An alternative membrane topology permits lipid droplet localization of peroxisomal fatty acyl-CoA reductase 1

Fatty acids are a fundamental requirement for the biogenesis of membranes, and their oxi­dation generates energy efficiently in the form of ATP. It is therefore not surprising that almost all cells have the capacity to store fatty acid derivatives in specialized organelles now widely termed lipid droplets (LDs; Murphy, 2012; Walther and Farese, 2012). LDs are highly dynamic organelles, with their size and number dependent on nutrient supply and hormonal cues. Neutral lipids, such as triglycerides and cholesteryl esters, are stored in the hydrophobic core, which is surrounded by a phospholipid monolayer with embedded pro­teins. LD proteomics of mammalian cells has revealed the presence of constitutively associated LD proteins (perilipins) and many enzymes of lipid metabolism (Bersuker et al., 2018; Brasaemle et al., 2004; Goodman, 2009; Liu et al., 2004). Hydrophobic domains are some­how required for targeting of proteins to LDs, but the mechanisms involved appear to be diverse and are not fully understood (Kory et al., 2016; Ohsaki et al., 2014; Thiele and Spandl, 2008).

Fatty acyl-CoA reductases (Far) are peroxisomal membrane proteins that catalyze the for­mation of fatty alcohols from fatty acyl-CoA and NADPH (Cheng and Russell, 2004), which in turn are utilized for ether lipid synthesis. In mammals, the Far family is represented by Far1 and Far2, which show a similarity of 58% (Cheng and Russell, 2004). Far1 expres­sion has been found in most of the tissues, whereas the distribution of Far2 is more restricted (Cheng and Russell, 2004). Patients with a homozygous knockout of the FAR1 gene suffer from severe epilepsy, cataracts and profound growth and mental retardation (Buchert et al., 2014), suggesting that Far2 cannot compensate for Far1 at the whole-organism level. Most peroxisomal membrane proteins are synthesized on free ribosomes in the cytosol and are inserted into the peroxisomal membrane either by direct targeting (class I proteins) or via the ER (class II proteins) (Smith and Aitchison, 2013).

Membrane protein topology describes the number and orientation of one or more hydropho­bic domains relative to the membrane. The vast majority of membrane-associated proteins possesses only one stable configuration. However, there is a small subset of proteins that show changes in the orientation of one, multiple or all transmembrane helices upon membrane insertion, resulting in complex topological variants (von Heijne, 2006). The phenomenon of multiple topologies of a single protein is very rare, and neither the evolutionary background nor the mechanisms of exhibiting multiple topological variants are understood satisfactorily. Here, we report that Far1 is not only targeted to peroxisomes but also to LDS, and suggest that this is due to the presence of two different topological orientations.

The biochemistry of long-chain fatty acids is at the heart of the complex lipid metabolic network. Here, we analyzed the localization and targeting of fatty acyl-CoA reductase 1 (Far1) primarily in human osteosarcoma U-2 OS cells to gain further insights into the topology and subcellular compartmentalization of fatty acid metabolism.

Fatty acyl-CoA reductases have not been widely studied, and there appeared to be no suitable antibodies for immunocytochemistry protocols available. Therefore, we cloned the cDNA of Far1 from the human sebaceous gland cell line SZ95 and designed a mCherry–Far1 fusion plasmid. Transient expression in U-2 OS cells showed that there were numerous dot-like structures throughout the cytoplasm, which overlapped with the signal from the peroxisomal membrane marker GFP–Pex11β (Fig. 1A), consistent with previous work (Honsho et al., 2013).

Far1 localized to LDs after oleate (oleic acid, OA) supplementation, as indicated by the colocalization with the LD marker perilipin-3 (Plin3, also known as TIP47) and the segregation from Pex11β (Fig. 1A). This surprising finding was further confirmed by neutral lipid staining demonstrating an abundance of LDs surrounded by mCherry–Far1 (Fig. 1A). A LD localization score was devised to account for cell-to-cell variation, and demonstrated that an average cell would show a predominant LD localization of Far1 with the remaining protein localizing to peroxisomes. The Pearson correlation coefficient between Far1 and Pex11β decreased from 0.70 to 0.31 after OA treatment, whereas the value for Far1 and the LD marker A3Nt–GFP (A3Nt denotes the N-terminus of ACSL3; Poppelreuther et al., 2012) increased from −0.07 to 0.40, respectively (Fig. 1B). The presence of Far1 on LDs was both time and OA concentration dependent (Fig. S1A). Decrease of LDs induced by starvation led to the disappearance of LD-associated Far1 even before the LDs were completely consumed (Fig. S1B), consistent with the situation for cells cultured at steady state that also harbor some LDs (Fig. 1A).

Palmitate or arachidonate supplementation also triggered the appearance of Far1 on LDs albeit with a lower efficiency (Fig. S2). Lipid droplets were generally much smaller and less abundant compared to after treatment with OA, suggesting these fatty acids were not as efficiently taken up and converted into triglycerides.

The 65 C-terminal amino acids of Far1 were sufficient to allow targeting of a fluorescent reporter to LDs in OA-treated U-2 OS cells (mCherry–Far1Ct; Fig. 1C). The same C-terminal region is also necessary for the targeting of Far1 to peroxisomes (Honsho et al., 2013).

To enable characterization at a higher resolution, we established U-2 OS cells stably expressing mCherry–Far1 together with the LD marker A3Nt–GFP by retroviral transduction. Analysis by correlative light electron microscopy (CLEM) showed that the fluorescent signal of both mCherry–Far1 and A3Nt–GFP corresponded to LDs, which were identified by transmission electron microscopy (TEM) as large circular homogenously electron-dense structures surrounded by a more electron-dense line (Fig. 1D). ER tubules were identified by EM but did not correspond to the circular fluorescence pattern of Far1 and A3Nt–GFP. Additionally, mCherry-positive structures were also associated with the electron-dense stratified organelles presumably corresponding to peroxisomes.

To enable the direct comparison of two peroxisomal membrane-associated proteins by subcellular fractionation, epitope-tagged variants of Far1 and Pex11β were stably co-expressed through retroviral transduction (denoted 3xFLAG-GFP-Far1.3xHA-mCherry-pex11β.U-2 OS cells). Lipid droplets were enriched by sedimenting most of the other cellular membranes for 1 h at 100,000 g. The supernatant was further subfractionated by sucrose density gradient centrifugation and analyzed by western blotting (Fig. 1E). Far1 was enriched 2.7-fold over Pex11β in the LD fractions, even if it did not reach the level of our LD marker protein (endogenous ACSL3).

Preliminary screening of several other cell lines for the expression of Far1 by western blotting suggested that A431 epidermoid carcinoma cells might be a suitable model system to investigate the localization of endogenous Far1. However, initial standard immunofluorescence protocols employing either chemical (paraformaldehyde, PFA) or solvent (methanol at −20°C) fixation only gave background signal. It was only after including a guanidine hydrochloride antigen denaturation step (Peranen et al., 1993) that the available commercial antibody recognized endogenous Far1 in U-2 OS and A431 cells, overlapping with endogenous ACSL3 or surrounding the BODIPY neutral lipid stain after treatment with OA (Fig. 2A). Further confirmation of endogenous Far1 associating with LDs was obtained by western blotting of subcellular fractions for both U-2 OS and A431 cells (Fig. 2B).

Ether lipid synthesis, as assessed through a quantification of radiolabeled ethanolamine plasmalogens, was decreased in OA-treated U-2 OS cells, even though the expression of endogenous Far1 was increased more than 2-fold and the total amount of peroxisomes was not significantly changed (Fig. S3).

Targeting of proteins to LDs appears to be quite conserved across evolutionary distant model systems. Therefore, we tested whether the LD localization of Far1 would be conserved in S. cerevisiae, even if yeast does not synthesize ether lipids itself. However, human Far1 localized neither to peroxisomes nor to LDs when expressed in S. cerevisiae (Fig. S4; see also Discussion).

On peroxisomes, Far1 displays a classical type I membrane topology with most of the protein, including the enzyme domain exposed to the cytosol, and a short luminal tail (Honsho et al., 2013). This topology, however, is not found on LD-associated proteins because the hydrophobic core does not allow penetration by charged amino acids (Kory et al., 2016). In addition, LDs are surrounded by a phospholipid monolayer that is too narrow to incorporate standard integral transmembrane bilayer domains.

To resolve this apparent conflict, we used double epitope tagging of Far1 (FLAG–Far1–HA) combined with selective permeabilization. Digitonin solubilizes the plasma membrane but not intracellular membranes, like the ER or peroxisomes, allowing access of antibodies only to cytoplasmically oriented epitopes. Antibodies against the C-terminal HA epitope tag stained Far1 only when it was present on LDs in OA-treated cells or if the non-discriminative detergent Triton X-100 (TX-100) was used (Fig. 3A). This suggested that the orientation of the C-terminus would be towards the cytosol on LDs but intralumenal on peroxisomes.

Protease protection assay using subcellular fractions from OA-treated U-2 OS cells was used as an independent method to verify the topology of Far1. Sedimented membranes containing peroxisomes retained proteolytic fragments recognized by anti-HA antibodies, suggesting that the C-terminus of Far1 is not accessible from the cytosolic side. Far1 recovered from the LD-containing supernatant, however, was efficiently digested (Fig. 3B).

Targeting of LD proteins starts either from the ER or from the cytosol (Kory et al., 2016). To distinguish between these possibilities, we undertook glycosylation tagging of Far1 with an opsin-derived sequence (Bañó-Polo et al., 2011; Favaloro et al., 2008; Nilsson and von Heijne, 1993). Cells stably expressing the N-glycosylation consensus sequence (Far1Ct_ops) were analyzed by western blotting and immunofluorescence (Fig. 3C). Partial glycosylation of Far1 constructs was observed only for cells with the opsin-derived sequence containing intact N-glycosylation motifs, as compared to the opsin control sequence containing two point mutations. Remarkably, glycosylation was not only present in OA-treated cells but also when cells underwent starvation to remove LDs. This suggested that, in our model system, the ER is a part of the trafficking itinerary towards peroxisomes. Both opsin constructs were partially found on LDs (Fig. 3C), likely corresponding to the non-glycosylated variants because N-glycoproteins are neither expected to be found on LDs nor has there been any report of this, to the best of our knowledge.

In silico transmembrane domain prediction by mean of TMHMM (Krogh et al., 2001) suggested that there are two C-terminal hydrophobic domains (HDs) in Far1, of which the first one [HD1, amino acids (aa) 466–484] reached the threshold for a bona fide transmembrane domain, whereas the second hydrophobic domain (HD2, aa 492–510) did not (Fig. 4A). Starting with mCherry–Far1Ct (Fig. 1C), either HD1 or HD2 were deleted (an overview of constructs used is shown in Fig. 4B) and analyzed for their localization. Both mCherry–HD1 and mCherry–HD2 localized to the ER under steady-state conditions (Fig. 4C,D). OA treatment did not change the localization of mCherry–HD1 (Fig. 4C), but mCherry–HD2 overlapped strikingly with the LD marker A3Nt–GFP in this condition (Fig. 4D). Pearson correlation coefficient analysis confirmed that mCherry–HD1 did not overlap with LDs whereas mCherry–HD2 showed high values for LD colocalization but reduced ER overlap upon treatment with OA (Fig. 4E). Neither of the constructs localized to peroxisomes, likely because the putative binding site for the peroxisomal sorting chaperone Pex19 (Honsho et al., 2013; Van Ael and Fransen, 2006) was compromised.

Both constructs were double-epitope tagged and retrovirally transduced into U-2 OS cells, followed by selective permeabilization analysis as above. The HD1 construct behaved like a type I integral membrane protein of the ER, with the C-terminal HA epitope oriented into the lumen and only accessible after using TX-100 for complete permeabilization of all subcellular membranes (Fig. 5A). Remarkably, the HD2 construct was stained efficiently when using digitonin, suggesting that both the N- and the C-terminus were facing the cytosol (Fig. 5B).

The HD1 and HD2 mCherry constructs still contained adjacent amino acids from the Far1 sequence. To further analyze the two HDs, only the hydrophobic domains themselves were used to replace the HD of the A3Nt–GFP LD marker (Fig. 5C,D). While the hydrophobic domain of ACSL3 allows both the N- and C-terminus to be oriented towards the cytosol at the ER and on LDs (Poppelreuther et al., 2012), there is no consensus yet whether this is due to the formation of an amphipathic helix or a hairpin motif (Kory et al., 2016; Pataki et al., 2018). Regardless, the corresponding A3Nt–HD2–GFP construct behaved virtually indistinguishably from A3Nt–GFP, overlapping extensively with LDs after incubation with OA (Fig. 5D), and with Sec61β at the ER under steady state conditions (Fig. S5). A3Nt–HD1–GFP localized to mitochondria, which was consistent with an in silico prediction that this construct contained a mitochondria-targeting peptide (http://www.cbs.dtu.dk/services/TargetP/; data not shown).

The sorting of Far1 to peroxisomes depends on the cytosolic trafficking chaperone Pex19 (Honsho et al., 2013). Recently, Pex19 has also been implicated in the targeting of UBXD8 towards LDs via the ER (Schrul and Kopito, 2016). To test a possible involvement of Pex19 in the targeting of Far1 to LDs, Pex19 was inactivated in U-2 OS cells by CRISPR/Cas9-mediated indel formation (Fig. 6A). As expected, peroxisomes were absent in these cells, as verified here by the cytosolic location of the peroxisomal SKL reporter protein (Fig. 6B). However, the localization of Far1 on LDs was not impaired (Fig. 6B). Re-expression of Pex19 restored peroxisomal sorting of BFP-SKL (Fig. S6).

ASNA1 (also known as TRC40) targets C-terminal anchored membrane proteins to the ER (Borgese and Fasana, 2011; Guna et al., 2018; Rabu et al., 2009). Since the membrane association domains of Far1 are also located at the C-terminus, we considered whether ASNA1 has a role in the targeting of Far1 (to the ER; further trafficking towards LDs would be mediated either by the HD2 domain or additional factors). For unknown reasons, single-cell-derived clonal lines could not be obtained after targeting of ASNA1 by CRISPR/Cas9. Instead, cells were selected for the transient presence of Cas9 and propagated as a pool of cells with an estimated overall knockdown efficiency of >95% as assessed by western blotting (Fig. 6C) and genome sequencing (Fig. S7). However, the localization of Far1 on LDs was not changed, (Fig. 6D). The knockdown of ASNA1 was also applied in the Pex19-knockout background, but again did not change the LD localization of Far1 (Fig. 6E).

We demonstrated the presence of exogenous and endogenous Far1 on LDs in U-2 OS cells by several independent methods, and showed that the C-terminal 65 amino acids are sufficient for this unusual localization of a peroxisomal membrane protein (Figs 1 and 2). LD association of endogenous Far1 was also demonstrated for a second model system (A431 cells; Fig. 2). Additionally, Far1 was also observed on LDs in OA-treated epithelial COS-7 cells by fluorescence microscopy, and the closely related Far2 protein was equally localized to LDs (data not shown). Consistent with our results, proximity labeling-based proteomics experiments have recently suggested that Far1 is the most abundant peroxisomal protein on LDs of U-2 OS cells (Bersuker et al., 2018). Endogenous Far1 was also detected in LD fractions prepared from OA-treated HeLa cells by mass spectrometry (B.S., unpublished results). Following the localization of Far1 over time (Fig. S1A) suggested that Far1 arrives late on LDs (by 3–6 h), unlike ACSL3 which is already present 2 min after OA addition (Kassan et al., 2013; Poppelreuther et al., 2012). Our data are also more consistent with a gradual appearance of Far1 on LDs, and would therefore be different from what is seen for GPAT4 which appears to transit spontaneously within a short time frame from the ER to LDs (Wilfling et al., 2013). Upon starvation, Far1 also disappears earlier than ACSL3 from LDs (Fig. S1B). It is currently unclear whether this is due to turnover/degradation of Far1, or if it is ‘crowded out’ early in the process (Kory et al., 2015), because the LD affinity may be lower than for other more tightly associated LD proteins. The subcellular fractionation data are also consistent with our interpretation that Far1 associates less avidly with LDs than ACSL3. Remarkably, we made the incidental observation that Far1 did not move to LDs in OA-treated HuH-7 human hepatoma cells but remained on peroxisomes (data not shown). This suggests that these particular cells may lack one of the putative required factors for movement of Far1 to LDs, which will be addressed in future studies.

Close proximity between peroxisomes and LDs has been observed previously in different model organisms, sometimes enhanced by the remarkable flexibility of peroxisomal protrusions and tubules (Binns et al., 2006; Gao and Goodman, 2015; Hayashi et al., 2001; Schrader, 2001; Schuldiner and Bohnert, 2017). Peroxisomal extensions were also suggested to transfer the Arabidopsis thaliana lipase SDP1 to LDs during early seedling growth (Thazar-Poulot et al., 2015). In our hands, tubular peroxisomes were apparent in COS-7 cells overexpressing a high level of Far1 (data not shown) but were only very rarely observed in U-2 OS cells.

Heterologous expression of human Far1 in the yeast Saccharomyces cerevisiae indicated that the localization to LDs is not universally conserved. Far1 was localized to the ER in yeast cells cultured in glucose-containing medium, and to the vacuole when OA was the main energy and carbon source (Fig. S4), but neither to peroxisomes nor to LDs. Since sorting of mammalian Far1 to peroxisomes is Pex19 dependent (Honsho et al., 2013), this suggests that the yeast Pex19 homolog does not recognize the sorting motif contained within human Far1. On a side note, yeast does not have an endogenous fatty acyl-CoA reductase (Far) activity, nor does it synthesize ether lipids.

Targeting of proteins to LDs can be broadly categorized by their respective original localizations, that is from the ER or the cytosol (Kory et al., 2016; Ohsaki et al., 2014; Thiele and Spandl, 2008). ER-derived membrane proteins appear to lack intralumenal domains so that they may be readily accommodated in the monolayer LD membrane; they also relocate to the ER if LDs are consumed during starvation (Zehmer et al., 2009). Far1 does not fit well into this class of LD proteins because it is neither an ER protein originally nor is it located at the ER under starvation. It also displays an intralumenal domain while on peroxisomes (Honsho et al., 2013) and is partially located at the ER (as evidenced by N-glycosylation; Fig. 3).

Integral transmembrane domains and the corresponding topology are usually well predicted by the TMHMM algorithm (Krogh et al., 2001). However, when it comes to LD-associated proteins we found it to be rarely consistent with experimental data (data not shown), presumably because of the obvious difficulty of squeezing a ‘standard’ trans-membrane domain into the phospholipid monolayer. Hairpin topologies and amphipathic helices, which are typical for LD proteins, pose a formidable future challenge to bioinformatics.

The role of the ER for peroxisomal targeting and biogenesis is controversial (Buentzel et al., 2015; Erdmann, 2016; Lodhi and Semenkovich, 2014; Mayerhofer, 2016; Smith and Aitchison, 2013). N-glycosylation of the Far1 reporter proteins (Fig. 3) indicated that at least half of Far1 associated transiently with the ER in our model system, presumably before further transport towards the peroxisome was accomplished.

Pex19 is a cytosolic chaperone implicated in the targeting of peroxisomal membrane proteins (Jones et al., 2004), and it is also required for the sorting of Far1 to peroxisomes (Honsho et al., 2013). This traditional view has been extended by recent findings that Pex19 is also involved in transporting the hairpin protein UBXD8 to the ER, from where it moves onwards to LDs (Schrul and Kopito, 2016). However, while our pex19-KO cells replicated the expected phenotype (loss of peroxisomes), the localization of Far1 on LDs or at the ER was not impaired (Fig. 6A,B).

Far1 is also a tail-anchored membrane protein (Honsho et al., 2013), which is at least transiently exposed to the N-glycosylation machinery at the ER (Fig. 3C). Several chaperone assisted targeting pathways have been described (Borgese and Fasana, 2011; Guna et al., 2018). The probably most prominent sorting mechanism involves ASNA1, which was also required for targeting peroxisomal proteins to the ER (van der Zand et al., 2006). However, depletion of ASNA1 did not affect Far1 LD localization to any significant extent (Fig. 6C,D). In hindsight, our most likely explanation would be that the targeting of tail-anchored proteins involves a rather redundant machinery (Rabu et al., 2009). We also tried a double Pex19 and ASNA1 knockout approach, but again Far1 was still found on LDs (Fig. 6E). Knocking out even more chaperones would be expected to show severe side effects (the translocon component Sec61β is also a tail-anchored protein), and was not tried here. We conclude that Far1 targeting to the ER and LDs is robustly supported by the cellular machinery of mammalian cells and is unlikely to be dependent on one single factor.

We propose that Far1 may be integrated into the ER membrane in two different ways (Fig. 7). The first topology features an integral transmembrane domain (first hydrophobic domain; HD1) and an amphipathic helix-like second domain (HD2), which associates primarily with the lumenal leaflet of the ER membrane. This appears also to be the only form competent for export/integration to peroxisomes and satisfyingly explains our glycosylation, selective permeabilization and protease sensitivity data.

The alternative second topology might be driven by a strong preference of the HD2 domain for the cytosolic leaflet of the ER under conditions of increased triglyceride synthesis. Molecular dynamics simulations indicate that the properties of the surrounding phospholipid monolayer are significantly affected by the underlying neutral lipids (Bacle et al., 2017; Prevost et al., 2018), leading to packaging defects that would be even more pronounced during rapid LD growth. Amphipathic helices containing bulky hydrophobic amino acids would show synergistic binding (Čopič et al., 2018; Prevost et al., 2018); the HD2 domain contains several of these amino acids, and wheel projections suggest a sidedness typical for amphipathic helices (Fig. 7). Consistent with this, HD2 appeared as efficient as the native ACSL3 hydrophobic domain in terms of LD targeting (Fig. 5D). ACSL3 is considered a bona fide LD protein (Fujimoto et al., 2004; Poppelreuther et al., 2012; Walther et al., 2017) and is already present on emerging LDs (Kassan et al., 2013). If HD2 associates first with the membrane, the short loop of intermittent amino acids would no longer allow an integral transmembrane orientation for HD1. This topology has the short C-terminal region exposed to the cytosol and is therefore in line with our permeabilization and protease sensitivity data for LD-localized Far1.

Summarizing speculatively, the targeting of Far1 may be determined by a kinetic competition between an insertion machinery recognizing HD1, and the affinity of HD2 for the cytoplasmic surface of the ER. Under conditions of increased triglyceride synthesis, binding of HD2 could simply be faster/stronger than the recognition of HD1. The binding affinity of HD2 for the phospholipid monolayer would then pull the Far1 molecule onto growing LDs. The incorporation of HD1 into the outer leaflet of the ER membrane might constitute an opposing force, delaying the transfer of Far1 to LDs.

Alternative scenarios may involve direct trafficking from peroxisomes to LDs as exemplified by the A. thaliana lipase SDP1 (Thazar-Poulot et al., 2015) or a retrograde transport from peroxisomes to the ER (Passreiter, 1998; Titorenko and Rachubinski, 1998). However, this would involve moving the lumenal short hydrophilic segments of the Far1 C-terminus across the hydrophobic phospholipid layer, which is considered energetically highly unfavorable.

Dual topology is a rare phenomenon, and most cases described so far involve multipass transmembrane domain proteins (more than three TMDs; Lambert and Prange, 2001; Rapp et al., 2006; Stockbridge et al., 2014; Wurie et al., 2011). Generally, the orientation of membrane proteins is considered to be determined during their co-translational insertion into the membrane, but post-assembly topology switching does occur (Bogdanov et al., 2014). We applied the term ‘dual topology’ to mean that one amino acid sequence occurs in two different topological orientations with respect to a phospholipid membrane (Granseth, 2010). However, dual topology has also been more narrowly defined as proteins that “insert into the membrane in two opposite orientations with an approximate 1:1 stoichiometry” (von Heijne, 2006). This would not encompass Far1, because the alternative topology of Far1 is restricted to the C-terminus and does not change the cytosolic orientation of the enzyme domain. Moreover, the stoichiometry of the two Far1 topologies is not fixed but appears to be dependent on the extent of triglyceride synthesis.

The topologies of Far1 appear to be strongly correlated to the subcellular localization. Far1 converts fatty acyl-CoA into fatty alcohols; on peroxisomes these are further metabolized by lumenal enzymes and are finally incorporated into ether lipids. Far1 on LDs may therefore reduce the production of ether lipids from exogenously supplied fatty acids, which would be consistent with our ethanolamine plasmalogen quantification (Fig. S3). This speculative scenario is reminiscent of the protein storage ‘depot’ function of LDs (Cermelli et al., 2006). However, clarification clearly requires future studies, for example, employing localization-fixed Far1 versions targeted exclusively to either the ER, LDs or peroxisomes. The metabolic fate of LD-synthesized fatty alcohols is currently unclear and beyond the scope of this study. They could enter a fatty alcohol cycle, which would be involved in cellular lipid homeostasis (Rizzo, 2014), balancing ether and non-ether lipid levels. In line with this, the vast majority of externally added hexadecanol is converted into palmitate rather than to ether lipids in cultured fibroblasts (Rizzo et al., 1987). Other metabolic routes are also conceivable, as many enzymes involved in lipid metabolism are found on LDs (Goodman, 2009).

Finally, tantalizing recent data suggest that LD and peroxisome biogenesis may occur at the same ER subdomains marked by mammalian MCTP2 (yeast Pex30) and seipin (Joshi et al., 2018; Wang et al., 2018). This spatial proximity may well be the mechanistic basis for the alternative sorting pathways of Far1, and may also be instrumental in allowing dual topologies upon insertion into the membrane.

Human osteosarcoma U-2 OS cells were cultured in McCoys 5A modified medium (Thermo Fisher Scientific, Waltham, MA) supplemented with 10% fetal calf serum (FCS; Life Technologies, Carlsbad, CA), 2 mM L-glutamine (GlutaMAX, Life Technologies) and 1% penicillin/streptomycin (Life Technologies). Cells were subcultured when reaching 80% confluence. Human immortalized SZ95 sebocytes (Zouboulis et al., 1999) were cultured in Sebomed^® basal medium (Merck, Kenilworth, NJ) supplemented with 10% FCS, 1 mM CaCl₂ and 5 ng/ml human EGF. Human A431 epidermoid carcinoma cells (ATCC CRL-1555) were grown in Dulbecco's modified Eagle's medium (DMEM) with 4.5 g/l glucose (Life Technologies, #41965) containing the same supplements as for U-2 OS cells. All cell lines were routinely tested for mycoplasma contamination and authenticated.

For transient expression, cells were grown on coverslips to 80–90% confluence and transfected with FugeneHD (Promega, Madison, WI). Briefly, the transfection mix for a 12-well plate with 0.8 ml medium included 40 µl of OptiMEM (Life Technologies), 0.8 µg total DNA and 4 µl FugeneHD (Promega). For six-well plates, the ratios were adjusted to 2 ml medium. After 4 h, the transfection mix was removed and cells were cultured overnight in antibiotic-free medium. LD growth was induced by adding 600 µM OA (from a 8.66 mM OA and BSA stock solution; molar ratio 6:1; oleic acid, O1383 from Sigma, fatty acid-free BSA A8806 from Sigma, St. Louis, MO). The OA medium was present on the cells overnight, if not indicated otherwise.

For stable expression, the constructs were integrated into the genome by retroviral transduction as described previously (Schuck et al., 2004). Briefly, phoenix-gp cells were transfected with the indicated fusion protein cDNAs cloned into the retroviral pRIJ vector (Küch et al., 2014) or pRIJ-neo vector. Viral particles contained in the medium supernatants were harvested and applied to the cells, followed by antibiotic selection of the transduced target cells (puromycin, 2 µg/ml, 48 h; neomycin, 2 mg/ml for 4 days). Cells were never allowed to reach densities above 90% during the selection process.

Full-length Far1 cDNA was obtained from a cDNA library of SZ95 sebocytes (Zouboulis et al., 1999) and was cloned with PacI and KpnI into a modified version of pEGFP-N1 (Clontech) containing a FLAG epitope at the 3′-end of GFP (Far1–GFP–FLAG). Sequencing confirmed the cDNA to be 100% identical to the GenBank database entry NM_032228.5.

For the N-terminally tagged version, the full-length coding region of Far1 was amplified, cloned into mCherry–C1 (Clontech) and subsequently subcloned yielding FLAG–mCherry–Far1. FLAG–mCherry–Far1 was the basis for all further cloned deletion mutants (Fig. 4B), double-epitope tag variants (Fig. 3 and Fig. 5) and the opsin-tagged variants (Fig. 3).

The opsin glycosylation tag (Bañó-Polo et al., 2011; Favaloro et al., 2008; Nilsson and von Heijne, 1993) consisted of the amino acid sequence AASGS­GPNFYVPFSNKTLEGPIL derived from the bovine sequence. The N-glycosylation motif NxT was changed to QKT for the opsin control construct. Both tags were preceded by amino acids 451–515 of Far1.

The N-terminus of ACSL3 (aa 1–135 of human ACSL3) fused to GFP/BFP was used as a LD marker protein (Poppelreuther et al., 2012). As an ER marker, a signal sequence (CD8)–GFP–KDEL cDNA sequence was transfected. Alternatively, the ER resident acyl-CoA-synthetase FATP4 (also known as ACSVL5) (Milger et al., 2006) was used. Fluorescent proteins containing a C-terminal SKL tag (GFP–SKL and BFP–SKL) were used as lumenal peroxisomal marker proteins. The full-length cDNA of human Pex11β fused to GFP or mCherry, respectively, was used as a marker for peroxisomal membranes (Koch et al., 2010). Details on the plasmid sequences, cloning strategies and primers are freely available upon request. All constructs created were sequenced for validation.

Immunofluorescence was performed with rabbit anti-CaBP1 (Fullekrug et al., 1994), rabbit anti-FLAG [F7425 Sigma; 1:1000 for immunofluorescence (IF)] and mouse anti-HA (SC-7392, Santa Cruz Biotechnology, Dallas, TX; 1:200 for IF). As a LD marker, anti-Plin3 antibody (cat. no. MAB76641. R&D Systems, Minneapolis, MN, USA, 1:600) was used. Guinea pig anti perilipin-2 (Plin2) antibodiy (Heid et al., PLoS 2013) was used for western blotting at 1:1000; 5% BSA was used as a blocking reagent. For western blotting, mouse anti-FLAG (F1804, Sigma; 1:4000), rabbit anti-HA (SC-805, Santa Cruz Biotechnology; 1:1000) and mouse anti-actin (A5441, Sigma; 1:40.000) were used. As markers for the ER, anti-calnexin antibody (cat. no PA1-30197, Thermo Fisher Scientific; 1:200) was used. Anti-ACSL3 antibody (cat. no H00002181-B01P, Abnova, Taipei, Taiwan, 1:3000) served as a marker for LDs. Anti-Pex19 antibody (cat. no PA5-22129, Thermo Fisher Scientific) was used in a 1:10.000 dilution for western blotting. The anti-Far1 antibody was raised against an N-terminal fragment (NBP1-89847, Novus Biologicals, Centennial, CO) and was used at a 1:100 dilution for immunofluorescence and 1:2000 for western blotting. Anti-Pex13 (#ABC143, Millipore, Burlington, MA) was used at a 1:5000 dilution for western blotting.

U-2 OS cells were grown on 10 mm coverslips and fixed with 4% PFA in PBS for 20 min. Cells that were not processed for immunofluorescence were mounted in Mowiol-glycerol solution (Mowiol 4-88; Calbiochem, San Diego, CA).

For conventional immunofluorescence, the cells were permeabilized-blocked with 0.1% saponin, 0.5% gelatin and 0.5% BSA (SGB) for 10 min at room temperature. Cells were incubated with the first antibody diluted in SGB for 1h. Coverslips were subsequently washed three times with 0.01% saponin, 0.2% gelatin (SG-buffer), and secondary antibodies were diluted in SGB and applied for 60 min. After two washes with SG-buffer and two washes in PBS cells were mounted as above.

Immunostaining of endogenous Far1 included a denaturation step with 6 M guanidine hydrochloride for 10 min before anti-Far1 antibodies (1:100) were applied. Sequential staining was used for double immunofluorescence; permeabilized-blocked cells were stained for ACSL3 first, followed by an extra 20 min 4% PFA fixation step and finally by the denaturation protocol to allow Far1 immunostaining.

For immunofluorescence under selective permeabilization conditions, cells were fixed as described above and permeabilized with 10 µM digitonin for 10 min on ice. Permeabilization with 1% Triton-X 100 on ice solubilized all membranes, and served as a staining control. Cells were subsequently blocked with 1% BSA for 1 h at room temperature and incubated with the first antibody diluted in 1% BSA for 1 h. Cells were then washed three times for 5 min with PBS and incubated with the secondary antibody diluted in PBS for 1 h at room temperature. After additional three washes for 5 min in PBS, the cells were embedded in Mowiol. The intraluminal ER protein CaBP1 (Fullekrug et al., 1994) was used to verify the inaccessibility of extracytoplasmic epitopes after digitonin permeabilization.

Neutral lipids were stained with 2 µg/ml BODIPY 493/503 (D-3922, Invitrogen, Waltham, MA) or 1 µM monodansylpentane (MDH; SM1000a, Abgent, San Diego, CA) diluted in PBS for 15 min at room temperature.

Raw images were acquired on an Olympus BX41 microscope, 60x oil immersion Plan S Apo NA 1.35, F-view II CCD camera with cell^D software (Olympus, Tokyo, Japan). Gamma values reached from 1.5 (ER structures) over 1.3 (peroxisomes) to 1.0 (LDs).

LD localization scoring was undertaken as follows. The subcellular distribution of mCherry–FAR1 was assessed manually by distinguishing five categories, yielding a score from 0 to 1. Researchers were not blinded for the presence of OA (Fig. 1B), or for time and concentration of OA (Suppl. Figs. S1A,B). Cells with mCherry–FAR1 being only localized to peroxisomes received a score of 0, whereas minimal but significant presence on LDs qualified for a score of 0.25. A score of 0.5 was given if the localization of mCherry–FAR1 on peroxisomes and LDs was close to being evenly distributed. When mCherry–FAR1 was mostly located on LDs but with a remaining peroxisomal localization, the cell was assigned a score of 0.75, whereas a score of 1 indicated the complete presence of FAR1 on LDs without any remaining peroxisomal localization.

Pearson correlation coefficient analysis was undertaken as follows. The images were first deconvolved using the ImageJ plugin ‘Diffraction PSF 3D’ (Dougherty, 2005) together with a point spread function matching the technical settings of the Olympus BX41 microscope. Colocalization of the marker protein and the respective Far1 construct was assessed with the Coloc2 plugin of ImageJ, using the Pearson correlation coefficient to quantify the colocalization in images (no thresholding was applied).

Quantification of HA/FLAG epitope tagging (Figs 3A and 5E) was achieved by measuring the fluorescence intensities of 75 randomly chosen cells from three independent experiments as assessed by ImageJ after thresholding (15,000–65,535 in 16-bit images in Fig. 3A, without thresholding in Fig. 5E). The ratio of the HA-derived fluorescence to the FLAG-derived fluorescence was calculated for each cell individually.

Cells were seeded at a low density in MatTek dishes (Ashland, MA), which have glass-gridded coverslips with an alphanumeric pattern to relocate cells or cell clusters. After treatment, cells were fixed with 4% PFA and 0.2% glutaraldehyde in PBS for 30 min. After fixation, the cells were washed three times with 150 mM glycine and twice with PBS. Subsequently, cells were imaged with a Nikon Ti Eclipse microscope equipped with an Ultraview VoX confocal spinning disc system (PerkinElmer, Waltham, MA). First, cells were imaged with a 20× objective (Plan Fluor, NA 0.75, oil), to record the positions of the cells of interest expressing fluorescent proteins. Then a 60× objective was used (Apo TIRF, NA 1.49, oil) to acquire higher magnification images to improve the correlation with the EM data. Post-fixation of the cells was carried out with 2% osmium tetroxide (OsO₄) in 50 mM Na-cacodylate buffer for 40 min on ice. Afterwards, the cells were washed three times with ddH₂O and stained with 0.5% uranyl acetate (UA) in H₂O for 30 min. After three additional washes with ddH₂O the cells were dehydrated in an increasing ethanol series at room temperature (40%, 50%, 60%, 70% and 80%) for 5 min each followed by 95% and 100% ethanol for 20 min each. The cells were then quickly immersed in 100% propylene oxide and embedded in an Araldite-Epon mixture (Araldite 502/Embed 812 kit, Electron Microscopy Sciences, Hatfield, PA). The samples were incubated for 48 h at 60°C for polymerization of the resin. The MatTek dishes and the gridded coverslips were subsequently removed from the polymerized resin blocks and the embedded cell monolayers were sectioned using a Leica Ultracut UCT ultramicrotome (Leica Microsystems) and a diamond knife (Ultra 35° from Diatome). Finally, the 70 nm sections were counter-stained with 2% UA in 70% methanol and lead citrate in CO₂-free H₂O and visualized under a JEM-1400 (JEOL) TEM equipped with a TemCam-F416 digital camera system (TVIPS) operated at 120 kV. Correlations between the light microscopy images and the electron micrographs were calculated using the non-rigid transformation of the Icy ec-CLEM plugin (http://icy.bioimageanalysis.org/; Paul-Gilloteaux et al., 2017). CLEM as applied here has been described in detail recently (Romero-Brey, 2018).

LDs were prepared by a sucrose step gradient (modified from Thul et al., 2017 and Krahmer et al., 2013). Cells were grown to subconfluency in two 180 cm² plates and treated with or without 600 µM OA overnight. The following steps were carried out at 4°C. Cells were washed three times with PBS, collected by scraping and were sedimented at 700 g for 10 min). The pellet was resuspended in buffer A1 (50 mM Tris-HCl pH 7.4, 20 mM sucrose, 1 mM EDTA, 1 mM β-mercaptoethanol) and homogenized using 10 strikes with a G22 needle and additional 23 strikes with a G25 needle. The postnuclear supernatant (PNS) was obtained by centrifugation at 1000 g for 10 min. The PNS was centrifuged at 100,000 g (TLA55 rotor) for 60 min to obtain a membrane pellet and the cytosolic supernatant including LDs. To further separate the cytosol from LDs, the supernatant was adjusted to 2.7 ml with buffer A1 and mixed with 2.7 ml of buffer A4 (50 mM Tris-HCl pH 7.4, 1.08 M sucrose, 1 mM EDTA, 1 mM β-mercaptoethanol). Together, it formed the bottom layer of the sucrose gradient. This was carefully overlaid with 1.8 ml of buffer A3 (50 mM Tris-HCl pH 7.4, 270 mM sucrose, 1 mM EDTA, 1 mM β-mercaptoethanol), buffer A2 (50 mM Tris-HCl pH 7.4, 135 mM sucrose, 1 mM EDTA, 1 mM β-mercaptoethanol) and buffer A1. This step gradient was centrifuged for 105 min at 150,000 g in an SW41TI rotor. Eight 1.35 ml fractions were collected from top to bottom. Fractions 1–8 were precipitated with chloroform:methanol and resuspended in 40 µl sample buffer [2% SDS (w/v), 62.5 mM Tris-HCl pH 6.8, 10% (v/v) glycerol, 100 mM β-mercaptoethanol]. The membrane pellet was directly dissolved in 320 µl sample buffer. 10 µl of each fraction were subjected to SDS-PAGE with subsequent western blot analysis. Densitometry values were normalized to 1.0 for each protein from all fractions, and the average relative ratio between Far1 and Pex11β in the LD fractions calculated as 2.7 (from three independent fractionations: 3.1, 2.2 and 2.6).

U-2 OS wild-type cells were seeded at ∼80,000 cells/well (12-well plate), and incubated for 18 h in either starvation medium (1 g/l glucose DMEM, 200 µM fatty acid-free BSA), standard growth medium, or oleation medium (600 µM bound to BSA). [¹⁴C]acetate (NEC084HSB; PerkinElmer) was added to a final concentration of 10 µM at 50 Ci/mol for 6 h. Lipids were extracted based on the Folch protocol as described in detail recently (Poppelreuther et al., 2018). Vacuum-dried lipids were resuspended in 3% sodium methoxide (Sigma) in methanol, and incubated for 20 min at 50°C at 300 rpm. The reaction was stopped by the addition of 13.5 µl glacial acetic acid and the samples placed on ice. Lipids were extracted as before, spotted onto a thin-layer chromatography (TLC) plate and developed in chloroform:methanol:glacial acetic acid (65:25:10, v/v) for 35 min. Incorporation of radioactivity was detected by phosphorimaging plates, scanned by the BAS-1500 system (Fuji, Tokyo, Japan) and quantified with ImageJ.

Stably expressing FLAG-FAR1-HA.U-2 OS were seeded in 180 cm² plates and treated with 600 µM OA bound to BSA for 24 h. The cells of each plate were scraped and resuspended in 600 µl homogenization buffer (50 mM Tris-HCl pH 7.4, 20 mM sucrose, 1 mM EDTA) and homogenized by passing the cells 20 times through a G22 needle. Postnuclear supernatants (PNS) were obtained by centrifugation at 1000 g for 10 min. Membranes were sedimented by centrifugation at 100,000 g in a TLA45 rotor. The supernatant containing LDs and cytosolic proteins was carefully removed and the pellet was resuspended in 600 µl homogenization buffer. Digestion with 1 µg of proteinase K (New England Biolabs, Ipswich, MA) per 200 µl sample was for 30 min at 4°C. Triton X-100 (1%) was used to solubilize all membranes as a control. The reaction was terminated by the addition of 600 µl methanol and 140 µl chloroform. Proteins were recovered by sedimentation and solubilized in 2× sample buffer [4% SDS (w/v), 125 mM Tris-HCl pH 6.8, 20% glycerol, 200 mM β-mercaptoethano] for SDS-PAGE.

To evaluate the localization of human Far1 in yeast, the cDNA coding for FLAG–mCherry–Far1 was cloned into pYX142 (R&D Systems) using PCR primers encoding XmaI and NheI restriction sites (s_BglII_XmaI_FLAG 5′-ACGTAGATCTCCCGGGACCATGGACTACAAGGACGACG-3′ and a_NheI_XbaI_NotI_Far1Ct 5′-ACGTGCTAGCTCTAGAGCGGCCGCTCAGTATCTCATAGTGCTGG-3′). For the microscopy analysis, the LD marker Erg6 (Pu et al., 2011) and the peroxisomal marker Pxa1 (Shani et al., 1996) in the yeast strains BY4741 GFP-Pxa1 (his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 hphΔn::URA3::SpNOP1pr-GFP-Pxa1) and BY4741 GFP-Erg6 (his3Δ1 leu2Δ0 met15Δ0 ura3Δ0 hphΔn::URA3::SpNOP1pr-GFP-Erg6) were used. Liquid cultures were grown overnight at 30°C in synthetic defined dextrose (SD) medium without uracil. Strains were diluted to an optical density at 600 nm (OD₆₀₀) of ∼0.2 with SD medium (6.7 g/l yeast nitrogen base and 2% glucose) or S-Oleate (6.7 g/l yeast nitrogen base, 0.2% OA and 0.1% Tween-80). Strains were incubated at 30°C for 4 h in SD medium or for 20 h in S-Oleate. The cultures in the plates were then transferred into glass-bottom 384-well microscope plates (Brooks Life Science Systems, Chelmsford, MA) coated with concanavalin A (Sigma). After 20 min, wells were washed twice with SD medium without riboflavin (for strains in glucose) or with double-distilled water (for strains in OA) to remove non-adherent cells and to obtain a cell monolayer. The plates were then transferred to the ScanR inverted fluorescent microscope system (Olympus, Tokyo, Japan). Images of cells in the 384-well plates were recorded in the same liquid as the washing step at 24°C using a 60× air lens (NA 0.9) and with an ORCA-ER charge-coupled device camera (Hamamatsu, Hamamatsu, Japan). Images were acquired in two channels: GFP (excitation filter 490/20 nm, emission filter 535/50 nm) and mCherry (excitation filter 572/35 nm, emission filter 632/60 nm). All images were taken at a single focal plane.

For generating the Pex19-knockout cell lines, plasmid eSpCas9(1.1), Addgene #71814 (Slaymaker et al., 2015) was modified to include a puromycin resistance gene. The guide for Pex19 (Schrul and Kopito, 2016) was generated by annealing sense (5′-CACCGTGTCGGGGCCGAAGCGGAC-3′) and antisense (5′-AAACGTCCGCTTCGGCCCCGACAC-3′) oligonucleotides and ligation into the backbone containing the Cas9 plasmid cut with BbsI. U-2 OS cells were transfected in six-well plates; a Cas9 vector without a guide sequence served as control. At 24 h post transfection, the cells were trypsinized and treated with 2 µg/ml puromycin for 48 h. After an additional 24 h, cells were trypsinized and a volume corresponding to 30 cells was seeded into 60 cm² dishes. At 2 weeks after transfection, single-cell clones were transferred to a 24-well plate. Single-cell clones were screened by eGFP-SKL transfection for the absence of peroxisomes and verified by western blotting.

Knockout of ASNA1 was mediated by transfection of a plasmid containing SpCas9 (Ran et al., 2013; Addgene #62988) and a guide RNA targeting exon 3 of the ASNA1 gene (sense 5′-CACCGCCTGACGAGTTCTTCGAGG-3′, antisense 5′-AAACCCTCGAAGAACTCGTCAGGC-3′). After puromycin selection (2 µg/ml) for 48 h, the cells were grown for 2 weeks with cell densities above 30%. No single-cell clones could be obtained. For the preparation of genomic DNA, ∼50,000 cells were resuspended in 17 µl ddH₂O, 2 µl of 10× Taq-PCR-Buffer (QIAGEN, Hilden, Germany) and 800 U proteinase K (New England Biolabs, Ipswich, MA). Incubation at 65°C for 60 min was followed by heat inactivation for 15 min at 95°C. The targeted genomic region was amplified by conventional PCR and sequenced. The knockout efficiency was estimated by the TIDE algorithm (Brinkman et al., 2014), comparing the sequence chromatograms derived from the wild-type locus and the guide-targeted locus. The knockout was further confirmed by western blotting and estimated to reduce ASNA1 expression by >95%. Antibodies used were rabbit anti-ASNA1 (1:1000, 15450-1-AP, Proteintech), and rabbit anti-β-tubulin (1:1000, 11224-1-AP Proteintech). The antibody incubation was carried out in PBS with 0.1% Tween-20 overnight at 4°C. Blots were imaged using an Odyssey Sa Infrared imaging system with IRDye LiCOR secondary antibodies.

Guides for ASNA1 were designed with CRISPOR (http://crispor.tefor.net/). Genomic alterations were estimated with the TIDE algorithm (Brinkman et al., 2014). Hydrophobicity scores were generated and evaluated by TMHMM 2.0 (http://www.cbs.dtu.dk/services/TMHMM/); (Krogh et al., 2001). The helical wheel projection was generated by HeliQuest (http://heliquest.ipmc.cnrs.fr/). Adobe Photoshop was used to arrange composite figures (Adobe Systems, San Jose, CA), and Adobe Illustrator was used to draw the cartoon.

We thank the Electron Microscopy Core Facility (EMCF) at Heidelberg University, especially Uta Haselmann and Simone Hoppe, for their technical assistance with the CLEM. Furthermore, we thank Vibor Laketa of the Infectious Diseases Imaging Platform (IDIP) at the Center for Integrative Infectious Disease Research Heidelberg (CIID) for light-microscopy support. Markus Kunze, Johannes Berger, Gunter Stier, Irmgard Sinning, Raphael Zoeller, Gabriele Dodt as well as Juan Liao, Simone Sander and Simon Gairing from our group kindly contributed plasmids for this study. Hans Heid (Heidelberg, Germany) kindly provided antibodies against perilipin family proteins. Willi Just (Heidelberg, Germany) is acknowledged for intellectual exchange on peroxisomal biology. T.E. was supported by the MD/PhD program of the Medical Faculty of the University of Heidelberg. Blanche Schwappach and Willi Just are acknowledged for intellectual exchange on targeting of tail-anchored proteins and peroxisomal biology, respectively.