GAP2 (R426C; Handley and Aligianis, 2012) complex, a specific GEF for Rab18 (Gerondopoulos et al., 2014), were also identified in patients suffering from WARBM. Surprisingly, abnormal large LDs accumulate in fibroblasts isolated from WARBM patients who harbor Rab18 or Rab3GAP1/2 mutations (Aligianis et al., 2005; Handley and Aligianis, 2012; Handley et al., 2013; Liegel et al., 2013). A recent study showed that mammalian TRAPPII complex may act as a GEF of Rab18 to activate and recruit it to LDs in several cell types (Li et al., 2017). Despite all these analyses, the precise mechanism by which Rab18 controls LD dynamics remains unclear.

The yeast Dsl1 tethering complex is stably associated with three ER-associated Q-SNAREs (Ufe1p, Sec20p, and Use1p that have a glutamine at the center of their coiled-coil domain; Diefenbacher et al., 2011) and believed to play a central role in the tethering and fusion of Golgi-derived COPI transport vesicles to the ER (Andag et al., 2001; Kamena and Spang, 2004; Kraynack et al., 2005; Tagaya et al., 2014). Structural analysis revealed that the Dsl1 complex forms a 20-nm-tall “tower” from the ER surface (Ren et al., 2009; Tripathi et al., 2009). The NRZ tethering complex, the mammalian homologue of Dsl1 that includes ZW10 (Dsl1p), NAG (Dsl3p), and RINT1 (Tip20p), is associated with ER-localized Q-SNAREs (Use1, Syntaxin18, and BNIP1; Tagaya et al., 2014) and plays a potential role in Golgi-ER retrograde vesicle trafficking and fusion (Hatsuzawa et al., 2000; Burri et al., 2003; Hirose et al., 2004). Interestingly, a systemic screening of Rab effectors in Drosophila melanogaster cells reveals that the ZW10 and RINT1 subunits of the NRZ complex and Syntaxin18 of the ER–SNARE complex may be effectors of Rab18, and overexpression of Rab18 enhances the LD association of ZW10 (Gillingham et al., 2014).

Here, we have identified Rab18 as an important regulator of LD dynamics as its deficiency results in a defective LD growth and maturation without affecting LD biogenesis. Rab3GAP1/2 complex, the GEF of Rab18, promotes LD growth by activating and targeting Rab18 to LDs. Activated Rab18 binds to NRZ and ER-associated Q-SNAREs and recruits them to the close proximity of LDs. The Rab18-NRZ-SNARE interaction and complex formation result in tethering ER to LD and generating a close contact between the ER and LDs, therefore facilitating lipid incorporation to promote LD growth.

To identify RabGTPases that control LD dynamics, we knocked down 20 RabGTPases that were reported to be potentially localized to LDs in 3T3-L1 preadipocytes using their specific siRNAs (Wu et al., 2014a; Fig. 1 A). As control wild type or Rab10 knocking down cells accumulated a large number of LDs, knocking down Rab18 in 3T3-L1 preadipocytes led to the accumulation of fewer but significantly larger LDs (Fig. 1 B and Fig. S1 A). Knockout (KO) of Rab18 by CRISPR/Cas9 in 3T3-L1 preadipocytes also resulted in the accumulation of significantly fewer but larger LDs (Fig. 1 C and Fig. S1 B). When Rab18 was reintroduced into Rab18-deficient cells, LD size was reduced and the number of visible LDs was increased, showing a restoration of normal LD morphology (Fig. 1 C and Fig. S1 B).

To quantitatively evaluate the role of Rab18 in controlling LD dynamics, we systematically measured the number and sizes of LDs and the total volume of lipids in Rab18-deficient cells using a confocal optical section in combination with 3D surface reconstitution (Materials and methods). In control cells, ∼99% of LDs had a diameter ranging from 0.25 µm (limits of the fluorescent microscope) to 2 µm when cells were treated with oleic acid (OA; Fig. 1 D). LDs larger than 2 µm in diameter were rarely observed. As nascent LDs generated from ER were shown to have a size ranging from 60–100 nm in diameter, these droplets had to subsequently grow and mature via continuous TAG incorporation (Zanghellini et al., 2010; Pol et al., 2014; Wilfling et al., 2014; Fig. 1 E). We therefore define LDs ranging from 0.25 to 2 µm in diameter as mature LDs, and define LDs larger than 2 µm in diameter as supersized LDs.

In Rab18-deficient cells, the number of mature LDs in each cell was dramatically decreased (117 ± 13 per cell) compared with that in control cells (938 ± 105 per cell), representing an 88% reduction in the number of mature LDs (Fig. 1, D and F). The number and percentage of supersized LDs in each cell were significantly increased in Rab18-deficient cells (Fig. 1, D and G). When the sizes of the largest LD in a cell were measured in wild-type or Rab18-deficient cells, the mean diameter of the largest LD in Rab18-deficient cells was nearly twofold larger than that in control cells, representing an eightfold increase in LD volume (Fig. 1 H). Quantitative analysis also confirmed that reintroduction of Rab18 into Rab18-deficient cells increased the number of mature LDs but decreased the number of supersized LDs (Fig. 1 H).

To monitor the dynamic LD growth process under OA treatment, cells were first cultured in serum-free medium overnight to deplete existing LDs and then fed with OA. The morphology of LDs was continuously recorded under a time-lapse microscope. In control cells, a large amount of small LDs accumulated in the presence of OA. In Rab18-deficient cells, only a few LDs were observed, and these LDs expanded rapidly and grew into supersized ones (Fig. S1 C and Videos 1, 2, 3, and 4). Quantitative analysis of the dynamic LD growth process indicated that the rate of increase in the number and the total volume of mature LDs in Rab18-depleted cells were significant lower than in control cells (Fig. S1, D and E). However, the expansion of supersized LDs in Rab18-deficient cells was markedly faster (Fig. S1 F). The reduction of mature LD numbers and the appearance of supersized LDs were not a result of the fusion/coalescence of small LDs as this was not observed during the live-cell imaging recording process (Video 2). Overall, these data strongly suggest that depletion of Rab18 disrupted normal LD morphology, leading to a significant reduction in the number of mature LDs but accumulation of a few supersized LDs.

Next, we measured the amount of total cellular TAG and its subcellular distribution in Rab18-deficient cells and observed significantly lower cellular TAG and reduced TAG levels in the LD fraction in Rab18-deficient cells after they were treated with OA for 20 h (Fig. 1, I and J). Similar rates of fatty acid (FA) uptake and TAG hydrolysis were observed between wild-type and Rab18-deficient cells (Fig. S1, G and H). However, Rab18-deficient cells had a decreased TAG synthesis (Fig. S1, I and J). As long-term exposure to excess FA results in a chronic ER stress in cells lacking the capacity to synthesize or store neutral lipids (Ota et al., 2008; Chitraju et al., 2017), we evaluated the role of Rab18 in FA overloading–induced ER stress. As shown in Fig. 1 K, expression levels of Bip, a specific ER stress marker, were significantly increased in the presence of thapsigargin (TG), a known reagent to induce ER stress. The expression levels of Bip were also increased in Rab18-deficient cells with a prolonged OA treatment. Splicing levels of XBP-1, another ER stress marker, was also significantly increased in Rab18-deficient cells (Fig. 1 L). Therefore, Rab18 plays an important role in regulating lipid storage, and its deficiency results in reduced lipid storage and increased ER stress in the presence of excess free FAs.

As Rab18 was previously shown to be highly expressed in differentiated adipocytes, we knocked down Rab18 in 3T3-L1 adipocytes and observed that these cells contained larger LDs compared with those in control adipocytes (Fig. 2, A and B). In contrast, overexpression of Rab18 in adipocytes resulted in a dramatic decrease in LD sizes (Fig. 2, C and D). In addition, depletion of Rab18 in adipocytes led to a 20% reduction in the cellular TAG levels (Fig. 2 E) and an ∼15% decrease in basal and stimulated lipolysis (Fig. 2 F). This minor decrease in lipolysis is probably a result of a secondary effect of increased LD sizes and decreased LD surface area in Rab18-depleted adipocytes. Rab18 expression is also observed in Leydig cells that contain large amounts of LDs and are responsible for the production of testosterone in testes (Yamaguchi et al., 2015; Shen et al., 2016). We generated Rab18-deficient TM-3 (a Leydig cell line) cells and observed that Rab18-deficient TM-3 cells contained less mature LDs but an increased number of supersized LDs after OA treatment (Fig. 2, G and H). Overall, these data indicate that Rab18 regulates the LD morphology and lipid storage in adipocytes and Leydig cells.

As Rab18 depletion dramatically reduced the number of mature LDs, we checked whether its deficiency would disrupt LD biogenesis and nascent LD formation by expressing GFP-LiveDrop, a nascent LD marker (Wang et al., 2016), in Rab18-deficient cells or in control wild-type cells. Cells were starved for 16 h to deplete the existing LDs. After short-term OA treatment (1 h), similar numbers of LiveDrop-positive LDs, representing nascent LDs, accumulated in both Rab18-deficient and control cells (Fig. 3, A and B). In control cells, LiveDrop-positive signals overlapped very well with LipidTOX (65.6%) signals that stain mature LDs. However, in Rab18-deficient cells, only a small percentage of LiveDrop positive signals (8.09%) overlapped with LipidTOX-positive mature LDs (Fig. 3 C). After cells were treated with OA for 4 h, a majority of LiveDrop-positive LDs were also positive for LipidTOX signals in control cells. On the contrary, in Rab18-deficient cells, most LiveDrop-labeled LDs remained small and did not overlap with LipidTOX staining (Fig. 3 D). Therefore, Rab18 does not appear to affect LD biogenesis but is responsible for the growth and maturation of nascent LDs.

To further confirm the role of Rab18 in regulating the growth of nascent LDs, we examined LD morphology of Rab18-deficient cells by EM analysis. The presence of nascent LDs within the sizes of 100–200 nm in diameter was observed in both control and Rab18-deficient cells 1 h after OA treatment (Fig. 3 E). Quantitative analysis indicated that the number of nascent LDs was slightly increased in Rab18-deficient cells compared with that in control cells (Fig. 3 F). After 8 h of OA treatment, the total number of LDs visible in EM analysis showed no significant difference between control and Rab18-deficient cells. However, LD size distribution changed dramatically (Fig. 3, E, G, and H). In wild-type cells, the majority of nascent LDs had grown larger than 150 nm in diameter. In contrast, LDs with diameter larger than 150 nm (except a few LDs larger than 2 µm in diameter) were rarely observed in Rab18-deficient cells. Instead, the majority of LDs remained smaller than 150 nm in Rab18-deficient cells (Fig. 3, E, G, and H). In summary, EM analysis confirmed that Rab18 deficiency did not affect LD biogenesis and nascent LD formation but controlled the growth of nascent LDs into mature LDs.

To evaluate the role of TAG synthesis enzymes in Rab18-mediated LD growth, we checked the expression levels and subcellular distribution of DGAT1/2 and GPAT3/4 in Rab18-deficient cells and found that Rab18 deficiency affected neither their expression (Fig. S2 A) nor their subcellular distribution (Fig. S2, B–G). Inhibition of DGAT1 activity led to the accumulation of a significantly lower number of mature LDs and decreased cellular TAG (Fig. 4, A and B; Wilfling et al., 2013) in wild-type cells and reduced LD sizes in Rab18-deficient cells. On the contrary, overexpression of DGAT1 in wild-type cells increased the number of mature LDs by about twofold (Fig. 4, C and D). However, overexpression of DGAT1 in Rab18-deficient cells did not restore the number of mature LDs (Fig. 4, C and D). Surprisingly, inhibition of DGAT2 did not affect the number of mature LDs in wild-type cells as well as the sizes of LDs in Rab18-deficient cells (Fig. 4, E and F). Knocking down both GPAT3 and GPAT4, which act as rate-limiting enzymes in the TAG synthesis pathway upstream of DGAT1 and DGAT2, reduced the number of mature LDs and the intracellular TAG level in wild-type cells (Fig. 4, G and H) and decreased LD sizes in Rab18-deficient cells (Fig. 4, G and H). These data suggest that TAG synthesis on ER that is controlled by DGAT1/GPAT3/4 was responsible for the formation of supersized LDs, whereas local TAG synthesis mediated by DGAT2 did not contribute to the formation of supersized LDs in Rab18-deficient cells. Consistent with Chitraju et al. (2017), we observed increased FA overloading–induced ER stress in wild-type cells when DGAT1 was inhibited (Fig. 4 I). In Rab18-deficient cells, ER stress was induced under the FA overloading condition and further enhanced when DGAT1 (but not DGAT2) was inhibited (Fig. 4 I). Overall, we demonstrated that Rab18-mediated LD growth and maturation are dependent on TAG synthesis on ER, and Rab18 may play a role in mediating the transfer of ER-synthesized lipids to nascent LDs.

To delineate the mechanism by which Rab18 controls LD growth, we checked its subcellular localization and observed that endogenous Rab18 was associated with LDs in 3T3-L1 preadipocytes (Fig. 5 A). Biochemical fractionation further confirmed that Rab18 was highly enriched in the LD fraction (Fig. 5 B). In serum-free medium, Rab18 was observed at specific loci on ER and associated with nascent LDs when cells were treated with FAs, overlapping with ACSL3, an early LD marker (Kassan et al., 2013; Fig. 5 C). Rab3GAP1/2 complex, a GEF of Rab18, was also localized to LD when Rab3GAP2 was coexpressed with Rab18 in 3T3-L1 preadipocytes (Fig. 5 D). Colocalization of Rab3GAP1, Rab3GAP2, and Rab18 was also observed at the newly synthesized LDs (Fig. S3 A). When Rab18 was introduced into Rab3GAP1- or Rab3GAP2-deficient cells, the LD localization of Rab18 was completely abolished in these cells (Fig. 5 E and Fig. S3 B). Biochemical fractionation also showed that the enrichment of Rab18 on LD fraction was abolished in cells deficient in Rab3GAP1 (Fig. 5 F).

We then checked the LD morphology in Rab3GAP1- or Rab3GAP2-deficient 3T3-L1 preadipocytes (Fig. 5 G; and Fig. S3, B and C) and observed that the number and percentages of mature LDs in these cells were dramatically reduced (41 ± 2 per cell in GAP1 KO cells or 33 ± 1 per cell in GAP2 KO cells vs. 612 ± 28 per cell in wild-type cells, representing a nearly 95% reduction; Fig. 5, G and H; and Fig. S3, C and D). Instead, both Rab3GAP1- and Rab3GAP2-deficient cells accumulated supersized LDs (Fig. 5, G and H; and Fig. S3, C, E, and F), similar to those in Rab18-deficient cells. Consistent with reduced mature LDs, the total amount of cellular TAG in these cells was also decreased (Fig. S3 G). In mature adipocytes, both Rab3GAP1 and Rab3GAP2 were markedly increased during the course of adipocyte differentiation (Fig. S3 H) and colocalized with Rab18 on LDs, consistent with the observation in preadipocytes (Fig. S3 I). Rab3GAP1 deficiency in adipocytes did not affect 3T3-L1 differentiation as the expression levels of FABP4, Plin1, and Fsp27 were similar between wild-type and Rab3GAP1-deficient cells (Fig. S3 J). Similar to the observation in Rab18-deficient adipocytes, Rab3GAP1-deficient adipocytes accumulated fewer but larger LDs (Fig. 5 I and Fig. S3 K). Overall, these data strongly indicate that Rab3GAP1/2 acts as a GEF to promote the LD localization and activation of Rab18 in controlling LD growth and maturation in both pre- and mature adipocytes.

Because NRZ tethering complex that comprises NAG, RINT1, and ZW10 has been reported to interact with Rab18 (Gillingham et al., 2014), we further confirmed their interaction by overexpressing FLAG-Rab18 and checked the presence of endogenous NRZ proteins in the Rab18–immunoprecipitation (IP) products. Endogenous NAG and ZW10 were detected in the precipitates (Fig. 6 A). The amount of coprecipitated NAG and ZW10 was increased in the presence of Rab3GAP1/2. RINT1, the third component in the NRZ complex, was also detected in the Rab18-IP product in the presence of Rab3GAP1/2 (Fig. 6 A). To test the interaction of Rab18 with individual subunits in the NRZ complex, we first coexpressed ZW10 and Rab18 in 293T cells, and observed that ZW10 and Rab18 were able to reciprocally pull down each other (Fig. 6, B and C). To ascertain the direct interaction between Rab18 and ZW10, we purified Flag-ZW10 from 293T cells and GST-Rab18 from Escherichia coli (Fig. S4 A) and performed in vitro binding assays. As shown in Fig. 6 D, GST-Rab18 was able to specifically pull down ZW10 in the presence of GTPγS that locked Rab18 in its GTP-bound form. We then generated serial truncations of ZW10 and identified that its C terminus was necessary for its interaction with Rab18 (Fig. S4, B and C). ZW10 truncation lacking the C-terminal region still bound to NAG and RINT1, consistent with previous observation (Arasaki et al., 2006) that the N-terminal region of ZW10 mediates its interaction with NAG/RINT1. NAG alone showed weak interaction with Rab18, and its association with Rab18 was dramatically increased in the presence of ZW10 (Fig. 6 E). Truncation analysis further revealed that the C-terminal region (aa 1355–2371) of NAG mediated its interaction with ZW10 and Rab18 (Fig. 6 E and Fig. S4 D). These data suggest that ZW10 in the NRZ complex interacts directly with the GTP-bound form of Rab18, and its C-terminal region is essential for this interaction.

To check whether Rab18 affects the subcellular distribution of NRZ complex, we first isolated various subcellular fractions from wild-type or Rab18-deficient TM-3 cells (Fig. 6 F). Components of NRZ complex including ZW10 and NAG were detected in both ER and LD fractions in wild-type cells. However, their presence in the LD fraction was completely abolished in the absence of Rab18. When Cherry-ZW10 and HA-Rab18 were coexpressed in 3T3-L1 preadipocytes, both endogenous NAG and Cherry-ZW10 were observed on the LD surface, colocalizing with Rab18 (Fig. 6 G). In addition, introduction of Rab18 into 3T3-L1 preadipocytes markedly enhanced the LD localization of endogenous NAG and ZW10 (Fig. 6 H). RINT1 was also observed on the LD surface (Fig. S4 E). When GFP-ZW10 was introduced into mature adipocytes, the LD association of ZW10 was observed only when it was coexpressed with wild-type Rab18 but not with Rab18-C203A (Fig. S4 F). These data indicate that Rab18, in its GTP-bound form, interacts with and recruits the NRZ complex to the LD surface.

To evaluate the function of the NRZ complex in regulating LD growth, we knocked out NAG or ZW10 in 3T3-L1 preadipocytes (Fig. S5 A). Depletion of NAG or ZW10 led to the accumulation of significantly fewer but larger LDs (Fig. 7 A). Quantitative analysis of LD morphology showed that NAG or ZW10 deficiency resulted in a dramatic reduction in the number and percentage of mature LDs after OA treatment (218 ± 6 per cell in NAG-deficient cells or 297 ± 19 per cell in ZW10-deficient cells vs. 557 ± 22 per cell in wild-type cells, representing a nearly 50% reduction; Fig. 7, B and C; and Fig. S5 B). On the contrary, numbers of supersized LDs were significantly increased in NAG- or ZW10-deficient cells (Fig. 7 B and Fig. S5, B and C). Similar to the observation in Rab18-deficient cells, time-lapse imaging showed that NAG-deficient cells accumulated a few large LDs that expanded rapidly (Fig. S5, D–G; and Videos 5 and 6). In addition, levels of total cellular TAG were significantly decreased in NAG- or ZW10-deficient cells (Fig. 7 D), and the susceptibility of long-term FA–induced ER stress was increased in NAG-deficient cells (Fig. 7 E). In addition, overexpressing Rab18 in NAG-deficient cells did not increase the number of mature LDs (Fig. 7, F and G) or decrease the size of supersized LDs (Fig. 7, F and H). Importantly, when truncated ZW10 that is defective in interacting with Rab18 was introduced into ZW10-deficient cells (Fig. S5 H), it did not restore its function in promoting LD growth and maturation (Fig. S5, I and J). These data indicate that NRZ complex acts downstream of Rab18, and the interaction between Rab18 and NRZ is crucial for its role in promoting LD growth and maturation.

To test whether NAG controls LD biogenesis, we expressed GFP-LiveDrop in wild-type or NAG-deficient cells. 4 h after OA treatment, the majority of LiveDrop-positive LDs overlapped with LipidTOX in wild-type cells. In NAG-deficient cells, a large number of LiveDrop-positive LDs accumulated, and they did not overlap with the LipidTOX signal (Fig. S5 K). EM analysis showed that nascent LDs <150 nm in diameter were present in both wild-type and NAG-deficient cells 1 h after OA treatment (Fig. 7 I), and the number of nascent LDs was slightly increased in NAG-deficient cells (Fig. 7 J). After prolonged OA treatment (8 h), the total number of LDs visible in EM analysis showed no significant difference between wild-type and NAG-deficient cells (Fig. 7, I and K). However, in NAG-deficient cells, the majority of LDs had a diameter less than 150 nm, and a few supersized LDs were observed (Fig. 7, I and L). These data indicate that NAG deficiency does not affect LD biogenesis but controls the growth of nascent LDs to form mature LDs, almost completely recapitulating the phenotype of Rab18 and Rab3GAP1/2 deficiencies.

As the NRZ complex has been reported to interact with ER-associated Q-SNAREs including Syntaxin18, Use1, and BNIP1 (Hatsuzawa et al., 2000; Hirose et al., 2004; Nakajima et al., 2004), we tested whether these proteins play roles in regulating LD growth by knocking them out in 3T3-L1 preadipocytes (Fig. 8 A). A loss of any member of the Q-SNAREs resulted in a reduction of the accumulation of mature LDs (110 ± 3 per cell in Stx18 KO cells, 166 ± 5 per cell in Use1 KO cells, and 161 ± 7 per cell in BNIP1 KO cells vs. 593 ± 24 per cell in wild-type cells, representing a nearly 75% reduction; Fig. 8, B–D), whereas the number of supersized LDs was significantly increased and levels of cellular TAG were markedly reduced in Stx18-, Use1-, or BNIP1-deficient cells after prolonged OA treatment (Fig. 8, B, E, and F). As three Q-SNAREs and one R-SNARE (that has an arginine at the center of its coiled-coil domain) are required to trigger typical membrane fusion, we isolated several potential R-SNAREs including Sec22b, Ykt6, and Vamp8 from LD fraction of the liver of ob/ob mice (Table S1). Knocking down Sec22b or Vamp8 did not affect LD morphology (Fig. 8, G and H). Knocking down Ykt6 led to a very minor effect in LD sizes and numbers (Fig. 8, G and H).

Consistent with its function in controlling LD growth, ER-associated Q-SNAREs were detected in the Rab18-IP complex (Fig. 8 I). Expression of Rab3GAP1/2 further enhanced the association of Rab18 with these SNAREs (Fig. 8 I). Biochemical fractionation showed that Stx18, Use1, and BNIP1 were all present in both ER and LD fractions in wild-type cells, but their presence in LD fraction was completely abolished in Rab18-deficient cells (Fig. 8 J). In addition, when Stx18, Use1, or BNIP1 was coexpressed with Rab18, they all showed in close proximity to LDs (Fig. 8 K). The LD association of ER-associated Q-SNAREs was dependent on the NRZ complex as NAG deficiency dramatically abolished their LD association (Fig. 9 A). GFP-Use1, when coexpressing with HA-Rab18, was recruited to LDs in 80% of wild-type cells, but its LD association was only observed in less than 20% NAG- or ZW10-deficient cells (Fig. 9, B and C). Surprisingly, Ykt6 interacted with neither Rab18 (Fig. 8 I) nor Stx18/Use1/BNIP1 Q-SNAREs (not depicted). In addition, it was not localized to LD, either (Fig. 8 J). Therefore, ER-associated Q-SNAREs are important in controlling LD growth and lipid storage by forming a complex with Rab18/NRZ tethering complex. The role of R-SNARE Ykt6 in Rab18-mediated LD growth remains unclear.

As Stx18, Use1, and BNIP1 are ER membrane–integrated proteins, we hypothesize that the recruitment of these SNAREs to the LD surface may result in the recruitment of ER membrane to the LD surface to create a close contact between ER and LDs. Indeed, when CB5, an ER-specific protein, was coexpressed with Rab18 and BNIP1 in 3T3L1-preadipocytes, BNIP1-positive signals were almost completely colocalized with CB5 and Rab18 on the LD surface in approximately 80% of cells (Fig. 9, D and E), but in NAG-deficient cells the colocalization of CB5/BNIP1 (ER-positive) signals and Rab18 (LD-associated) signals was significantly abolished (observed in less than 20% of the cells; Fig. 9, D and E). The colocalization between ER and LD was also observed when another ER-specific protein, RFP-KDEL, was coexpressed with GFP-Use1 and HA-Rab18 (Fig. 9 F).

To directly visualize ER–LD contact, we coexpressed GFP-Stx18 and APEX-GBP (GFP binding protein) in wild-type, Rab18-overexpressing, and Rab18-deficient 3T3-L1 preadipocytes. As APEX catalyzes the formation of a high-electronic-density substance in the presence of DAB and OsO4 (Ariotti et al., 2015), the ER-localized Stx18 was observed as a higher density signal. Consistent with the fluorescent imaging analysis (Fig. 9, D and F), EM images showed that APEX-Stx18–positive ER cisternae were clearly observed in Rab18-overexpressing cells (Fig. 10 A). Moreover, multiple APEX-Stx18–positive signals were observed on the LD surface, indicating multiple direct contacts between ER and LD (Fig. 10 A, red arrows). Quantitative analysis showed that nearly 80% cells overexpressing Rab18 had at least one LD that was labeled with multiple condensed APEX-Stx18 signals (Fig. 10, A and B). Approximately 10% cells showed a single APEX-Stx18 positive signal on the LD surface in control wild-type cells, indicating a single ER–LD contact (Fig. 10, A and B). In Rab18-deficient cells, less than 2.5% cells contained such ER–LD contact (Fig. 10, A and B).

Furthermore, we also measured the presence of ER membrane near the LD surface (within the distance of 20 nm) as an indirect criterion of a potential ER–LD. In Rab18-overexpressing cells, nearly 60% LDs were apposed to ER membrane, and 33% of total LD surface area had ER membrane apposed to it. In control wild-type cells, ∼30% LDs showed ER apposition and 5% total LD surface area was apposed to ER membrane. In Rab18-deficient cells, less than 5% LDs showed ER apposition, and 1% of total LD surface area was apposed to ER membrane (Fig. 10, C and D). These data demonstrate that Rab18 controls a direct ER–LD contact and LD apposition to ER membrane.

Overall, our data indicate that Rab18 is able to recruit the NRZ protein complex and its associated ER–SNAREs to the close proximity of LD to form a close LD-ER contacting site that potentially enhances lipid transfer from ER to LD to promote LD growth.

LDs serve as lipid storage depots, and their dynamic changes, including biogenesis and growth, are important in regulating lipid homeostasis. Here, we show that Rab18, a LD-associated RabGTPase, plays a key role in controlling LD growth and maturation by interacting with its effector, the NRZ/SNARE tethering protein complex, resulting in the initiation of close contact between ER and LD and facilitation of the incorporation of ER-synthesized lipids to LDs for their growth.

One of the most prominent features in Rab18-deficient fibroblast cells is the disappearance of normal-sized LDs and the appearance of a few supersized LDs. A similar role for Rab18 is also observed in mature adipocytes and in Leydig cells. Rab18 deficiency in 3T3-L1 preadipocytes did not affect nascent LD biogenesis, lipolysis, FA uptake, and LD coalescence. TAG synthesis in Rab18-deficient 3T3-L1 preadipocytes was reduced. As we did not observe a difference in their expression levels and subcellular distribution of TAG synthesis enzymes in Rab18-deficient cells, the lower TAG synthesis may be a result of a secondary effect of defective LD growth. Rab18 was previously shown to control lipolysis in adipocytes (Pulido et al., 2011). We also observed a slight decrease in both basal and stimulated lipolysis in Rab18-deficient adipocytes. This is likely a result of a secondary effect of a decreased surface-to-volume ratio of LDs and the cellular TAG level. The formation of supersized LDs in Rab18-deficient cells may be a result of redirection of lipids from ER to a subset of LDs. Change of LD morphology was not observed in Rab18-deficient AML12, HeLa, Cos7, and 293T cells (unpublished data), indicating a cell type–specific role of Rab18 in controlling lipid storage and LD growth. Interestingly, although we observed similar expression levels of Rab18, Rab3GAP1/2, and NRZ/SNAREs in these cells to that in 3T3-L1 preadipocytes, the LD association of Rab18 was greatly enhanced in 3T3-L1 pre- and differentiated adipocytes and in Leydig cells (unpublished data). The distinct localization of Rab18 may account for its role in controlling LD growth and maturation. Alternatively, other RabGTPases may play a functionally redundant role in controlling LD growth in these cells.

The LD localization and activity of Rab18 is controlled by its GEFs, Rab3GAP1/2, as Rab3GAP1/2 is colocalized with Rab18 on LDs and their depletion abolishes the LD association of Rab18 and results in the loss of mature LDs and reduction in lipid storage. As Rab3GAP1/2 is also shown to be responsible for the ER localization of Rab18 in COS7 cells and may regulate ER tubulation (Gerondopoulos et al., 2014), the localization and function of Rab18 and its GEF could be cell type dependent.

Importantly, we found that the GTP-bound form of Rab18 interacts directly with the C-terminal region of ZW10, one of the subunits of the NRZ tethering factors. In addition, NRZ complex and its associated Q-SNAREs form a protein complex with Rab18. Therefore, ZW10 acts as a bridge to recruit the NRZ tethering complex and its associated Q-SNAREs to Rab18. Functionally, NRZ and its associates, Q-SNAREs, act downstream of Rab18 to control LD growth, lipid storage, and FA-overloading–induced ER stress, almost a complete recapitulation of the role of Rab18. Unlike the membrane fusion mediated by Q-SNAREs and R-SNARE, thus far, a typical R-SNARE in Rab18-mediated LD growth has not yet identified. Ykt6 appears to have a very minor effect in Rab18-mediated LD sizes, but it was not localized to LDs and did not interact with Rab18 either (Fig. 8, I and J). This minor effect of Ykt6 may be a result of its regulatory role in controlling the assembly and association of ER-associated SNAREs with the NRZ complex. A similar role of Ykt6 has been observed in Golgi-ER retrograde transport (Meiringer et al., 2011).

Mechanistically, interaction between LD-associated Rab18, NRZ tethering complex, and ER-associated Q-SNAREs is able to bring LD and ER within close proximity. We have several pieces of evidence to support this. First, our biochemical fractionation analysis showed that NRZ proteins and ER-associated Q-SNAREs are present in LD fraction in the presence of Rab18. Second, fluorescent imaging analysis indicates that NRZ and Q-SNAREs are colocalized with Rab18 on the LD surface. Finally and most important, EM analysis using the APEX staining technique clearly demonstrated that in the presence of Rab18, ER membrane is in close contact with LDs to form special ER–LD contact sites. This is consistent with previous observation that overexpression of Rab18 causes an enwrapping of LD by ER membrane (Ozeki et al., 2005).

The cross talk between ER and LD in regulating LD growth and maturation has been observed in yeast, where nascent LDs seem to be constantly connected with the ER (Jacquier et al., 2011) in Caenorhabditis elegans (Xu et al., 2012b) and in hepatocytes where ApoB is shown to tightly associate with the LDs (Ohsaki et al., 2008). ER–LD contact is also observed in seipin-mediated LD growth (Salo et al., 2016; Wang et al., 2016). But the precise molecular nature of such contact remains unclear.

What is the precise mechanism of Rab18-NRZ-SNAREs in controlling lipid incorporation from ER to LDs for their growth? According to the crystal structure of Dsl1 and the speculated structure of NRZ tethering complex (Ren et al., 2009), we postulate that when Rab18-GTP binds to the ZW10 subunit that is localized on the top of the NRZ “rod,” it will bind to ER-associated SNAREs, resulting in the assembly of Rab18-NRZ-SNARE complex. Assembly of Rab18-NRZ-SNARE complex results in the tethering of ER membrane to LDs to create a direct ER–LD contact. In addition, Rab18-NRZ-SNARE complex assembly may provide energy for the subsequent deformation of ER bilayer and LD monolayer membrane, resulting in a direct transfer of TAG synthesized from ER to LDs at the ER and LD interface to promote nascent LD growth. Alternatively, Rab18-NRZ-SNARE complex may control phospholipid dynamics or composition in ER–LD interface to facilitate the distribution of TAG from ER to LDs (Fei et al., 2011; Ben M’barek et al., 2017). Finally, Rab18-NRZ-SNARE complex assembly may allow the recruitment of other proteins to the ER–LD contact site to form a special channel or pore to allow the incorporation of lipids from ER to LDs (Fig. 10 E). The delineation of the Rab18-NRZ-SNARE regulatory loop holds great promise for them to be molecular targets for developing therapeutic drugs for lipid storage–related diseases.

Full-length cDNAs encoding various mouse proteins were amplified by PCR from cDNAs of 3T3-L1 preadipocytes or mouse liver. cDNA encoding mouse Rab18 was cloned into pCMV5-HA (Clontech), pCMV5-Flag (Clontech), pEGFP-C1 (Clontech), pmCherry-C1 (Clontech), or pGEX-6P-1 (GE Healthcare) vectors. Point mutations of Rab18 were introduced by a PCR-based site-directed mutagenesis. LiveDrop LD marker was generated by cloning sequences encoding the hairpin domain of mouse GPAT4 into pEGFP-N1 (Clontech) vector, according to previous description (Wang et al., 2016). GFP-CB5 was generated by cloning sequences encoding the C-terminal transmembrane domain of mouse Cytb5 into pEGFP-C1 vector. cDNA encoding mouse GPAT4 was cloned into pEGFP-N1 vector. cDNAs encoding mouse GPAT3, DGAT1, and DGAT2 were constructed into pEGFP-C1 vector. cDNAs encoding mouse Rab3GAP1 and Rab3GAP2 were cloned into pCMV5-HA or pCMV5-Flag vectors. cDNAs encoding mouse ZW10 and RINT1 were cloned into pEGFP-C1, pCMV5-HA, or pCMV5-Flag vectors. Full-length cDNA encoding human NAG was amplified from cDNA libraries provided by Jiahuai Han (Xiamen University, Xiamen, China) and cloned into pCDNA 3.1(-)-Flag (Invitrogen), pCMV5-HA, or pEGFP-C1 vectors. Truncations of NAG encoding amino acids 734–1355 (Sec39 domain) and 1356–2371 were subcloned into pCDNA3.1(-)-Flag vector. cDNAs encoding Stx18 and Use1 were cloned into pEGFP-C1 vector. cDNA encoding mouse BNIP1 was cloned into pEGFP-N1 vector.

DGAT1 inhibitor (PF-04620110) was purchased from Selleck Chemicals. OAs, TG, free glycerol reagent, triglyceride reagent, isoproterenol, IBMX, DGAT2 inhibitor (PF-06424439), M2-agarose beads, glutathione, and 3×Flag peptide were purchased from Sigma-Aldrich. Protease inhibitor cocktail tablet and protein G agarose beads were purchased from Roche. Glutathione agarose beads were from GE Healthcare. Free fatty acid (FFA) uptake kit was from Abcam. Bodipy 493/503, Bodipy C12, LipidTOX neutral red, and LipidTOX deep red were from Life Technologies. Monodansylpentane (MDH) was obtained from Abgent.

Antibodies against Fsp27 were described previously (Sun et al., 2013a; Wu et al., 2014a). Antibodies against HA (mouse, sc-7392), Flag (mouse, sc-807), GFP (rabbit, sc-8334), GRP94 (rat, sc-32249), RINT1 (goat, sc-19404), Syntaxin18 (mouse, sc-293067), and BNIP1 (goat, sc-1713) were purchased from Santa Cruz Biotechnology. Antibodies against Rab10 (rabbit, 4262), FABP4 (rabbit, 3544) and Bip (rabbit, 3177) were obtained from Cell Signaling Technology. Antibodies against NAG (rabbit, ab122370; used for Western blot) and ZW10 (rabbit, ab21582; used for immunofluorescence [IF]) were obtained from Abcam. Antibodies against Rab3GAP1 (rabbit, HPA034495); Rab3GAP2 (rabbit, HPA026273), NAG (rabbit, HPA036817; used for IF), Use1 (rabbit, HPA047562), and Ykt6 (rabbit, HPA030817) were obtained from Human Protein Atlas. Antibodies against Rab18 (mouse, 60057–1-Ig; used for IF) and ZW10 (rabbit, 24561–1-AP; used for Western blot) were obtained from Proteintech. Antibody against ACSL3 (mouse, H00002181-B0) was obtained from Abnova. Antibodies against Rab18 (rabbit, SAB4200173, used for Western blot), Calnexin (rabbit, C4731), β-tubulin (mouse, T0198), and actin (mouse, A5441) were purchased from Sigma-Aldrich. Antibodies against ADRP (guinea pig, 20R-Ap002) and Plin1 (guinea pig, 20R-pp004) were purchased from Fitzgerald Industries. Antibody against GM130 (mouse, 610822) was obtained from BD Bioscience. Antibody against GST (mouse, AB101) was obtained from Tiangen. HRP-conjugated goat anti–mouse or rabbit IgG was obtained from GE Healthcare. Alexa Fluor 488/568/647–conjugated donkey anti–mouse or rabbit IgG was purchased from Life Technologies.

293T cells (CRL-3216; ATCC) and 3T3-L1 preadipocytes (CL-173; ATCC) were cultured in high-glucose DMEM (Life Technologies) containing 10% FBS (Life Technologies), 2 mM l-glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin at 37°C in a humidified incubator containing 5% CO₂. TM-3 Leydig cells were obtained from ATCC (CRL-1714) and cultured in high-glucose DMEM supplemented with 5% FBS and 2.5% horse serum. Monolayers of 3T3-L1 preadipocytes were induced to differentiate into mature adipocytes as described by the protocol of ATCC.

Plasmid DNAs were transfected into 293T cells using Lipofectamine 2000 according to the manufacturer’s instructions (Invitrogen). Electroporation of 3T3-L1 preadipocytes and mature adipocytes was performed using Amaxa Nucleofector II (Lonza) with program A-033 according to the manufacturer’s instructions.

siRNAs were introduced into 3T3-L1 preadipocytes by electroporation as described in the Cell culture and transfection section. Cells were then cultured for 48 h and harvested. The knockdown efficiency for each protein was evaluated by semiquantitative real-time PCR analysis or Western blot analysis. siRNAs used in all experiments are listed in Table S2.

Generation of KO cell lines was performed according to previous studies (Ran et al., 2013; Cao et al., 2016). In brief, single guide RNAs (sgRNAs) were designed using CHOPCHOP (https://chopchop.rc.fas.harvard.edu/) and constructed into pLentiCRISPRv1 vector (49535; AddGene). Lentivirus were produced by cotransfecting pLentiCRISPRv1 with packaging plasmids pMD2.G (12259; AddGene) and psPAX2 (12260; AddGene) into 293T cells. Purification of Lentivirus was performed as previously described (Nian et al., 2010). 3T3-L1 preadipocytes or TM-3 Leydig cells were infected with Lentivirus for 24 h, and were treated with 2 µg/ml of puromycin for 4–5 d to select positive clones. The drug-resistant cells were then diluted and plated on 96-well dishes to form a single colony. KO efficiency of each cell line was determined by Western blot analysis. Genomic edition was finally validated by sequencing the genomic sequence flanking the targeting sites that were amplified using PCR and subcloned to pEASY-T1 (Transgene). sgRNAs used for knocking out various genes are shown in Table S3.

For immunostaining, cells cultured on coverslips were washed twice with PBS, fixed with 4% paraformaldehyde for 20 min, and permeabilized with 0.1% saponin for 20 min. Cells were blocked with 10% goat serum in PBS supplemented with 0.1% saponin for 1 h and incubated with primary antibodies (1:50–1:200 diluted) in PBS supplemented with 0.1% saponin overnight at 4°C and then with secondary antibodies (1:500 diluted) for 1 h at room temperature. LDs were stained with Bodipy 493/503 (1:200 diluted), LipidTOX (1:300 diluted), or MDH (1:500 diluted) in PBS for another 20 min.

Images were collected on a Nikon A1R+ laser scanning confocal microscope with CFI Plan Apo 100× oil immersion objective (NA 1.45) and 405/488/561/640 laser. GaAsP multidetector photomultiplier tube was used to obtain images. For fixed cells, images were acquired at room temperature. Images with 1024 × 1024 pixels were obtained with 4 frames mean. For z-stack images, images (1,024 × 1,024 pixels) with 0.3 µm per section were obtained. Living cell imaging was performed according to a previous study (Sun et al., 2013b). Cells were cultured in a living cell station (Oko Lab) at 37°C, 5% CO₂, and 80% humidity. After loading with 200 µM OA (with 1:1,000 diluted Bodipy 493/503 or LipidTOX neutral red), images (512 × 512 pixels) were taken in 2-min intervals for 8 h. For 3D time-lapse imaging, optical sections were obtained with 0.3-µm thickness in every 30 min. NIS-Element software (Nikon) was used to obtain images.

Structured illumination microscopy (SIM) images were obtained using a Nikon N-SIM superresolution microscope with SR Plan Apo 100× oil immersion objective (NA 1.49) and Andor iXon 897 EMCCD at room temperature.

Cells were fixed in 2.5% gluteraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, for 1 h and washed with 0.1 M sodium cacodylate buffer. Cells were then postfixed with 1% OsO₄, 1.5% potassium ferrocyanide, and 1 mM CaCl₂ in 0.1 M sodium cacodylate buffer. Next, the cells were dehydrated in a graded series of ethanol, substituted with propylene oxide, and infiltrated with the SPON12 resin mixture followed by polymerization at 60°C for 24 h. Ultrathin sections were prepared using Lecia Ultracut R. Samples were observed using a Hitachi H-7650 transmission electron microscopy at 80 kV.

APEX-EM was performed according to the method described in Ariotti et al. (2015) and Lam et al. (2015). In brief, cells were washed 3 times with 0.1 M sodium cacodylate buffer, and then washed with 1 mg/ml DAB/cacodylate mixture for 2 min. Sodium cacodylate buffer supplied was 1 mg/ml DAB and 5.88 mM H₂O₂ were added to the cells for 30 min at room temperature. After that, the cells were washed three times with 0.1 M sodium cacodylate buffer and postfixed with 1% OsO₄ in 0.1 M cacodylate buffer for 2 min.

For quantitative analysis of the percentage of cells containing APEX-Stx18–labeled ER–LD contact, 22–39 cells/genotype were randomly selected. A cell that had at least one LD containing APEX-Stx18 signal on its surface was counted as positive.

For quantitative analysis of the percentage of LDs apposing to ER, 11–17 regions of interest (ROIs) from 11–17 cells were randomly selected, and 143–363 LDs were evaluated for wild-type, Rab18-overexpressing, and Rab18-deficient cells. An LD that was closely apposing to ER cisternae (within a distance of less than 20 nm) was counted as positive. For quantitative analysis of the percentage of LD surface area apposing to ER, 11–17 ROIs were randomly selected from 11–17 cells per genotype (143–363 LDs). The length of LD surface apposing to ER and LD circumference was measured using ImageJ-FIJI. For each LD in a ROI, the ratio of the length of LD-ER apposition to the LD circumference was calculated and averaged.

Images were processed in NIS-element or ImageJ. Deconvolution of fluorescent images in Fig. 5 C; Fig. S3 (A and I); Fig 6 (G and H); Fig. S4 D; Fig. 8 K; and Fig. 9 (B, D, and F) was performed in NIS-element. The diameters of the largest LD per cell were measured through NIS-element or ImageJ. LD numbers per field and the diameter of LDs in EM images were determined using NIS-element. 3D surface reconstitution of LDs was performed in Imaris 8.0 (Bitplanet). LDs stained with fluorescent dyes were detected in surface mode using background subtraction method, and the numbers, sizes, and volumes of LDs were automatically measured.

Cells were loaded with 200 µM OA for 20 h and harvested. Cells were scraped from dishes, washed with PBS, and lysed in lipid extraction solution (hexane/isopropanol 3:2, vol/vol). After centrifugation at 1,000 g for 5 min, the upper organic phase was collected and dried with nitrogen gas. Lipids were resuspended in chloroform containing 1% Triton X-100 and then dried with nitrogen gas. Lipids were finally resuspended in water. TAG level was measured using TAG reagent (Sigma-Aldrich) and free glycerol reagent (Sigma-Aldrich).

FFA uptake rate was measured using the Free Fatty Acid Uptake Assay kit (Abcam). In brief, 3T3-L1 preadipocytes were serum-starved for 1 h, and then incubated with FA dye-loading solution for 10 min. Fluorescence signals were measured using a microplate reader with FITC filter (PerkinElmer).

3T3-L1 preadipocytes were loaded with 200 µM OA and 1 µCi/ml [³H]-OA overnight. Cells were rinsed twice with opti-MEM and cultured in opti-MEM containing 1% FA–free BSA for 2 h at 37°C. FA released was extracted from medium and measured using a MicroBeta Jet (PerkinElmer) liquid scintillation counter.

For lipolysis assay of mature adipocytes, 8-d differentiated 3T3-L1 adipocytes were washed three times with PBS and then incubated in phenol red–free, serum-free DMEM (Gibco) containing 1% fatty acid–free BSA (Sigma-Aldrich) in the presence or absence of 10 mM isoproterenol and 10 mM IBMX for 4 h. The amount of glycerol released into culture medium was measured by free glycerol reagent (Sigma-Aldrich).

3T3-L1 preadipocytes were serum starved overnight, and then incubated with 200 µM OA and 1 µg/ml Bodipy C12 for 0, 30, 60, and 120 min before harvesting. Lipids were extracted using chloroform/methanol (2:1, vol/vol), loaded onto TLC plates, and separated with hexane/diethyl ether/acetate (70:30:1, vol/vol). The TLC plate was scanned by Typhoon trio+ (GE Healthcare) and the fluorescent intensity of TAG spots was measured in ImageJ.

GST-Rab18 was purified from E. coli strain BL21 (DE3). Bacteria that were transformed with pGEX-6p-1-Rab18 were grown in LB medium containing 100 µg/ml ampicillin to OD 0.6, and induced with 0.01 mM IPTG for 12 h at 16°C. Cells were harvested, resuspended, and then lysed in lysis buffer (25 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100, 1 mM PMSF, 1 mM DTT, and 10% glycerol) by sonication. Recombinant protein was first purified using Glutathione Sepharose 4B (GE Healthcare) according to the manufacturer’s instructions. Proteins were concentrated and further purified by gel filtration chromatography (Superdex 200 10/300 GL; GE Healthcare) in 20 mM Tris, pH 7.4, 100 mM NaCl, 1 mM MgCl₂, and 1 mM DTT.

Flag-ZW10 was purified from 293T cells. 8 µg of pCMV5-flag-ZW10 plasmid was introduced into each 10-cm dish of 293T cells for 24 h. Cells were scraped from dishes and washed with PBS. Cells were resuspended in lysis buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 1 mM EDTA, and 0.4% CHAPS), supplemented with protease inhibitor cocktail (Roche), and then broken by sonication. Cell lysates were spun at 12,000 g, and the supernatants were incubated with M2 beads for 3 h at 4°C. The beads were washed with washing buffer five times (20 mM Tris, pH 7.4, 700 mM NaCl, 5 mM EDTA, and 0.4% CHAPS), and the protein was eluted with elution buffer (20 mM Tris, pH 7.4, 100 mM NaCl, and 100 µg/ml 3×flag peptide).

Protein concentration was determined using the Bradfold method (Bio-Rad). The purity of protein was evaluated with Coomassie Brilliant Blue (Sigma-aldrich) staining or silver staining.

For IP, cellular protein was extracted with IP buffer (20 mM Tris, pH 7.4, 100 mM NaCl, 1 mM MgCl₂, and 0.4% CHAPS), supplemented with protease inhibitor cocktail (Roche). After being centrifuged at 12,000 g, the supernatants were incubated with M2 beads to pull down flag-tagged protein for 3 h at 4°C. The beads were washed with IP buffer three times before adding 1.2× laemmli loading buffer. Proteins co-IP were analyzed by Western blot. For IP of endogenous protein, antibody or preimmune serum premixed with protein G beads (Roche) was used.

For in vitro binding assay, 1.5 µg GST protein or 2.5 µg GST-Rab18 was loaded with 1 mM GTP-γS or 1 mM GDP in nucleotide loading buffer (20 mM Tris, pH 7.4, 100 mM NaCl, 1 mM MgCl₂, 5 mM EDTA, and 0.4% CHAPS) for 1 h at 4°C, followed by adding 10 mM MgCl₂ to stop reaction for 10 min on ice. Next, 2 µg flag-ZW10 prebound to M2 beads was added to form protein complex with GST-Rab18 for 2 h at 4°C. The beads were then washed with washing buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 10 mM MgCl₂, and 0.4% CHAPS) three times before the addition of 1.2× laemmli loading buffer.

Subcellular fractionation was performed according to the method described in Ding et al. (2013) and Gong et al. (2011) with slight modifications. TM-3 cells were harvested, washed with PBS, and resuspended in hypotonic buffer (10 mM Hepes, pH 7.8, 1 mM EGTA, and 25 mM KCl) for 20 min on ice. Cells were spun down at 600 g for 5 min, and the supernatants were discarded. Cells were resuspended with TES buffer (20 mM Tris, pH 7.4, 1 mM EDTA, and 250 mM sucrose) and homogenized through Dounce homogenizer (Kimble Chase) with loose pestle 25 times. The postnuclear supernatants (PNSs) were obtained after centrifugation at 1,000 g for 10min. PNS was further centrifuged at 12,000 g for 15 min to get postmitochondrial fraction. Floating buffer (100 mM NaCl, 20 mM Tris, pH 7.4, and 1.5 mM MgCl₂) was laid on postmitochondrial fraction and centrifuged at 200,000 g for 1 h using a Beckmann SW41 rotor. The pellet fraction and middle gradient were collected as crude microsomal fraction and cytosol, respectively. The LDs on the top were collected and washed three times with floating buffer. All procedures were performed at 4°C, and protease inhibitors were added. Protein concentration was determined using BCA assay (Thermo Fisher Scientific). Relative amounts of protein on each fraction were analyzed by Western blot.

Data were subjected to statistical analysis and plotted using Graphpad Prism 5.0. Statistical calculations to predetermine required sample size were not performed. The D’Agostino and Pearson omnibus normality test was used to determine the normal distribution of data. For data with normal distribution, single comparisons were performed using the two-tailed Student’s t test, whereas multiple comparisons were assessed by one-way ANOVA with the Tukey post hoc test. Dataset that did not follow normal distribution were analyzed using nonparametric tests (Mann-Whitney test for single comparisons or Kruskal-Wallis test for multiple comparisons). For all analyses, a p-value <0.05 was considered to be statistically significant. Results are reported either by mean ± SEM or mean ± SD as indicated in the figure legends. The methods for calculating p-values are indicated in the figure legends.

Fig. S1 shows defective LD growth and maturation in Rab18-deficient cells (related to Fig. 1). Fig. S2 shows that TAG synthesis on ER is required for Rab18-controled LD growth (related to Fig. 4). Fig. S3 shows that Rab3GAP1/2 controls the activity and LD localization of Rab18 (related to Fig. 5). Fig. S4 shows that Rab18 binds to NRZ tethering complex (related to Fig. 6). Fig. S5 shows that NRZ complex acts as a downstream effector of Rab18 to control LD growth (related to Fig. 7). Video 1 shows the time course of LD accumulation in control cells. Video 2 shows the time course of LD accumulation in Rab18 knocked down cells. Video 3 showstime-lapse 3D images of LD accumulation in control cells. Video 4 shows time-lapse 3D images of LD accumulation in Rab18 knocked down cells. Video 5 shows time-lapse 3D images of LD accumulation in control cells. Video 6 shows time-lapse 3D images of LD accumulation in NAG-deficient cells. Table S1 shows SNAREs identified from LD fraction isolated from hepatocytes of ob/ob mice. Table S2 shows siRNAs used in this work. Table S3 shows sgRNAs used in this work.

We thank members of the P. Li Laboratory at Tsinghua University for helpful discussion, and the Center of Biomedical Analysis at Tsinghua University and the Centre for Microscopy and Microanalysis at the University of Queensland for scientific and technical assistance.

This work was supported by grants from the National Natural Science Foundation of China (31690103 and 31430040) to P. Li, the National Key R&D Program of China (2016YFA0502002) to P. Li, the National Health and Medical Research Council, Australia (569542, 1045092, and 1037320), to R.G. Parton, and the National Natural Science Foundation of China (31621063) to P. Li. H. Yang is a Senior Research Fellow of the National Health and Medical Research Council, Australia.

The authors declare no competing financial interests.

Author contributions: D. Xu and L. Wu designed experiments. D. Xu, Yuqi Li, and L. Wu performed the majority of experiments with the help of D. Zhao, J. Yu, and T. Huang. D. Xu, Ying Li, C. Ferguson, and R.G. Parton performed the EM analysis. H. Yang and R.G. Parton participated in results discussion and manuscript preparation. P. Li supervised the entire project and was responsible for finalizing and submitting the manuscript.