The CYTOLD and ERTOLD pathways for lipid droplet–protein targeting

Abstract

Lipid droplets (LDs) are the main organelles for lipid storage, and their surfaces contain unique proteins with diverse functions, including those that facilitate the deposition and mobilization of LD lipids. Among organelles, LDs have an unusual structure with an organic, hydrophobic oil phase covered by a phospholipid monolayer. The unique properties of LD monolayer surfaces require proteins to localize to LDs by distinct mechanisms. Here we review the two pathways known to mediate direct LD protein localization: the CYTOLD pathway mediates protein targeting from the cytosol to LDs, and the ERTOLD pathway functions in protein targeting from the endoplasmic reticulum to LDs. We describe the emerging principles for each targeting pathway in animal cells and highlight open questions in the field.

Lipid Droplets as dynamic cellular organelles

Lipid droplets (LDs) are ubiquitous organelles, found in eukaryotic and some prokaryotic cells, that are hubs of cellular lipid metabolism and act primarily to store neutral lipids. They consist of a core of triglycerides (TGs) or sterol esters (SEs) that is bounded by a monolayer of phospholipids. Through the action of lipases acting at the LD surface, neutral lipid components are released to generate membrane lipids or metabolic fuel. Other functions of LDs include storing proteins, preventing the toxic accumulation of lipids, providing reservoirs of lipid-soluble drugs and vitamins, and serving as cellular hubs for the replication of some viruses [1–5]. These broad functions of LDs underlie the increasing identification of LD functional roles in different biological processes, including for instance energy metabolism and roles in innate immunity [6].

LD proteins number in the tens to hundreds and vary among species and cell types (see Box 1). Many are important enzymes involved in cellular lipid metabolism (see below). Increasingly, variations in LD proteins have been recognized as risk factors for metabolic and other human diseases. For example, variations in PNPLA3 or HSB17B13, which encode metabolic enzymes that localize to LDs, modify the risk for human non-alcoholic fatty liver disease (NAFLD) (see Glossary) and/or non-alcoholic steatohepatitis (NASH) [7, 8].

TEXT BOX 1: Methods and criteria to identify lipid droplet (LD) proteins.

LD proteomes have been characterized by a variety of techniques in many different cell types and in different species [14, 46, 47, 97, 98]. However, the interpretation of many of these studies is complicated by the copurification of contaminants as LD proteins. This can be due to the unspecific binding of non-LD proteins to the LD oil phase during cell fractionation and to the close association between LDs and other organelles, such as the ER and mitochondria. Additionally, microscopy studies that claim the LD localization of proteins based on their overexpression should be approached with caution. This is because overexpression can saturate a protein’s normal localization sites within the cell, leading to their aberrant binding to the LD oil-water interface.

Ideally, the classification of a protein as a bona fide LD protein should be supported by multiple lines of evidence. These might include assessing the LD localization of the endogenous protein in cells by fluorescence microscopy, identifying it in LD cellular fractions by biochemistry methods, and when possible, performing super-resolution or electron microscopy to show that the protein directly binds to the LD surface [15, 46, 47, 60, 99, 100].

How proteins recognize and localize to LD surfaces has emerged as a fascinating and important aspect of cell biology with biomedical relevance. Here, focusing on animal cells, we review some of the major LD proteins and the principles of how they target to this organelle.

LD proteins and their functions

The discovery of specific LD proteins helped to define LDs as cellular organelles. In particular, the identification of perilipins as LD proteins suggested these are regulated organelles harboring specific proteins [9]. Perilipins regulate lipolysis in response to extracellular signals [10]. Analogously, in plants, oleosins decorate LDs, particularly in seeds, where they likely perform similar structural and regulatory functions [11, 12].

The majority of bona fide LD proteins are involved in lipid metabolism. Examples of such proteins include enzymes that act in neutral lipid synthesis, such as fatty acyl-CoA synthetases ACSL3 and ACSL5 [13, 14], glycerol-3-phosphate acyltransferase 4 (GPAT4) [15], or the yeast squalene monooxygenase Erg1, which is involved in sterol synthesis [16]. In Drosophila cells, targeting neutral lipid synthesis enzymes (e.g., GPAT4) to the LD surface enables LD-localized TG synthesis, which is catalyzed in an endoplasmic reticulum (ER)-LD compartment by diacylglycerol acyltransferase 2 (DGAT2), and thus the expansion of initial LDs [15, 17]. Although little is known about the relative activities of many lipid metabolic enzymes at the LD surface compared with the ER, in yeast, the DGAT2 orthologue Dga1p exhibits different activities in the corresponding subcellular fractions [18].

Enzymes of phospholipid synthesis and transport also localize to LDs. These include the rate-limiting enzyme controlling phosphatidylcholine (PC) synthesis, CTP:phosphocholine cytidylyltransferase (CCT; [19]), and enzymes of the Lands cycle that remodel phospholipid acyl chains (LPCATs; [20]), which are found on the LDs of some cell types [20, 21]. In the case of CCT, LD binding occurs in response to PC deficiency on the growing LD surface and activates the enzyme to increase PC synthesis to meet demands [19, 22, 23]. In humans, mutations in the LD- and membrane-binding motif of the CCT isoform encoded by PCYT1A are associated with lipodystrophy [24, 25]. Several lipid transfer proteins, such as PC transfer protein [26] and ORP5 [27], have also been localized to LDs, where they might be involved in the transfer of specific phospholipids or sterols to or from the LD monolayer.

LD proteins also include lipases that degrade neutral lipids, including the major TG lipase adipose TG lipase (ATGL/PNPLA2; [28]), its activator comparative gene identification-58 (CGI-58), and the major diacylglycerol (DAG) lipase hormone-sensitive lipase (HSL; [29]). These proteins act in concert to degrade TG to fatty acids (FAs) and glycerol, a process best understood in adipocytes (reviewed in [30, 31]). Deficiency of ATGL or CGI-58 results in human neutral lipid storage disease [32] characterized by the accumulation of fat in various tissues of the body.

Another putative lipase, PNPLA3, is a LD protein with direct connections to human metabolic disease. A common variant of PNPLA3 (I148M) is a major risk factor for the development of NAFLD/NASH [7, 33]. PNPLA3 has been suggested to transfer fatty acyl chains between TGs and phospholipids [34]. The PNPLA3 (I148M) variant appears to be catalytically compromised and accumulates on the surface of LDs [35], likely due to a reduction in ubiquitylation and protein degradation [36]. A current model posits that the accumulation of PNPLA3 (I148M) on LD surfaces results in a defect in TG mobilization from LDs, perhaps via diminished ATGL activity by sequestering its activator CGI-58, leading to excessive LD accumulation and ultimately NAFLD/NASH [37]. Reducing the amount of PNPLA3 (I148M) on the surface of LDs might thus provide a therapeutic avenue to treat these metabolic conditions [38].

Similarly, genetic studies revealed that loss-of-function alleles for HSD17B13, encoding the LD protein 17-β-hydroxysteroid-dehydrogenase 13 (HSD17B13), is associated with protection from NASH in humans [8]. Although the enzymatic function of HSD17B13 is poorly defined, its N-terminal domain has similarity to perilipin proteins and is required for its LD targeting and activity [39]. How HSD17B13 promotes NASH is unknown. The protective allele of HSD17B13 is associated with changes in PC levels in liver [40].

Unexpectedly, some transcription factors were found to be LD proteins [41, 42]. These include the transcription factor Max-like protein X (MLX) and MLX family members (MLXIP, MLXIPL/ChREBP; [43]), which are master regulators of glucose and lipid metabolism [44]. The binding of these transcription factors to LDs provides a potential additional layer of regulation of their activities in gene expression, where studies in cells have shown that LD binding acts in a feedback mechanism to downregulate their transcriptional activity [41]. A current model suggests that glucose metabolites activate these transcription factors, upregulating de novo lipogenesis and lipid storage in LDs, among other responses. This triggers binding of these transcription factors to LDs in the cytosol, thereby attenuating the transcriptional response. Conversely, the LD protein perilipin 5 appears to promote expression of PGC-1α/SIRT target genes, including mitochondrial proteins [42], further suggesting a relationship between LD metabolism and metabolic gene expression.

LDs can also act as protein storage depots. For example, LDs temporarily sequester a subset of histones in Drosophila embryos (e.g., H2A and H2B) through a LD adaptor protein, Jabba [1, 45]. This allows for the storage and proper balance of the massive amounts of histones needed during rapid embryonic nuclear divisions, and it prevents excess histones from interfering with normal embryonic development [1].

Also found on LDs are the enzymes of dolichol synthesis (e.g., Rer2; [46]) and other enzymes involved in the proximal steps of the pathway generating glycans on dolichol for N-linked glycosylation in the ER (e.g., ALG14; [46, 47]). The significance and function of the LD localization of these proteins are unknown, although dolichols stored as dolichol esters are likely found in LDs [48]. Similarly, UBX domain-containing protein 8 (UBXD8), a protein of the ubiquitin-proteasome system, is consistently identified as a LD protein [49, 50]. In the ER, UBXD8 functions in ER-associated protein degradation (ERAD) by recruiting the large-protein-extraction ATPase VCP/p97 [51]. It also has been reported to be a sensor of unsaturated FAs, regulating their incorporation into neutral lipids [52]. The function of LD-localized UBXD8 is unclear.

A number of other proteins localize to the interface between LDs and other organelles: Mdm1 localizes to LD-vacuole contact sites in yeast [53], and Snazarus localizes to LDs near the plasma membrane in Drosophila larvae lipid deposits [54]. Little is known about the LD targeting mechanisms of these proteins, and they are reviewed elsewhere [55].

Some proteins of important intracellular pathogens also localize to LDs. For instance, flaviviruses, such as hepatitis C, Dengue, and Zika virus, use the surfaces of LDs as part of their replication site in cells [5]. The hepatitis C core protein localizes to LDs, and its overexpression is sufficient to cause steatosis in mice [56]. Notably, viperin, an innate immune response protein, also localizes to LDs [57], suggesting that these organelles serve as battlegrounds during viral infection. Further, a number of proteins involved in innate immunity in response to bacterial infections have been found in LD fractions of cells in which LDs are induced by lipopolysaccharide [6]. Thus, an important function of LDs in immune cells is being increasingly recognized and is reviewed elsewhere [58].

Principles of protein targeting to LDs

The formation of LDs from the ER [59, 60] results in the generation of an organic phase in the cytosol, effectively converting cells into emulsions. The LD oil phase is covered by a monolayer of phospholipids that makes up the interface with the aqueous cytosol and acts as a surfactant to lower surface tension (Figure 1, left). As a consequence, LD proteins must recognize and localize to an interface that has unusual biophysical properties among cellular organelles. How proteins interact with this phase boundary is governed by the principles of emulsion physics [61].

Figure 1. Unique features of the lipid droplet (LD) surface influence protein targeting.

Left: LDs possess a unique structure consisting of a core of neutral lipids [e.g., triglycerides (TGs) or sterol esters] surrounded by a monolayer of phospholipids. Middle: Due to their dynamic character, LDs undergo constant expansion and shrinkage, which can affect the phospholipid packing density of the LD surface. Inset: Properties, such as phospholipid packing defects, limited surface area, and the neutral lipids that underlie the phospholipid monolayer are all features that affect the mechanisms of LD protein targeting. For instance, LDs can be rich in TGs or sterol esters, which can influence their surface properties and the proteins that target to LDs. In the case of TG-rich LDs, some TG molecules can be located at the LD surface, at sites where phospholipid packing defects exist, and align themselves with the phospholipid monolayer plane. These are known as surface-oriented TG molecules (SURF-TG, magenta). SURF-TG behave as structural components of the LD monolayer, while also altering its surface tension to promote the LD targeting of specific proteins.

Studies of protein targeting to LDs have yielded a number of important insights. First, LD protein targeting entails mechanisms that differ markedly from those for targeting proteins to bilayer membrane-bound organelles, such as the ER, mitochondria, or peroxisomes. For example, there appears to be no dedicated protein machinery on LDs that mediates the direct insertion of proteins into the LD surface, nor have specific lipids (such as different phosphoinositides employed by other organelles) been identified as markers for directing proteins to the LD surface.

Second, although the phospholipid monolayer covering LDs has a composition similar to the ER membrane [62, 63], its biophysical properties are unusual (Figure 1, inset). Similar to bilayer membranes, the monolayer surface includes phospholipid packing defects, which are sites where the phospholipid tails are dynamically exposed to water in the cytosol. Molecular modeling suggests that packing defects are more frequent, larger, and more persistent on the LD surface than in bilayer membranes [64, 65]. Such modelling studies also indicate that LD packing defects are chemically distinct: the TG acyl chains are intercalated into the phospholipid tail region, and in some instances, entire TG molecules are integrated so that their glycerol backbones align with the phospholipids and act as monolayer components [65]. These surface-oriented TG molecules (SURF-TG) (Figure 1, inset) are unique to LD packing defects and slightly increase LD surface tension relative to bilayer membranes. Additionally, the types of exposed neutral lipids may vary (i.e., sterol esters or TGs) and influence the surface properties of LDs, resulting in different proteins binding to specific types of LDs [41, 66].

Third, the LD surface changes dynamically during the LD lifecycle in ways that may affect protein targeting (Figure 1, left and middle). For example, during LD expansion, the LD surface stretches and the integration of new SURF-TG molecules into the monolayer decreases the induced surface tension, enabling new proteins to target altered surface properties [67] (see details below). Similarly, when LDs shrink, weakly bound proteins become displaced due to molecular crowding [68] and competition for the fewer available packing defects. This provides a mechanism to modulate the protein composition of LDs, based on the biophysical properties of the LD surface and the protein binding to the interface.

Given these unique features of the LD monolayer surface, how do proteins detect and localize to LDs? At least two major and distinct protein targeting pathways appear to operate in cells. In one, which we term CYTOLD (for “cytosol to LD”) targeting, proteins are initially translated in the cytosol and subsequently bind to the LD surface. In the second, ERTOLD (for “ER to LD”) targeting, proteins are initially inserted into the ER membrane and then relocalize from there to the LD surface.

CYTOLD targeting pathway

The CYTOLD targeting pathway (Figure 2) begins with the translation of proteins in the cytosol. Examples of proteins that use CYTOLD targeting are listed in Table 1. Many CYTOLD-targeted proteins contain amino acid sequences with the potential to form amphipathic helices. Some sequences are folded regions of a protein, but others are unstructured in the cytosol until they associate with a bilayer membrane or LD monolayer, which facilitates their folding into amphipathic helices with hydrophobic and polar residues on opposing faces (Figure 2, left and middle). Perilipins, for instance, are CYTOLD-targeted proteins that contain many of these sequences, with each corresponding to an 11-amino acid repeat [69]. Perilipin 3 (PLIN3) appears to bind LDs early after their formation [60, 70]. These amphipathic sequences might be shielded from aggregation in the cytosol by binding chaperones. The cytosolic pool of perilipin 1 (PLIN1) is rapidly degraded by the ubiquitin-proteasome system if the protein is not bound to LDs. Thus, the LD pool of PLIN1 seems to be protected from degradation [71].

Figure 2. CYTOLD pathway for lipid droplet (LD) protein targeting.

Proteins that access and bind the surface of LDs from the cytosol via the “cytosol to LD” (CYTOLD) pathway are soluble and often characterized by containing one or more amphipathic helix (AH) motifs for LD binding. Left: A model for the CYTOLD pathway starts with the synthesis of proteins in the cytosol. Middle: Once fully released from the protein synthesis machinery, AH-containing proteins are thought to be unfolded in the cytosol and interact with packing defects at the surface of LDs (which form within the endoplasmic reticulum (ER) membrane at assembly complexes containing seipin). Inset: The AH motifs of these proteins mediate their binding to the LD monolayer surface through a thermodynamic mechanism: i) first, the unfolded AH motif is in a high free energy (E) state in the cytosol; ii) large hydrophobic residues in its sequence bind to phospholipid packing defects on the LD surface, anchoring the protein to it; iii) as more large hydrophobic residues bind to packing defects, the AH motif folds on the surface of LDs. This renders the protein in a lower free energy state and stabilizes its binding to the surface of LDs. Right: As the metabolic needs of the cell change and neutral lipids are mobilized, LDs may shrink in size leading to an increase in protein density at their surfaces. As a consequence, weakly bound proteins, such as AH-containing proteins, may be displaced from the LD surface due to protein crowding, releasing them back into the cytosol.

Table 1.

Examples of CYTOLD-targeted proteins

Protein name	Function	References	Cell types (examples)
Perilipins (1–5)	Coat the surface of LDs early during their biogenesis, preventing their breakdown by lipases and regulating their metabolism	[9, 42, 69, 73]	3T3-L1 adipocytes, mouse Leydig tumour cells (MLTC-1), HeLa, COS-7, Drosophila S2, S. cerevisiae (yeast)
Comparative gene identification-58 (CGI-58/AB5HD)	Co-activator of the lipase ATGL/PNPLA2	[101, 102]	3T3-L1 adipocytes
CTP:phosphocholine cytidylyltransferase (CCT)	Catalyzes the key rate-limiting step in the CDP-choline pathway for phosphatidylcholine biosynthesis	[19]	Raw267.2 and bone marrow-derived macrophages (BMDMs), Drosophila S2
Max-like protein X (MLX) and members of the MLX family of transcription factors	Activate the transcription of glycolytic genes in a glucose-responsive manner	[41]	THP-1 macrophages, mouse primary hepatocytes, HEPG2, U2OS, COS-7, SUM159

Much of our current understanding of the CYTOLD targeting pathway comes from the analysis of model peptides (e.g., peptides derived from perilipins or CCT) for which biochemical, biophysical, and molecular dynamics simulation studies have been performed [19, 64–66, 69, 72, 73]. Experimental studies of CCT indicated that multiple large hydrophobic residues (e.g., leucines, phenylalanines, or tryptophans) were required to bind LDs; complementary simulations demonstrated initial association of these residues with phospholipid packing defects [64] (Figure 2, inset). Subsequent simulations showed that a single tryptophan is critical for the initial association of the peptide with the glycerol moiety in SURF-TG packing defects, and that three additional phenylalanines require neighboring defects to be present in expanding LDs [67]. Large tryptophan residues also appear to be required for the binding of CGI-58 to LDs [74]. The resulting model of large hydrophobic residues binding to LD surface packing defects predicts that parameters that affect the abundance of packing defects on LDs (e.g., phospholipid composition, protein abundance, or the underlying oil expansion or shrinkage) impact the LD binding of amphipathic helical proteins [64, 65, 67].

Once LD binding is initiated, proteins that are unstructured in the cytosol, such as CCT, fold into an α-helix [75, 76], stabilized by hydrophobic interactions with the acyl chains of phospholipids and TGs at the LD surface and polar/charged interactions with phospholipid head groups and water. Thus, CYTOLD targeting is often energetically coupled to protein folding at the surface of LDs, which leads to very slow unbinding rates for strongly associated proteins, such as perilipins.

The principles of CYTOLD targeting may play a large role in governing the protein composition of LDs. Studies have shown that the amphipathic helical regions of many cellular proteins can bind to LD monolayer surfaces [64, 68], thus prompting the question of how specificity is maintained for genuine LD proteins. One feature that may contribute to determining specificity is the competition between proteins for available LD surface. For instance, increasing protein density at the LD surface, which occurs during lipolysis, leads to macromolecular crowding and the displacement of more weakly bound proteins [68] (Figure 2, right). Similarly, competition for packing defects may limit LD adsorption and binding of bona fide LD proteins and it may be a key factor governing LD protein composition.

Another mechanism that potentially regulates the dynamic changes in LD protein composition is the post-translational modification of LD proteins. CCT is known to undergo phosphorylation [77], which may affect its localization. Further, high fat diet-induced hepatic lipid accumulation in mice causes changes in the phosphoproteome that include phosphorylation events correlating with the localization of proteins to LDs [78].

ERTOLD targeting pathway

Examples of proteins that target LDs via the ERTOLD pathway (Figure 3) are shown in Table 2. ERTOLD-targeted proteins commonly contain hydrophobic, membrane-embedded sequences, often disrupted in the middle by prolines, which suggests they adopt a hairpin conformation in the ER membrane [79]. Proteins that completely span the ER bilayer and have luminal loops or domains will not target to LDs [15], as embedding of the luminal hydrophilic segments in neutral lipids would be energetically unfavorable.

Figure 3. ERTOLD pathway for lipid droplet (LD) protein targeting.

Proteins that access and bind the surface of LDs from the endoplasmic reticulum (ER) via the “ER to LD” (ERTOLD) pathway usually contain hydrophobic, membrane-embedded motifs that are predicted to acquire a hairpin conformation in bilayer membranes. Left: A model for the ERTOLD pathway starts with proteins being synthesized and inserted into the ER membrane. Middle: During LD formation at ER sites defined by a LD assembly complex (which includes seipin), some hairpin-containing proteins, such as LiveDrop, relocalize to nascent LDs. Insets: Sequence features of these hairpin motifs mediate their LD targeting and accumulation: i) in the ER bilayer membrane, specific sequence features of the hairpin motif (i.e., positively charged residues, red, “+”) anchor the protein to the cytosolic and luminal leaflets of the membrane; ii) this positions large hydrophobic residues, such as tryptophans (purple, “W”), in an energetically unfavorable position within the middle of the hydrophobic plane of the ER membrane; iii) upon LD localization, some positively charged residues are released and reorient towards the monolayer interface (simulations suggest that others may remain embedded below the LD surface, interacting with water molecules that access the LD core); iv) this also allows the reorientation of tryptophans towards the phospholipid packing defects, lowering the free energy (E) state of the protein and further stabilizing the LD surface. Right: At later stages, specific hairpin-containing proteins, such as GPAT4, can be recruited to mature LDs from the ER, presumably via membrane bridges, to mediate LD-localized triglycerides (TG) synthesis.

Table 2.

Examples of ERTOLD-targeted proteins

Protein name	Function	References	Cell types (examples)
Glycerol-3-phosphate acyltransferase 4 (GPAT4)	Catalyzes the conversion of glycerol-3-phosphate and long-chain acyl-CoA to lysophosphatidic acid (LPA), which is the first and rate-limiting step of the de novo TG synthesis pathway	[15]	Drosophila S2, COS-7
Acyl-CoA synthetase long Chain family member 3 (ACSL3) and 5 (ACSL5)	Catalyze the esterification of FAs with CoA in order to chemically “activate” them for further TG synthesis	[13, 14]	COS-7, HepG2, mouse enterocytes
Adipose TG lipase (ATGL) or PNPLA2	Catalyzes the initial step of TG hydrolysis in adipocyte and non-adipocyte LDs	[28]	HeLa
Oleosins	Coat the surface of oil bodies (i.e., LDs from plant cells) to stabilize them, especially in seeds that undergo desiccation	[11]	Maize seeds
UDP-N-acetylglucosaminyltransferase subunit (ALG14)	Catalyzes the second step of the dolichol-linked oligosaccharide pathway	[46, 47, 103]	Drosophila S2, HeLa, S. cerevisiae (yeast)
Acyl-CoA:lysophosphatidylcholine acyltransferase (LPCAT)	Catalyzes the conversion of lysophosphatidylcholine (LPC) into phosphatidylcholine (PC), as part of the phospholipid remodeling pathways	[20, 47, 104]	Drosophila S2, human A431
Thioredoxin reductase-like selenoprotein or SelT-like protein	Probably has thioredoxin reductase-like oxidoreductase activity	[47]	Drosophila S2
1-acylglycerol-3-phosphate O-acyltransferase 3 (AGPAT3)	Catalyzes the conversion of LPA into phosphatidic acid (PA), as part of the de novo biosynthesis pathways of diacyl phospholipids and TGs	[47]	Drosophila S2
Multiple C2 and transmembrane domain-containing protein 2 (MCTP2)	Defines ER domains that serve as formation sites for preperoxisomal vesicles and LDs	[47, 105]	Drosophila S2, COS-7, S. cerevisiae (yeast)
UBX domain-containing protein 8 (UBXD8)	Recruits the ATPase p97/VCP to facilitate the extraction or dislocation of ERAD substrates from the ER membrane	[49]	Huh7, HeLa
Spastin	When localized to the surface of LDs, it tethers LDs to peroxisomes and supports the traffic of FAs between both organelles	[100]	HeLa
Fatty acid transport protein (FATP)	Mediates the vectorial acylation of FAs, which drives their movement across membranes and results in the production of acyl-CoA	[68]	Drosophila S2
Lipid droplet-associated hydrolase (LDAH)	Serine hydrolase	[106]	Drosophila S2, HeLa
Associated with LD protein 1 (ALDI)	Putative methyltransferase	[107]	COS-7, rat hepatocytes
Methyltransferase-like protein 7A (METTL7A) or AAM-B	Putative methyltransferase	[108]	HeLa, COS-7, S. cerevisiae (yeast)
Ancient ubiquitous protein 1 (AUP1)	Involved in ERAD and, when localized to LDs, it binds the E2 conjugase Ube2g2 which links LDs to the ubiquitination machinery	[109, 110]	COS-7, human Huh7, human A431, dog MDCK
Caveolin-1 (Cav1) and caveolin-2 (Cav2)	Associate with each other to form a complex that targets membrane lipids and drives the formation of caveolae	[111]	Human A431, Fisher rat thyroid (FRT), baby hamster kidney (BHK)
NAD(P)H steroid dehydrogenase-like (NSDHL)	Sequentially removes two C-4 methyl groups from the cholesterol precursor lanosterol during cholesterol biosynthesis	[112, 113]	HeLa, COS-7, Chinese hamster ovary (CHO)
Lipid droplet assembly factor 1 (LDAF1) or promethin	Interacts with seipin in the ER to form the LD assembly complex (LDAC), which defines the sites of LD formation	[60, 114]	SUM159 breast cancer cells, MCF7 breast cancer cells

The biogenesis of ERTOLD-targeted proteins begins with their translation and insertion into the ER membrane. The initial ER insertion step may occur through the canonical translocon-mediated pathway. UBXD8 and possibly other ER proteins use an alternative, post-translational pathway requiring the cytosolic protein Pex19 and the ER protein Pex3, which are best known for their roles in peroxisome biogenesis [80].

Once in the ER, proteins following the ERTOLD pathway appear to target the LD surface by moving from the ER bilayer membrane to the LD monolayer surface via ER-LD membrane connections [15, 81] (Figure 3, left). Early studies with fluorescence recovery after photobleaching in the yeast Saccharomyces cerevisiae suggested that yeast LDs are always physically connected to the ER, allowing proteins to travel to the LD surface [82]. In yeast, fly, and mammalian cells, physical connections between LDs and the ER appear to be maintained at sites of LD formation via LD assembly complexes (LDAC) that contain seipin [60, 81, 83–87]. In Drosophila cells, a few ER-localized proteins (e.g., the membrane-embedded region of GPAT4 (known as LiveDrop; [79]), LDAF-1 [60], or ACSL3 [13]) have been shown to target LDs during their formation. Many other ERTOLD-targeted proteins, including full-length GPAT4, do not localize to LDs during their biogenesis but instead target mature LDs well after their formation [15] (Figure 3, right). What determines which proteins localize to LDs during their formation or at later stages is unclear.

The ERTOLD targeting of proteins to mature, established LDs appears to happen via ER-LD membrane bridges that form after LD biogenesis. In Drosophila cells, where the LD targeting of GPAT4 has been well characterized, ER-LD membrane bridges can be visualized by electron tomography [15]. Fluorescence recovery after photobleaching experiments suggest that these membrane bridges are used by GPAT4 to rapidly relocalize from the ER to the LD surface [15].

How ER-LD membrane bridges are established and maintained remains a mystery. Genome-wide screens searching for LD phenotypes identified members of the Arf1/COPI membrane trafficking complexes as required for normal LD morphology and the LD targeting of some ER proteins [88, 89]. The Arf1/COPI machinery is best known for its function in the formation of coated vesicles that mediate retrograde vesicular trafficking in the secretory pathway [90]. How Arf1/COPI participates mechanistically in ER-LD bridge formation in cells is unknown, but it might be by modulating the properties of the LD surface. Specifically, in vitro studies have shown that Arf1/COPI proteins can mediate the budding of nano-LDs from LD surfaces [91], thereby removing phospholipids and increasing the surface tension of LDs, possibly rendering them in a fusion-competent state that allows them to connect to the ER [92].

Many ERTOLD-targeted proteins not only localize to LDs but also accumulate there. This accumulation requires an energy difference that prevents proteins from equilibrating between the ER and LDs. Otherwise, a high proportion of most proteins would stay in the massively more abundant ER. How then do some ERTOLD-targeted proteins accumulate at LDs? Although this question has not been examined in the context of full-length proteins, recent studies of short, membrane-embedded hairpin motifs, performed in cells and using molecular simulations, have shed light on the mechanism of ERTOLD targeting and LD accumulation (Figure 3, insets). Studies of the membrane-embedded domains of GPAT4 and ALG14, which likely adopt a hairpin conformation in the ER membrane, suggest that a specific combination and distribution of sequence features are important for their accumulation at LD surfaces. Each of these hairpin motifs possess positively charged residues within their hydrophobic segments, which appear to anchor them to the luminal leaflet of the ER bilayer membrane. This forces some residues into regions of the bilayer membrane that are energetically less favorable. For example, tryptophans, which are most stable in the phospholipid glycerol region where they satisfy both hydrophobic interactions and hydrogen bonding, are positioned in the middle of the phospholipid tails region [79]. Upon relocalization to the LD surface, this strain appears to be relieved as tryptophans and positively charged residues reorient towards the interface and surface of the LD phospholipid monolayer, resulting in an energetically more favorable conformation of the hairpin motifs (Figure 3, insets). Thus, the current model posits that the free energy difference for a hairpin motif to be in the ER and at a LD surface provides the driving force for its LD accumulation [79].

This thermodynamic model does not rule out additional factors that might contribute to the selective accumulation of ER proteins on LDs. For instance, differential oligomerization states in the different compartments may contribute. Also, some ERTOLD-targeted proteins may be degraded faster in the ER than on the LD surface, potentially contributing to the apparent accumulation of such proteins on LDs [93]. However, this mechanism does not appear to be sufficient to explain the LD accumulation of the LiveDrop sequence from GPAT4 [79].

Are there other pathways for protein localization to LDs?

Whether pathways other than CYTOLD and ERTOLD exist for LD protein targeting remains to be determined. For instance, one could imagine that lipid anchors, such as palmitate, myristate, or prenyl groups, found covalently attached to some proteins, could mediate targeting to LDs. Indeed, proteomic studies routinely identify a large number of lipidated Rab proteins on LDs, and some of them are associated with LD phenotypes when depleted from specific cell types [94–96]. However, microscopy evidence showing endogenous Rab proteins localizing specifically to LDs remains sparse. Therefore, at least some Rab proteins seem to partition to LDs nonspecifically during the biochemical purification of LDs.

Concluding remarks

Since the discovery of perilipins as specific LD proteins, new LD proteins have been identified, and the molecular understanding of how these proteins target to LDs via the CYTOLD and ERTOLD pathways has grown enormously. These studies have yielded fascinating cell biology insights. Perhaps more importantly, the associations between many LD proteins (e.g., ATGL, PNPLA3, HSD17B13, and CCT) and human metabolic diseases highlight the need to better understand the principles of LD protein targeting. Yet many important and compelling aspects of LD protein targeting remain mysterious (see Outstanding Questions). We anticipate that additional studies employing new proteomic, microscopy, structural biology, and biophysical approaches will rapidly lead to further insights into the mechanisms of LD protein targeting and its functional consequences.

Outstanding questions.

• What is the composition of the lipid droplet (LD) proteome in different cell types (e.g., brown adipocytes, cardiac myocytes, neutrophils)?

• What are the functional consequences of having different populations of LDs in the same cell?

• How does the LD proteome change with, and influence, the LD lifecycle under different cellular conditions?

• Are there other LD targeting pathways besides CYTOLD and ERTOLD?

• What factors control the specificity of the LD protein composition?

• Will the emergence of sequence-specific features for proteins that target to LDs enable computational predictions of LD localization?

• Are chaperones involved in the initial steps of the CYTOLD pathway?

• How are the ER-LD membrane bridges used for late ERTOLD targeting established? What factors control this process?

• How do the core neutral lipids affect the targeting of proteins to LDs?

• How does the phospholipid composition of the LD monolayer surface affects LD protein targeting?

• How are proteins selectively removed from LDs? Can ERTOLD-targeted proteins migrate back to the ER for ERAD removal?

• What machinery mediates selective autophagy (lipophagy) of LDs?

• Are LDs important for detoxifying the ER from specific proteins or lipids?

• How does the LD targeting of specific proteins facilitate the replication of some viruses?

• How does aberrant LD protein targeting leads to metabolic diseases?

Highlights.

• Lipid droplet (LD) organelles contain specific proteins, including metabolic enzymes, that differ among cell types.

• Proteins targeting the monolayer surface of lipid droplets do not follow the same principles as proteins targeting bilayer membranes.

• The targeting of proteins to lipid droplet surfaces is governed by the biophysical features of their unique oil-water interface.

• LD protein targeting occurs via two major pathways, CYTOLD and ERTOLD targeting.

• Elucidating the principles driving these protein targeting pathways has implications for further understanding physiology and metabolic diseases.

Glossary

Dolichol

membrane lipid that corresponds to isoprenoid polymers, also known as polyisoprenoid alcohols. They are synthesized on the cytosolic side of the ER membrane and exist as a mixture of species of different chain lengths. During N-linked protein glycosylation, they function as lipid carriers of the glycan molecules destined to be conjugated to proteins.

LD assembly complex (LDAC)

protein complex of ~600 kDa that forms in the ER membrane by oligomers of seipin and accessory proteins such as LDAF1 (Ldb16 and Ldo protein in yeast). LDACs define the sites of LD formation in the ER membrane and are thought to promote the nucleation and phase separation of TG lenses.

Lipid anchor

lipids covalently attached to proteins and also integrated into membranes, allowing the stable interaction of proteins with membranes. There are three main types of lipid anchors: isoprene polymers (e.g., farnesyl, geranylgeranyl), FAs (e.g., saturated myristic acid and palmitic acid), and glycosylphosphatidylinositol (GPI).

Lipodystrophy

conditions in which patients are unable to produce and maintain healthy fat tissue, resulting in metabolic complications (e.g., fatty liver, insulin resistance). They are caused by preventing the differentiation, survival, or functionality of adipocytes. Lipodystrophies can be classified based on their origin (i.e., congenital vs acquired) or on the extent the fat tissue is compromised (i.e., generalized vs. partial).

Lipolysis

the catabolism or degradation of the TGs stored in LDs. It occurs in three sequential steps, mediated by the enzymes ATGL, HSL, and monoglyceride lipase (MGL), and produces FAs and glycerol.

Non-alcoholic fatty liver disease (NAFLD)

liver condition characterized by the excessive accumulation of fat in the liver (i.e., hepatic steatosis) that is caused by metabolic disease, often with insulin resistance, rather than excessive alcohol intake, medication, or hereditary disorders. Specifically, it’s defined by the accumulation of large and heterogeneously sized LDs, consisting mainly of TGs, in more than 5% of hepatocytes.

Non-alcoholic steatohepatitis (NASH)

condition that results from the progression of NAFLD to more serious conditions, such as hepatocyte inflammation and liver tissue fibrosis. Chronic liver fibrosis and damage can lead to cirrhosis and liver failure.

Oleosins

proteins of around 15–26 kDa that are characteristic of LDs from plant cells, particularly, of seed LDs. They play a structural role as they coat the surface of LDs, providing stability and preventing them from coalescing with other LDs. Oleosins contain a long, hydrophobic hairpin motif (72 amino acids) which is involved in its targeting to LDs from the ER.

Seipin

protein encoded by the Berardinelli-Seip congenital lipodystrophy type 2 (BSCL2) gene, which can harbor mutations causing lipodystrophy. It forms oligomers in the ER membrane and it is involved in defining the sites of LD formation (as part of the LDAC) and in stabilizing the ER-LD connections.