Qa-SNARE syntaxin 18 mediates lipid droplet fusion with SNAP23 and SEC22B

Lipid droplets (LDs) are dynamic lipid storage organelles that can sense and respond to changes in systemic energy balance. The size and number of LDs are controlled by complex and delicate mechanisms, among which, whether and which SNARE proteins mediate LD fusion, and the mechanisms governing this process remain poorly understood. Here we identified a SNARE complex, syntaxin 18 (STX18)–SNAP23–SEC22B, that is recruited to LDs to mediate LD fusion. STX18 targets LDs with its transmembrane domain spanning the phospholipid monolayer twice. STX18–SNAP23–SEC22B complex drives LD fusion in adiposome lipid mixing and content mixing in vitro assays. CIDEC/FSP27 directly binds STX18, SEC22B, and SNAP23, and promotes the lipid mixing of SNAREs-reconstituted adiposomes by promoting LD clustering. Knockdown of STX18 in mouse liver via AAV resulted in smaller liver and reduced LD size under high-fat diet conditions. All these results demonstrate a critical role of the SNARE complex STX18–SNAP23–SEC22B in LD fusion.


Introduction
Lipid droplets (LDs) are ubiquitous organelles in eukaryotic cells.As cellular energy hubs, they play an important role in the maintenance of cellular homeostasis and their malfunction is highly related to fatty liver, liver fibrosis, diabetes, obesity, and other metabolic diseases [1][2][3] .Fatty acids as energy source for cells, are stored as triglycerol (TG) in LDs, as excess free fatty acids are toxic for the cells.Upon starvation, the fatty acids stored as TG are mobilized by lipolysis and released from LDs to provide energy, which leads to dynamic changes in the size, number, distribution, and mobility of LDs.
All LDs have very similar structures, but are unique from most other organelles wrapped by a phospholipid bilayer.LDs consist of a hydrophobic core of TG and sterol esters, which are enclosed by a phospholipid monolayer, with the hydrophilic phosphate ester head facing the cytoplasm, and the hydrophobic chain fatty acid tail facing the core 4 .There are a variety of proteins decorating the surface of LDs.They are attached to or inserted into the phospholipid monolayer through hydrophobic hairpins, amphipathic helices, or fatty acid modifications.The LD proteome is tightly regulated and closely related to the dynamics and functions of LDs 5,6 .
LDs are highly dynamic organelles, alternating between periods of growth and consumption.The biogenesis of LDs starts from the endoplasmic reticulum (ER) membrane, including the following steps: (1) neutral lipid synthesis and oil len formation; (2) LD budding from ER; (3) LD growth and maturation.LDs can expand through droplet-droplet fusion/coalescence, transfer of TG from ER or adjacent LDs, or TG synthesis on the LD surface 7 .
The fusion/coalescence of LDs facilitates the storage efficiency of TG.It also reduces the surface area of LDs and the risk of TG degradation by lipases.The cell deathinducing DFFA-like effector (CIDE) family of proteins, including CIDEA, CIDEB, and CIDEC (also known as FSP27), were reported to be enriched at LD-LD contact sites to mediate lipid transfer between LDs 8 .Homozygous mutation in the CIDEC/FSP27 has been found in human patients with fatty liver, hypertension, and insulinresistant diabetes 9 .CIDEC/FSP27 can form stable transorganelle oligomer to facilitate lipid transfer from small LDs to large ones in adipocytes 8,10 .Especially, for CIDEC/ FSP27, its gel-like condensation at LD-LD contact sites generates lipid-permeable plates for LD fusion 11 .However, so far reconstitution experiments with in vitro synthesized LDs and recombinant full-length CIDEC/FSP27 were still lacking and the detailed mechanism of how lipid-permeable plates were formed due to the condensation of CIDEC/FSP27 is unknown.
The process of membrane fusion is highly conserved in evolution.According to the classic theory, membrane fusion is driven by the zipping of SNAREs (soluble N-ethylmaleimide sensitive factor attachment protein receptors) into a four-helix bundle with the help of other proteins, such as Rab GTPase, SM proteins (Sec1/SM family proteins), tethering proteins and others 12 .SNAREs are membrane-anchored proteins via transmembrane domains (TMDs) or post-translationally modified hydrophobic tails (palmitoylation, etc.).All SNARE proteins contain evolutionarily conserved coiled-coil SNARE motifs.According to the amino acid that is present in the zero ionic layer of the SNARE domain, SNARE proteins are divided into R-SNAREs (with arginine in the zero ionic layer) and Q-SNAREs (with glutamine in the zero ionic layer) 13 .There are three Q-SNARE families: Qa, Qb and Qc.Usually, R-, Qa-, Qb-, and Qc-SNAREs form a four-helix bundle by spontaneously zippering from the N-terminus to the C-terminus of the SNARE domains.During this process, the trans-SNARE complex pulls and distorts the two membranes on which the proteins are anchored.With the help of other proteins, the membrane lipids are reorganized, and the membranes are eventually fused.
Despite being a core machinery of membrane fusion, whether SNAREs mediate LD fusion is not clear.Previously, it was reported that at the contact site between LDs and ER, ER-resident SNARE complex (syntaxin 18 (STX18)-USE1-BNIP1) can function together with NRZ complex (NAG-RINT1-ZW10) and Rab18 as a bridging complex to maintain this contact site 14 .In another study, it was reported that SNARE complex (STX5-SNAP23-VAMP4) can drive LD fusion 15 , but this hypothesis was under debate as no in vitro fusion evidence directly supports it.Additionally, SNAP23 was also reported to be involved in LD formation 16 .Recently, it was reported that SNAREs are involved in protein targeting from ER to LD, probably through a fused membrane bridge between LD and ER, but whether SNAREs mediate the LD-ER fusion is not demonstrated 17 .Besides, how SNARE proteins with TMD are located on phospholipid monolayer is elusive.Whether SNAREs are critical fusion machinery for LD fusion remains to be an important fundamental question to be answered.
In this work, we identified a new set of SNAREs, STX18-SNAP23-SEC22B, that mediates LD fusion.They are located on LDs and form a stable SNARE complex both in vivo and in vitro.The TMD of STX18 spans the phospholipid monolayer twice, with its C-terminal amino acids facing the cytosol.Both the FRET-based lipid mixing assay and content mixing assay indicate that STX18-SNAP23-SEC22B mediates the fusion between LDs.Besides, CIDEC/FSP27 interacts directly with STX18, SEC22B, and SNAP23, promoting LD fusion driven by SNAREs.Furthermore, mouse liver with STX18 knockdown shows smaller LDs upon highfat diet (HFD) feeding.Altogether, we demonstrate STX18-SNAP23-SEC22B as a key SNARE complex mediating LD fusion.

STX18, SNAP23, and SEC22B mediate LD fusion in cells
Distinct sets of SNARE complexes mediate different types of membrane fusion [18][19][20][21][22][23][24] .To identify the specific set of SNARE proteins that mediates LD fusion in liver cells, we utilized siRNA screening to knock down various SNAREs in HepG2 cells.We found that compared to the knockdown of other SNAREs, the knockdown of STX18 or SNAP23 significantly reduced the population of LDs bigger than 2 μm (Fig. 1a, b).Interestingly, STX5 and VAMP4, which were reported to mediate LD fusion in a previous study 15 , showed minimal effects on the population of large LDs (> 2 μm) in our siRNA screening performed in HepG2 cells (Fig. 1a; Supplementary Fig. S1a).BNIP1 and USE1 have also been reported to interact with STX18 in the tethering complex between the ER and LDs in 3T3-L1 cells 14 ; however, knockdown of BNIP1 or USE1 (Supplementary Fig. S1b) did not result in significant changes in the number of small LDs (0-0.5 μm and 0.5-1 μm), medium LDs (1-2 μm) and large LDs (> 2 μm) in HepG2 cells treated with oleic acid (OA) (Supplementary Fig. S1c, d), suggesting that STX18 might interact with a new set of SNAREs to mediate LD fusion.
To identify which R-SNARE forms a complex with STX18 (Qa-SNARE) and SNAP23 (Qbc-SNARE), we performed immunoprecipitation-mass spectrometry (IP-MS) of Flag-STX18 to search for its binding R-SNARE.According to the MS results, SEC22B (R-SNARE) and SNAP23 (Qbc-SNARE) were two of the top candidates as STX18 (Qa-SNARE) binding proteins (Supplementary Number Fig. S1e).Besides, the reverse tandem affinity purificationmass spectrometry (TAP-MS) of ZZ-Flag-SEC22B also indicated interaction among these SNAREs and their possibility to be a SNARE complex (Supplementary Fig. S1f).Furthermore, we verified the effect of STX18, SNAP23, and SEC22B depletion on LD size distribution in a histogram assay.Knockdown of STX18, SNAP23, or SEC22B by siRNAs significantly reduced the size of LDs upon OA treatment, as the population of small LDs increased while the population of large LDs decreased (Fig. 1b-d).
In addition to OA treatment, we also examined whether the depletion of these SNAREs affects the LD growth induced by different mechanisms.Previous study has indicated that overexpression of CIDEC/FSP27 promotes LD growth by facilitating lipid transfer between LDs 8 .We examined whether the depletion of these SNAREs would affect LD growth in this situation.As expected, the size of LDs in CIDEC/FSP27-expressing cells was big, ~0.5-3 μm, while much fewer LDs of ~1-3 μm were observed upon knockdown of STX18, SNAP23, or SEC22B even with overexpression of CIDEC/FSP27 (Fig. 1e, f).Altogether, these data suggest that STX18, SEC22B, and SNAP23 play an important role in LD size control.
LD size is not only influenced by LD fusion, but also likely by many other factors, including LD biosynthesis, growth and turnover, neutral lipolysis in neutral cytoplasm, and LD engulfment in acidic lysosomes.Next, we investigated whether the knockdown of STX18 reduces the LD size by affecting lipid biogenesis, lipolysis, and fatty acid uptake in HepG2 cells.We found that depletion of STX18 had no effect on mRNA levels of crucial genes involved in fatty acid synthesis (fasn) and LD biogenesis (dgat1, dgat2) (Fig. 1g; Supplementary Fig. S1g).Besides, we also measured the TG synthesis levels upon the depletion of STX18, SNAP23, or SEC22B in HepG2 cells.TG content is a reflection of both TG synthesis and lipolysis.To evaluate TG synthesis, we treated cells with both neutral lipase inhibitor (ATGLi) and lysosomal lipase inhibitor (LALi) to block TG degradation.TG levels were increased in siNC-treated HepG2 cells after the addition of ATGLi and LALi as expected, but knockdown of STX18, SNAP23, or SEC22B could not cause further changes in TG levels (Fig. 1h), suggesting that the TG synthesis levels remained unaltered when any of these SNARE proteins were depleted.These results indicate that the decreased large-sized LD population upon STX18, SNAP23, or SEC22B depletion, was likely caused by defects in LD fusion rather than decreased TG synthesis.Moreover, in a prior study by Xu et al. 14 , a decrease in TG levels was observed upon STX18 depletion in 3T3-L1 cells without inhibiting lipase activity.This finding supports the essential role of STX18 in lipid metabolism, which may depend on distinct cellular contexts and genetic backgrounds.
Moreover, neither neutral lipolysis (Fig. 1i) nor fatty acid uptake (Fig. 1j) was affected upon STX18 depletion.These results suggest that the knockdown of STX18 has little if any effect on lipid synthesis, neutral lipolysis, and fatty acid uptake.Furthermore, to exclude the possibility that small LDs are caused by autophagosomal engulfment of large LDs, we depleted the autophagy essential gene ATG7 in HepG2 cells.The inhibition of autophagy by ATG7 siRNA knockdown in STX18 KD HepG2 cells also showed decreased number of large LDs (> 2 μm), indicating that small LDs did not come from the degradation of large ones by autophagy (Fig. 1k, l).
Next, as the biogenesis of LDs starts from the ER membrane, we determined whether ER stress is activated in STX18-depleted cells.Our data showed that (see figure on previous page) Fig. 1 Identification of SNAREs STX18, SNAP23, SEC22B for LD size control.a HepG2 cells were transfected with control siRNA or siRNAs targeting SNARE genes for 48 h and then treated with 0.2 mM oleic acid-modified bovine serum albumin (OA-BSA) for 12 h.LDs were stained by BODIPY 558/568 C 12 for additional 1 h before fixing the cells.The graph shows the quantification of large LDs (> 2 µm) by analyzing the average LD number in 50 cells.b HepG2 cells were transfected with control, STX18, SEC22B or SNAP23 siRNAs for 48 h and then cells were analyzed via western blot.c HepG2 cells were transfected with control, STX18, SEC22B or SNAP23 siRNAs for 36 h and then treated with 0.2 mM OA-BSA for 12 h.LDs were stained by BODIPY 558/568 C 12 for additional 1 h before cell fixing.Cells were analyzed via fluorescence.Scale bars, 10 μm.d Quantification of LDs with different sizes per cell (> 50 cells) in c.Student's t-test; ***P < 0.001, ****P < 0.0001.e U 2 OS cells were transfected with control, STX18, SNAP23 or SEC22B siRNA for 24 h, and then cells were further transfected with GFP-FSP27 for another 24 h.LDs were stained by BODIPY 558/568 C 12 for additional 1 h before cell fixing."+" labeling in cells indicates overexpression of GFP-Fsp27; "-" labeling in cells indicates no expression of GFP-Fsp27.Scale bars, 10 μm (main images).f Quantification of GFP-FSP27-positive LD number per cell (> 50 cells).Student's t-test; ***P < 0.001.g The mRNA levels of dgat1, dgat2, and fasn in STX18 WT or KO HepG2 cells were analyzed via RT-qPCR.Student's t-test; NS, not significant.h TG content in HepG2 cells.Cells were transfected with siNC, siSTX18, siSNAP23 or siSEC22B.After 24 h, 30 μM ATGLi plus 50 μM LALi or DMSO was added into the cells.12 h later, 0.2 mM OA-BSA was added into the cells and kept for another 12 h.The concentration of TG was measured by the commercial kit, and then normalized to protein amount.Unpaired t-test; NS, not significant; ***P < 0.001.i The neutral lipolysis levels were analyzed in STX18 WT or KD HepG2 cells.Cells were transfected with siNC or siSTX18 for 45 h.After that, the cells were treated with isoproterenol for 3 h, followed by measurement of free glycerol released into the culture medium.NS not significant.Paired t-test.j The fatty acid uptake levels were analyzed in STX18 WT or KD HepG2 cells.Cells were transfected with siNC or siSTX18 for 48 h, followed by fatty acid uptake measurement.NS not significant.Unpaired t-test.k ATG7 WT or KD HepG2 cells were transfected with control or STX18 siRNA for 36 h and then treated with 0.2 mM OA-BSA for 12 h.LDs were stained by BODIPY 558/568 C 12 for additional 1 h before fixing the cells, and cells were analyzed via fluorescence.Scale bar, 10 μm.l Quantification of large LDs (≥ 2 µm) per cell (> 50 cells) in k.Student's t-test; ***P < 0.001.knockdown and knockout of STX18 in HepG2 cells failed to induce ER stress as evidenced by lack of induction of CHOP and Xbp1s expression, respectively (Supplementary Fig. S1h-j).Next, we examined the ER structure upon STX18 knockdown in HeLa cells by investigating the distribution and morphology of two ER markers, ER exit site marker Sec31A and ER membrane protein Bap31.The distribution and morphology of both ER proteins displayed no obvious difference between wild-type (WT) and STX18 KD cells (Supplementary Fig. S1k, l).Furthermore, as the perturbed ER homeostasis might change the synthesis of LDs, we examined the synthesis of nascent LDs by LiveDrop probe upon STX18 depletion.
LiveDrop is a widely used probe to label nascent LDs 25 .
The results showed that neither the area nor the number of LiveDrop puncta changed significantly in STX18 KD HepG2 cells (Supplementary Fig. S1m-o).Therefore, the depletion of STX18 did not change the synthesis of nascent LDs.All these results suggest that the reduction in LD size upon depletion of STX18 is not due to change in ER homeostasis.According to the above results, we presumed that STX18, SEC22B, and SNAP23 play an important role in LD size control likely through LD fusion.

STX18, SNAP23, and SEC22B form a SNARE complex
It is well known that Qa-, Qb-, Qc-and R-SNAREs form a four-helix bundle to drive membrane fusion; therefore, we next sought to determine whether STX18, SNAP23, and SEC22B form a SNARE complex both in vivo and in vitro.Both IP-MS of Flag-STX18 (Supplementary Fig. S1e) and TAP-MS of ZZ-Flag-tagged SEC22B (Supplementary Fig. S1f) indicated the interaction among STX18, SNAP23 and SEC22B.We first verified their interaction in a co-immunoprecipitation (co-IP) experiment, and the western blot results showed that overexpressed Flag-STX18 co-immunoprecipitated with overexpressed SEC22B and SNAP23 (Fig. 2a, b).Besides, we performed the endogenous co-IP without overexpression of these SNAREs, and the results showed that in HepG2 cells, the endogenous STX18 interacted with endogenous SNAP23 and endogenous SEC22B (Fig. 2c).As STX18 is composed of one TMD and one SNARE domain 26 (Fig. 2d), we next performed deletion mapping to determine which domains of STX18 are required for its binding with SEC22B and SNAP23 in a co-IP assay.STX18 mutants with deleted SNARE domain significantly reduced its interaction with SNAP23 or SEC22B (Fig. 2a, b), suggesting that SNARE domain interaction is required for the assembly of SNAREs in cells.Besides, deletion of TMD also reduced the interaction between STX18 and SEC22B or SNAP23 (Fig. 2a, b), which is probably due to its important role in the membrane localization of STX18.Furthermore, to verify the direct binding among STX18, SEC22B, and SNAP23, we performed pull-down assays with purified SNARE proteins.GST-STX18 directly bound SEC22B (Fig. 2e) and SNAP23 (Fig. 2f).Meanwhile, GST-SNAP23 interacted directly with SEC22B (Fig. 2g).Next we used more rigorous assays to evaluate the assembly and stoichiometry of the STX18-SNAP23-SEC22B SNAREpin complex.We co-expressed and co-purified four SNARE domains of STX18 (243-305), SEC22B (134-194), and SNAP23 (14-76 and 146-208) in a mini-SNAREpin complex.The size exclusion chromatography (SEC) showed that the co-purified proteins were eluted as a compact complex with molecular weight close to 43 kDa, which was about the sum of the molecular weight of these four SNARE domains, 40 kDa (Fig. 2h, i).The MS results of the purified complex also verified the presence of these four SNARE domains, and the coverage also fits the SNARE length in the whole primary sequence (Fig. 2j).We also tested whether the individually purified SNAREs assemble into a complex in vitro.When expressed individually, not every SNARE domain polypeptide is soluble, we have attempted different fragments and combinations.In the best combination, we used full-length SNAP23 instead of its SNARE domains.The SEC was applied to examine the assembly of STX18-SNARE (243-305), SEC22B-SNARE (134-194) and SNAP23.The SDS-PAGE of SEC fractions showed that at least a portion of (see figure on previous page) Fig. 2 STX18, SNAP23, and SEC22B form a SNARE complex.a HEK293T cells transfected with GFP-SEC22B and different Flag-STX18 constructs (Flag-STX18 WT, STX18 ΔTMD, or STX18 ΔSNARE) for 36 h, and then cells were subjected to Flag IP followed by western blot analysis.b HEK293T cells transfected with GFP-SNAP23 and different Flag-STX18 constructs (Flag-STX18 WT, STX18 ΔTMD, or STX18 ΔSNARE) for 36 h, and then cells were subjected to Flag IP and analyzed via western blot.c Interaction between STX18 and SNAP23, SEC22B verified by endogenous IP experiment in HepG2 cells.d Boundaries of different domains in STX18 (red), SNAP23 (green), and SEC22B (blue).The primary sequences of the C-termini of STX18 (331-DWYDS-335) and SEC22B (212-FWWL-215) were indicated.e Interaction between purified SEC22B and GST-Flag-tagged STX18 using an in vitro GST pull-down assay followed by immunoblot (IB) and Coomassie blue staining.f Interaction between purified SNAP23 and GST-Flag-tagged STX18 using an in vitro GST pull-down assay followed by IB and Coomassie blue staining.g Interaction between purified SEC22B and GST-Flag-tagged SNAP23 using an in vitro GST pull-down assay followed by IB and Coomassie blue staining.h Purification of mini-SNARE complex with His-SEC22B-SNARE, STX18-SNARE, His-SNAP23-SNARE1, and SNAP23-SNARE2 from E. coli.The His-tag affinity purification protein mixture was subjected to Superdex75 increase 10/300 GL and eluted as a single peak.i The elution fraction of purified mini-SNARE complex in h was assessed by SDS-PAGE.j The mini-SNARE complex bands excised from SDS-PAGE in h were assessed by MS. k The purified SNAP23, STX18-SNARE, SEC22B-SNARE or their in vitro incubated mixture were subjected to Superdex75 10/300 GL, followed by SDS-PAGE analysis.l Huh7 cells were transfected with Flag-STX18 for 36 h, treated with or without 0.2 mM OA-BSA for another 2 h, 6 h, 12 h, and then subjected to Flag IP followed by western blot analysis.
STX18-SNARE and most of SEC22B-SNARE changed their elution volume and co-migrate with SNAP23 to a fraction with high molecular weight (Fig. 2k).These results demonstrate that STX18, SNAP23, SEC22B are capable of assembling into a SNAREpin complex in vitro.Furthermore, we tested whether the assembly of this SNARE complex can respond to rapid TG biogenesis induced by OA treatment.The IP results showed that the interaction among STX18, SNAP23, and SEC22B can be obviously observed after OA treatment for 2 h and enhanced with time (Fig. 2l).Taken together, these data suggest that STX18, SNAP23, and SEC22B assemble into a SNARE complex, in response to LD biogenesis induced by OA treatment.

Targeting of SNAREs to LDs
We wondered whether STX18 executes its potential fusogenic activity on the surface of LDs.The localization of STX18 on LDs was visualized by confocal microscopy analysis.Upon OA treatment in HepG2 cells, GFP-tagged STX18 co-localized with LDs stained by Nile Red (Fig. 3a), and GFP-tagged SEC22B and SNAP23 also co-localized with LDs labeled by perilipin 2 (Plin2), a specific LDassociated protein (Supplementary Fig. S2a, b).Using electron microscope, we found that GFP-labeled STX18 decorated on the membrane of LDs upon OA treatment (Fig. 3b).
It is well-accepted that LDs comprise a hydrophobic core of TG and sterol esters, enveloped by a phospholipid monolayer.Notably, sequence analysis reveals that SEC22B's C-terminal region terminates with hydrophobic amino acids (212-FWWL-215) at the end of its TMD (Fig. 2d), fitting well with the hydrophobic LD environment.In contrast, STX18 has several hydrophilic amino acids (331-DWYDS-335) following its TMD (Fig. 2d), the topology of which is the next question we aimed to address.
To figure out the topology of STX18 on phospholipid monolayer of LDs, we first predicted the structure of TMD in STX18 by I-TASSER 27 .The results showed that its TMD can be one intact α-helix or two split short αhelices with a linker in between (Supplementary Fig. S2c).The length of both split α-helices is enough to span the phospholipid monolayer.To verify whether the TMD spans the phospholipid monolayer and to figure out how it spans the phospholipid monolayer, we designed a cysteine accessibility assay with adiposome (artificial LD) associated STX18 (Fig. 3c).The cysteine residues can be specifically labeled with fluorescence through reaction of their thiol groups with electrophile maleimide-linked fluorescence reagents if the cysteines are exposed but not buried in the membrane or the hydrophobic core of adiposomes.
Adiposomes are biochemically reconstituted structures with neutral lipid core covered by phospholipid monolayer membrane, resembling native LDs in terms of structure and composition 28 .We constructed adiposomes following previously published method 28 .Initially, we assessed the size of the adiposomes using Dynamic Light Scattering (DLS), revealing a diameter distribution centered around 180 nm (Supplementary Fig. S2d).Subsequently, we examined the freshly prepared adiposomes using both ultrathin transmission electron microscopy (TEM) and cryo-electron microscopy (cryo-EM) at lower magnifications.Both TEM (Supplementary Fig. S2e, f) and cryo-EM (Supplementary Fig. S2g, h) consistently demonstrated that the average size of the adiposomes was ~160 nm.It is worth noting that the quality and uniformity of these artificially constructed LDs closely resembled those previously reported in the literature 28 .
As STX18-SNARE-TMD (a.a.243-335) contains only one cysteine residue at the second predicted short α-helix in the TMD (Cys323) that could be labeled with fluorescence, to facilitate mapping, we also generated a mutant with an additional cysteine residue (Cys336) at the extreme C-terminus (Fig. 3c).If the TMD would not span the phospholipid monolayer, both fragments should be labeled with fluorescence dye.If the TMD only spans the phospholipid monolayer once, both cysteines should be buried and neither of them could be labeled.If the TMD spans the phospholipid monolayer twice with both N-terminus and C-terminus of STX18 facing cytosol, Cys323 would be buried and Cys336 would be exposed and labeled with fluorescence (Fig. 3c).STX18 Cys336 mutation imposes no effect on its secondary structure (Supplementary Fig. S2i).We then reconstituted these two STX18 fragments, STX18-SNARE-TMD and STX18-SNARE-TMD-Cys336 on adispomes, respectively, and their adiposome targeting was confirmed by co-flotation assay (Fig. 3d).The top layers containing the STX18-reconstituted adiposomes were collected (Fig. 3d, red box), and then subjected to the reaction with maleimide-linked Alexa Fluor 488 in the presence or absence of 2% Triton X-100.We verified the labeling efficiency by treating adiposomes with 2% Triton which can lyse adiposomes completely.Both STX18 fragments were labeled with Alexa Fluor 488 in the presence of 2% Triton X-100 (lanes 2 and 4 in Fig. 3e), indicating that the labeling reactions worked well.With the intact adipososmes in the absence of detergent, only STX18-SNARE-TMD-Cys (323 and 336 Cys) fragment was labeled (lane 3 in Fig. 3e), whereas STX18-SNARE-TMD was not labeled (lane 1 in Fig. 3e).These results support the model that TMD of STX18 spans the phospholipid monolayer twice, with its C-terminal tail exposed to the outside of adiposomes and the interval cysteine was buried in the membrane (Fig. 3c).This model also explains the topology of two hydrophilic residues (D334 and S335) at the C-terminal end of STX18.Collectively, these results indicate that STX18, SEC22B, and SNAP23 localize on LDs, and the TMD of STX18 spans the phospholipid monolayer twice.

STX18-SNAP23-SEC22B mediates lipid mixing and content mixing between adiposomes
To obtain direct evidence that the STX18-SNAP23-SEC22B complex mediates LD fusion, we first performed the SNARE-dependent adiposome lipid mixing assay in vitro (Fig. 4a).We reconstituted NBDlabeled adiposome with SEC22B-SNARE-TMD (134-215) and Rhodamine (Rhod)-labeled adiposome with STX18-SNARE-TMD (243-335) and SNAP23.The SDS-PAGE analysis indicated the successful incorporation of these proteins into adiposomes (Fig. 4b).The cryo-EM imaging displayed the intact morphology of fluorescence-labeled adiposomes as a sphere-shaped structure with a single electron-dense line around (Supplementary Fig. S3a) 29 .Lipid mixing was measured as the fluorescence emission (580 nm) of Rhod acceptor dyes arising from FRET upon excitation of NBD dyes with 460 nm light.The results revealed that STX18-SNARE-TMD, SNAP23, and SEC22B-SNARE-TMD together drove the lipid mixing between adiposomes, but without SNAP23, this lipid mixing was greatly decreased (Fig. 4c, d).Besides, both cryo-EM (Supplementary Fig. S3a, b) and DLS (Fig. S3c) showed that adiposome size was increased after mixing the SNAREsreconstituted adiposomes together, which is in accordance with their roles in LD fusion.Furthermore, we also examined the lipid mixing between adiposomes decorated by full-length STX18, SNAP23, and SEC22B.The SDS-PAGE analysis indicated that these proteins were incorporated into adiposomes (Fig. 4e).The protein-to-lipid ratios were measured: STX18:phospholipid is ~1:474, SEC22B:phospholipid is ~1:427 and SNAP23:phospholipid is ~1:215.The FRET assay showed that the lipid mixing between these adiposomes can be driven by full-length SNAREs (Fig. 4f,  g).Besides, this lipid mixing was abolished in the absence of any of these SNAREs (Fig. 4f, g).Taken together, these results demonstrate that STX18, SNAP23 and SEC22B together directly drive the lipid mixing of adiposomes.
As the lipid mixing assay does not show the core TG exchange between adiposomes directly, we further performed the in vitro FRET-based content mixing assay of adiposomes (Fig. 4h).We reconstituted STX18-SNAP23adiposomes encapsulated with Cy5-labeled DAG, and SEC22B-adiposomes encapsulated with Cy3-labeled DAG.The content mixing between adiposomes can be monitored by FRET built up between Cy5-DAG and Cy3-DAG.The SDS-PAGE showed successful reconstitution of these SNARE proteins on the adiposomes (Fig. 4i).The content mixing assay showed that in the presence of a full set of STX18, SEC22B and SNAP23, the adiposomes fused, but not in the reaction missing SNAP23 (Fig. 4j, k).
Altogether, our experiments including the lipid mixing assay and content mixing assay demonstrate that STX18-SNAP23-SEC22B drives membrane fusion of adiposomes.

CIDEC/FSP27 promotes lipid mixing between adiposomes driven by STX18-SEC22B-SNAP23
Previously, it was reported that CIDEC/FSP27 localized at LD contact sites and facilitated the LD fusion/coalescence.Its C-terminal sequence (a.a.179-217) is responsible for LD targeting 8 , while its CIDE-N domain (a.a.39-119) forms homodimers and defect of this homodimerization reduced CIDEC/FSP27's activity in promoting the large LD formation 30 .We investigated the interplay between CIDEC/FSP27 and LD SNAREs in LD fusion.We first examined the interaction between CIDEC/FSP27 and SNAREs.Flag-STX18 interacted with GFP-FSP27 in HepG2 cells and this interaction was significantly enhanced upon OA treatment (Fig. 5a).Next, we tested whether CIDEC/FSP27 co-localized with STX18 on LDs in HepG2 cells.Without OA treatment, FSP27-mCherry localized on LDs, but STX18-GFP did not; thus we barely detected the co-localization between STX18 and CIDEC/FSP27.However, upon OA treatment, STX18-GFP was enriched on LDs, and co-localized well with FSP27-mCherry on LDs stained by LipidTox Deep Red (Fig. 5b).Moreover, we purified CIDEC/FSP27 to examine its direct binding with SNAREs.The pull-down (see figure on previous page) Fig. 3 Targeting of SNAREs to LDs. a HepG2 cells were transfected with STX18-GFP and then treated with 0.2 mM OA-BSA for 6 h.LDs were stained by Nile Red for 1 h after cell fixing.Cells were analyzed via fluorescence.Scale bar, 10 μm.b HepG2 cells transiently expressing STX18-GFP were treated with 0.2 mM OA-BSA for 12 h, and cells were analyzed via immunogold electron microscopy.The anti-GFP primary antibody and goldconjugated secondary antibody were used to detect GFP signals.Red arrowheads indicate STX18-GFP.Scale bar, 200 nm.c Cysteine accessibility of STX18 with predicted orientations on adiposomes, where the TMD spans phospholipid monolayer twice.The cartoon showed that the buried inherent Cys323 (blue hexagon) was not labeled by Alexa Fluor 488 C5 maleimide dye but the exposed artificially engineered Cys336 (cyan hexagon) can be labeled by Alexa Fluor 488 C5 maleimide dye.However, the treatment of 2% Triton X-100 made both Cysteine residues exposed and labeled by Alexa Fluor 488 C5 maleimide dye.Successful labeling of cysteine by AF488 was indicated by green stars.d Purified STX18-SNARE-TMD (Cys323) and STX18-SNARE-TMD-Cys (Cys323, Cys336) were reconstituted on adiposomes followed by co-flotation analysis.experiments showed that the full-length CIDEC/FSP27 bound to GST-Flag-STX18, GST-Flag-SEC22B, and GST-Flag-SNAP23 (Fig. 5c).Furthermore, we purified CIDEC/ FSP27 to assess its function in lipid mixing assay.CIDEC/ FSP27 alone failed to drive lipid mixing of naked adiposomes, but CIDEC/FSP27 made lipid mixing of SNAREsreconstituted adiposomes more efficient (Fig. 5d, e).By analyzing CIDEC/FSP27 purified from E. coli, we found that the purified CIDEC/FSP27 fraction is a mixture of full-length CIDEC/FSP27 and a truncated form which turns out to be its N-terminal fragment (1-150) by MS analysis (Supplementary Fig. S4a).To investigate the possibility that CIDEC/FSP27 (1-150) plays a role in lipid mixing of adiposomes, we cloned and purified the CIDEC/FSP27 (1-150) from E. coli.The results showed that no significant enhancement effect was observed when we used purified CIDEC/FSP27 (1-150) protein in the lipid mixing assay (Fig. 5f, g).We used DLS to examine the tethering function of CIDEC/FSP27 purified from E. coli on adiposomes, and interestingly, we found that adiposomes clustered without SNAREs, and this clustering effect was reversed with the addition of proteinase K (Supplementary Fig. S4b, c).Together, these results suggest that CIDEC/FSP27 could promote LD fusion driven by STX18-SNAP23-SEC22B by tethering adiposomes.

STX18 mediates LD growth in mouse liver
STX18-SNAP23-SEC22B SNARE complex displays fusogenic activity in vitro, which prompted us to investigate their roles as a fusion machinery for LD coalescence in vivo.We first used live-cell imaging to document the fusion events in WT HepG2 cells, STX18 KO cells, SEC22B KD cells, and SNAP23 KD cells, respectively.Our results showed that compared to WT cells, fusion events were much less in cells depleted of STX18, SNAP23, or SEC22B upon OA treatment up to 12 h (Fig. 6a, b).Next, we investigated whether depletion of STX18 affects LD acumulation in mouse liver.Mice were injected with AAV carrying shRNA against STX18 or mock-treated for 4 weeks to knock down STX18 in the liver (Fig. 6c), and the mice were further fed with HFD for another 7 weeks.No significant difference was observed in the body weight between the control and STX18-knockdown group (Fig. 6d), but knockdown of STX18 reduced the liver weight (Fig. 6e) and TG level (Fig. 6f).Importantly, the knockdown of STX18 resulted in smaller LDs under HFD condition (Fig. 6g, h).Given that STX18 depletion in HepG2 cells has no effect on lipid synthesis (Fig. 1h), the decrease of TG in mouse liver is likely due to defects in LD fusion.Besides, previous reports have suggested that LDs with smaller sizes are more susceptible to lipophagic internalization 31 .Consequently, the depletion of STX18 in the mouse liver resulted in the reduction of LD size, liver weight and TG levels, possibly due to increased lysosomal uptake and degradation of these smaller LDs following STX18 depletion.All these results indicated a critical role of STX18 in LD fusion, and STX18 depletion antagonizes LD growth upon HFD feeding.

Discussion
SNARE proteins are proven to be the core of fusion machinery in many membrane fusion events; however, the fusion machinery in LD coalescence was not fully demonstrated before.In the present study, we identified a new set of SNAREs, STX18-SNAP23-SEC22B, to be essential for LD fusion.This notion is supported by the following evidence: (1) STX18, SNAP23, and SEC22B were identified in unbiased RNAi screening or MS proteomic (see figure on previous page) Fig. 4 STX18-SNAP23-SEC22B mediated adiposome fusion.a Scheme of FRET-based adiposome lipid mixing assay.b SDS-PAGE analysis of Rhod-labeled adiposome reconstituted with STX18-SNARE-TM and SNAP23, and NBD-labeled adiposome reconstituted with SEC22B-SNARE-TM, followed by Coomassie blue staining.c Lipid mixing between adiposomes reconstituted with the indicated SNAREs was measured from the development of FRET between NBD-PE and Rhod-PE.The NBD-adiposomes were reconstituted with SEC22B-SNARE-TM, while the RhoD-adiposomes were reconstituted with STX18-SNARE-TM and SNAP23.In the complete reaction set, the two adiposomes were mixed with extra SNAP23.For the incomplete reaction set, RhoD-Adi-STX18-SNARE-TM was used to replace RhoD-Adi-STX18-SNARE-TM&SNAP23.No extra SNAP23 was included in the incomplete reaction set.Both reactions were recorded at 37 °C.d Quantification of lipid mixing experiments of c. Bars represent averages of the FRET signal observed after 3000 s in three independent experiments.Error bars represent SDs.Paired t-test; *P < 0.05.e SDS-PAGE analysis of Rhod-labeled adiposome reconstituted with STX18 and SNAP23, and NBD-labeled adiposome reconstituted with SEC22B, followed by Coomassie blue staining.f Lipid mixing between adiposomes reconstituted with the indicated SNAREs was measured from the development of FRET between NBD-PE and Rhod-PE.The NBD-adiposomes were reconstituted with or without SEC22B, while the RhoD-adiposomes were reconstituted with or without STX18 and SNAP23.The reactions were recorded at 37 °C.g Quantification of lipid mixing experiments of f.Bars represent averages of the FRET signal observed after 3000 s in three independent experiments.Error bars represent SDs.Unpaired t-test; **P < 0.01.h Scheme of FRET-based adiposome content mixing assay.i SDS-PAGE analysis of reconstituted STX18-adipsomes encapsulated with Cy5-DAG, STX18-SNAP23-adiposomes encapsulated with Cy5-DAG, and SEC22B-adiposomes encapsulated with Cy3-DAG, followed by Coomassie blue staining.j Content mixing between adiposomes reconstituted with the indicated SNAREs was measured from the development of FRET between Cy3-DAG and Cy5-DAG.Both STX18-adipsomes and STX18-SNAP23-adiposomes were encapsulated with Cy5-DAG, while SEC22B-adiposomes were encapsulated with Cy3-DAG.In the complete reaction set, extra SNAP23 was added into the reaction.The reactions were recorded at 37 °C.k Quantification of content mixing experiments of j.Bars represent averages of the FRET signal observed after 7200 s in three independent experiments.Error bars represent SDs.Unpaired t-test; **P < 0.01.We observed a pivotal role of STX18-SNAP23-SEC22B in LD size control, LD targeting, and LD fusion in HepG2 cells, but we still cannot exclude the possibility that SNAREs other than this set are also involved in LD fusion.Besides, SNAREs might regulate different aspects of LD dynamics depending on the cell type.For instance, in STX18 KO 3T3-L1 preadipocytes, both small and medium/ large LDs are reduced, implying that STX18, in conjunction with BNIP1 and USE1, might play a role in LD biogenesis in these cells 14 .Moreover, previous studies reported that depletion of Sec22b has moderate effect on LD morphology in 3T3-L1 cells 14 and HeLa cells 32 .In our study in HepG2 cells, STX18 translocates to LDs where it complexes with SEC22B and SNAP23 to mediate LD fusion.Whether STX18-SNAP23-SEC22B mediates LD fusion in other tissues apart from the liver still needs further investigation.
STX18 was reported as a ER-resident t-SNARE involved in targeting and fusion of Golgi-derived retrograde transport vesicles with the ER 33 and also as component of the tethering complex between ER and LDs 14 .We found that a significant portion of STX18 also localizes on LDs.Different from most organelle membrane, LD membrane is composed of single leaflet with the lipid hydrophobic tail embedded in the hydrophobic TG core of LD.How STX18 folds on LD membrane is important for its fusogenic activity.We figured out the topology of STX18 on in vitro reconstituted adiposomes.The cysteine accessibility assay suggests that STX18 spans the phospholipid monolayer twice, with its N-terminal soluble part and C-terminal hydrophilic residues facing cytosol.It is well known that the thickness of phospholipid bilayer is ~3-4 nm, and the thickness of phospholipid monolayer is therefore ~1.5-2 nm.There are 23 residues in TMD of STX18 (311-333) and two hydrophilic residues (334D and 335S) following the TMD.The I-Tasser program predicted that the TMD of STX18 can be two split short α-helices with a flexible linker in between.As each amino acid residue in an α-helix corresponds to a translation of 1.5 Å along the helical axis, these short α-helices with 10 amino acids predicted by I-Tasser, should be long enough to cross the phospholipid monolayer.The cysteine accessibility results (Fig. 3e) proved our hypothesis that TMD of STX18 spans the phospholipid monolayer twice, with both N-terminal and C-tail hydrophilic residues facing the cytosol.Besides, STX18 with C-terminal fused GFP is efficiently targeted to LD (Fig. 3a, b), which also suggests that the tail of STX18 is exposed to cytosol rather than facing the hydrophobic core of LD in cells.
Although SNARE proteins constitute the minimum fusion machinery, the tethering factors and the accessory proteins are also crucial to bridge two membrane compartments and boost fusion efficiency and precision.Previously, CIDEC/FSP27 was reported to locate at LD contact sites and promote LD growth.Its C-terminal sequence (a.a.179-217) is responsible for LD targeting 8 , while its CIDE-N domain (a.a.39-119) forms homodimers 30 .Therefore, CIDEC/FSP27 possibly serves as a membrane tether in LD fusion.In the present study, we showed that CIDEC/FSP27 interacted with LD SNAREs (Fig. 5a, c), and co-localized with STX18 upon OA treatment (Fig. 5b).Besides, CIDEC/FSP27 promoted LD lipid mixing driven by STX18-SNAP23-SEC22B (Fig. 5d, e).
(see figure on previous page) Fig. 5 CIDEC/FSP27 promotes lipid mixing between adiposomes driven by STX18-SEC22B-SNAP23. a HEK293T cells were transfected with Flag-STX18 and GFP-FSP27 for 36 h, and then treated with or without 0.2 mM OA-BSA for 12 h.The cells were subjected to Flag IP followed by western blot analysis.b HepG2 cells were transfected with STX18-GFP and mCherry-FSP27 for 48 h, and then treated with 0.2 mM OA-BSA for 6 h.The LDs were stained by LipidTox Deep Red for 1 h after fixing the cells and the cells were analyzed via fluorescence.Scale bars, 10 μm (main images), 5 μm (inset).c Interaction between purified CIDEC/FSP27 and GST-Flag-tagged SNAREs using an in vitro GST pull-down assay followed by western blot.d Lipid mixing between adiposomes was measured from the development of FRET between NBD-labeled lipids and Rhod-labeled lipids.The NBD-adiposomes were reconstituted with SEC22B or not, while the RhoD-adiposomes were reconstituted with STX18, SNAP23 or not.The assays were performed in the presence of CIDEC/FSP27 or not.The reactions were recorded at 37 °C.e Quantification of lipid mixing experiments of d.Bars represent averages of the FRET signal observed after 3000 s in three independent experiments.Error bars represent SDs.Unpaired t-test; *P < 0.05, NS, not significant.f Lipid mixing between adiposomes was measured from the development of FRET between NBD-labeled lipids and Rhod-labeled lipids.The NBD-adiposomes were reconstituted with SEC22B or not, while the RhoD-adiposomes were reconstituted with STX18, SNAP23 or not.The assays were performed in the presence of CIDEC/FSP27 (1-150) or not.The reactions were recorded at 37 °C.g Quantification of lipid mixing experiments of f.Bars represent averages of the FRET signal observed after 3000 s in three independent experiments.Error bars represent SDs.Unpaired t-test; NS, not significant.Moreover, CIDEC/FSP27 has no fusogenic activity by itself (Fig. 5d, e) but can promote adiposome clustering without SNAREs (Supplementary Fig. S4b, c).Furthermore, CIDEC/FSP27-promoted LD fusion could be abolished by depletion of STX18, SNAP23, or SEC22B in vivo (Fig. 1e,  f).Therefore, we propose a model in which CIDEC/FSP27 dimer tethers LDs/adiposomes to close proximity and promotes STX18-SNAP23-SEC22B-mediated membrane fusion (Fig. 7a).It is worth noting that we did not exclude the possibility that CIDEC/FSP27 might have functions other than membrane tether in LD fusion.Li's group recently proposed that the gel-like condensation of CIDEC/FSP27 generates lipid-permeable plates for LD fusion 11 .Whether CIDEC/FSP27 provides a hub to recruit SNAREs and other regulators to this plate and function together for fusion will be an attractive model to test.
In addition to the fusion machinery on the LD surface, the lipid compositions may also contribute to LD fusion.It is reported that PC inhibits fusion by lowering surface tension, and PA promotes fusion by increasing membrane curvature 6,34 .How STX18-SNAP23-SEC22B cooperates with lipid components on LDs will be further investigated.Furthermore, we showed that STX18 not only affects LD fusion on synthetic adiposomes, but also significantly influences LD size in cultured cells and mouse livers.Targeting this fusion machinery could be a promising strategy to counter against liver with extra lipid accumulation as well as many other diseases or pathological conditions with aberrant lipid metabolism.So far, we only included SNAREs and tethering proteins in the adiposome membrane fusion system.However, for classic membrane fusion theory, Rab GTPase, tethering proteins, SM proteins, NSF/αSNAP, lipid compositions, and membrane curvature all play a role in guaranteeing the specificity and efficiency of the membrane fusion.With this basic fusion machinery in place, we can start to dissect the complex and delicate regulatory mechanism of LD fusion.

RT-qPCR
Total RNA was isolated from cells and frozen tissues by homogenization in Trizol.Reverse transcription was performed using M-MLV reverse transcription reagents (Promega).cDNA was subjected to qPCR analysis. The

Immunofluorescence analysis
Cells were rinsed in PBS, fixed with 4% paraformaldehyde (PFA) for 15 min, permeabilized with 0.1% saponin in PBS for 10 min, blocked with 10% FBS in PBS for 30 min, followed by incubation with primary antibody for 2 h, washed three times with PBS, and incubated with fluorescently labeled secondary antibody for 1 h.Cells were then washed with PBS three times.ProLong TM Diamond Antifade Mountant with DAPI (Thermo Fisher Scientific, P36962) was used to stain the nucleus.Images were acquired on a confocal laser microscope (Zeiss LSM880 Airyscan) using a 60× oil-immersion objective lens.Images were processed with ImageJ software.The microscopy data are quantified in a blinded manner.

LD staining
Cells were treated with 0.2 mM OA-BSA for the indicated time, and then LDs were stained by BODIPY 493/ 503 or BODIPY 558/568 C 12 for 1 h at 37 °C with 5% CO 2 before being fixed with 4% PFA.Cells were also stained by Nile red or lipidTox Deep red for 0.5 h if required.

Ultrathin section
The fresh adiposomes were quickly mixed with gelatin and solidified on ice.The solidified samples were cut into blocks of ~1 mm 3 .The blocks of adiposomes were fixed with 2.5% glutaraldehyde at room temperature (RT) for 1 h.After the blocks were washed three times with PB buffer, they were fixed with 1% osmic tetroxide at RT for 2 h.After washing three times with PB buffer, the fixed blocks were dehydrated with ethanol and acetone, and then embedded in resin.The blocks were polymerized at 37 °C for 8 h, then at 65 °C for 48 h, and subjected to untrathin slicing at 70 nm.Finally, ultrathin sections were stained with uranyl acetate for 7 min.Micrographs were recorded on a transmission electron microscope (Thermo Fisher/FEI Talos L 120 C).

Immunoelectron microscopy
The HepG2 cells transfected with plv-STX18-GFP were cultured on 3 mm sapphire discs.A Sapphire disc was placed with cells facing up on a flat aluminum planchette and another aluminum planchette with 25-µm depth inner space was used as a cover.The spaces between the two aluminum planchettes were filled with 1-hexadecane.Then the samples were frozen immediately using the EM ICE high-pressure freezing machine (Leica) and rapidly transferred into liquid nitrogen for storage.After all of the samples were frozen, the samples were transferred into the EM ASF2 (Leica) for substitution.Samples were incubated for 48 h in acetone containing 0.2% UA at -90 °C.Then the temperature was raised to -50 °C for 4 h.After incubation in acetone containing 0.2% UA for another 12 h, the temperature was raised to -30 °C for 4 h.After 2 h incubation at -30 °C, the samples were rinsed three times with pure acetone (15 min each).Then the samples were gradually infiltrated in HM20 resin with grades of 25%, 50%, 75% and pure resin (1 h each) at -30 °C.After being infiltrated in pure resin overnight, the samples were embedded in gelatin capsules.The samples were polymerized under UV light for 48 h at -30 °C and 12 h at 25 °C.After polymerization, the samples were trimmed and ultrathin sectioned with a microtome (Leica UC7).Serial thin sections (100 nm thick) were collected onto formvarcoated nickel grids.The formvar-coated nickel grids with sections were incubated in 0.01 M PBS containing 1% BSA, 0.05% Triton X-100, and 0.05% Tween 20 for 5 min.Then the sections were incubated with the primary antibody (rabbit anti-GFP) diluted in 0.01 M PBS containing 1% BSA and 0.05% Tween 20 at 4 °C overnight.After being washed 6 times (2 min each) with 0.01 M PBS, the sections were incubated with the secondary antibody (goat anti-rabbit conjugated with 10 nm gold) diluted in 0.01 M PBS containing 1% BSA and 0.05% Tween 20 (1:50) for 2 h at RT.After being washed 6 times (2 min each) with 0.01 M PBS and 4 times (2 min each) with distilled water, the section was dried at RT and examined under a transmission electron microscope (Thermo Fisher/FEI Talos L 120 C).

Lipolysis (3T3-L1) colorimetric assay kit
In 6-cm culture dish, the HepG2 cells were treated with siNC or siSTX18 for 45 h.Then cells were washed three times with PBS and incubated in phenol red-free, serumfree F12 medium (Gibco) containing 1% fatty acid-free BSA in the presence or absence of 500 nM isoproterenol (Sigma-Aldrich, 1351005) for 3 h.The amount of glycerol released into culture medium was measured by commercial kit, Free Glycerol Reagent (Sigma-Aldrich, F6428).

Fatty acid uptake assay
The HepG2 cells were treated with siNC or siSTX18 for 48 h.At least 10,000 cells per well were cultured in 96well plate.Fatty acid uptake level was analyzed via Fatty Acid Uptake Assay Kit (MAK156) from Sigma.

Measurement of LD fusion rates in HepG2 cells
HepG2 cells were grown on glass-bottom confocal dishes, and were transfected with siNC, siSEC22B, or siSNAP23 for 48 h.Cells were further treated with 0.2 mM OA-BSA for 2 h and stained with BODIPY 493/ 503 for another 30 min.Cells were imaged and captured by Olympus FV3000 Confocal Microscope with cell culture system.The dishes were monitored and addressed by the Z drift compensator system.Images were acquired using a 60×/1.4oil objective with Z series and time-series.Time lapses was performed with 30-s interval for 120 cycles.Cells were imaged by using the 488 nm for BOD-IPY 493/503.Acquired images were processed with software FV31S-DT to analyze the fusion events.The LD fusion events were quantified from 25 cells of three biological replicates.

Animal experiments
Male C57BL/6 mice (8 weeks of age) were housed under a 12 h light/dark cycle with access to standard rodent chow diet (2916, Teklad) and water.AAVs (serotype 8, transduce liver) were custom packaged at SignaGen Laboratories.The mouse STX18 shRNA sequence for AAV preparation is TTCTTGTCCAC-CACGCGCT. STX18-shRNA or GFP-shRNA AAV was delivered by tail vein injection with vector titers at 1 × 10 10 genomic copies/μL.Mice injected with 10 μL AAV vectors were studied at least 4 weeks after AAV injection to achieve maximal knockdown efficiency of STX18.Mice were further fed with HFD (60% kcal from fat, Bio-Serv) for 7 weeks.All animal studies and experimental procedures were approved by the Animal Care and Use Committee of the animal facility at UT Southwestern Medical Center.

Immunohistochemistry
Livers were excised and fixed overnight in 10% PBSbuffered formalin.Then tissues were rinsed with 50% ethanol three times, embedded in paraffin blocks by the University of Texas Southwestern Medical Center Molecular Pathology Core, and sliced 5 μm each section.The slices were rinsed 3 times for 5 min each in PBS, pH 7.4, and trichrome staining was processed by University of Texas Southwestern Medical Center Molecular Pathology Core via standard procedure.
Cloning, expression, and purification of full-length STX18, SEC22B, and SNAP23 Full-length STX18 was cloned into pGEX-4T-1 vector between BamHI and NotI, with a GST-TEV-3×FLAG tag at its N-terminal sequence.The construct was transformed into E. coli BL21 (C43) cells.The target protein expression was induced by 0.1 mM IPTG at 16 °C for 16 h.The harvested cells were lysed at lysis buffer (20 mM HEPES, pH 7.4, 500 mM NaCl, 1 mM EDTA) supplemented with Protease Inhibitor Cocktail (APExBIO) and centrifuged at 8000 rpm in JA 14 rotor (Beckman Coulter) for 25 min to remove inclusion bodies and cell debris.The supernatant was collected and re-centrifuged at 40,000 rpm in Ti45 rotor (Beckman Coulter) for 2 h.The pellet containing the membrane was homogenized by Dounce homogenizer (KONTES GLASS) in lysis buffer.The membrane proteins were extracted by solubilizing the membranes with 2% Dodecylmaltoside (Anatrace) and then the samples were centrifuged at 40,000 rpm in Ti45 rotor for 60 min.The supernatant was incubated with Glutathione Sepharose 4B (GE Healthcare) overnight, and then the resins were washed by wash buffer (20 mM HEPES, pH 7.4, 500 mM NaCl, 1 mM EDTA, 0.1% Fos-Choline-12).The protein was treated with TEV protease on the beads at 4 °C for 2 h, and then eluted by elution buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 1 mM EDTA, 0.1% Fos-Choline-12).The full-length SEC22B was cloned into pGEX vector between EcoRI and XhoI with a GST tag at its N-terminus and a TEV cleavage site following GST tag.The expression and purification of SEC22B are the same as that of full-length STX18.The full-length SNAP23 was cloned into pGEX vector between EcoRI and XhoI with a GST tag at its N-terminus and a TEV cleavage site following GST tag.The construct was transformed into E. coli BL21 (DE3) cells.Overexpression of the protein was induced by the addition of 0.1 mM IPTG at 16 °C for 16 h.The harvested cells were lysed by high pressure homogenizer in lysis buffer (20 mM HEPES, pH 7.4, 500 mM NaCl, 1 mM EDTA) supplemented with Protease Inhibitor Cocktail (APExBIO) and centrifuged at 40,000 rpm in Ti45 rotor for 60 min.The cleared supernatant was incubated with Glutathione Sepharose 4B (GE Healthcare) at 4 °C overnight.The resins were washed with lysis buffer.The target protein was treated with TEV protease on the beads at 4 °C for 2 h, and then eluted by elution buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 1 mM EDTA).The protein was concentrated to the desired concentration if needed.
Cloning, expression, and purification of STX18-SNARE and SEC22B-SNARE STX18-SNARE (243-305) was cloned into pGEX-4T-1 vector between BamHI and NotI, with GST-TEV-FLAG tag at its N-terminal sequence.The construct was transformed into E. coli BL21 (DE3) cells.Overexpression of the protein was induced by addition of 0.1 mM IPTG at 16 °C for 16 h.The harvested cells were lysed with lysis buffer (20 mM HEPES, pH 7.4, 500 mM NaCl, 1 mM EDTA) supplemented with Protease Inhibitor Cocktail (APExBIO) and centrifuged at 40,000 rpm in Ti45 rotor for 60 min.The cleared supernatant was incubated with Glutathione Sepharose 4B (GE Healthcare) at 4 °C overnight.The beads were washed with lysis buffer.The protein was treated with TEV protease on the GST agarose at 4 °C for 2 h, and then eluted by lysis buffer.Gel filtration chromatography (Superdex 75 increase 10/300 GL, GE Healthcare) was applied to further purify the protein in S75 buffer (20 mM HEPES, pH 7.4, 150 mM NaCl, 1 mM EDTA).The protein fractions were collected and concentrated to the desired concentration.SEC22B-SNARE (134-194) was cloned into pET28 vector between NdeI and EcoRI, with 6×His tag at its N-terminal sequence.The construct was transformed into E. coli BL21 (DE3) cells.Overexpression of the protein was induced by the addition of 0.1 mM IPTG at 16 °C.The cells were harvested after induction for 16 h, and resuspended in lysis buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 10 mM Imidazole) supplemented with Protease Inhibitor Cocktail (APExBIO).The cells were lysed by high pressure homogenizer (ATS Engineer Inc., China) and centrifuged at 40,000 rpm for 60 min.The pellet was resuspended in urea buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 6 M urea, 1 mM β-mercaptoethanol, 20 mM Imidazole) at RT for 2 h and centrifuged again at 20,000 rpm for 45 min.The cleared supernatant was applied to Ni-NTA agarose beads (Qiagen).The resins were washed by urea buffer.The target protein was refolded on Ni-NTA agarose beads by gradually reducing the urea concentration to 0 M. Finally, the protein was eluted from Ni-NTA agarose beads by elution buffer (20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 500 mM Imidazole) and concentrated to the desired concentration.

Cloning, expression and purification of the SNARE complex
The SNARE domains of STX18 (243-305) and SEC22B (134-194) were cloned into pETDuet-1 vector, with SEC22B (134-194) containing N-terminal His tag inserted between BamHI and NotI restriction sites, and STX18 (243-305) inserted between NdeI and XhoI restriction sites, respectively.The SNARE domains of SNAP23 (14-76 and 146-208) were cloned into pACYCDuet-1 vector, with SNAP23 (14-76) containing a N-terminal His tag inserted between BamHI and SalI restriction sites, and the SNAP23 (146-208) fragment inserted between NdeI and XhoI restriction sites, respectively.The two plasmids were cotransformed into E. coli BL21 (DE3) cells under the selection of both Ampicillin and Chloramphenicol.Expression of target proteins was induced by 0.1 mM IPTG at 16 °C for 16 h.The cells were harvested and resuspended in lysis buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 20 mM Imidazole) supplemented with Protease Inhibitor Cocktail (APExBIO).Then the cells were lysed and centrifuged at 40,000 rpm for 60 min.The cleared supernatant after centrifugation was loaded onto Ni-NTA agarose beads (Qiagen).The resin was washed with wash buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 50 mM Imidazole), and the target proteins were eluted with elution buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 500 mM Imidazole).Fractions containing the SNARE complex were pooled and concentrated using an Amicon Ultra-15 centrifugal filter with 10-kDa molecular mass cut off (Millipore) to reduce the volume.The complex was further purified by gel filtration chromatography (Superdex increase 75 10/300 GL, GE Healthcare) with SEC buffer (50 mM HEPES, pH 7.4, 150 mM NaCl, 0.5 mM TCEP).Fractions were analyzed by SDS-PAGE and MS.

Naked adiposome preparation
Adiposomes were prepared following the previously published method with some modifications 28 .Two milligrams of total phospholipids (66.6% DOPC, 33.3% POPE) in chloroform was added to a 1.5 mL microcentrifuge tube, and the solvent was dried by nitrogen gas. 100 μL RB buffer (20 mM HEPES, pH 7.4, 100 mM KCl, 2 mM MgCl 2 ) was added into the tube and 20 μL of TG extracted from mouse fat tissue was added on the top of the RB buffer before vortex.The tube containing lipids and buffer was vortexed for 24 cycles of 10 s on and 10 s off.Then, the milky lipid mixture was centrifuged at 20,000× g at 4 °C for 10 min.The fraction containing adiposomes formed a floating white band at the top of the tube.The underlying solution and pellet were removed.A total of 100 μL RB buffer was added to the remaining fraction containing adiposomes which were resuspended by vortex.The sample was centrifuged again at 20,000× g for 5 min, and the solution underneath and pellet, if present, were removed.These procedures were repeated for three times, and then the white band containing adiposomes was resuspended in 100 μL RB buffer.After centrifugation at 1000× g for 5 min, the adiposomes as milky solution underneath a floating white band were collected for further application.The phospholipid concentration can be measured according to the method described before 35 .

Reconstitution of adiposomes for ensemble lipid mixing assay
Donor adiposomes with full-length STX18 or STX18-SNARE-TMD and SNAP23 contained 66.6% DOPC, 23.3% POPE and 10% NBD PE.Acceptor adiposomes with full-length SEC22B or SEC22B-SNARE-TMD contained 66.6% DOPC, 31.3%POPE and 2% Liss Rhod PE.Adiposomes with corresponding lipid compositions were made freshly as described above.Purified SNARE proteins were added to adiposomes to make STX18:SNAP23:Lipid = 1:1.9:83for donor adiposomes, and SEC22B:Lipid = 1:36 for acceptor adiposomes.The concentration of β-OG was kept ~0.8%, and the mixtures were incubated at 4 °C for 60 min.Later, the mixtures were diluted to make concentration of β-OG around 0.4% and then dialyzed against the reaction buffer (25 mM HEPES, pH 7.4, 150 mM KCl, 0.5 mM TCEP) with 1 g/L Biobeads SM2 (Bio-Rad) overnight.The samples were centrifuged at 20,000× g for 5 min, and then the adiposomes floating on the top were taken for lipid mixing assays.To make control adiposomes, the SNAREs were replaced with reaction buffer.

Lipid mixing assay with adiposomes
For lipid mixing assays, donor adiposomes (0.15 mM phospholipids) were mixed with acceptor adiposomes (0.15 mM phospholipids) with the addition of SNAP23 (20 μM) or CIDEC/FSP27 if required in a total volume of 200 μL.The FRET signal was measured by exciting NBD-PE fluorescence probe at 460 nm and observing the emission signal from Rhod-PE at 580 nm with a Spectrofluorometer (Varian Eclipse).All experiments were performed at 37 °C.All the experiments were repeated with three independent preparations.For quantification, we calculated the average fluorescence at 3000 s and the corresponding standard deviation (SD).

Reconstitution of adiposomes for ensemble content mixing assay
We used in-house synthesized Cy3-DAG and Cy5-DAG as the content dye for adiposomes in FRET-based content mixing assay (see Supplementary information for synthesis procedures of Cy3-DAG and Cy5-DAG).The synthesized Cy3-DAG and Cy5-DAG were dissolved in TG to make 1% (m:v) working stock separately.
Adiposomes encapsulated with Cy3-DAG or Cy5-DAG were made similarly to adiposomes with TG, except that half of the TG volume was replaced by Cy3-DAG or Cy5-DAG working stock.Theoretically, the reconstituted Cy3labeled adiposomes and Cy5-labeled adiposomes should end up with surrounding phospholipid layer at 66.6% DOPC, 33.3% POPE and encapsulated content at 0.5% Cy3-DAG or Cy5-DAG.

Content mixing assay with adiposomes
For content mixing assay, Cy5-labeled STX18-SNAP23-adiposomes (0.15 mM phospholipids) were mixed with Cy3-labeled SEC22B-adiposomes (0.15 mM phospholipids) with addition of SNAP23 (20 μM) in a total volume of 200 μL.For control group, Cy5-labeled STX18-adiposomes (0.15 mM phospholipids) were mixed with Cy3-labeled SEC22B-adiposomes (0.15 mM phospholipids) without addition of SNAP23 in a total volume of 200 μL.The FRET signal was measured by exciting Cy3 at 553 nm and observing the emission signal from Cy5 at 670 nm with a Spectrofluorometer (Varian Eclipse).All experiments were performed at 37 °C.All the experiments were repeated in three independent experiments.For quantification, we calculated the average fluorescence at 7200 s and the corresponding SD.

Measurement of SNAREs:lipids ratio
We used the Stewart Method to measure the phospholipid concentration of adiposomes.To avoid fluorescence dye to interrupt the 472 nm absorbance reading, we made adiposomes as how we made the adiposomes in FRET assay except that there are no fluorescence dye included.On the other side, we made a standard curve by liposomes with the same phospholipid compositions (66.6% DOPC, 33.3% POPE), to get a function of absorbance reading at 472 nm in terms of phospholipid amount.Based on this function and the 472 nm reading of the adiposomes we freshly made, we can get its phospholipid amount.Furthermore, the protein amounts of the adiposome samples were estimated by comparing the corresponding SNARE protein bands with BSA bands on SDS-PAGE.Then the ratio between phospholipids and proteins can be calculated.

Cryo-EM of adiposomes
For cryo-EM, 2 μL of fresh adiposome was applied to grids and blotted for 3 s in 100% humidity using a FEI Vitrobot Mark IV and then vitrified by quickly plunging into liquid ethane precooled with liquid nitrogen.Micrographs were recorded using a ceta 4 K × 4 K CCD in an Talos Arctica cryo-electron microscope operating at 200 kV.

Cysteine accessibility assay with adiposomeassociated STX18
The cysteine labeling assay was performed following the previously published method with some modifications 36 .Purified STX18-SNARE-TMD (243-335) or STX18-SNARE-TMD-Cys was reconstituted on freshly made adiposomes (66.6% DOPC, 33.3% POPE).The STX18decorated adiposomes were isolated from free proteins by a co-floatation assay on a Histodenze gradients (40%:35%:30%).The top fractions containing STX18decorated adiposomes were collected and analyzed by SDS-PAGE and Coomassie blue staining.Then these adiposomes were labeled with excess of Alexa Fluor 488 C5 maleimide dye (Thermo Fisher Scientific) in the absence or presence of 2% Triton X-100 at 4 °C overnight.The samples were separated by SDS-PAGE and analyzed by fluorescent imaging and Coomassie blue staining.

Measurement of the secondary structure of protein by circular dichroism
Both the purified STX18 (243-335) and STX18 (243-335)-Cys were dissolved in stock buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, 1 mM EDTA, 1% β-OG).The protein samples were diluted with ddH 2 O 15 times, and then diluted samples were applied to circular dichroism, using the diluted stock buffer as the baseline.

Quantification and statistical analysis
Statistical parameters including the definition and exact values of n, distribution and deviation are reported in the figure legends.Data are expressed as means ± SD.The significance of the variability between different groups was determined by two-way analyses of variance using GraphPad Prism software.Error bars, SDs of two or three independent experiments.Student's t-test, a P value of < 0.05 was considered statistically significant and a P value of > 0.05 was considered statistically non-significant (NS).

Fig. 5 (
Fig. 5 (See legend on next page.) N e g A A V -s h -S T X 1 8