Identification of motifs and mechanisms for lipid droplet targeting of the lipolytic inhibitors G0S2 and HIG2

Lipid droplets (LDs) are composed of a neutral lipid core coated by a phospholipid monolayer, and are dynamic organelles critically involved in triglyceride (TG) storage and breakdown (Kory et al., 2016; Walther et al., 2017). In response to fasting or nutrient deprivation, TG stores in LDs undergo a hydrolytic process termed lipolysis to release fatty acids (FAs) (Grabner et al., 2021; Zechner et al., 2017). The highest activities of intracellular lipolysis are observed in adipose tissue, followed by oxidative tissues, such as striated muscles and liver. The biochemical process is likewise conserved in adipocytes and non-adipocyte cells, where adipose tissue triglyceride lipase (ATGL; also known as PNPLA2) is the rate-limiting enzyme acting at the LD surface to catalyze hydrolysis of TG to one free FA and one diglyceride (DG) (Bezaire et al., 2009; Girousse and Langin, 2012; Jenkins et al., 2004; Villena et al., 2004; Zechner et al., 2009, 2012; Zimmermann et al., 2004). Structurally, the N-terminal region of ATGL contains a predicted three-layer α/β-hydrolase fold and an overlapping patatin-like domain, which harbors an unconventional serine-aspartate catalytic dyad similar to the one found in cytosolic phospholipase A₂ (cPLA₂) (Dessen et al., 1999). The C-terminal region of ATGL, on the other hand, contains a hydrophobic lipid-binding domain (LBD) of 45 amino acids that are critical for LD targeting (Lu et al., 2010; Schweiger et al., 2008). Deletion of this LBD, as often observed in ATGL mutants associated with neutral lipid storage disease with myopathy (NLSDM), produces truncated proteins that lose LD-associated lipase activity in vivo (Fischer et al., 2007; Kobayashi et al., 2008; Schweiger et al., 2009).

Aside from its LD localization, the ability of ATGL to hydrolyze TG at the LD surface is tightly controlled by its interaction with other regulatory proteins (Cerk et al., 2018). The lipolytic activity of ATGL is coactivated by comparative gene identification-58 (CGI-58; also known as ABHD5) (Lass et al., 2006), ATGL-mediated lipolysis can be inhibited by G0/G1 switch gene 2 (G0S2) (Yang et al., 2010) and the recently identified hypoxia-inducible gene 2 (HIG2) (also known as hypoxia-inducible lipid droplet-associated; HILPDA) (Padmanabha Das et al., 2018; Zhang et al., 2017b). G0S2 is a 103-amino-acid protein primarily expressed in hepatocytes and adipocytes in response to elevation in intracellular FA levels. G0S2 is a basic protein that contains a central hydrophobic domain (HD). Through direct interaction with the patatin-like domain of ATGL, the sequence encompassing the G0S2 HD domain mediates the noncompetitive inhibition of ATGL with an IC₅₀ in the nanomolar range (Cerk et al., 2014; Padmanabha Das et al., 2018). Functional studies using various mouse models have demonstrated that the interplay between ATGL and G0S2 is critical for balancing tissue-specific TG storage versus mobilization as well as for partitioning of metabolic fuels between adipose tissue and liver (Choi et al., 2014; El-Assaad et al., 2015; Heckmann et al., 2014; Heier et al., 2015; Ma et al., 2014; Nielsen and Moller, 2014; Shin et al., 2014; Sugaya and Satoh, 2016; Wang et al., 2013; Zhang et al., 2017a, 2014).

In comparison to G0S2, HIG2 is a 63-amino-acid protein that contains a HD domain similar to that of G0S2 (Povero et al., 2020). Like G0S2, HIG2 inhibits ATGL through direct physical association (Padmanabha Das et al., 2018; Zhang et al., 2017b). In various types of cells, hypoxia upregulates HIG2 expression and promotes TG-containing LD accumulation through HIG2-mediated ATGL inhibition (van Dierendonck et al., 2020, 2022; VandeKopple et al., 2019; Zhang et al., 2017b). In hypoxic cancer cells, lipolytic inhibition is a defense mechanism against hypoxia-induced oxidative stress as it triggers downregulation of PPARα target genes for mitochondrial activity and FA β-oxidation, leading to decreased reactive oxygen species (ROS) generation and apoptotic cell death (Zhang et al., 2017b). Although it is a weaker ATGL inhibitor than G0S2 in in vitro biochemical assays (Padmanabha Das et al., 2018), the function of HIG2 appears to be ATGL-specific in vivo as co-deletion of ATGL was able to completely reverse the phenotypes caused by HIG2 deficiency in both cancer cells and macrophages (van Dierendonck et al., 2022; Zhang et al., 2017b).

On the cellular level, HIG2 and G0S2 both inhibit basal and CGI-58-activated ATGL at the LD surface (Lu et al., 2010; Padmanabha Das et al., 2018; Yang et al., 2010). Whereas ATGL localization to LDs is mediated by the Arf1–coat protein complex I (COPI) machinery (Beller et al., 2008; Ellong et al., 2011; Soni et al., 2009; Suzuki et al., 2015), the mechanisms and structural basis for targeting G0S2 and HIG2 to LDs remain undefined. In this study, we show that although LD localization of HIG2 is fully dependent on ATGL, G0S2 can be targeted from endoplasmic reticulum (ER) to LDs via both ATGL-independent and -dependent mechanisms. Moreover, we use mutagenesis and cell imaging experiments to demonstrate that a hairpin loop formed by two helices at the N-terminal region of G0S2 is essential for anchoring the protein to ER prior to translocation to LDs. Furthermore, we found that basic residues within the hinge sequence of the hairpin play an important role in sorting of ER-localized G0S2 to LDs.

G0S2 plays an important role in regulating ATGL-mediated intracellular lipolysis in liver and adipose tissue (Zhang et al., 2017a). To determine whether the LD localization of G0S2 is dependent on its physical interaction with ATGL, we expressed mouse G0S2 in primary hepatocytes isolated from ATGL-knockout mice via adenoviral infection. A FLAG epitope tag fused at the C-terminus of G0S2 (G0S2–FLAG) allowed examination of its localization using immunofluorescence staining. As shown in Fig. 1A,B, G0S2 was mainly localized at honeycomb-like structures that resemble ER tubules and cisternae when cells were maintained in regular medium without extra FA. However, after oleic acid was added to induce LD formation, G0S2 became highly enriched at the surface of LDs, which were stained with the neutral lipid dye BODIPY 493/503. Interestingly, coexpression of Myc-tagged ATGL resulted in the enrichment of FLAG–G0S2 with ATGL at the LD surface regardless of oleic acid treatment (Fig. 1C,D). These results suggest that G0S2 is targeted to LDs via ATGL-independent and -dependent mechanisms, the former of which coincides with the induction of new LD formation.

G0S2 is known to localize at both ER and LDs (Yang et al., 2010; Zandbergen et al., 2005). To investigate the dynamic distribution of G0S2 between these two subcellular loci, we transiently transfected HeLa cells with mouse G0S2–FLAG and then examined its localization by immunofluorescence staining. Given their capacity to form and breakdown LDs, HeLa cells have been widely used as a cell biology model for studying LD association of ATGL and its regulatory proteins (Ellong et al., 2011; Keenan et al., 2021; Lu et al., 2010; Smirnova et al., 2006; Soni et al., 2009; Yang et al., 2010; Zhang et al., 2017b). To reveal the ER structure, cells were co-transfected with GFP-tagged Sec61, a general ER marker. To visualize intracellular LDs, cells were stained with LipidTOX dye following fixation. Initially, transfected cells were incubated in the presence of lipoprotein-deprived serum to reduce LDs (referred to as lipid starvation). In response to lipid starvation, G0S2 was found in small punctate structures apparently connected to the ER (at time 0 min, Fig. 2) whereas mature LDs were not visible. After 5 min of oleic acid loading, G0S2 had already accumulated in rounded structures consistently connected to the ER. LipidTOX staining revealed these structures as newly formed LDs, as they contained a core of neutral lipids. When starved cells were loaded with oleic acid for 15 min, some G0S2-coated LDs had enlarged and detached from ER. By the 45-min time point post oleic acid loading, a majority of G0S2 protein was localized at the surface of LDs that were detached from ER. G0S2-coated LDs were almost completely separated from ER after 15 h of OA loading (Fig. 2). These results suggest that lipid arrival promotes segregation of G0S2 within the ER into newly formed LDs.

Many proteins appear to access the LD surface through one of two major pathways – from the ER or from the cytosol (Olarte et al., 2022). G0S2 appears to belong to the Class I LD proteins that have a dual localization in the ER and LDs. According to the AlphaFold structure prediction (Jumper et al., 2021; Varadi et al., 2022), the N-terminal portion of G0S2 consists of two α-helices separated by a positively charged hinge sequence (Fig. S1). Together, they form a V-shaped hairpin loop structure that might be capable of embedding in the phospholipid monolayer. To determine their individual involvement in ER and LD localization, we deleted each of the three sequences in the hairpin of mouse G0S2 as indicated in Fig. 3A and co-expressed the derived mutants in HeLa cells with GFP–Sec61. After oleic acid treatment, immunofluorescence staining revealed that the full-length wild-type (WT) protein was localized to the rim-like structures that resemble the surface of LDs and are separate from the ER (Fig. S2; Fig. 3B,C). In comparison, deletion of any of the three sequences in the hairpin rendered the protein incapable of localizing to the LD surface rims (Fig. S2). A majority of G0S2Δ1-15, which lacks the first α helix, existed as puncta that partially overlapped with the ER (Fig. 3B,C). Deletion of the second α helix (Δ27-42/ΔHD) led to a tubule-like distribution that also only partially overlapped with ER. Strikingly, deletion of the hinge sequence (residues 16–26) resulted in a complete ER colocalization (Fig. 3B,C). Together, these results indicate the importance of an intact hairpin structure for targeting G0S2 to ER as well as to LDs, and a critical role of the hinge sequence within the hairpin in mediating the exit of G0S2 from the ER.

To further determine the role of the positive charges in the hinge sequence, we mutated three basic residues, individually or in combination, to the neutrally charged alanine. The FLAG-tagged mutant proteins were overexpressed in HeLa cells, which then were treated with oleic acid to promote LD formation. As shown in Fig. 4A,B, both the K25A single mutant and K22A/K25A double mutant retained the ability to traffic to LDs, whereas the R20A/K22A/K25A triple mutant did not display LD localization. As a result, the triple mutant was completely restricted at the ER as revealed by its colocalization with the ER marker Sec61 (Fig. 4C). These results suggest that the positive charges in the hinge sequence are a critical part of the LD targeting motif in G0S2 that sorts G0S2 from ER to LDs.

The HD domain (residues 27–47; i.e. the second putative α-helix) in G0S2 is essential for both ATGL interaction and LD localization. To determine whether this ATGL-interacting domain is involved in LD localization, we co-expressed mouse ATGL tagged N-terminally with Myc (Myc–ATGL) with a series of deletion mutants of mouse G0S2 tagged C-terminally with FLAG, each lacking a 4-residue sequence in the HD domain (Fig. 5A). The G0S2–ATGL interaction was then determined by co-immunoprecipitation analysis. As shown in Fig. 5B, all G0S2–FLAG variants were expressed at comparable levels. When compared to the full-length protein, the G0S2 mutant lacking the entire HD domain (G0S2Δ27-42) showed an almost complete loss of interaction with ATGL. Interestingly, deletion of each of the three 4-residue sequences in the C-terminal 12 residues of the HD, namely Δ31–34, Δ35–38 and Δ39–42, failed to significantly affect the binding to ATGL. In comparison, the ATGL interaction with G0S2Δ27-30, the mutant that lacks first 4-residue sequence YVLG, was largely disrupted (Fig. 5B). This result is in agreement with our previous findings showing that the YVLG motif is critically needed for mediating inhibition of ATGL activity (Zhang et al., 2017b). To determine whether loss of ATGL interaction was due to G0S2 being unable to localize to the LD surface, we overexpressed WT G0S2 or deletion mutants of the G0S2 HD domain in ATGL^−/− HeLa cells we generated previously (Varadi et al., 2022). Whereas deletion of the entire HD domain abolished the presence of G0S2 at the LD surface (Fig. 3), lack of the ATGL-interacting YVLG motif in the HD domain did not affect the LD localization of G0S2Δ27-30 (Fig. 5C). Neither did deletion of any other 4-residue sequences in the HD. The results clearly demonstrate that, although it requires the HD, the LD localization of G0S2 does not involve the ATGL-interaction motif.

Our results indicate that the positively charged hinge sequence facilitates the sorting of G0S2 from the ER to LDs. To determine whether the basic residues involved are also involved in mediating ATGL inhibition, we performed a TG hydrolase activity assay by using extracts of HeLa cells overexpressing ATGL as a lipase source. As shown in Fig. 6A, addition of cell extracts containing WT G0S2 or the R20A/K22A/K25A triple mutant both led to substantial inhibition of ATGL, indicating that the positive charges proximal to the N-terminal end of the ATGL-interacting HD domain are not required for the lipolytic inhibitory activity of G0S2. When expressed alone in HeLa cells, the ER-localized R20A/K22A/K25A mutant showed no noticeable presence at the LDs in response to oleic loading (Fig. 6B). Interestingly, when co-expressed with ATGL, a significant amount of R20A/K22A/K25A mutant was detected along with ATGL at the LD surface.

A hydrophobic LBD (residues 259–337) in its C-terminal region is known to anchor ATGL to the surface of LDs. As described previously, deletion of this LD-anchoring LBD (as seen in ATGL/ΔLBD) prevented ATGL from localizing to LDs, a phenotype that was rescued when the WT G0S2 was co-expressed (Lu et al., 2010). Intriguingly, although individually they were both deficient in LD localization, ATGL/ΔLBD and G0S2/R20A/K22A/K25A mutants were able to traffic to LDs together when co-expressed in HeLa cells (Fig. 6C). The result suggests that interaction with ATGL, but not the ability of ATGL to anchor to LDs, stimulates the translocation of G0S2/R20A/K22A/K25A mutant to LDs. Indeed, the LD recruitment by both WT and ΔLBD mutant of ATGL was abolished upon further deletion of the ATGL-binding motif in the triple mutant of G0S2 (R20A/K22A/K25A/Δ27-30) (Fig. 6B,C).

HIG2 inhibits ATGL using a similar HD domain to that in G0S2. However, based on the prediction by AlphaFold (Fig. S3) (Jumper et al., 2021; Varadi et al., 2022), HIG2 lacks the first α-helix and a positively charged hinge sequence that form the putative hairpin-like loop structure in G0S2. To further examine how HIG2 localization would be impacted by LD formation, we expressed mouse HIG2 tagged at the C-terminus with a FLAG epitope tag (HIG2–FLAG) in ATGL^−/− HeLa cells with the ER marker GFP–Sec61. Under the lipid starvation conditions, HIG2 exhibited a punctate localization at the ER (Fig. 7A). Upon 15 min of oleic acid to induce LD formation, HIG2 was concentrated in the dot-like structures at the ER with some of those dots flanking the newly formed LDs. After an overnight oleic acid treatment, a majority of HIG2 protein was found to localize at the interface between enlarged LDs and ER instead of the LD surface rims as exhibited by G0S2. Interestingly, co-expression of ATGL was sufficient to promote complete LD localization of HIG2 (Fig. 7B). As observed with the G0S2/R20A/K22A/K25A triple mutant, such an effect of ATGL co-expression did not require the LD-anchoring LBD of ATGL. These data suggest that the full LD localization of HIG2, which lacks a putative hairpin structure and a positive charged sequence proximal to its HD, is ATGL dependent. Furthermore, addition of the G0S2 hinge sequence to the N-terminus of full-length HIG2 (G0S2/HIG2 fusion 1) yielded a hybrid protein that partially localizes to LDs attached to ER (Fig. 7C). In comparison, addition of both the first helix and the hinge sequence of G0S2 was able to fully drive the G0S2/HIG2 hybrid protein to LDs that are detached from ER. The data are in support of the concept that a complete hairpin is required for sorting the protein from ER onto the LD surface.

Understanding mechanisms controlling FA and TG metabolism are essential for developing ways to counteract the negative effects of their dysregulation. ATGL, with its role as the rate-limiting enzyme catalyzing intracellular lipolysis and FA release from TG-containing LDs, is central to this understanding. G0S2 and HIG2 are two small protein inhibitors of ATGL, and their conserved HD domain mediates the inhibitory interaction with ATGL. At cellular level, how these inhibitory proteins make their ways onto LDs, where they meet with ATGL, is still under investigation. In this work, we identified unique targeting mechanisms for localizing these inhibitor proteins to LDs. We found that G0S2 contains a putative hairpin-like structure for initial insertion to the ER, followed by sorting from the bulk of ER to LDs. The latter process is mediated by the positively charged hinge sequence connecting the two α-helices in the hairpin. This positively charged hinge sequence, however, is dispensable for LD localization when G0S2 binds to ATGL. In comparison, the predicted structure of HIG2 lacks the first α-helix and the hinge sequence that form the hairpin in G0S2. The structural difference likely explains why HIG2 is incapable of trafficking to LDs on its own and instead, is dependent on ATGL for its full LD localization.

G0S2 appears to belong to a general class of proteins that initially localize to the ER and then migrate to the LD surface during LD formation. Ingelmo-Torres et al. were the first to identify a localization signal for sorting proteins with a hydrophobic domain within the ER into LDs (Ingelmo-Torres et al., 2009). The presence of a positively charged sequence is the main feature of this signal that correlates with the subsequent targeting from ER to LDs. A recent study by Olarte et al. further identified minimal motifs for ER to LD (ERTOLD) targeting as a hydrophobic hairpin-like structure containing two phospholipid embedded domains and a connecting hinge region with positively charged residues (Olarte et al., 2020). We found that the N-terminal portion of G0S2 shares these structural features, and deletion of the entire positively charged hinge region in hairpin abolished the LD localization of G0S2. Instead, the mutant protein adopted an even distribution completely overlapping the ER network, and was excluded from the LD-forming sites. The same effect was recreated upon concurrent mutation of three basic residues in the hinge sequence to neutrally charged alanine residues, suggesting that the positive charges play an active role in driving the LD migration of G0S2. Given that mutating individual basic residues had little effect on LD targeting compared to mutating all three, the sum of positive charges in the hinge region seems to be more crucial for the preference of G0S2 localization at the LD surface. Despite the well-defined role of positively charged sequence in LD targeting, the underlying biophysical mechanisms remain to be determined. In the ER membrane, those charged residues are thought to be stabilized by interacting with the phosphate group of the luminal phospholipid leaflet (Olarte et al., 2020, 2022). When embedded in the phospholipid monolayer of LD surface, these charged residues can be then stabilized by water in the LD core. Notably, the hydration properties of simulated LD systems recently demonstrated ∼10 times more water in the LD core than previously reported (Kim and Swanson, 2020).

As also suggested by simulations, tryptophan and tyrosine residues can help ERTOLD proteins achieve a lower free energy state in the oil phase of the LD due to hydrogen bonding with the head groups of phospholipids, the glycerol moieties of TG or coordinating water molecules (Olarte et al., 2020, 2022). However, this is unlikely the case for G0S2 as its LD targeting hairpin contains no tryptophan. Moreover, tyrosine-27 (Y²⁷) within the ATGL-interacting HD domain (i.e. the second α-helix) is the lone tyrosine in the G0S2 hairpin. As mapped by the deletion mutagenesis, ATGL inhibition requires the Y²⁷VLG motif, the first four residues of the HD. Consistent with this, loss of this conserved sequence motif also abolished the ATGL interaction and inhibitory activity of HIG2 (Zhang et al., 2017b). A very recent study further showed that the single Y²⁷G mutation in G0S2 resulted in complete loss of ATGL inhibition (Riegler-Berket et al., 2022). Despite its importance of this conserved tyrosine residue in regulating ATGL, however, Y²⁷ does not seem to be required for LD targeting because deletion of the entire YVLG motif elicited no effect on the ATGL-independent LD localization. Furthermore, G0S2 does not contain a central proline in its hairpin hinge region as commonly found in other ERTOLD proteins (Olarte et al., 2022). Taken together, our observations suggest that the hairpin conformation of G0S2 might differ considerably from the one contained in other ERTOLD proteins, and additional mechanisms might contribute to its accumulation at the LD surface.

Our data indicate that ATGL binding enables a second mechanism for targeting G0S2 to LD. The fact that ATGL was able to recruit the G0S2/R20A/K22A/K25A triple mutant to LDs suggests that such a mechanism does not involve the positively charged ERTOLD sorting signal. However, such recruitment seems to require a direct interaction, as deletion of the ATGL-interacting motif in the G0S2 triple mutant abolished its ATGL-mediated LD localization. To this end, we propose two potential scenarios for the interplay between G0S2 LD localization and ATGL interaction. One is for G0S2 and ATGL to traffic to LDs separately and form inhibitory complex after their arrival at the LD surface. Previous studies have shown that ATGL might partially exist in the cytosol (Villena et al., 2004; Zimmermann et al., 2004), but most ATGL is likely to exist in a membrane-bound form (Soni et al., 2009). Transport of ATGL to LDs requires both the Arf1–COPI machinery (Beller et al., 2008; Ellong et al., 2011; Soni et al., 2009; Suzuki et al., 2015) and association of its C-terminal LBD with the LD phospholipid monolayer (Lu et al., 2010; Schweiger et al., 2008). Thus, whereas G0S2 follows the ERTOLD pathway via its hairpin-like structure, ATGL translocates to LDs through vesicular trafficking and via its LBD. Alternatively, the second scenario is for ATGL to bind to G0S2 at the ER, and the two proteins subsequently migrate to LDs as a complex. In this case, ATGL binding to the G0S2 HD domain might trigger conformational change in the G0S2 hairpin, resulting in LD relocalization independently of the positively charged hinge sequence. Of note, we showed that the R20A/K22A/K25A triple mutant still retained the full capacity to inhibit ATGL, suggesting that these charged residues proximal to the HD domain are not essential for ATGL interaction. A somewhat surprising finding is that the LD translocation is not mediated by the LD localization of ATGL, given that the ATGL mutant lacking the LBD was still able to drive the G0S2/R20A/K22A/K25A triple mutant to the LD surface. Interestingly, a previous study demonstrated that C-terminally truncated ATGL mutants, which lack the LBD, translocate to the LD upon co-expression with G0S2 (Schweiger et al., 2012).

Our data are in support of a similar ATGL-dependent mechanism for targeting HIG2 to LDs. Compared with G0S2, HIG2 is a shorter peptide with a lower content of basic residues. Maier et al. showed that endogenous HIG2 is present at points of contact between ER and LDs, and that knockout of the HIG2 gene prevents hypoxia-induced LD formation in macrophages (Maier et al., 2017). In agreement with this previous observation, our data showed that in ATGL-knockout cells, HIG2 is only partially localized to ER during lipid starvation and mostly to the LD–ER interface upon LD formation. This is consistent with its lack of the first α-helix and a charged hinge sequence that form the predicted LD-targeting hairpin in G0S2. As a result, the full LD localization of HIG2 is dependent on ATGL, as evidenced by the fact that HIG2 only relocalizes to LDs upon ATGL coexpression. Similar to what is seen for the G0S2/R20A/K22A/K25A triple mutant, such a LD relocalization of HIG2 does not seem to require the C-terminal LBD of ATGL. Mechanistically, how such a mutual LD recruitment occurs remains unclear. Given that HIG2 and G0S2 interact with the N-terminal patatin-like domain of ATGL, it is reasonable to assume that ATGL undergoes a so-called ‘folding-upon-binding’ process upon interaction with these inhibitory proteins, which enables anchoring of ATGL to the phospholipid surface of the LDs without involving its C-terminal LBD. If so, it would be tempting to determine whether binding to the inhibitor proteins would induce dissociation of the LBD of ATGL from the LD phospholipids. If this occurred, whether such dissociation would prevent the physical contact of ATGL with the inner TG core of LDs, thereby impeding the hydrolytic action of ATGL.

The complete open reading frames of mouse ATGL, G0S2 and HIG2 were cloned into pRK vector with a FLAG or Myc epitope tag as previously reported (Yang et al., 2010; Zhang et al., 2017b). Deletion and point mutations were generated by using the QuikChange Site-Directed Mutagenesis Kit (Agilent Technologies) according to the manufacturer's guidelines. To generate G0S2/HIG2 hybrid expression constructs, we designed duplex oligonucleotides flanked by BamHI sites and encoding either the G0S2 hinge sequence alone or the first helix and the hinge sequence of G0S2. Upon BamHI digestion, these oligonucleotides were ligated into the BamHI site at the start of HIG2 cDNA sequence in the HIG2-FLAG vector. The correct insert orientation was verified by DNA sequencing and subsequent protein expression in transfected HeLa cells.

Recombinant adenovirus encoding N-terminally Myc-tagged ATGL and C-terminally FLAG-tagged G0S2 under the control of a cytomegalovirus promoter were custom generated by Vector Biolabs. A cytomegalovirus-null virus [null adenovirus (Ad-Null)] was also obtained for use in control experiments. Primary mouse hepatocytes were isolated from male whole-body ATGL-knockout mice (C57BL/6 background) as described previously (Zhang et al., 2019) and in accordance with an approved protocol by the Institutional Animal Care and Use Committee at Mayo Clinic. Following isolation, primary hepatocytes were seeded into six-well plates in the DMEM with low glucose and 1% fetal bovine serum (Thermo Fisher Scientific) at 37°C under 5% CO₂. After 3 h, cells were infected with 2×10⁷ pfu per well of adenovirus for 40 h in the presence of 2.5 μg/ml of polybrene.

WT or ATGL^−/− HeLa cells as generated previously (Zhang et al., 2017b) were cultured in high-glucose DMEM (Invitrogen) containing 10% FBS (Invitrogen) supplemented with 100 U ml⁻¹ penicillin/streptomycin (Invitrogen). Cells at 70–80% confluence were transfected with plasmids expressing Myc–ATGL, G0S2–FLAG and/or HIG2–FLAG fusion proteins using Lipofectamine 2000. After overnight incubation with transfection mixtures, cells were lysed accordingly for downstream biochemical analyses. For immunoblotting, preparation of tissue and cell lysates and subsequent immunoblotting analysis were performed as described previously (Zhang et al., 2014). Briefly, following SDS-PAGE and transfer to nitrocellulose membranes, individual proteins were blotted with rabbit anti-FLAG antibody at 1:2000 (MilliporeSigma, cat. #F7425) and rabbit anti-Myc antibody (Cell Signaling, cat. #2272) at 1:1000 dilutions. After an overnight incubation, peroxide-conjugated secondary antibodies were incubated with the membrane at a dilution of 1:5000. The signals were then visualized by chemiluminescence (Supersignal ECL, Pierce). For immunoprecipitation, cells were lysed in cell lysis buffer containing 50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% Triton X-100, 1 mM DTT and protease tablet inhibitors (1 mini tablet per 10 ml of buffer). Anti-Flag M2-conjugated gel (MilliporeSigma, cat. #A2220) was incubated with the lysates for 3 h at 4°C. The beads were then washed four times with lysis buffer, and the bound proteins were eluted in SDS sample loading buffer and analyzed by immunoblotting.

Cells were incubated with 400 μM oleic acid with 0.2% BSA for varying periods of time to promote LD formation as described previously (Brasaemle and Wolins, 2006). Following fixation with 4% paraformaldehyde in PBS for 15 min, cells were permeabilized with 0.25% Triton X-100 for 5 min, quenched with 100 mM glycine in PBS for 20 min, and then blocked with 2% BSA in PBS for 1 h. The cells were then exposed to primary antibody (MilliporeSigma mouse anti-FLAG, cat. #F1804; Cell Signaling rabbit anti-Myc, cat. #2272) at 1:500 dilution for 2 h at room temperature. Following three washes with PBS, the cells were treated for 1 h with Alexa Fluor secondary antibodies (Thermo Fisher Scientific, cat. #A11008, #A11001, #A11004, #A32733) at 1:1000 dilution. To visualize lipid droplets, 1 μg/ml of BODIPY 493/503 (Thermo Fisher Scientific, cat. #D3922) was added during the incubation with secondary antibodies. Alternatively, the cells were incubated in Deep Red LipidTOX dye (Thermo Fisher Scientific, cat. #H34476) at 1:2000 dilution in PBS for 15 min after the secondary antibodies. Samples were mounted on glass slides with Vector Shield mounting medium and examined under a Zeiss LSM 980 inverted confocal microscope with Airyscan. Acquired images were processed and quantified manually with Zen Blue and ImageJ FIJI software.

Colocalization analysis was performed using JACoP plugin embedded in ImageJ. LD colocalization of immunofluorescently stained proteins was determined quantitatively as described recently (Olarte et al., 2020). First, the LD-occupied region in each cell was identified based on BODIPY 493/503 staining. Specifically, an automatic threshold was applied to the raw images of BODIPY 493/503 channel. For each of the quantified cells, a mask was created via the BODIPY 493/503 channel that focuses on the LD-occupied region. The mean intensity of a given protein signal on LDs was divided by the mean intensity of the protein signal in the whole cell excluding the LD-occupied region. Thus, the ‘LD enrichment factor’ of individual proteins represents the ratio of mean pixel intensity and area of a protein signal (red) in the LD-occupied region relative to those in the whole cell excluding LDs.

For quantitative ER colocalization analysis, the colocalization threshold plugin was used to calculate the Pearson coefficient of colocalization with the ER marker GFP–Sec61 (Addgene, plasmid #121159). Each image was analyzed by creating a mask via the GFP–Sec61 channel that focuses on the region of interest (ROI) to the ER structures. The estimation of Pearson's correlation coefficient is applied for matching the fluorescence pixel intensities from a protein channel (red) to the ROI in order to describe the degree of overlap between a protein and ER, taking into account the environment of each pixel. A value of 1.0 is the result of complete colocalization between two channels. The Pearson coefficient was calculated for 20 cell images per condition. Bars represent the mean±s.d. from each image.

In vitro transcription–translation was carried out by using the TNT SP6 High-Yield Protein Expression System (Promega) according to the manufacturer's instructions. Specifically, reactions consisting of 30 μl TNT SP6 High-Yield Wheat Germ Master Mix and 5 μg vector DNA, made up to 50 μl with molecular biology grade water were incubated for 120 min at 25°C. Then the reaction mixture was used for TG lipase activity or immunoblotting analysis.

A total of 35 μl of in vitro transcription–translation reaction mixture was prediluted with 65 μl of cell extraction buffer (0.25 M sucrose, 1 mM EDTA, 1 mM DTT). The TG hydrolase activity against ³H-labeled triolein was measured as described previously (Zhang et al., 2013) by mixing 0.1 ml of diluted reaction mixture with 0.1 ml of substrate solution. Reactions were terminated by adding 3.25 ml of methanol/chloroform/heptane (10:9:7) and 1.0 ml of 0.1 M potassium carbonate, 0.1 M boric acid (pH 10.5). Following centrifugation at 800 g for 15 min, radiolabeled FAs in 1 ml of upper phase were measured by liquid scintillation counting.

When evaluating the statistical difference between the mean values of two datasets, we used an unpaired two-tailed Student's t-test with Welch's correction. A one-sample t-test was used in the case of normalized data. No statistical method was used to predetermine sample size.

We would like to thank Dr Xiaodong Zhang, Alicia Saarinen and Joseph Webb for their excellent technical support. We are grateful that Dr Rudolf Zechner kindly shared the ATGL-knockout mice, from which primary hepatocytes were derived. All confocal microscopy was performed at the Microscopy and Cell Analysis Core of Mayo Clinic.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.260236.reviewer-comments.pdf