Targeting surface voids to counter membrane disorders in lipointoxication-related diseases

Eukaryotic cells are sub-divided into highly specialized organelles, thereby allowing specific functions to occur efficiently and simultaneously in different compartments of the cell. To allow the optimal function of these subcellular compartments, cells have developed complex and highly regulated processes to maintain specific phospholipid compositions within membranes, such as phospholipid remodeling (Yamashita et al., 2014), to regulate the acyl chain composition, and selective lipid transport from their site of synthesis to the accepting membrane (Gillon et al., 2012; Tamura et al., 2014). As a consequence, when a single phospholipid type is considered, its acyl chain composition varies greatly depending on the organelle; the saturation rate of the acyl chains increases progressively along the secretory pathway to reach its maximum at the plasma membrane (Schneiter et al., 1999). The fatty acyl content of a phospholipid molecule contributes to its overall shape, which is defined by the cross-sectional area of the headgroup relative to the cross-sectional area of its acyl chains (de Kroon, 2007). For example, an oleoyl chain (C18:1) occupies a larger volume than a palmitoyl chain (C16:0), because the double bond induces a ‘kink’ in the middle of the chain (Deguil et al., 2011). Therefore, inserting a kinked acyl chain within a phospholipid species tends to shift its shape from rather cylindrical to conical. Conical lipids have major impacts on the biophysical properties of biological membranes; when inserted into a bilayer they tend to generate voids at the membrane surface (Vamparys et al., 2013).

The heterogeneity in the fatty acyl composition of phospholipids along the secretory pathway led Bigay and Antonny to propose the coexistence of two broad territories within the cell (Bigay and Antonny, 2012): the territory of loose lipid packing and large and deep surface voids [i.e. from the endoplasmic reticulum (ER) to the trans-Golgi Network (TGN)], where weakly charged monounsaturated phospholipids are overrepresented, and the territory of electrostatics, which is defined by the presence of saturated, highly charged lipids (i.e. the plasma membrane and endosomal compartments). In the territory of large surface voids, cytosolic proteins take advantage of the surface clefts generated by the membrane-packing defects to be recruited to the organelle membranes, and thereby regulate central cellular processes, such as vesicular budding (Bigay and Antonny, 2012; Bigay et al., 2005).

Not surprisingly, perturbations of the fatty acid composition of phospholipids have dramatic effects of organelle function. For example, exogenously supplied saturated fatty acids (SFAs), which efficiently incorporate within phospholipids, have a major impact on the integrity of the organelles from the loose-lipid packing territory. Early in the secretory pathway, SFAs have been shown to alter vesicular budding in the ER, promote the accumulation of unfolded proteins within this compartment and induce a so-called ER stress, which can ultimately lead to cell death by apoptosis (Borradaile et al., 2006; Dhayal and Morgan, 2011; Guo et al., 2007; Gwiazda et al., 2009; Laybutt et al., 2007; Pineau et al., 2009; Preston et al., 2009). Later in the pathway, SFAs have also been demonstrated to affect vesicle formation at the Golgi in a process that could be directly connected to increased lipid packing (Payet et al., 2013). Indeed, we have demonstrated that saturated phospholipids alter the recruitment of proteins of the ArfGAP family to nascent vesicles in vivo, which emerge from this compartment both in yeast and human cells (Payet et al., 2013; Vanni et al., 2014). Given that ArfGAP and mechanistically related proteins are required for vesicle transport and contain motifs that sense lipid packing (Bigay et al., 2005), it has been proposed that this process could account for the general disruption of the secretory pathway under conditions of SFA accumulation (Payet et al., 2013).

Such observations could be very pertinent for understanding several physiopathological states in humans. Indeed, SFAs, which are abundant in the western diet, have been directly related to the etiology of several high-occurrence ‘way-of-life’-related diseases, including type 2 diabetes, hepatic steatosis and atherosclerosis (Akazawa et al., 2010; Alkhateeb et al., 2007; Cnop et al., 2001; Miller et al., 2005).

In the present study, we isolated new molecules able to counter SFA-related lipointoxication. Taken together, the results presented here validate the method as an efficient high-throughput screening approach to identify new lipointoxication inhibitors. Moreover, they point at the modulation of surface voids as a central parameter to consider in order to efficiently counter the impact of SFA on cell function in certain human diseases.

The first screening step used in this study relies on a Saccharomyces cerevisiae knock-out mutant of the Δ-aminolevulinate (δ-ALA) synthase (hem1Δ), which brings together most of the trademarks of the lipid-induced toxicity observed in pancreatic β-cells (Pineau et al., 2009) (Fig. 1A). When grown in the absence of δ-ALA, this strain fails to synthesize heme, the prosthetic group of Ole1p, the fatty acid desaturase. As a consequence, the hem1Δ cells stop growing as early as 5 h after a shift to non-δ-ALA supplemented medium [YPD+ergosterol (Erg), denoted Ø] and this arrest correlates with an accumulation of SFA [mainly myristic (C14:0), palmitic (C16:0) and stearic (C18:0) acids] at the expense of unsaturated fatty acids [UFAs; palmitoleic (C16:1) and oleic (C18:1) acids] (Ferreira et al., 2004; Pineau et al., 2008). Interestingly, cell growth can be fully recovered and all the trademarks of SFA lipointoxication alleviated if selective UFAs, such as the mono-unsaturated fatty acid oleic acid (Ole), are added to the medium [(Deguil et al., 2011); Table S1 and Fig. 1A].

However, Ole presents an important limitation as a potential agent to alleviate SFA lipointoxication. Indeed, the inability to incorporate excess exogenous Ole into triacylglycerols (TAGs) in a yeast mutant lacking the acyltransferases Lro1p, Dga1p, Are1p and Are2p, which contribute to TAG synthesis (hereafter referred to as the quadruple mutant strain), results in dysregulation of lipid synthesis, massive proliferation of intracellular membranes and ultimately cell death, in a process that appears largely independent of the unfolded protein response (UPR) pathway (Petschnigg et al., 2009; Fig. 1A). Importantly, the same observations can also be reproduced in mammalian models (Listenberger et al., 2003). This probably explains why Ole becomes toxic under lipointoxicated states, when the overall capacity to buffer excess fatty acids within TAG pools is exceeded or, under normal conditions, to cells with low TAG synthesis capacity, such as islet non-β cells (Cnop et al., 2001).

The second step of our screening strategy therefore consisted in evaluating the whether the molecules identified with the first screen were non-toxic (i.e. for their ability to restore the growth of hem1Δ cells under SFA accumulation) towards the quadruple mutant strain. Based on our previous work, 64 fatty acids of various chain length and levels of unsaturation were first tested (Table S1). Interestingly, in contrast to Ole, among the 12 UFAs which had proven to be efficient in countering SFA toxicity in the hem1Δ model (Deguil et al., 2011), nine appeared to be less toxic to the quadruple mutant cells than Ole (Table S1). However, it appeared that low toxicity systematically correlated with a lower efficiency to counter the SFA-deleterious effects on growth (compare Ole and cis-vaccenic acid, for example). Therefore, none of the fatty acids identified in this test proved to be as efficient as Ole in relieving SFA intoxication.

These results led us to initiate a broader screening process and to apply this two-step yeast-based assay to commercial molecule libraries (see the Materials and Methods section). Among the 1470 molecules tested in this study, only two compounds appeared to counter SFA toxicity as efficiently as Ole in the hem1Δ model, with low toxicity to the quadruple mutant strain and the wild-type strain (Fig. 1B–F).

Both of them, namely 1-oleoyl-2-acetyl-sn-glycerol (OAG) and 1-oleoyl lysophosphatidic acid (OLPA) shared very similar structures (Fig. 1C), with an oleic acyl chain at the sn-1 position. Consistent with the lack of toxicity in the quadruple mutant strain, and in contrast to Ole, OLPA and OAG did not induce the accumulation of TAG in a normal, TAG-synthesis competent strain (Fig. S1).

Because it has been shown that the essential features required to promote viability reflects the ability of fatty acids to alleviate ER stress both in yeast and mammalian cells (Deguil et al., 2011; Dhayal and Morgan, 2011; Pineau et al., 2009), the capacity of OAG and OLPA to counteract the SFA-related UPR-induction was tested (Fig. 2). In yeast, induction of the UPR pathway was determined through the activity of the β-galactosidase reporter gene, set under the control of four UPR elements (UPREs; Cox and Walter, 1996). According to our expectations, OAG and OLPA relieved the SFA-induced ER stress to an extent very similar to Ole (Fig. 2A).

OAG and OLPA also potently protected BRIN-BD11 cells (a rat hybrid pancreatic β-cell line) against palmitate-induced cytotoxicity (Fig. 2B) in a way very similar to what had already been reported for Ole (Dhayal and Morgan, 2011). Modulation of the UPR pathway was evaluated in this cell line by monitoring the phosphorylation of eIF2α because this protein represents an appropriate surrogate for induction of ER stress through the pathway mediated by protein kinase RNA-like endoplasmic reticulum kinase (PERK, also known as EIF2AK3) in fatty-acid-treated β-cells (Dhayal and Morgan, 2011; Fig. 2C). Culture of BRIN-BD11 cells with palmitate resulted in increased phosphorylation of eIF2α, consistent with the induction of ER stress. Both OAG and OLPA markedly reduced the extent of eIF2α phosphorylation observed in cells cultured with palmitate (Fig. 2C).

Overall, these data validate our yeast-based screening assay as an efficient tool to identify new lipointoxication inhibitors with potent effects in mammalian cells.

Given that both OLPA and OAG bear an oleic acyl chain at the sn-1 position, a possible explanation for their suppressive effect could be that they directly act as Ole donors. To test this hypothesis, we used the property of Saccharomyces cerevisiae cells to grow on a variety of carbon sources, including fatty acids (Hiltunen et al., 2003). Therefore, hem1Δ cells were grown on a complete medium deprived of glucose (YP medium) and the ability of Ole, OLPA or OAG to support growth under these restrictive conditions was evaluated. As shown in Fig. 3A, neither OLPA nor OAG proved to be as potent in sustaining hem1Δ growth as Ole. However, in contrast to OAG, some residual growth could be observed with OLPA (Fig. 3A). Therefore, in terms of fatty acid supply, the three lipid species followed the order: Ole>>OLPA>OAG.

In order to identify potential common pathways for Ole, OLPA and OAG actions on SFA-induced cytotoxicity, we then performed a genomic analysis on hem1Δ cells by comparing their transcriptomic profiles under growth in minimum medium supplemented with Erg and Ole, OLPA or OAG and with Erg alone (SFA accumulation, denoted Ø). Differential expression analysis, conducted with a cut-off of two for the fold-change and of 0.0025 for the P-value, identified a total of 46 genes with changes in expression levels after Ole addition as compared to the SFA accumulation conditions (Fig. 3B; Table S2). Among this list, 14 genes were directly related to sterol and fatty acid or phospholipid metabolism (Table S2). OLPA addition resulted in the dysregulation of 17 genes, 13 of them being shared with those found with Ole. This overlap between Ole and OLPA suggests that the regulation of common metabolic pathway could be shared by the two molecules to account for resistance to SFA. A prime example was the induction of FAA1 (Table S2), which encodes a long-chain fatty acyl-CoA synthetase displaying a strong preference for C18:1 fatty acids (Knoll et al., 1994). By contrast, the addition of OAG to the culture was not associated with any changes in gene expression as compared to the SFA-accumulating condition (Fig. 3B).

Other elements in favor of different modes of action for OLPA and OAG came from lipidomic approaches. Fig. 3C shows the relative distribution of phosphatidylcholine species under the various culture conditions, obtained by comparison of positive ion mode spectra. Growth in YPD+Erg medium (SFA accumulation; Ø) resulted in decreased amounts of phosphatidylcholine species bearing long unsaturated chains (36:2, 36:1, 34:2, 34:1, 32:2, 32:1) in favor of species with saturated acyl chains (32:0, 30:0, 28:0, 26:0, 24:0), as compared to in normal growth conditions (δ-ALA) (Fig. 3C; Fig. S2A). Such rearrangements of the fatty acyl chain content were also observed in other phospholipid species (Fig. S3). Addition of Ole resulted in the enrichment in the diunsaturated (two double bonds) 36:2 species (dioleoylphosphatidylcholine, DOPC), which corresponds to a phosphatidylcholine bearing two Ole (18:1) chains (Fig. 3C,E; Fig. S2A). A similar C18:1 enrichment within phosphatidylcholine was also observed when the cells were incubated with OLPA, showing that this molecule can act as a potent oleic chain donor and acceptor (Fig. 3C; Fig. S2A). By contrast, OAG supplementation resulted in an increase of the monounsaturated species phosphatidylcholine 34:1 (palmitoyl oleoyl phosphatidylcholine, POPC; Fig. 3C,E; Figs S2A and S3). As a consequence, the saturation rate of phosphatidylcholine remained remarkably high under OAG treatment (50%; Fig. 3D). Given that a 40% saturation ratio is enough to preclude yeast cell growth (Pineau et al., 2008), we conclude from these experiments that, in contrast to OLPA, the effects of OAG on alleviating SFA lipointoxication cannot be accounted for by a simple restoration of the phospholipid content. However, the enrichment of 34:1 lipid species revealed that some of the OAG Ole chains are hydrolyzed and redistributed within phospholipids to form monounsaturated (one double bond) species (Fig. 3C–E; Figs S2A and S3). Therefore, in terms of C18:1 distribution, the following order was observed: Ole>OLPA>OAG, with Ole and OLPA preferentially forming species with two double bonds, and OAG phospholipids species with one double bond.

As mentioned above, the accumulation of SFA within phospholipids also alters the late secretory pathway: it precludes the delivery of a model plasma membrane transporter, the uracil permease Fur4p, to its final destination, in a process that correlates with a reduction of vesicular budding efficiency at the TGN (Payet et al., 2013). Saturated phospholipids appear to do this by increasing lipid order within intracellular membranes and therefore reducing surface voids which, in turn, alters the recruitment of Gcs1p, a void-sensing protein of the ArfGap family required for the cycling of nascent vesicles emerging from the trans-Golgi (Payet et al., 2013). Both OLPA and OAG efficiently restored Fur4p delivery to the plasma membrane (Fig. 4A) and Gcs1p recruitment to the TGN (Fig. 4B).

Gcs1p binds to bilayers thanks to its amphipathic-lipid-packing sensor (ALPS), which recognizes surface voids by peptide partitioning driven by the insertion of hydrophobic residues into large surface clefts that are formed in the bilayer (Vanni et al., 2013). Such voids can be obtained by introducing conical lipids within flat bilayers (Vamparys et al., 2013). In line with this hypothesis, we demonstrated that Gcs1p cannot bind to liposomes obtained with phosphatidylcholine species purified from hem1Δ cells grown under SFA accumulation, because of increased lipid packing related to high levels of saturated phospholipids (Payet et al., 2013). This process could therefore account for Gcs1p mislocalization in vivo (Payet et al., 2013; Fig. 4B) and is likely to participate in the observed reduction in vesicular budding from the TGN. Given that OAG failed to restore the minimal diunsaturated phospholipid ratio required for Gcs1p recruitment, we therefore evaluated whether this molecule could generate surface voids by itself when incorporated within a packed or highly ordered membrane. Relevant to this hypothesis is the finding that OAG could be detected in its free form in hem1Δ cells when supplemented to the medium at a molar percentage ratio of 1.6±0.4% OAG to total phospholipids and 5.7±0.6% OAG to phosphatidylcholine (Fig. S2B).

To address the effect of OAG on the activity of an ALPS-containing protein, we performed a GTPase-activating protein (GAP) assay using Gcs1p and liposomes (Fig. 4C). As previously shown (Antonny et al., 1997; Bigay et al., 2005), the GAP activity of Gcs1p was very low in the case of Golgi-mix liposomes containing mainly 16:0-C18:1 monounsaturated phospholipids (see Materials and Methods). This mix was chosen because 16:0-C18:1 phosphatidylcholine was the most prominent species under OAG supplementation (Fig. 3C–E; Figs S2A and S3). This result demonstrated that the formation of monounsaturated species related to partial OAG metabolization is not sufficient to account for the Gcs1p recruitment to the TGN observed in vivo. By contrast, the GAP activity of Gcs1p was increased by 15-fold by the addition of the conical lipid dioleoyl glycerol (DOG; Fig. 4C), which creates lipid packing defects or surface voids (Vamparys et al., 2013; Vanni et al., 2014, 2013). As shown in Fig. 4C, the addition of OAG also increased the activity of Gcs1p. A significant induction of Gcs1p activity could be observed in vitro with as low as 7 molar percent OAG, a molar ratio that falls in the same order of magnitude as the ratio measured in vivo (see above). However, a dose–response curve showed that the effect of a 33 molar percent of OAG was comparable to the effect of a 13 molar percent of DOG. Collectively, these results confirm that OAG can create surface voids, albeit less efficiently than DOG.

To test further the hypothesis of OAG acting as a potent ‘nanospacer’, molecular dynamics simulations were performed on bilayers composed either of a disaturated (DMPC) or a monounsaturated phosphatidylcholine species (POPC) in the presence or in the absence of OAG (Fig. 5). Such compositions were chosen because disaturated and monounsaturated phosphatidylcholine forms were the predominant species in yeast under OAG supplementation (Fig. 3E). For comparison, a pure DOPC bilayer, mimicking the situation encountered under Ole and OLPA supplementation conditions (Fig. 3E), was also simulated. Moreover, additional simulations were also performed with DOG to compare its impacts relative to OAG on POPC bilayers (Vamparys et al., 2013). To evaluate the size and probability of the surface voids under these different conditions, we used a recently developed method of membrane surface analysis (Vamparys et al., 2013). This method allows the identification of chemical defects, where hydrocarbon chains are accessible to the solvent, and geometrical defects (i.e. voids deeper than the glycerol backbone). The main results of these simulations are presented in Fig. 5. Panels A and B show side and top views of a snapshot for a 67% POPC, 33% OAG (molar percentage) bilayer.

The results in terms of surface voids are presented in Fig. 5C,D. These figures show that incorporating 10% OAG has a slight but substantial effect both on DMPC and POPC, for which we observed an increase of the size of surface voids, as monitored by the packing defect constant (pdc), of 0.7 and 0.5 Ų, respectively. Of importance, pdc increased in a synergistic manner when saturated acyl chains were replaced by monosaturated ones and when OAG or DOG were introduced into the bilayer; the pdc followed the order: DMPC<90% DMPC, 10% OAG<POPC<90% POPC, 10% OAG ≤67% POPC, 33% OAG<87% POPC, 13% DOG<DOPC (numbers represent molar percentages). These simulation results nicely fit our experimental data. First, the impact of OAG on DMPC and POPC bilayers can fully account for the efficiency of this molecule to improve Gcs1p recruitment to the TGN under conditions of SFA accumulation in vivo (Fig. 4B) and to monounsaturated liposomes in vitro (Fig. 4C). Second, the fact that POPC and OAG combinations fail to induce surface voids as efficiently as DOPC can also account for the lower capacity of this molecule to promote Gcs1p recruitment to the TGN, as compared to Ole and OLPA (Fig. 4B). Finally, these simulations also confirmed that OAG, with its single monounsaturated aliphatic tail, is less efficient than DOG for promoting surface voids, as observed in vitro (Fig. 4C).

In summary, these simulations suggest that OLPA and OAG can induce surface voids compatible with Gcs1p adsorption, albeit according to different mechanisms. Whereas OAG likely acts as a nanospacer, OLPA preferentially generates voids at the membrane surface by favoring the formation of DOPC species (Fig. 5E).

Given that OAG and OLPA share very similar structures, their common molecular features were used to generate a pharmacophore for the in silico screening of 6.5 million commercially available molecules (Fig. 6; see Materials and Methods for details). Using this virtual screen, 23 different compounds were identified and were individually assayed for their ability to counter SFA lipointoxication in the hem1Δ model, and for whether they were non-toxic towards the quadruple mutant strain. Among these 23 candidates, six proved to efficiently counter lipointoxication without significant impacts on the quadruple mutant growth.

Interestingly, this virtual screening revealed a strict requirement for the presence of an Ole chain within the molecular structure because all the non-Ole containing molecules which matched, at least partly, the pharmacophore, failed to counter SFA lipointoxication in our yeast model (Fig. 6; Table S3). Lipidomic analyses allowed separating the six positive hits into two different categories, that is, OLPA-like molecules, leading to the accumulation of DOPC, and OAG-like candidates, which failed to restore normal diunsaturated phospholipids amounts (see the example of CM22 on Fig. 6 and Fig. S3). Interestingly, the hits tested in this study proved to potently protect BRIN-BD11 cells against palmitate-induced cytotoxicity (Fig. S4).

CM22 appeared as a very interesting candidate because it behaved as a fully non-hydrolyzable analog of OAG (Fig. S3: compare CM22 and ø), but still counteracted all the traits associated to SFA toxicity. These observations reinforce the idea that restoring the phospholipid content is not an absolute prerequisite to counter the impact of SFA: as an alternative suppressive mechanism, OAG-related candidates likely act as nanospacers to restore the surface voids compatible with cell survival and growth restoration under SFA lipointoxication.

The two-step screening assay used in this study led us to identify, among 1470 molecules of various structures, two lipid compounds, namely OLPA and OAG, sharing very similar structural features. These two molecules appear to counter all the trademarks of SFA lipointoxication: they relieve the ER stress (Fig. 2) and restore normal protein trafficking in the late secretory pathway (Fig. 4).A very important conclusion of the present study is that both molecules restore surface voids within cellular membranes, albeit according to different mechanisms, as monitored by the recruitment of the void-sensing protein Gcs1p to the TGN in vivo (Fig. 4B) and to liposomes in vitro (Fig. 4C).

OLPA treatment correlates with the formation of the diunsaturated phospholipid species such as DOPC (Fig. 3C and Fig. S2). This is not a surprising observation because OLPA is a central molecule both in the remodeling and de novo synthesis of phospholipids, as a substrate for acyltransferases and transacylases (Yamashita et al., 2014). In this context, OLPA can be used both as an Ole donor and acceptor, therefore resulting in the formation of the DOPC that bears two oleic acyl chains.

Diunsaturated phosphatidylcholine such as DOPC, thanks to their two monounsaturated chains, generate more packing defects and surface voids in the bilayer than the cylindrical disaturated phosphatidylcholine DMPC (Vamparys et al., 2013; Fig. 5C,D). We have already demonstrated that Gcs1p can bind avidly to liposomes made of DOPC, but also to liposomes elaborated with phosphatidylcholine purified from hem1Δ cells grown in the presence of δ-ALA (Payet et al., 2013; 55% diunsaturated species, Fig. 3E). By contrast, Gcs1p recruitment was low to liposomes elaborated with DMPC or phosphatidylcholine species isolated from hem1Δ cells grown under SFA-accumulation (Payet et al., 2013; 5% diunsaturated species, Fig. 3E). These observations are very consistent with the mechanism for membrane recruitment of Gcs1p in yeast and its mammalian homologue ArfGAP1. These two proteins sense membrane curvature via their common ALPS motif (Bigay et al., 2005), which inserts hydrophobic residues into large surface voids/clefts into the bilayer (Vanni et al., 2014, 2013). The data reported in this study both in vivo (Fig. 4B), in vitro (Fig. 4C) and in silico (Fig. 5C,D) suggest that OLPA restores normal vesicular budding, and therefore plasma-membrane protein delivery to their final destination, by favoring the production of diunsaturated phospholipids species.

The hypersensitivity of Gcs1 and ArfGAP1 to lipid packing makes them acute reporters of changes in the packing properties of cellular membranes, but this does not imply that these changes affect only these proteins. In fact, any protein that uses hydrophobic membrane insertion should be sensitive to the level of unsaturated lipids. However, the degree of sensitivity depends on the balance between different modes of membrane interaction. For example, the amphipathic protein α-synuclein is not only sensitive to lipid packing but also to the presence of anionic lipids owing to its highly charged character (Pranke et al., 2011), which complicates the analysis. This is why we focused here on the ALPS protein Gcs1p.

One could consider that OLPA is also a bioactive lipid, as it is known to be involved in several physiological and physiopathological pathways. Relevant to lipointoxication, OLPA has been shown to activate the peroxisome proliferator activated receptor-γ (PPARγ) in mammalian cells (Stapleton et al., 2011). PPARγ is involved in adipocyte differentiation and fatty acid metabolism (Stapleton et al., 2011). However, such a system is not present in yeast, which lacks both nuclear hormone and LPA_x receptors (McIntyre et al., 2003). Therefore, alhough we cannot exclude that OLPA could help to counter SFA intoxication in mammalian cells through PPARγ induction, production of diunsaturated species is likely a prominent suppressive mechanism.

By contrast, OAG failed to induce the production of diunsaturated phosphatidylcholine species, which remained in remarkably low amounts (Fig. 3C,E). OAG is a permeable analog of DOG. Pioneering work in mammalian cells has suggested that, unlike DOG, OAG might directly intercalate into the phospholipid bilayer and regulate downstream pathways [i.e. activation of protein kinase C (PKC)] without interaction with specific cell receptors (Kaibuchi et al., 1983). Kaibuchi et al. have suggested that DOG, when exogenously supplied, is ineffective in inducing PKC activation, due to the steric hindrance related to its two long fatty acyl chains, which might preclude its intercalation within the phospholipid bilayer (Kaibuchi et al., 1983). These observations nicely fit our data. Indeed, DOG [supplied as part of the Bioactive lipid library and identified by the virtual screening (CM26, Fig. 6)] proved to be inefficient in countering SFA-related toxicity suggesting that the membrane-permeable properties of OAG are a prerequisite for its action. Once intercalated within the membrane, OAG behaves in a way very similar to DOG, by generating surface voids compatible with the recruitment of Gcs1p (Figs 4 and 5).

As in the case of OLPA, we cannot exclude that OAG might also exert its suppressive effects in mammalian cells by the regulation of downstream signaling pathways, notably the PKC pathway. However, given that Pkc1p, the main PKC yeast isoform, has been shown to be completely independent of DOG for its activation (Antonsson et al., 1994), such a mechanism could not explain the extreme efficiency of OAG in counteracting SFA toxicity in this model organism. This hypothesis was reinforced by the fact that the DOG and OAG homolog 12-O-tetradecanoylphorbol-13-acetate (TPA), a potent inducer of PKC (Niedel et al., 1983), which displays a saturated acyl chain instead of an Ole and does not match the pharmacophore developed in this study, failed to counter SFA lipointoxication in our cellular model (our unpublished data, R.F.-C and T.F.).

Several bulk properties of membranes, including their resistance to stretching (tension) and to bending (bending modulus), are important parameters for cell physiology and also depend on the level of lipid unsaturation (Rawicz et al., 2000). Therefore, the effects of C18:1-containing molecules might not exclusively occur through the formation of surface voids. However, it is interesting to note that polyunsaturated fatty acids, which are efficiently incorporated within phospholipids but poorly restore yeast growth and Fur4p trafficking to the plasma membrane (Deguil et al., 2011; Table S1), make lipid membranes very flexible but are less prone to create deep surface voids than monounsaturated lipids (Pinot et al., 2014; Rawicz et al., 2000). Therefore, the capacity to form surface voids seems to be at the root of the effects reported here.

To conclude, the present data point at modulation of membrane surface voids as a central parameter to consider in order to efficiently counter the impact of SFA on cell function in lipointoxication-related human diseases. To our knowledge, this study is the first evidence that targeting ubiquitous membrane biophysical properties with highly specific exogenously supplied molecules could be of therapeutic relevance.

Saccharomyces cerevisiae strains used in this study are listed in Table S4. The hem1Δ cells were grown aerobically at 28°C in YPD medium [1% yeast extract (w/v), 1% peptone (w/v) and 2% glucose (w/v)] supplemented with 80 µg ml⁻¹ δ-aminolevulinate (δ-ALA). Accumulation of endogenous saturated fatty acids (SFA) was induced as previously described (Pineau et al., 2009) by inoculating 2×10⁶ cells/ml, previously grown in YPD+δ-ALA, in YPD supplemented with 80 µg ml⁻¹ ergosterol (Erg; denoted Ø), or alternatively, by seeding 3500 cells on the equivalent solid medium containing 2% (w/v) agar. Compounds – such as oleic acid (Ole), 1-oleoyl-2-acetyl-sn-glycerol (OAG; Cayman, Bertin Pharma, Montigny le Bretonneux, France) or 1-oleoyl-lysophosphatidic acid (OLPA; Santa Cruz Biotechnology, Dallas, TX) – were supplemented, after cell transfer into or onto YPD+Erg (Ø) liquid or solid medium, to assay their anti-SFA properties.

For Fur4p biogenesis studies, cells were transformed with the plasmid pFl38-Fur4p-green fluorescent protein (GFP), bearing a FUR4 fusion gene, encoding Fur4p with a C-terminal GFP tag under the control of the inducible GAL10 promoter (Marchal et al., 2002). Localization of Gcs1p was conducted using a single-copy plasmid expressing Gcs1p tagged at the N-terminus with GFP (pRS315-GFP-GCS1; Robinson et al., 2006). Most effects of SFA accumulation were studied 7 h after transfer to YPD+Erg (Ø).

All products and reagents were purchased from Sigma-Aldrich, excepted where mentioned.

For the screening process, the Prestwick Chemical Library^® (Prestwick, IllKirch, France), which contains 1280 small molecules, 100% approved drugs and the Screen-Well^® Bioactive lipid library (Enzo^® Life Science, Farmingdale, USA), consisting of 190 lipid compounds of various structures, were used.

To evaluate the anti-SFA propensities of these molecules, hem1Δ cells were seeded on the surface of YPD+Erg (Ø) plates as described above, and 5 µl drops of 1, 10 or 100 mM stock solutions [diluted in dimethyl sulfoxide (DMSO) or ethanol (EtOH) solvents] were spotted prior to incubation at 28°C for 3 days. Colony formation at a specific spot reflects the ability of the corresponding compound to rescue SFA-blocked hem1Δ proliferation (see Deguil et al., 2011). To evaluate the effects of the candidate molecules in liquid cultures, the compounds were added at a 200 µM final concentration to YPD+Erg medium. Proliferation kinetics were followed by measuring the optical density of the culture at 600 nm (OD_600 nm) every single hour (1 OD_600 nm unit corresponding to 2×10⁷ cells/ml).

In parallel, the toxicity of the various compounds to a strain unable to store excess fatty acids within lipid droplets was also evaluated. To achieve this, WT and quadruple mutant strains were grown in YPD medium, under gentle shaking at 28°C and in aerobic conditions, prior to seeding 3500 respective cells/cm² onto YPD+agar 2% (w/v) plates. 1 µl drops of 1, 10 and 100 mM compound solutions (into EtOH or DMSO), were spotted onto agar prior to incubation at 28°C for 3 days. The diameters of growth inhibition halos were used as indicators of compound toxicity.

hem1Δ cells (inoculated at a 2×10⁷ cells/ml concentration) were grown in YPD+δ-ALA or YPD+Erg, supplemented or not with 200 µM of the candidate molecules, under gentle shaking for 7 h, at 28°C and in aerobic conditions. Quantification of the various lipid species was performed by the mean of internal standards (phosphatidylcholine, phosphatidylethanolamine, phosphatidylinositol, phosphatidylserine and phosphatidic acid 25:0, http://avantilipids.com/), which were added before extraction. Phospholipids were extracted and purified from 10⁸ cells as described previously (Payet et al., 2013). Purified phospholipids were evaporated and reconstituted in 100 µl 2-propanol, methanol and water (2:1:1, v:v:v) with 0.1% (v/v) formic acid for positive mode (phosphatidylcholine analysis) and in 100 µl 2-propanol, methanol and water (2:1:1, v:v:v) with 0.1% (v/v) triethylamine for negative mode (phosphatidylethanolamine, phosphatidylinositol, phosphatidylserine and phosphatidic acid analysis). The MS analysis was performed on a Waters Synapt G2 HDMS with a nanoESI source operated in negative and positive ionization modes. The scan range was from 150 to 1200 m/z. Data analysis was performed by using the LipidXplorer software (https://wiki.mpi-cbg.de/lipidx/Main_Page).

Because OAG is a relatively nonpolar compound, its quantification was assayed by atmospheric pressure chemical ionization (APCI)-LC-MS/MS rather than electrospray ionization (ESI)-MS. To this aim, dried lipid samples obtained as described above were dissolved with 250 µl of CHCl₃, CH₃OH and H₂O (60:30:4.5, v:v:v) and 2 µl were injected on a 1200 6460-QqQ LC-MS/MS system equipped with an APCI source (Agilent Technologies). Separation was achieved on a Poroshell C8 2.1×100 mm, 2.7 µm column (Agilent Technologies) and acquisition was performed in APCI-positive single ion monitoring (SIM) mode. A calibration curve was obtained using known quantities of an authentic standard (0.4 up to 2.5 nmol on a column). The amount of free OAG per 10⁸ hem1Δ cells was determined after 7 h of growth in YPD+Erg supplemented with 200 µM OAG and extensive washing of the cells with water. OAG amounts were finally expressed as the molar percentage ratio of OAG to total phospholipids and OAG to phosphatidylcholine.

hem1Δ cells transformed with the plasmid pPW344 (2 µ URA3 4×UPRE-LacZ; Cox and Walter, 1996) were grown into YPD+δ-ALA or YPD+Erg, supplemented or not with 200 µM of the compound of interest, under gentle shaking at 28°C and in aerobic conditions for 7 h. Then, 10⁸ cells were harvested in order to quantify the β-galactosidase (β-gal) activity resulting from the UPR-triggered LacZ transgene expression, as described previously (Pineau et al., 2009).

Palmitic acid stock was prepared in 50% (v/v) ethanol by heating to 70°C and then bound to fatty-acid-free bovine serum albumin (BSA) by incubation at 37°C for 1 h.

BRIN-BD11 cells were cultured in RPMI-1640 medium containing 11 mM glucose and supplemented with 10% (w/v) fetal bovine serum, 2 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. Cells were seeded at a density of 0.5×10⁵ cells/ml in six-well plates in complete medium for 24 h at the beginning of each experiment. Then, the medium was removed and replaced by a serum-free version containing appropriate reagents complexed to BSA. Relevant vehicle (BSA plus ethanol) controls were included in each experiment. Cell viability under the various treatments was determined by propidium iodide staining as described previously (Dhayal and Morgan, 2011).

To evaluate UPR induction, BRIN-BD11 cells were seeded at a density of 5×10⁵ cells/ml in T25 flasks in complete medium for 24 h. The medium was then removed and replaced by a serum-free version containing appropriate BSA-complexed compounds for 6 h. After treatment, whole-cell proteins were extracted and the ratio of phosphorylated eIF2α to total eIF2α was determined as described previously (Dhayal and Morgan, 2011).

RNAs were extracted from 2×10⁷ hem1Δ cells using a commercial kit (RNeasy Mini Kit, Qiagen). For each growth condition, samples were prepared from three individual biological samples. Agilent's Yeast oligonucleotide microarray contains 60-mer oligonucleotide probes to 6256 annotated ORFs for the S288C strain of Saccharomyces cerevisiae. The ORF list was downloaded from the Saccharomyces Genome Database (http://www.yeastgenome.org/) at Stanford University. For statistical confidences and determination of differential gene expression, data files were analyzed using the free software R associated to packages of the ‘Bioconductor’ project. Developed workflows begin by data normalization using the robust multichip average (RMA) method (Irizarry et al., 2003) followed by a data analysis workflow consisting of selecting differentially expressed genes. The statistical algorithm significance analysis of microarrays (SAM) (Tusher et al., 2001) was used to compute the P-value and fold change for each gene. Selected genes were obtained using a cut-off both for fold change (of 2) and P-value (of 0.0025).

Samples were examined by confocal laser scanning microscopy using a Bio-Rad MRC 1024 equipped with a 15-mW argon-krypton gas laser. BodiPY and GFP fusion proteins were visualized by excitation at 488 nm and with a 522-nm band-pass filter. Sec7p–RFP was visualized by excitation at 568 nm and with a 605-nm band-pass filter. Quantification of the amount of Fur4p–GFP at the plasma membrane was performed according to the procedure described by Nalaskowski et al. (2011). For each experiment, a minimum of 60 cells were randomly selected and examined. To determine the plasma membrane to intracellular ratio, GFP fluorescence intensities both over the plasma membrane and in intracellular areas, in each cell, were averaged to calculate the ratio of plasma membrane over intracellular intensity. For the GFP–Gcs1p [expressed from a single-copy plasmid encoding the Gcs1p protein tagged at the N-terminus with GFP (pRS315-GFP-GCS1; Robinson et al., 2006)] and Sec7p–RFP colocalization assay, a four-step analysis was carried out. First, from the microscopy images, only GFP-expressing cells were selected and retained. Then, among these selected cells, RFP fluorescence spots were defined. Finally, the GFP fluorescence intensities were quantified in these specific areas. To determine the co-localization scores, the RFP-dot-specific GFP intensity and the global GFP intensity were averaged. For evaluation of data, an unpaired Student's t-test was used using the Graphpad Prism 5 software. Mean±s.e.m. values are given.

Liposome preparation and a GAP assay were carried out as previously described (Bigay et al., 2005) with some modifications. Briefly, all lipids in chloroform were purchased from Avanti Polar Lipid. The standard composition of Golgi-mix liposomes was (molar percentage given in brackets) 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) (50), 1-palmitoyl-2-oleoylphosphatidylethanolamine (POPE) (19), 1-palmitoyl-2-oleoylphosphatidylserine (POPS) (5), liver phosphatidylinositol (10) and cholesterol (16). When indicated, OAG (7, 13, 23 or 33 molar percent) or 1,2-dioleoylglycerol (DOG) (13 molar percent) were further added to the Golgi mix lipids in chloroform. Tryptophan fluorescence at 37°C was measured at 340 nm (bandwidth 20 nm) upon excitation at 298 nm (bandwidth 1 nm) using a JASCO FP-8300 fluorimeter. The fluorescence quartz cuvette initially contained the liposomes (200 µM) in HKM buffer (HK+1 mM MgCl₂, 1 mM DTT). Thereafter, Arf1-GDP (0.4 µM) at 60 s, GTP (50 µM) at 120 s, EDTA (2 mM) at 180 s, MgCl₂ (2 mM) at 780 s and Gcs1p (50 nM) at 840 s were added sequentially and the fluorescence was monitored in real time. The fluorescence intensities at 840 s were normalized to 100. The apparent half-time (t_1/2) of Arf deactivation was determined graphically as previously described (Antonny et al., 1997).

Molecular dynamics (MD) were performed to understand the effect of adding OAG in terms of surface voids (i.e. what we call ‘surface voids’ are polar head packing defects that promote the adhesion of peripheral proteins). Previous 200-ns simulations of pure DMPC, pure POPC and pure DOPC (Vamparys et al., 2013) were extended to 400 ns in this work. Additional simulations of 90% DMPC and 10% OAG, 90% POPC and 10% OAG, 67% POPC and 33% OAG, and 87% POPC and 13% DOG (all in molar ratio) were performed. Each lipid system consisted of a patch of 256 lipids hydrated with 40 to 43 water molecules per lipid (according to the different lipid compositions).

All molecular dynamics simulations were performed with GROMACS 4.5.3 (Hess et al., 2008). The Berger force field (Berger et al., 1997) was used for DMPC, POPC and DOPC, and we used our previous model for DOG (González-Rubio et al., 2011; Vamparys et al., 2013; Vanni et al., 2013). The topology of OAG was copied from DOG. The systems were simulated for 400 ns and the analysis was performed on the last 300 ns. Each simulation was analyzed in terms of packing defects as previously described (Pinot et al., 2014; Vamparys et al., 2013; Vanni et al., 2014, 2013). Briefly, we used a grid (with a granularity of 1 Å) parallel to the plane of the bilayer. We scanned each grid point vertically starting from outside of the bilayer towards its interior (water is excluded from the analysis) up to 1 Å below the sn-2 carbon atom of the nearest glycerol. If only lipid aliphatic atoms are encountered the grid point is considered as an elementary packing defect of 1 Ų; if any lipid polar atom is encountered that grid point is not considered as a defect. Adjacent elementary defects are then clustered using a connected component algorithm, and the area of each cluster is calculated (a packing defect is thus a cluster of elementary defects). This analysis is performed for each leaflet separately. The obtained distributions are then fit to a mono-exponential decay: p(A)=b e^−A/d, where p(A) is the probability of finding a defect of area A Ų, d is the exponential decay in units of Å⁻², and b a constant. We then define the packing defect constant pdc in units of Ų as the inverse of the decay (pdc=d⁻¹). The higher this constant, the higher the probability of finding large defects. The error bar reported for each pdc corresponds to the standard error of regression coefficient estimated during the fitting procedure.

The present study was carried out using LigandScout software (version 4.03). LigandScout is a software tool that extracts and interprets ligands and their macromolecular environment to create 3D pharmacophore models (Wolber and Langer, 2005).

Starting from two well-known molecules (1-oleoyl-2-acetyl-sn-glycerol, OAG, and 1-oleoyl-lysophosphatidic acid, OLPA), the ligand-based modeling method was used. A maximum of 200 conformations were generated for each molecule with OMEGA-best settings. Then, a set of ten pharmacophores was generated, using the shared feature pharmacophore method. After pharmacophore optimization, the pharmacophores were validated. A database of 80 fatty acid derivatives containing 14 tested active and 66 inactive compounds on the desired target (Deguil et al., 2011) was screened against these pharmacophores, without taking into account the exclusion volumes. The quality of the pharmacophore models was quantitatively evaluated by calculating the selectivity, specificity and accuracy for each model separately (Braga and Andrade, 2013). Virtual screening was performed on the selected pharmacophore using LigandScout on different databases that represent around 6.5 million commercially available compounds coming from the Zinc database Drug-now subset and various commercial suppliers.

Except where mentioned, P values were calculated by a two-tailed t-test using the Graphpad Prism 5 software. Means±s.e.m. at least representative of triplicates are indicated with n.s., non-significant, ***P<0.001, **P<0.01 and *P<0.05.

We are very grateful to Anne Cantereau (Confocal and Electron Microscopy facilities, University of Poitiers, Poitiers, France) for her precious advice regarding confocal microscopy imaging, and Lidwine Trouilh (Biochips Platform, LISBP, Toulouse, France) for performing microarray preparations and analyses. We are also grateful to Jean-Paul Pais de Barros (Plateforme de Lipidomique, INSERM UMR866, Dijon, France) for his help concerning OAG quantification and to Patrick Bazzini (Prestwick Chemical, Illkirch, France) for the virtual screening reported in this study.