Targeting of the Drosophila protein CG2254/Ldsdh1 to a subset of lipid droplets

Cells store energy-rich neutral lipids in specialized organelles, the so-called lipid droplets (LDs). While LDs have been considered simple and static lipid inclusions for a long time, they are now recognized as true cellular organelles (for example, reviewed in Farese and Walther, 2009). LDs consist of a hydrophobic core of storage lipids, which is surrounded by a phospholipid monolayer to which proteins are attached. While LDs generally share this apparently simple structure, LDs from different cell types and tissues show a remarkable plasticity in terms of their individual size or size distribution (Beller et al., 2010a). Similarly, the collection of proteins attached to the LD phospholipid monolayer (Martin et al., 2005; Straub et al., 2008, 2010; Thiele and Spandl, 2008; Wilfling et al., 2013; Wolins et al., 2006) or the LD lipid content (Hsieh et al., 2012) can vary remarkably between cell types but also within cells. It is unknown how cells realize these different subsets of LDs (hereafter LD subsets), whether the different protein compositions of LD subsets are of functional significance and whether distinct subcellular LD populations are also present in vivo under non-pathological conditions.

Here, we initiated an analysis of the subcellular LD diversification using the annotated Drosophila melanogaster short-chain dehydrogenase/reductase (SDR) CG2254 as a LD subset marker. CG2254 was previously identified in an unbiased proteomics screen geared to systematically identify the proteins associated with the LDs of the cells of the fat body, the fat storing organ of Drosophila (Beller et al., 2006). SDRs are a subtype of dehydrogenases and represent a large and ancient protein family with more than 160,000 members in sequence databases (Persson and Kallberg, 2013). SDRs act on various substrates and it is difficult, if not impossible, to predict the substrate specificity on the basis of their primary amino acid sequence (Oppermann et al., 2003). Cellular dehydrogenases, such as the ER-localized 11-β-hydroxysteroid dehydrogenase type 1 (Morton, 2010; Staab and Maser, 2010; Tomlinson et al., 2004) or the mitochondrial medium- and short-chain L-3-hydroxyacyl-CoA dehydrogenase (Schulz et al., 2011), have been identified as important regulators of lipid metabolism. This suggests that SDRs represent suitable drug targets for the treatment of metabolic diseases including diabetes, atherosclerosis and obesity (Kushner, 2008; Staab and Maser, 2010).

Previous proteomics screens have identified multiple dehydrogenases in LD preparations from cells and tissues of diverse species (Bartz et al., 2007; Beilstein et al., 2013; Beller et al., 2006; Bouchoux et al., 2011; Brasaemle et al., 2004; Cermelli et al., 2006; Deisenroth et al., 2011; Du et al., 2013; Hodges and Wu, 2010; Krahmer et al., 2013; Larsson et al., 2012; Wang et al., 2015; Zhang et al., 2012). In murine samples, for example, at least 18 dehydrogenase enzymes have been identified so far (Table S1). This finding suggests a prominent role of LD-associated SDRs in LD function or general lipid metabolism. For five out of the 18 murine dehydrogenases, the subcellular localization has been investigated. The NAD(P)H steroid dehydrogenase-like protein (Nsdhl) (Ohashi et al., 2003), the short-chain dehydrogenase/reductase 3 (DHRS3) (Deisenroth et al., 2011), the 17-β-hydroxysteroid dehydrogenase type 13 (17-βHSD13; also known as HSD17B13) (Horiguchi et al., 2008a) as well as the 17-β-hydroxysteroid dehydrogenase type 11 (17-βHSD11; also known as HSD17B11) (Horiguchi et al., 2008b; Yokoi et al., 2007) are localized to the ER in the absence of LDs and translocate to the LD surface once LDs become visible in the cytoplasm. The retinol dehydrogenase RDH10 (Jiang and Napoli, 2013), however, was found in association with mitochondria and translocated upon lipid storage induction to LDs. Notably, all previous studies focused on a general LD localization of the respective protein without taking a possible LD subset localization into account.

We have shown that transgene-derived EGFP-tagged CG2254 serves as a LD subset marker within tissue culture and fat body cells (Beller et al., 2006). However, the cause and mechanism of this localization, as well as the nature of the CG2254-decorated LD subsets, remained unresolved. Here, we report that the CG2254 protein decorates metabolically active growing LDs. Localization of CG2254 depends on two sequence motifs, one for the general LD targeting and the second for restricting CG2254 association to a subset of LDs. This second motif is sufficient to restrict the general LD-associated protein CG9186 (Thiel et al., 2013) to LD subsets.

In vivo analyses of endogenous CG2254 protein localization revealed the existence of LD subsets in a whole-organism context and show that the pattern of CG2254 localization is highly dynamic. In the fat body of Drosophila larvae, as well as in different regions of the midgut, CG2254 is associated with all or at least the vast majority of LDs when the larvae were consuming food, and it becomes restricted to LD subsets when larvae stop the food intake prior to pupariation.

The Drosophila CG2254 gene is located at cytogenetic position 7D5 of the X-chromosome (Gramates et al., 2016). It encodes a single transcript which gives rise to a 320-amino-acid-long polypeptide (Fig. 1A,B). The protein harbors a characteristic Rossmann-type fold NAD(P)-binding motif (Gramates et al., 2016), which is diagnostic for short-chain dehydrogenases/reductases (SDRs), and an N-terminal hydrophobic region (residues 19–41) predicted to form a transmembrane helix (in silico analyses with TMHMM Server v. 2.0, data not shown). Initially, CG2254 was identified in a proteomics screen geared to identify the LD-associated proteins in Drosophila third-instar larval fat body cells, and an overexpressed EGFP-tagged CG2254 protein was indeed associated with LDs of third-instar larval Drosophila fat body cells (Beller et al., 2006). Endogenous CG2254 is not present in tissue culture cells, as transcripts were not detected in commonly used Drosophila tissue culture cell lines such as Kc167 or S2 according to the modENCODE database (Gramates et al., 2016) nor was the transcribed protein detected in Kc167 cells (Fig. S1). Nevertheless, the EGFP-tagged CG2254 protein was localized on LDs in embryonic Kc167 cells (Fig. 1C; Fig. S2) and neuronal Drosophila ML-DmBG3-c2 cells of larval origin (Fig. 1C). Intriguingly, the protein does not localize to all cellular LDs, but only a subset of them. The available data therefore suggest that CG2254 is a LD subset-associated short-chain dehydrogenase/reductase. Based on these findings, we propose the name Lipid droplet subset dehydrogenase 1 (Ldsdh1) for CG2254.

In order to analyze the subcellular localization of the endogenous Ldsdh1 protein in vivo, we raised a specific antiserum against a peptide within the Ldsdh1 amino acid sequence (see Fig. 1B and Materials and Methods). The specificity of the antiserum was tested by western blot analysis using overexpression and knockdown conditions as controls (Fig. S1).

To rule out the possibility that Ldsdh1 is only in the proximity of rather than associated with LDs, we purified LDs from third-instar larval fat body cells by discontinuous sucrose gradient centrifugation using established protocols (Beller et al., 2006; Brasaemle and Wolins, 2016) and examined the distribution of Ldsdh1. As controls, we used marker proteins across the different collected fractions by Western blot analysis employing the corresponding specific antibodies. The protein alcohol dehydrogenase (Adh) was mainly found in the cytosolic fractions of the gradient with little protein present in the LD-containing fractions. Ldsdh1, as well as the LD-associated marker protein Perilipin 2 (PLIN2; also known as Lsd-2 in files) (Cermelli et al., 2006; Fauny et al., 2005; Grönke et al., 2003; Teixeira et al., 2003), was highly enriched in the LD fractions (Fig. 1E). While PLIN2 was mostly limited to these fractions, Ldsdh1 was also detected in the microsomal membrane fraction of the gradient and in small amounts in the cytosolic fractions. Thus, Ldsdh1 localization is highly enriched but not limited to LDs.

To establish the sites of Ldsdh1 expression in the organism, we performed RNA in situ hybridizations of whole-mount embryos at various developmental stages with Ldsdh1 antisense probes. Fig. 2A shows an enrichment of transcripts in embryonic fat body cells. During the postembryonic stages, Ldsdh1 expression is reported to be more widespread based on published microarray (Chintapalli et al., 2007) and next-generation sequencing data (Gelbart and Emmert; FlyBase high-throughput expression pattern data; http://flybase.org/reports/FBrf0221009.html). Ldsdh1 transcripts appear to be enriched in multiple lipid-storing tissues such as the midgut or the salivary gland. In adult stages, Ldsdh1 expression is fairly widespread and strong in tissues that have a lipid storage function, such as the fat body and the midgut, as well as in tissues that have not so far been shown to have a clear-cut lipid storage function, such as the eye, the brain and the spermatheca. These results suggest a prominent role of Ldsdh1 within bona fide as well as non-classical lipid-storing tissues.

The distribution of the Ldsdh1 protein in embryos as detected by antibody staining corresponded to the distribution of the RNA message (Fig. 2A,B), and shows dispersed Ldsdh1-positive circular-shaped signals in the fat body. Since LDs are usually tightly packed in the embryonic fat body (see, for example, Beller et al., 2010b), our findings suggest the presence of endogenous Ldsdh1 protein at LDs and the existence of Ldsdh1-decorated LD subsets in fat body cells of fly embryos.

Ldsdh1 was initially identified in a LD proteomics screen carried out with third-instar larval fat bodies (Beller et al., 2006). Accordingly, we identified Ldsdh1-positive LDs in fat bodies of wandering third-instar larval pre-pupae (Fig. 2C). The Ldsdh1 signal in third-instar fat body cells clearly decorated only a fraction of the cellular LDs. This observation demonstrates different subsets of LDs in the organism, of which one is marked by Ldsdh1 association.

In order to further investigate the presence of Ldsdh1-associated LD subsets in vivo, we examined the localization pattern of Ldsdh1 in digestive tracts dissected from early L3 larvae where the Ldsdh1 gene is prominently expressed (Gelbart and Emmert; FlyBase high-throughput expression pattern data; http://flybase.org/reports/FBrf0221009.html). We chose early L3 larvae because LDs are more abundant in the midgut at this stage, and therefore easier to visualize than in mature pre-pupa L3 larvae (see below). We detected numerous ring-shaped structures with the anti-Ldsdh1 antiserum in the m2 region of the gut (region description based on Murakami et al., 1999), which is located close to the proventriculus (Fig. 2D). Thus, the localization pattern of Ldsdh1 in cells is therefore different in diverse tissues of the organism since in the fat body of mature pre-pupae L3 larvae, Ldsdh1 is localized at LD subsets (Fig. 2C), whereas in the midgut of early L3 larvae, Ldsdh1 targets the vast majority of the LDs (Fig. 2D).

We asked next whether the overexpressed Ldsdh1 fusion protein shows the same localization pattern as endogenous Ldsdh1. To test for a possible influence of the metabolic status and/or the age of the animal on the subcellular localization of Ldsdh1, we expressed EGFP-tagged Ldsdh1 both in the midguts and fat bodies of early as well as mature pre-pupae L3 larvae. In the midguts of early L3 larvae, we detected the overexpressed protein on most, if not all, cellular LDs (Fig. 2E), whereas in the guts of mature pre-pupae L3 larvae Ldsdh1 was found on LD subsets (Fig. 2F). At the same time, the overall number of LDs appeared to be strongly reduced at the later stage. Intriguingly, in the guts of early L3 larvae, LDs were localized in a polarized fashion with a clear-cut enrichment at the apical side of the gut cells facing the gut lumen (Fig. 2E). We found the same EGFP–Ldsdh1 localization pattern in fat bodies of both early (Fig. 2G) and mature pre-pupae L3 larvae (Fig. 2H). Thus, the patterns of endogenous Ldsdh1 and the overexpressed EGFP-tagged Ldsdh1 protein were similar and dynamic. This suggests that the localization of Ldsdh1 could depend on the age and/or metabolic status of the animals, since early L3 animals are constantly feeding, whereas the mature L3 pre-pupae have stopped eating and left the food.

In order to test whether the localization of Ldsdh1 to LD subsets is evolutionary conserved, we expressed the Drosophila Ldsdh1 protein in human U-2 OS bone sarcoma cells. YFP-tagged Ldsdh1 indeed targets to LD subsets following overexpression in U-2 OS cells provided 100 µM oleic acid (OA) for LD deposition (Fig. 3). In order to investigate whether the presence of LD subsets depends on the amount of LD-inducing OA, and thus the size and amount of LDs, we also overexpressed YFP-tagged Ldsdh1 in U-2 OS cells fed with 200 and 800 µM OA, respectively (Fig. 3). With 800 µM OA, U-2 OS cells developed more and/or bigger LDs than cells induced with lower OA amounts. In contrast to the cells fed with lower OA doses, almost all LDs appeared to be decorated by Ldsdh1, although with variable amounts.

In the absence of LDs, fluorescently tagged Ldsdh1 protein localizes in a reticular pattern (Fig. 4A). Since a growing number of LD-associated proteins are known to reside in the ER in the absence of LDs (e.g. Klemm et al., 2011; Thiel et al., 2013; Wilfling et al., 2013; Zehmer et al., 2009), we tested whether Ldsdh1 localizes in the ER in the absence of LDs. For this purpose, we co-expressed a fusion protein composed of Ldsdh1 and the red fluorescent TdT protein (Shaner et al., 2008) together with the GFP-tagged ER marker protein Sec61β (Wilfling et al., 2013). Fig. 4A shows an overlap of the signals indicating that Ldsdh1 is localized in the ER in the absence of LDs. This observation is consistent with the microsomal pellet signal of Ldsdh1 found in the subcellular fractionation experiments (Fig. 1E). Following the induction of lipid stores by OA, Ldsdh1 is found at LDs and the ER signal had strongly decreased (Fig. 4A). In order to follow the translocation process with temporal resolution, we used spinning disc confocal microscopy. Within 20 min after addition of OA, we detected the appearance of circular-shaped Ldsdh1 signals, which suggest a translocation of the protein from the ER to LDs during or shortly after LD biogenesis (Fig. 4B; Movie 1).

To further establish the dynamics of ER- and LD-localized Ldsdh1, respectively, we tagged the protein with the photoconvertible protein Dendra2 (Chudakov et al., 2007). The emission fluorescence of Dendra2 can be irreversibly switched from the green to the red spectrum by UV irradiation. Thus, Dendra2 allows one to follow the fate of the switched protein pool over time without the detection of confounding newly synthesized protein. We transiently expressed the Dendra2-tagged Ldsdh1 fusion protein in Drosophila Kc167 cells and analyzed its mobility over time. Upon UV induction, the initially ER-localized red-labeled Dendra2–Ldsdh1 quickly redistributed all over the cell resulting in a considerable overlap between red and green fluorescence (Fig. 4C). Additionally, the red fluorescence intensity quickly dropped within the region where the switch took place (Fig. 4D).

When we monitored LD-localized Dendra2–Ldsdh1 protein following photoconversion, we detected a slower decrease of red fluorescence levels (Fig. 4D) and no yellow fluorescence in the neighboring area or on close-by LDs, which would have been indicative of the redistribution of switched Dendra2–Ldsdh1 (Fig. 4C).

Three different types of LD subsets have been reported to date: (1) LD subsets characterized by differential localization of perilipin proteins in differentiating mammalian adipocytes (Wolins et al., 2006) and in disease states such as fatty liver (Straub et al., 2008) or liver cancer (Straub et al., 2010), (2) LD subsets containing different lipid classes in mammalian cells (Hsieh et al., 2012; Khor et al., 2014), and (3) LD subsets representing various physiological states as suggested by the differential association of lipogenic enzymes to growing LDs (Wilfling et al., 2013) or the Rab18 protein to lipolytic LDs (Martin et al., 2005; Pulido et al., 2011).

We first asked whether Ldsdh1 preferentially associates with growing or shrinking LDs. For this purpose, we followed the localization of EGFP-tagged Ldsdh1 on LDs in cells fed with 400 µM OA (Fig. 5A). After 6 h, we removed the OA-containing medium and replaced it with medium lacking OA to induce the remobilization of lipid stores that were deposited during the OA feeding period. After 30 min of OA feeding, prominent EGFP–Ldsdh1-labeled LD subsets were observed (Fig. 5A), thus suggesting that the Ldsdh1 localization observed in the time-lapse movies (Fig. 4B shows still images; whole movie is shown as Movie 1) represents localization of the protein to a distinct set of LDs. In the course of the 6 h OA induction period, the number of LDs decreased while the size of persistent LDs increased (Fig. 5A). During the complete 24 h timecourse (therefore including the lipid remobilization phase), EGFP–Ldsdh1 was present on a subset of LDs, which appeared to be, on average, larger than the non-decorated ones.

This observation prompted us to measure the size of EGFP–Ldsdh1-covered LDs and to follow the EGFP–Ldsdh1-decorated LD subsets over a time period of 72 h. For this purpose, we recorded confocal images and used computer-based image segmentation to classify and quantitatively characterize LDs that were decorated and not decorated with EGFP–Ldsdh1, respectively (Fig. 5B). Within this timeframe, lipid storage amounts were increasing (for details of the measurements, see the Materials and Methods section; Fig. 5C). At 4 h after OA addition, the sizes of Ldsdh1-decorated and non-decorated LDs were almost the same (Fig. 5D). After 24 h, however, two clearly separable LD populations could be distinguished; that is, the EGFP–Ldsdh1-decorated LDs were larger than non-decorated LDs (Fig. 5D).

To test whether Ldsdh1 can target preformed LDs in the absence of additional LD biogenesis, we pre-fed Drosophila Kc167 cells with 800 µM OA for 24 h. Subsequently, we replaced the OA-containing medium with serum-reduced medium containing TriacsinC, which prevents the re-esterification of liberated free fatty acids (Igal et al., 1997). Afterwards, we transfected the cells with an EGFP–Ldsdh1-encoding plasmid and visualized the localization of the fusion protein 24 h later. In these cells, EGFP–Ldsdh1 could not be found in association with LDs but only in the ER and presumptive protein aggregates (Fig. 6A). These results suggest that Ldsdh1 localization depends on the biogenesis and growth of LDs.

To test whether the Ldsdh1 protein localization depends on the biogenesis of LDs enriched for a given lipid species, we induced cholesterylester-enriched LDs by providing green fluorescent NBD-cholesterol to the cells (Fig. 6B). We compared the localization of Strawberry-tagged Ldsdh1 with TdT-tagged PLIN2 (an LD-associated marker protein; Grönke et al., 2003) (Fig. 6B). While PLIN2 targeted all cholesterylester-enriched LDs, Ldsdh1 decorated LD subsets only (Fig. 6B).

Wilfling et al. previously described a LD subpopulation consisting of growing LDs that are on average larger than non-growing LDs (Wilfling et al., 2013). The growth of these LDs is not due to LD fusion or reduced lipolysis, but is characterized by de novo synthesis of triacylglycerol (TAG). TAG-synthesizing enzymes such as glycerol-3-phosphate acyltransferase 4 (GPAT4) were found to associate with growing LDs and are therefore a marker for anabolically active LDs (Wilfling et al., 2013). EGFP–Ldsdh1 also associates with LDs, which are on average larger than the non-decorated LDs (Fig. 5D). We therefore asked whether Ldsdh1 and GPAT4 localize to the same LDs and decided to co-express EGFP–Ldsdh1 with an mCherry-tagged GPAT4. Fig. 6C shows that the two proteins indeed colocalize to the same subset of LDs. Thus, Ldsdh1 associates with anabolically active and growing LDs.

To investigate which sequence intervals of Ldsdh1 are responsible for its association with LDs, we deleted parts of Ldsdh1 (Fig. 7A) and assayed the localization of the corresponding EGFP fusion proteins in Drosophila Kc167 cells. Fig 7B and Fig. S3A show the subcellular localization of representative constructs. Deletion of the first 48 amino acids of Ldsdh1 (aa 49–320) abrogated the LD localization completely, resulting in a cytoplasmic localization of the Ldsdh1 fusion protein (Fig. 7B; Fig. S3A). Furthermore, truncation of the N-terminal 20 amino acids (Ldsdh1 aa 21–320) had no effect on the LD localization of the protein (Fig. 7B, Fig. S3A) indicating a role for aa 21–48 in LD targeting. Indeed, a minimal construct spanning amino acids 21–48 of Ldsdh1 localized to all LDs and thus this sequence is both necessary and sufficient for association with LDs and is henceforth called the LD-targeting sequence (denoted LDT). Importantly, overexpression of the truncated Ldsdh1 protein variant spanning amino acids 21–48 of Ldsdh1 does not interfere with the presence of LD subsets as shown by a co-expression with the full-length Ldsdh1 protein, which still targets LD subsets (Fig. S3B). Amino acids 21–48 encode an amphipathic helix, which might mediate LD binding (see below). In order to test this hypothesis, we generated a quadruple point mutation (Ldsdh1 QM) where four hydrophobic amino acid residues were exchanged for the polar amino acid serine residues (I29S, V30S, I33S and W37S). Indeed, these amino acid replacements completely abolished LD binding (Fig. 7B). We further tested whether a presumptive enzymatic activity might be needed for LD binding. For this purpose, we mutated the amino acid residues S194 and Y207 to an alanine residue. Both amino acids should participate in binding of NADP based on homology modeling (see below). Both mutant Ldsdh1 variants, however, localized to LD subsets (Fig. 7B), indicating that enzymatic activity is not a strict requirement for or is distinct from LD binding.

To identify sequences necessary for the LD subset localization, we extended the LDT-carrying variant (aa 1–48) C-terminally. Constructs spanning amino acids 1 to 60, 67, 73 (data not shown, Fig. 7A) or 82 (Fig. 7A,B; Fig. S3A) resulted in an association of the fusion proteins with LDs, but failed to target to LD subsets. Thus, amino acids 49–90 of Ldsdh1 play a critical role in subset targeting and accordingly contain a LD subset-targeting sequence (denoted LDST). In addition, various deletions within the LDST as well as other truncations such as amino acids 49–238 (Fig. 7A) revealed a second LDST likely to reside within the amino acids 91–238 of Ldsdh1 (a putative LDST, denoted pLDST).

In order to test for differences in the binding to LDs mediated by the LDT and LDST we performed in vitro binding experiments of the various sequences to determine their level of binding to bilayer and monolayer membranes (Fig. S3C–F). Our results suggest that the construct spanning amino acids 1–90 shows a higher affinity towards bilayer membranes as compared to the LDT, which targets equally well to monolayer and bilayer membranes. The localization mechanism facilitated by the LDT and LDST sequences is evolutionary conserved, as the sequence stretches were also sufficient to mediate their localization to all LDs (Ldsdh1 aa 1–48) or LD subsets (Ldsdh1 aa 1–90) in human U-2 OS cells (Fig. S3G).

In order to test whether the newly identified LDST of Ldsdh1 is sufficient to restrict proteins to LD subsets, we generated chimeric LD-associated proteins. For this purpose, we fused the Ldsdh1 LDST sequence (amino acids 49–90) to the LD-associated proteins PLIN2 and CG9186, respectively (Fig. 8A). We have chosen these two proteins as PLIN2 has so far only been found to bind to LDs and not to other cellular membranes such as the ER (Fig. 1D; Beller et al., 2010b; Cermelli et al., 2006; Welte et al., 2005) and because CG9186 targets to LDs after a previous transient localization to the ER (Thiel et al., 2013). Both proteins target all cellular LDs (Fig. 8B,C). When we assayed the localization of an EGFP-tagged PLIN2:Ldsdh1_LDST chimeric protein we found localization to all cellular LDs (Fig. 8D) as observed with wild-type PLIN2 (Fig. 8C), but no LD subset localization. When we placed the LDST at the C-terminus of CG9186 we observed the same localization as for wild-type CG9186 (compare Fig. 8B and E).

Based on these observations, we concluded that a C-terminal LDST fusion is not functional. We then hypothesized that the LDST needs to be placed in close proximity of the general LD-targeting sequence as observed with Ldsdh1. To test this hypothesis, we first identified the PLIN2 LD targeting sequence by localization analyses (Fig. S4). Amino acids 1–120 of PLIN2 were sufficient to target LDs in a manner that was indistinguishable from that of the wild-type protein. We decided to extend this construct up to amino acid 152 as this sequence completely covers the PAT domain (Arrese et al., 2008), which is shared by all perilipin family proteins (Bickel et al., 2009; Kimmel et al., 2010; Sztalryd and Kimmel, 2014). A chimeric protein consisting of amino acids 1–152 of PLIN2 and the LDST of Ldsdh1 still targeted all cellular LDs (Fig. 8F; PLIN2 aa 1–152:Ldsdh1_LDST), which was again indistinguishable from wild-type PLIN2 localization (Fig. 8C).

However, when the Ldsdh1 LDST was fused in direct proximity to the demonstrated LD-targeting sequence of CG9186 (Thiel et al., 2013) (CG9186aa1-195:Ldsdh1_LDST), the chimeric protein associated only with LD subsets (Fig. 8G) similar to the localization of Ldsdh1. These results indicate that the LDST sequence of Ldsdh1 is sufficient to limit protein localization to LD subsets, provided that the LDST is in juxtaposition with the LDT and LD localization occurs after a previous transient localization to the ER.

In this study, we investigated the targeting of the Drosophila protein Ldsdh1 to a subset of LDs and characterized the nature of these LD subsets. The targeting of endogenous Ldsdh1 to LD subsets in the fly fat body (Fig. 2B,C) presents first evidence for LD diversification in a whole-organism context under non-pathogenic conditions. Ldsdh1 will therefore not only provide a valuable marker for a subset of LDs but also provide a tool to investigate LD diversification in a genetically highly accessible model organism.

Only few LD subset marker proteins have been identified so far. They include the Ancient ubiquitous protein 1 (AUP1; Thiele and Spandl, 2008), the OSBP-related protein 2 (ORP2, also known as OSBPL2; Thiele and Spandl, 2008), Rab18 (Martin et al., 2005), different perilipins (Hsieh et al., 2012; Wolins et al., 2005), the phosphatidylinositol transfer protein Sfh3 (also known as PDR16) (Ren et al., 2014), as well as AGPAT3 and GPAT4 (Wilfling et al., 2013). The specific properties of the LDs they are associated with include for example differences in the maturation status or lipid composition (perilipins), or a catabolic (Rab18) or anabolic (GPAT4) metabolic state, respectively.

Ldsdh1 colocalizes with GPAT4 (Fig. 6C). GPAT4 decorates growing LDs and participates in localized lipogenesis (Wilfling et al., 2013). The localization of Ldsdh1 to the largest LDs within cells (Fig. 5D) is fully compatible with a localization to growing LDs. While it is also possible that the Ldsdh1-decorated LDs are growing due to a block in lipolysis, the colocalization with the lipogenic GPAT4 protein rather supports a lipogenic nature for the LD subset. Intriguingly, while GPAT4 targets pre-existing LDs to support localized lipogenesis and thus LD growth (Wilfling et al., 2013), Ldsdh1 does not move to pre-existing LDs (Fig. 6A). Additionally, Ldsdh1 is also not removed from the surface of LDs once lipid remobilization is initiated (Fig. 5A, 24 h). This localization suggests that Ldsdh1 does not directly participate in the growth or remobilization of LDs. Indeed, we did not observe an effect of Ldsdh1 on overall TAG storage amounts when the protein is overexpressed (in vivo and in vitro) or reduced by siRNA in vivo (data not shown).

The translocation of Ldsdh1 from the ER to LDs depends on amino acids 21–48. Based on secondary structure predictions, these amino acids can adopt an α-helical structure that covers several hydrophobic amino acid residues. The charged and polar amino acid residues D28 and K35 contribute to an amphipathic character of the sequence stretch. As these residues are spatially located on one face of the helix they likely mediate the attachment of the Ldsdh1 protein to the LD hemimembrane (Fig. 8H,I). The general LD targeting of Ldsdh1 thus follows the common theme where amphipathic helices mediate LD binding (Ingelmo-Torres et al., 2009; Kory et al., 2016; Zehmer et al., 2008). This is supported by the cytoplasmic localization of EGFP–Ldsdh1, when the amphipathic characteristic of the helix is removed by mutating I29, V30, I33 and W37 to polar serine residues (EGFP–Ldsdh1 QM) (Figs 7B and 8I).

Little is known about the mechanism(s) restricting proteins to LD subsets. The targeting of GPAT4 to LD subsets depends on the activity of the coat protein complex I (COPI) machinery on pre-existing LDs (Wilfling et al., 2014). In contrast to GPAT4, however, Ldsdh1 appears rapidly on LD subsets (Figs 4B and 5A) and is not able to target pre-existing LDs (Fig. 6A). These results suggest that two different pathways exist which mediate LD subset targeting.

Here, we identified a 42 amino-acid-long sequence (LDST motif) that is sufficient to restrict Ldsdh1 (Fig. 7B) and a chimeric protein (Fig. 8G) to LD subsets. Since a mutated Ldsdh1 protein which lacks the LDST sequence still targets LD subsets, it appears that Ldsdh1 contains additional sequences with a redundant LD subset-targeting function. The N-terminal LDST of Ldsdh1 (amino acid residues 49–90) consists of two distinct regions, namely a most likely unstructured linker region (R49–K59) and a second region comprising V60–V90, which adopts defined secondary structure elements within the core domain. According to a BLAST search, the region is conserved in different organisms other than Drosophila but is yet poorly characterized. The sequence by itself does not provide an obvious explanation for the observed LD-subset targeting experimentally observed in this work. Still, candidate amino acid residues for a potential membrane interaction exist between the at least partially exposed M70 and L81 residues, which contain hydrophobic and aromatic amino acids.

Based on our chimeric protein localization experiments (Fig. 8), the identified LDST is sufficient to guide or stabilize general LD-localized proteins to LD subsets. Our results suggest that LDST-mediated LD subset targeting depends on: (1) the juxtaposition of the LDST sequence and the general LD-targeting sequence, and (2) that the protein reaches LDs after a previous transient localization to the ER. How does the LDST restrict localization to LD subsets? For the sorting of proteins and the formation of LD diversity, various mechanisms are possible (Thiam and Beller, 2017). For example, it is conceivable that the LDST sequence acts as a recognition motif for a dedicated sorting machinery or that it provides an interaction surface for factors, such as proteins, specific fatty acids or other chemical compounds, that are already localized to the given LD subsets. Alternatively, the LDST sequence could also ‘sense’ a biophysical LD parameter, such as the curvature or tension of the nascent LD membrane, which both depend on the protein and phospholipid composition (Fei et al., 2011a,b; Kassan et al., 2013; Krahmer et al., 2011; Thiam et al., 2013; Yang et al., 2012). Our in vitro experiments indeed suggest that the LDST shows a higher binding affinity to bilayer membranes as compared to the LDT sequence, which partitions equally well to monolayer and bilayer membranes (Fig. S3C–F).

In summary, our findings demonstrate that LD subsets exist in cells of different organs in vivo and that these LD subsets are highly dynamic. Furthermore, we identified a sequence motif sufficient to restrict proteins to LD subsets. Taken together, our results pave the way to better understand the role of LD subsets, as well as their formation and regulation.

Gene coding sequences of CG2254/Ldsdh1 and homologs were PCR-amplified with flanking NotI/AscI restriction sites for cloning in the pENTR/D-TOPO vector (Invitrogen, Carlsbad, CA). The expressed sequence tag (EST) clone CG2254:RH47744 served as template for CG2254/Ldsdh1 (with an amino acid exchange from N to K at position 13). Deletion constructs of the Ldsdh1-coding sequence were created by site-directed mutagenesis PCR. Subsequently, the Gateway Recombinase System (Invitrogen, Carlsbad, CA) was used to transfer the protein-coding sequences into destination vectors for expression in either fly tissue culture cells [pUbiP-eGFP-rfA, pUbi-TdT-rfA (gift of Alf Herzig, Department of Molecular Developmental Biology, Max Planck Institute for Biophysical Chemistry, Göttingen, Germany), pUbiP-Strb-rfA, pUbiP-Dendra2-rfA (gifts of Stefan Hell, Department of NanoBiophotonics, Max Planck Institute for Biophysical Chemistry, Göttingen, Germany)] or mammalian tissue culture cells (pH-eYFP-rfA). The pA-mCherry-GPAT4 and pA-mCherry-DGAT2 plasmids were a kind gift by Tobias Walther (Department of Genetics and Complex Diseases, Harvard Medical School, Boston, USA).

Drosophila Kc167 (a gift from the Harvard RNAi screening center), S3 and ML-DmBG3-c2 cells (both from the Drosophila genomics resources center) were cultured in Schneider's medium (Gibco/Life Technologies, Carlsbad, and PAN-Biotech, Aidenbach, Germany) supplemented with 10% heat-inactivated fetal calf serum (‘Gold’ from PAA Laboratories, Pasching, Austria, or ‘South Africa’ from PAN-Biotech, Aidenbach, Germany) and 1% streptomycin-penicillin (Gibco Life Technologies). The medium for ML-DmBG3-c2 cells additionally contained 10 µg ml⁻¹ insulin (Sigma-Aldrich). The identity and homogeneity of the cell cultures were confirmed by regular visual inspection. Cells were transfected with Effectene (Qiagen, Hilden, Germany) or FuGENE6 (Promega, Madison, WI) according to the manufacturer's instructions. For the generation of the Kc167 cell line with stable expression of EGFP–CG2254, cells were co-transfected with the pCoBlast resistance plasmid (Invitrogen) and selected with 30 µg ml⁻¹ Blasticidin (Invitrogen).

We induced LDs by the addition of OA bound to fatty acid-free bovine serum albumin (BSA) for the time spans and with the concentrations given in the respective figure legends. To test the localization to cholesterylester-enriched LDs, we transfected Drosophila S3 cells with the given expression construct using the Effectene (Qiagen, Hilden, Germany) transfection reagent and incubated for 48 h. We subsequently incubated the cells with 1 µg ml⁻¹ NBD-cholesterol (Invitrogen/Molecular Probes) in tissue culture medium containing 5% serum for 3 h before they were washed with PBS, fixed with 5% paraformaldehyde in PBS containing 25 mM EGTA for 5 min and mounted in Mowiol after several additional PBS washes.

Human bone sarcoma-derived U-2 OS cells were obtained from the American Type Culture Collection and verified by regular visual inspection. We cultivated the cells at 37°C in a 5% CO₂ humidified environment in McCoy's 5A modified medium (Sigma-Aldrich) supplemented with 10% fetal bovine serum (Sigma-Aldrich) and 1% L-glutamine (Sigma-Aldrich).

A custom antibody detecting Ldsdh1 was raised by Peptide Specialty Laboratories GmbH (Heidelberg, Germany). Guinea pigs were immunized with a peptide covering amino acids 90–118 (VNEQTNNQTVKEIKNNGGK) of Ldsdh1 and the serum was purified on a peptide column.

Drosophila microscope images were recorded with a Leica TCS SP2 AOBS (Leica Microsystems, Solms, Germany) (Figs 1C and 2C and 6A), a Zeiss LSM780 microscope (Carl Zeiss Microscopy, Jena, Germany) (all other confocal laser scanning images), a Zeiss Axio Imager.M2 with an AxioCam MRm (Fig. 2B) or a Zeiss Axiophot upright microscope with a Kontron camera and a 20× air objective (Fig. 2A). Fluorescence images of U-2 OS cells were acquired with a Leica SP5 confocal microscope (DM6000CS; Leica Microsystems, Solms, Germany) equipped with a 63× HCX PL APO 1.40 oil CS objective (Fig. 3). The time-lapse images in Fig. 4B were recorded with an Olympus-based Ultraview VOX (Perkin Elmer, Waltham, MA) imaging system.

For the microscopy of fixed tissue culture cells, the cells were either transferred after a given amount of time post transfection to coverslips where they were allowed to adhere for 60 min, or were directly incubated in eight-well chamber slides on coverslips (Sarstedt, Nümbrecht, Germany) for the desired amount of time. Subsequently they were fixed with 5% paraformaldehyde in PBS containing 25 mM EGTA for 5 min. Following this, the cells were washed with PBS and then counterstained with HCS LipidTOX Deep Red (1:250 in PBS; Invitrogen/Molecular Probes) before they were mounted in Mowiol.

For U-2 OS sample preparation, cells were seeded in 96-well glass-bottom plates (Sensoplate Plus, Greiner Bio-One) coated with fibronectin (Sigma-Aldrich), and transfected with the Lipofectamine 2000 Transfection Reagent (Thermo Fisher Scientific) according to the manufacturer's instructions. At 24 h post transfection, we exchanged the medium with medium containing OA with the concentrations given in the respective figure legends. After 24 h incubation time, PBS-washed cells were fixed in 4% paraformaldehyde (PFA) in growth media supplemented with 10% FBS for 15 min and washed three times with PBS. Then, LDs were counterstained with HCS LipidTOX DeepRed (1:200, Invitrogen/Molecular Probes) and nuclei with DAPI (1:500, Sigma-Aldrich). After washing with PBS, the wells were completely filled with PBS and sealed. Fluorescence images were acquired with a Leica SP5 confocal microscope (DM6000CS) equipped with a 63× HCX PL APO 1.40 oil CS objective (Leica Microsystems, Solms, Germany).

For the Dendra2 photoswitching experiments shown in Fig. 4C, Drosophila Kc167 cells were transfected with a plasmid encoding Dendra2-tagged Ldsdh1 by means of the Effectene (Qiagen, Hilden, Germany) transfection reagent and grown in eight-well chamber slides on coverslips (Sarstedt, Nümbrecht, Germany) either in the absence or presence of 800 µM OA for 72 h. Images were recorded with a Zeiss LSM780 confocal microscope (Carl Zeiss Microscopy, Jena, Germany) in 2 s intervals. After 15 images, a region of interest was illuminated with a 405 nm laser for 20 iterations to change the emission fluorescence of the Dendra2-tag and, after a total of 100 images, the series was ended. Analysis of the red fluorescence intensity values of the time series was carried out with the Fiji ImageJ software (Schindelin et al., 2012) and numeric values were further analyzed with the Excel spreadsheet software (Microsoft).

For fluorescent antibody staining, wandering third-instar larvae were opened and fixed with 5% paraformaldehyde in PBS containing 25 mM EGTA for 20 min, permeabilized with 0.1% Triton X-100 for 20 min and incubated with purified antibodies against Ldsdh1 (4 µg ml⁻¹) or pre-immune serum in BBT (10 mM Tris, 55 mM NaCl, 40 mM KCl, 7 mM MgCl₂, 20 mM glucose, 50 mM Sucrose, 0.1% BSA, 0.1% Tween-20, pH 6.95) overnight at 4°C. For detection, a secondary antibody conjugated to Alexa Fluor 568 (1:500; Molecular Probes/Invitrogen) was used. LDs were counterstained with HCS LipidTOX Deep Red (1:500 in PBS, Invitrogen/Molecular Probes, Carlsbad, CA, USA). The tissues were dissected and mounted in Mowiol.

Larval tissues expressing GFP-tagged Ldsdh1 protein were dissected in PBS, fixed for 20 min with 5% paraformaldehyde/25 mM EGTA in PBS, stained with HCS LipidTOX Deep Red (1:500 in PBS, Invitrogen/Molecular Probes) and mounted in Mowiol.

For colorimetric antibody stainings of embryos, paraformaldyde- and heptane-fixed embryos were incubated with the antibody specific to Ldsdh1 (0.1 µg ml⁻¹) in BBT overnight. The secondary antibody was biotinylated (1:500; Biomol, Hamburg, Germany) and the VECTASTAIN Elite ABC Kit (Vectorlabs, Burlingame, CA, USA) was used to detect and enhance the signal. Following development of the precipitate, the embryos were washed with PBS with 0.1% Tween 20 (PBT), dehydrated and subsequently mounted in Canada balsam (Sigma-Aldrich).

For the whole-mount in situ hybridization, digoxigenin-labeled antisense RNA probes of Ldsdh1 were generated by T3 polymerase-mediated in vitro transcription of the XhoI-linearized EST clone CG2254:RH47744. The generation of control RNA probes (sense probe) was carried out with T7 polymerase and the BamHI-linearized CG2254:RH47744 clone. Whole-mount in situ hybridization was performed with 1–20-h-old white^– embryos as described previously (Grönke et al., 2003). Embryos were incubated with alkaline-phosphatase-coupled anti-DIG antibodies (1:2000; Pierce Biotechnology, Rockford, IL) and digoxygenin activity was detected using the colorimetric substrate Nitro Blue Tetrazolium/5-bromo-4-chloro-3-indolylphosphate (NBT/BCIP). Embryos were subsequently washed with PBT, dehydrated and finally mounted in Canada balsam.

Subcellular fractionation of fat body cells by sucrose gradient fractionation was performed as described previously (Beller et al., 2006). In brief, 30 fat bodies of late wandering third-instar larvae were dissected and collected in 100 µl FBB [10 mM HEPES pH 7.6, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 1 mM DTT, EDTA-free complete protease inhibitors (Roche, Basel, Switzerland)] and disrupted by sonication for 5 min (BioRuptor, low setting; Diagenode, Liège, Belgium). Nuclei and cell debris were removed by centrifuging at 3500 rpm (∼1200 g) for 8 min. The supernatant was adjusted to a 3 ml volume, mixed with 3 ml 1.08 M sucrose solution in FBB and then transferred into a 12 ml SW41 polyallomer ultracentrifuge tube (Beckman Coulter, Brea, CA), where it formed the bottom layer of the gradient. It was overlaid with 2 ml each of 0.27 M sucrose, 0.135 M sucrose and FBB only. The gradient was centrifuged for 1 h 45 min at 30,000 rpm at 4°C (SW41 Ti rotor; Beckman Coulter). After the run, eight 1.5 ml fractions were collected by pipetting from top to bottom. The fractions were precipitated with methanol–chloroform and the resulting protein pellet was resuspended in 2× protein sample buffer; the microsomal pellet was directly resuspended in 2× protein sample buffer. Protein concentrations were measured with the RC/DC protein assay (Bio-Rad Laboratories, Hercules, CA). The following primary antibodies were used: guinea pig anti-Ldsdh1 (1:5000; this study), rabbit anti-PLIN2 (1:3000; Grönke et al., 2003), mouse anti-Actin (1:500; Developmental Studies Hybridoma Bank clone JLA20), and goat anti-ADH/dG20 (1:250; sc-22676, Santa Cruz Biotechnology, Dallas, TX). Secondary antibodies (all from Perbio Sciences, Erembodegem, Belgium) conjugated to horseradish peroxidase (HRP) were used as following: anti-guinea-pig HRP (1:10,000), anti-rabbit HRP (1:40,000) and anti-goat HRP (1:40,000). Antibodies were removed from the blot membrane by Restore Western Blot Stripping Buffer (Thermo Fisher Scientific, Waltham, MA) according to the manufacturer's instructions.

Flies were raised on standard complex cornflour–soyflour–molasses food at 25°C. The transgenic EGFP–Ldsdh1 line was as described previously (Beller et al., 2006). CG2254 RNAi lines were obtained from the Vienna Drosophila RNAi Stock Center (VDRC101149 and VDRC5470) or the Bloomington Stock Center (Bl36731) with the w¹¹¹⁸ control lines VDRC60000 and Bl36303, respectively, as control reference strains. For transgene-derived fat body expression, we used the Gal4-activator line Fb-Gal4 (Grönke et al., 2003). Npc1b-Gal4 (Voght et al., 2007) served for transgene expression in the gut.

Polyclonal Drosophila Kc167 cells stably expressing EGFP-tagged Ldsdh1 protein were seeded in Sarstedt chamber slides (Sarstedt, Nümbrecht, Germany) and incubated for 4, 24, 48 or 72 h with 800 µM OA to induce LD formation. After the indicated times, cells were fixed and stained for DNA with Hoechst 33258 (Invitrogen/Molecular Probes) and HCS LipidTOX Deep Red (Invitrogen/Molecular Probes) for LDs. Images were acquired with a Zeiss LSM780 confocal microscope (Carl Zeiss Microscopy, Jena, Germany).

The separate channels were segmented with a custom routine developed for the CellProfiler software package (Soliman, 2015). Different parameters of the segmented nuclei, LDs and Ldsdh1 protein, such as area and maximum axis length, were measured. Data analysis was performed using the KNIME data analysis platform (Stöter et al., 2013). First, we normalized the amount of lipids (measured by the area of LDs per image) to the size of the nuclei (area per image) to detect the time-dependent lipid storage increase. Second, we selected cells expressing Ldsdh1 and sorted their LDs into two classes based on the presence or absence of colocalization of the Ldsdh1 protein. Then, the axis length of all LDs of the different classes was used to calculate densities within the distributions to generate the plots presented. CellProfiler and Knime analysis routines are available upon request.

The binding of Ldsdh1 aa 21–48 and Ldsdh1 aa 1–90 to phospholipid monolayer or bilayer membranes was investigated as previously published (Wang et al., 2016). In brief, we utilized buffer-in-TAG oil droplets decorated by a phospholipid monolayer, which can adhere to each other to form a bilayer (Thiam et al., 2012). In this system, the phospholipid monolayers are continuous with the bilayer, similar to the LDs connected with the ER. For the actual experiments, we separately expressed EGFP-tagged Ldsdh1 aa 21–48 and Strawberry-tagged Ldsdh1 aa 1–90 in Drosophila Kc167 cells, which were oleate-loaded to induce LD formation. We next purified LDs bound by the proteins by sucrose ultracentrifugation and mixed them in buffer droplets contained in TAG. The LD protein content subsequently relocated to the newly reconstituted buffer–TAG interface as reported previously (Kory et al., 2015). Next, the buffer droplets were mixed to form a bilayer as shown in Fig. S3C and reported recently (Wang et al., 2016). Images were recorded by confocal microscopy and the respective protein signal on the bilayer and monolayer was quantified by making sections across the droplets and measuring the intensity with the ImageJ software package. For Ldsdh1 aa 21–48, we found that the bilayer-to-monolayer signal ratio was two; which means that this fragment evenly distributes to the two types of membranes (the bilayer is indeed made of two monolayers). For Ldsdh1 aa 1–90, the ratio was around 3.25, which means that this fragment is enriched in the bilayer.

We thank Julia Goerigk and Ewa Maj for their assistance during the early stage of the studies, Ronald Kühnlein for fly stocks, Tobias Walther for the dmGPAT-4 expression constructs and the group of Stefan Hell for providing the Dendra-2 expression vector. Further reagents were obtained from the Bloomington Stock Center and the Drosophila Genomics Resources Center (supported by NIH grant 2P40OD010949-10A1).