Physiological and pathological roles of FATP-mediated lipid droplets in Drosophila and mice retina

Increasing evidence suggests that dysregulation of lipid metabolism is associated with neurodegeneration in retinal diseases such as age-related macular degeneration and in brain disorders such as Alzheimer’s and Parkinson’s diseases. Lipid storage organelles (lipid droplets, LDs), accumulate in many cell types in response to stress, and it is now clear that LDs function not only as lipid stores but also as dynamic regulators of the stress response. However, whether these LDs are always protective or can also be deleterious to the cell is unknown. Here, we investigated the consequences of LD accumulation on retinal cell homeostasis under physiological and stress conditions in Drosophila and in mice. In wild-type Drosophila, we show that dFatp is required and sufficient for expansion of LD size in retinal pigment cells (RPCs) and that LDs in RPCs are required for photoreceptor survival during aging. Similarly, in mice, LD accumulation induced by RPC-specific expression of human FATP1 was non-toxic and promoted mitochondrial energy metabolism in RPCs and non-autonomously in photoreceptor cells. In contrast, the inhibition of LD accumulation by dFatp knockdown suppressed neurodegeneration in Aats-metFB Drosophila mutants, which carry elevated levels of reactive oxygen species (ROS). This suggests that abnormal turnover of LD may be toxic for photoreceptors cells of the retina under oxidative stress. Collectively, these findings indicate that FATP-mediated LD formation in RPCs promotes RPC and neuronal homeostasis under physiological conditions but could be deleterious for the photoreceptors under pathological conditions.

cycle. An inability of RPCs to perform this role leads to photoreceptor death and is associated with diseases such as age-related macular degeneration (AMD), the most prevalent cause of blindness in developed countries [1].
Injury to RPCs, whether through disease or as part of the normal aging process, is accompanied by an accumulation of lipid deposits, named drusen, in the RPC itself or in the adjacent Bruch's membrane [2][3][4]. A common cause of injury is reduced blood flow to the retina, which causes hypoxia and a subsequent metabolic shift that results in accumulation in RPCs of intracellular lipid droplets (LDs) composed of neutral lipids, such as triacylglycerides (TAGs) and sterol esters. In AMD, lipids are also deposited extracellularly, further compromising the functions of RPCs and photoreceptors [5]. However, it is not clear whether lipid accumulation is a cause or a consequence of AMD pathology.
Work in Drosophila models has contributed greatly to our understanding of the mechanisms of LD biosynthesis and physiological functions [6,7]. For example, LDs have been shown to play an antioxidant role in the developing Drosophila nervous system by protecting polyunsaturated fatty acids from peroxidation under conditions of hypoxia [8]. In contrast, in Drosophila carrying mutations with mitochondrial dysfunction, oxidative stress induces glial cells to accumulate LDs that are toxic to the neighboring photoreceptor cells [9]. Thus, LDs are not simply cytoplasmic lipid storage organelles with critical roles in energy metabolism; they also have dynamic functions in regulating the response to stress of many cell types, including those in the nervous system [10][11][12].
LDs are synthesized on the endoplasmic reticulum membrane by a protein complex composed of diacylglycerol acyltransferase (DGAT-2) and fatty acid transport protein (FATP) [12,13]. The FATP family (also known as solute carrier family 27 [SLC27]) [14,15] are acyl-CoA synthetases involved in the cellular import of FAs [16] as well as other processes. FATPs interact physically and functionally with DGAT-2, an enzyme that catalyzes the conjugation of a fatty acyl-CoA to diacyl glycerol to produce TAGs, which are then incorporated into expanding LDs in C. elegans [13]. An important role for FATP in TAG storage has also been demonstrated in mammals. In Fatp1 knockout mice, the uptake of FA and the size of LDs in the brown adipose tissue are reduced compared with wild-type (WT) mice, resulting in defects in non-shivering thermogenesis [17]. The FATP1 gene is conserved from yeast to humans, where there are six proteins with different tissue distributions and substrate preferences. Drosophila Fatp (dFatp) shows a high degree of sequence similarity to human (h)FATP1 and hFATP4, which have broad expression profiles that include the brain and retina [16,18,19].
The Drosophila retina is composed of approximately 800 ommatidia, each of which comprises eight photoreceptors surrounded by a hexagonal lattice of nine RPCs (six secondary and three tertiary pigment cells, S1 Fig) [20]. Drosophila RPCs (dRPCs) differentiate during the pupal stage and superfluous cells are eliminated by apoptosis [21,22]. The survival of dRPCs requires cues that are provided by cell-to-cell communication from neighboring cone and primary pigment cells during pupal development [23,24]. In adult flies, dRPCs have similar functions to mammalian RPCs; that is, they are also juxtaposed to photoreceptor cells and supply metabolites for energy metabolism and other functions [25,26,27].
In Drosophila, dFatp is expressed in both photoreceptors and dRPCs [28], but its physiological role in LD formation in dRPCs and its impact on photoreceptor homeostasis remains to be investigated. Here, we examined the role of FATP in LDs formation and in cellular retinal homeostasis in Drosophila and in mice. We find that, although the architecture of the mammalian and insect retina are different, RPCs in both organisms play conserved roles in lipid and photoreceptor homeostasis. We show that FATP plays a crucial role in regulating LD production in RPCs and, in turn, RPC-derived LDs affect differentially photoreceptor viability in physiological and pathological conditions.

dFatp is necessary and sufficient for LD expansion in Drosophila retina under physiological and pathological conditions
To investigate the role of dFatp in LD formation and expansion in WT flies and in flies overexpressing dFatp (using the pan-retinal GMR-Gal4 driver), we first measured the uptake of BOD-IPY 500/510 -C 12 , a fluorescent long-chain FA analog. We observed an accumulation of LDs (visible as green foci) in the retina of 1-day old WT flies and this accumulation was greatly increased when dFatp was overexpressed (Fig 1Aa, 1Ab and 1C). Importantly, expression of brummer, a gene encoding a TAG lipase (referred to here as Bmm-lipase) [29], reduced BOD-IPY 500/510 -C 12 accumulation further demonstrating the lipid nature of the detected droplets in both WT and dFatp-overexpression conditions (Fig 1Ac, 1Ad and 1C). To overexpress dFatp, we also used the 54C-Gal4 driver, which was previously used as RPCs specific driver in several publications [9,24,27]. dFatp overexpression with the 54C driver induced an important accumulation of LDs similar to the GMR driver (Fig 1Ae, 1Af and 1C). We verified the expression pattern of the 54C driver (54C-Gal4, UAS-GFP) and confirmed that 54C driver is strongly expressed in dRPCs (primary, secondary and tertiary pigment cells), although low levels of GFP expression were also observed in one or two photoreceptors per ommatidium, rarely in cone cells and not in bristle cells at 42h after puparium formation (APF) (S2A and S2B Fig). The developmental expression of the 54C driver suggests that overexpression of dFatp during pupal development could account for the observed accumulation of LDs in adult fly retina. To confirm that the accumulation of LDs induced by 54C-dependent overexpression of dFatp is truly RPC specific, we concomitantly expressed Gal80 the inhibitor of Gal4 with Repo, a panglial driver that is expressed in RPCs [30]. Repo-Gal80 completely abolished the accumulation of BODIPY 493/503 induced by the overexpression of dFatp with the 54C driver (54C-Gal4, UAS-dFatp) confirming that the induction of LD by 54C is RPC-specific (S2C and S2D Fig). In this experiment, we used the lipophilic probe BODIPY 493/503 on fixed tissue instead of the BOD-IPY 500/510 -C 12 on live tissue. Because we see increased dFatp-induced BODIPY staining in both cases (compare Fig 1Ae and 1Af with S2Ca and S2Cb Fig), this shows that dFatp overexpression mediates both the active uptake of BODIPY 500/510 -C 12 into LD ( Fig 1A) and the increase total LD stores revealed with BODIPY 493/503 . Similar to the GMR driver, the accumulation of BODIPY 500/510 -C 12 , induced by overexpression of dFatp with the 54C driver was suppressed by Bmm-lipase concomitant overexpression (Fig 1Ag, 1Ah and 1C). We finally examined if the accumulation of LD due to dFatp expression requires Vitamin A, a precursor of retinal that is important for dFatp function in the regulation of Rh1 in photoreceptors [28]. As previously shown, we observed that the Vitamin A-deficient diet rescued photoreceptor degeneration in dFatp whole eye mutant clone (  (A) LD labeled with BODIPY 500/510 -C 12 (green) were revealed using confocal microscopy in horizontal sections of whole mount retinas from one-day-old flies expressing UAS-LacZ (control), UAS-dFatp and/or UAS-Bmm-lipase under the control of a pan-retinal (GMR-Gal4) (a-d) or a dRPC-specific (54C-Gal4) (e-h) driver. Photoreceptors were counterstained with phalloidinrhodamine (red). (B) BODIPY 500/510 -C 12 (green) uptake in horizontal sections of whole eye wild-type (FRT82B-wild-type) and mutant (FRT82B-Aatsmet FB ) clone generated with the GMR-hid/FLP-FRT technique [49] from one-day-old flies in the absence (a, b) or presence of dRPC-specific (driven by 54C-Gal4) UAS-dFatp-RNAi [GD16442] (c, d). Photoreceptors were counterstained with phalloidin-rhodamine (red). Scale bar, 25 μm. (C) Quantification of BODIPY 500/510 -C 12 uptake into lipid droplets from the images shown in (A) and (B). Data are presented as the fold change in fluorescence intensity (dots/μm 2 ) compared with the FRT82B-wild-type flies. The boxes represent the median and lower and upper quartiles, and the whiskers represent the 1.5 interquartile range. N = 6-41 retinas per condition. Blue and red statistical stars indicate significant differences between wild type and retina overexpressing dFatp or Bmm with GMR-Gal4 or 54C-Gal4, respectively. We then examined the requirement of dFatp in LD accumulation in retina of WT and Aatsmet FB mutant flies, which accumulate LD due to high mitochondrial ROS levels. As previously observed [9], BODIPY 500/510 -C 12 uptake into LDs was increased in the Aats-met FB mutant compared with the WT retina (Fig 1Ba, 1Bb and 1C). Expression of dFatp-specific dsRNA (RNAi) with the 54C-Gal4 driver abolished BODIPY 500/510 -C 12 uptake in the WT flies and strongly reduced it in the Aats-met FB mutants (Fig 1Bc, 1Bd and 1C). Furthermore, BOD-IPY 500/510 -C 12 staining was completely abolished in dFatp whole eye mutant clones, confirming the specificity of dFatp dsRNA and the role of dFatp in LD formation (S3D Fig). Collectively, these data demonstrate that dFatp is necessary for LD accumulation in the Drosophila retina under physiological and pathological conditions.
To examine the subcellular localization of LDs in the retina, we performed transmission electron microscopy (TEM) of flies overexpressing dFatp using GMR or 54C drivers. LDs were typically visible as homogeneous structures surrounded by a monolayer membrane in dRPCs and, to a much lesser extent, in photoreceptor cells (Fig 2A and 2B). Interestingly, dFatp overexpression significantly increased the size, but not the number, of LDs in both cell types (Fig  2A-2C). Finally, the expression Bmm-lipase with 54C-Gal4 caused a striking reduction in the number of LDs, not only in dRPCs but also in photoreceptors ( Fig 2B and S4 Fig). This indicates that the monolayer membrane-encapsulated vesicles observed by TEM are indeed LDs.
The TEM images also indicated the expansion of LDs induced by dFatp overexpression concomitantly caused a significant decrease in the size of mitochondria in the photoreceptors ( Fig 2D). We confirm that this reduction of mitochondrial size was correlated with LDs accumulation in a dFatp-dependent manner since co-expression of dFatp and Bmm-lipase prevented the decrease in photoreceptor mitochondrial size compared with expression of dFatp alone ( Fig 2D). Notably, despite the reduction in size, the morphological integrity of mitochondria was preserved, suggesting that their function may remain intact and that accumulation of LD is not toxic to photoreceptors. Collectively, these results show modulating LD levels by increasing LD biogenesis (54C>dFatp) or LD lipolysis (54C>Bmm) in dRPCs affects LD and mitochondrial sizes in photoreceptor cells.

Evidence of vesicle transfer from RPCs to photoreceptors
TEM images of retina showed various stages suggestive of the transfer of vesicles from dRPCs to photoreceptors in both wild-type and retina overexpressing dFatp with the 54C driver. Vesicles can be seen within the dRPCs (i), in the process of being internalized into the photoreceptor by a process resembling endocytosis (ii), and fully internalized into the photoreceptor, where they appear surrounded by a double membrane (iii) (Fig 3). These vesicles were comparable in size to LDs (~0.25 μm 2 ) but appeared slightly more electron dense. Due to their double membrane and localization at the proximity of the plasma membrane, these vesicles were distinguished from LD present in the cytoplasm of photoreceptors. These vesicles were seen juxtaposed to one or more mitochondria, which suggests that they contribute to or benefit from mitochondrial energetic function (Fig 3A and 3A'). These results suggest that the non-autonomous transfer of material contributes to the relay of information between dRPCs and photoreceptors.

Expression of hFATP1 in mRPC increases lipid storage and energy metabolism in the mouse retina
To determine whether the function of dFatp is conserved, we first asked whether the loss of photoreceptor cells observed in dFatp -/mutant retinas [28], could be rescued by the expression of the human homolog hFATP1 in Drosophila. Indeed, photoreceptor-specific expression To investigate the role of hFATP1 in lipid storage in the mammalian retina, we employed our previously described transgenic mice, in which hFATP1 is overexpressed using the mammalian RPC-specific VDM2 promoter (referred to as hFATP1TG mice) [31]. LDs in the retinas of WT (C57BL/6J) and hFATP1TG (Fig 4A and 4B) mice were detected by staining of neutral lipids with Nile Red. Notably, transgenic expression of hFATP1 significantly increased neutral lipid staining, visible as red foci, in the mRPCs of hFATP1TG mice compared with WT mice (Fig 4A-4C). Moreover, lipid profiling of dissected mRPCs by gas-chromatography revealed that levels of sterol esters and TAGs, both of which are components of LDs, were higher in the hFATP1 mRPCs compared with WT mRPCs (Fig 4D). We also observed that Nile Red staining largely co-localized with perilipin, a represent the median and lower and upper quartiles, and the whiskers represent the 1.5 interquartile range. N = 4 to 5 flies, from which we analyzed >150 fields of view. Log adjusted values: Ã p<0.05, ÃÃÃ p<0.001 by Tukey's HSD paired sample comparison tests.
https://doi.org/10.1371/journal.pgen.1007627.g002   (Fig 4E-4E"), supporting the notion that the Nile Redstained foci were indeed neutral lipid-containing LDs. The presence of LDs was also observed in mRPCs of hFATP1TG mice by TEM (Fig 4F and 4G). Taken together, these data indicate that hFATP1 expression in mouse RPCs leads to accumulation of LDs, consistent with our findings in Drosophila RPCs.
Immunofluorescence staining for the mitochondrial marker ATP synthase revealed that LDs and mitochondria were in close proximity in the RPCs of both WT and hFATP1TG mice ( Fig 4H-4I'), suggesting the possibility that LDs serve as an energy source in mRPCs. To test this hypothesis, we measured the mitochondrial respiratory rate in permeabilized mRPCs by monitoring O 2 consumption after activation of different pathways (β-oxidation and Cx I + II). We found that hFATP1 increased all respirations tested after addition of ADP to stimulate ATP production (EIIImo β-ox and EIIImpgso CxI+CxII) ( Fig 4J). Consistent with this, the activities of the TCA cycle enzymes isocitrate dehydrogenase, 2-oxoglurarate dehydrogenase (OGDH), and fumarase were all significantly increased in retinal extracts from hFATP1TG mice compared with WT mice (Fig 4L-4N). This increased respiration rate in transgenic mRPCs is unlikely to be due to an increase in mitochondrial mass because the activity of citrate synthase was unaffected by hFATP1 overexpression (Fig 4K). Taken together, these data demonstrate that hFATP1 overexpression in mRPCs increases the size of LDs and suggests that they are substrates for energy production in the mitochondria.
Next, we asked whether hFATP1 expression and enhanced lipid storage in mRPCs could non-autonomously affect the physiology of juxtaposed neural retinal cells, which includes photoreceptor cells. mRPCs and the neural retina can be dissociated during dissection, which enables the effects of mRPC-specific hFATP1 expression on the isolated photoreceptor cells to be examined. Lipid profiling of the neural retina revealed that sterol ester and TAG levels were higher while phospholipids levels remained unchanged in the neural retina of transgenic mice compared with WT mice (Fig 5A and 5B). Moreover, the magnitude of the increase was similar to that seen in isolated mRPCs (compare Fig 5A with Fig 4D). These results suggest a nonautonomous effect of mRPC-specific hFATP1 expression similar to the observed phenotype in Drosophila. Next, analyses of the mitochondrial respiratory rate and TCA cycle enzyme activities in neural retinas showed a similar enhancement of O 2 consumption ( Fig 5C) and enzyme activities (Fig 5D-5G) in tissue isolated from hFATP1TG mice compared with WT mice. Taken together, these results indicate that hFATP1 expressed in mRPCs not only increased lipid storage and mitochondrial respiration in the mRPCs themselves, but also in the neural retinal cells.

Genetic modifications of LD accumulation perturb retinal homeostasis
We next examined the consequences of LD accumulation caused by overexpression of FATP for the viability of RPCs in mice and Drosophila. TEM analyses showed that mRPC-specific expression of hFATP1 induced vacuolization of the mRPCs and thickening of both mRPC and Bruch's membranes in 3-month-old mice (Fig 6A-6D). Similarly, pan-retinal or dRPC-specific expression of dFatp caused a moderate loss of dRPCs (one to two missing dRPCs in most ommatidia), resulting in perturbation of the ommatidia organization in Drosophila retina ( Fig  6E and 6F). In contrast to dRPCs, photoreceptor cell survival (visualized by expression of Rh1-GFP) was unaffected by dFatp overexpression or by the knockdown of the Bmm-lipase which induced the accumulation BODIPY 493/503 similar to dFatp overexpression (Fig 6G and  6H and S6 Fig). These results in Drosophila retina are in agreement with our previous findings in mouse retina showing that photoreceptor layers were unaffected when hFATP1 was overexpressed in mRPCs [31]. Therefore, FATP-mediated LD accumulation in RPCs does not appear to be deleterious to photoreceptor cells under physiological conditions.
In contrast, knockdown of dFatp with the 54C driver, which is associated with a diminution of LD content, resulted in late onset photoreceptor degeneration in the Drosophila retina   6Gc, 6Gd and 6H). Importantly, this photoreceptor degeneration induced by expression of dFatp-RNAi with the 54C driver was suppressed by Gal80 expression in dRPCs with the Repo driver ( Fig 6J). In contrast, blocking the expression of dFatp-RNAi specifically in photoreceptors (elav-Gal80) did not inhibit photoreceptor degeneration (Fig 6J). This supports the fact that dFatp-dependent accumulation of LD in dRPC is non-autonomously required for photoreceptor viability. In agreement with the idea that LD are required for photoreceptor viability, we also observed a late onset photoreceptor degeneration similar to dFatp knockdown, by forcing LD lipolysis by overexpressing the Bmm-lipase or inhibiting LD biogenesis by expressing a RNAi against midway (mdy), a diacylglycerol acyltransferase (DGAT) that functions in the formation of TAG (Fig 6I and S6 Fig).
Next, we examined the effect of Vitamin A-deficient diet on photoreceptor degeneration induced by 54C>dFatp-RNAi. As previously shown [28], the Vitamin A deficient diet strongly suppressed photoreceptor degeneration in dFatp mutant (S3A Fig). In contrast, we only observed a modest rescue of photoreceptor degeneration in flies expressing dFatp-RNAi in dRPCs with this diet supporting the idea that a Vitamin A independent dRPC role of dFatp in photoreceptor degeneration is predominant (Fig 6K). Collectively, these results indicate that dFatp-dependent LDs support photoreceptors viability during the normal aging process.
Finally, we asked whether LD accumulation in dRPCs was beneficial or deleterious to photoreceptors under stress conditions. For this, we examined young (5-day-old) Aats-met FB mutant flies, in which dFatp RNAi prevents LD accumulation (see Fig 1B and 1C). Importantly, dFatp RNAi alone does not induce photoreceptor degeneration in 5-day-old flies [28], but the mutation Aats-met FB itself causes some degeneration, allowing us to examine the consequence of dFatp RNAi in Aats-met FB mutant on photoreceptor survival. We analyzed the apical/distal position of photoreceptor nuclei, which are normally clustered apically in the WT retina but are mislocalized under conditions associated with photoreceptor death [28]. Monitoring of mislocalized nuclei is a useful readout of cell death in situations where photoreceptor rhabdomeres are difficult to count due to their rapid degradation, as is the case for Aats-met FB mutants. We observed an increase in the number of nuclei mislocalized between the proximal and distal part of the retina in Aats-met FB mutants compared with WT flies (Fig 7Aa, 7Ac and 7B), consistent with previous reports for this mutant [9]. However, the mislocalization was completely reversed by dRPC-specific dFatp RNAi (Fig 7Ad and 7B). These results indicate that FATP-mediated LD accumulation in dRPCs is harmful to photoreceptors under conditions of pathological oxidative stress (Aats-met FB mutants), in stark contrast to the beneficial effect under physiological conditions.

Discussion
Here, we studied the consequences of dysregulated lipid storage in RPCs on photoreceptor structure and function in both Drosophila and mouse retinas. Our results show that the FATP dependent-mechanisms of lipid storage and communication between RPC and photoreceptor layers are largely conserved in flies and mice. We made several novel observations in this study. We showed that FATP expression is required for LD formation in mouse and Drosophila RPCs. Interestingly, FATP has an acyl-CoA synthetase activity that is thought to facilitate cellular uptake of FAs (reviewed in [16]). FATP also interacts with DGAT-2 on the endoplasmic reticulum, where it has been show to function in LD expansion in C. elegans [13]. The fact that dFatp overexpression increases LD size but not number also suggests that the role of FATP in LD expansion is conserved in Drosophila. Interestingly, retinal LD accumulation induced by FATP overexpression resembles one of the key hallmarks of AMD pathology; that is, the accumulation of lipids in RPCs and of drusen in Bruch's membrane [32]. We also found that mRPC-specific hFATP1 expression increased non-autonomously neutral lipids, β-oxidation, TCA enzyme activity, and mitochondrial respiration in photoreceptors. The fact that overexpression of FATP promotes both catabolic (mitochondrial respiration, TCA cycle, βoxidation) and anabolic (increase of TAG and SE) functions that are distinct processes could be interpreted as followed. First, FATP promotes the import of fatty acids, which are used for different metabolic function including catabolism and anabolism within the cell. And second, FATP carries an acyl-coA synthase activity that is central in lipid metabolism for lipid elongation, unsaturation but also β-oxidation [16]. In flies, we did not evaluate the mitochondrial respiration rate due to the technical difficulty of dissociating RPCs from photoreceptor cells. However, mitochondria were intact by TEM and frequently observed juxtaposed to LDs in both flies and mice. The proximity between LD and mitochondria suggests that release of TAGs from LDs mediated by lipases could provide an energy source for the mitochondria. This hypothesis is supported by studies showing that physical proximity between LDs and mitochondria is important for β-oxidation of FAs [33]. In flies, we used the 54C driver, which is strongly expressed in dRPCs to examine the nonautonomous role of dFatp on adjacent photoreceptor as previously described [27]. Because the 54C driver also showed a low level of expression in few photoreceptors during pupal development, we took advantage of the Gal80 system to show that photoreceptor degeneration observed in flies carrying 54C>dFatp-RNAi is due to the expression of dFatp-RNAi in dRPCs but not in photoreceptors. These results support a non-autonomous role of dFatp in fly RPCs as it is observed in the mouse. Altogether, we have shown that both the function of FATP in RPCs and its effect on photoreceptor cells are largely conserved between flies and mammals.
Although our data demonstrate that FATP-mediated LD accumulation in RPCs provides a metabolic signal to photoreceptors, it is not clear whether the signal itself is a lipid or another signaling molecule. Recent work proposed that in mutants that carry dysfunctional mitochondria, the accumulation of LDs in RPCs requires transfer of lactate from RPCs to photoreceptors [27]. The increased lactate levels in photoreceptors enhanced the activity of the TCA cycle and synthesis of FAs, which then underwent apolipoprotein-mediated transfer back to RPCs and induced LD formation [27]. A similar mechanism was described in mouse neuronal/glial co-cultures, raising the possibility that it may also be operating in the retina of hFATP1TG mice [27]. Our ultrastructural analysis supports the possibility of an exchange of electron dense material between RPCs and photoreceptor cells in Drosophila. Indeed, electron dense vesicles-similar in appearance and size to LDs-were observed in contact with dRPCs, as invaginations in the photoreceptor membrane, and finally as double-membraned structures in close proximity to the mitochondria in photoreceptors. Associations between LDs and mitochondria have also been observed in skeletal and heart muscle, where it was proposed to facilitate β-oxidation [34,35,33]. Although further studies will be required, our ultrastructural findings provide a hint that vesicle transfer could be a route of communication between RPCs and photoreceptors. Collectively, our results demonstrate that crosstalk between RPCs and photoreceptors is crucial for normal photoreceptor homeostasis and that its breakdown may contribute to retinal pathologies.
In our previous work, we showed that dFatp is expressed at higher levels in dRPCs than in photoreceptors, and that loss of dFatp led to a cell-autonomous photoreceptor degeneration [28]. We had proposed that degeneration of photoreceptors in the dFatp mutant is due to a failure to degrade Rh1, which accumulates to toxic levels and triggers apoptosis. Indeed, photoreceptor loss was rescued by down-regulating Rh1 levels in a Vitamin A-deficient diet [28]. We now show that photoreceptor undergo degeneration in flies expressing dFatp-RNAi in dRPCs even if fed in a Vitamin A-deficient diet. This indicates that dFatp expression in dRPCs also contributes to photoreceptor viability via its role in LD biosynthesis. This is further supported by our results showing that forcing LD lipolysis or inhibiting LD biogenesis induced late onset photoreceptor degeneration similar to dFatp knockdown. Therefore, dFatp would have two distinct functions in photoreceptors and dRPCs. In photoreceptors, dFatp is required for optimal Rh1 metabolism, presumably due to its role in phosphatidic acid synthesis and Rh1 trafficking [16,36,37]. In dRPCs, dFatp is required for expansion of LDs, which can then be metabolized for energy production. Finally, hFATP1 has been shown to directly interact with visual cycle enzymes in RPC to regulate retinyl-ester-dependent RPE65 isomerase necessary for production of 11-cis-retinal and rhodopsin [38,39]. In Cubizolle et al., we have also shown by HPLC that retinyl esters accumulate in mRPC of transgenic mice expressing hFATP1 [31]. We had proposed that the accumulation of retinyl ester could be due to either the inhibition of RPE65 or to an increase of long-chain fatty acids (LCFA) induced by hFATP1 expression, which results in increased retinyl esters. Indeed, incorporated LCFA can form phosphatidyl choline (PC), which serves in the esterification of all-trans-retinol to form alltrans-retinyl esters in a reaction catalyzed by the Lecithin retinol transferase (LRAT) [40].
Together, our results show that in addition to the role of FATP in the regulation of visual cycle precursors in mouse RPC [31] and in the regulation of Rh1 levels in Drosophila photoreceptor [28], FATP has a conserved RPC-specific function in lipid storage that promotes the increase of neutral lipids in photoreceptors and the overall energy metabolism by the mitochondria in RPC and photoreceptor layers.
Our results show that the ectopic accumulation of LDs in RPCs induced by overexpression of FATP is not toxic to photoreceptors in either Drosophila (this study) or mice [31], and that physiological levels of LD are protective for Drosophila photoreceptors during aging. The loss of photoreceptors following depletion of LDs may therefore be due to the dwindling supply of energy, especially since photoreceptors have high-energy demands [5]. This hypothesis is supported by the proximity of LD and mitochondria that is observed in both mouse and fly retina that overexpress FATP. It is also supported by the increased mitochondrial respiration and βoxidation rates measured in both RPC and PR layers of mice retina that overexpress hFATP1. Another non-exclusive hypothesis supported by recent reports, is that de novo LD biogenesis protects against lipotoxicity under conditions of low nutrient supply or high-energy demand as it was shown in adipocytes and mouse embryonic fibroblasts [41,42]. Collectively, these results indicate that lipid storage in LDs is physiologically required for photoreceptor health during aging. In addition to acting as an energy source, LDs may also reduce the cellular sensitivity to ROS. The latter possibility is supported by the finding that ROS induced by light exposure, which promotes the accumulation of photo-damaged proteins and lipids [43], enhances photoreceptor degeneration to a greater extent in dFatp mutant flies than in WT flies [28]. Similarly, LD accumulation in glia protects against ROS-induced damage of the developing nervous system of Drosophila upon exposure to hypoxia [8]. Moreover, LDs are not uniform structures but exist in various forms with differing potential functions, such as storage of phospholipids, vitamin E, and cellular toxins [10].
Our data establish that LDs are not toxic to retinal cells under physiological conditions, but that they can contribute to neurodegeneration under some stress conditions as shown here with the Aats-met FB mutants (model in Fig 8). In these mutants, suppression of LD accumulation in dRPCs is neuroprotective, suggesting that the LDs non-autonomously promote photoreceptor cell death under conditions of oxidative stress. This is consistent with studies of flies with mitochondrial defects that cause high ROS levels in neurons, which demonstrated that removal of LDs by dFatp knockdown rescued retinal degeneration in sicily (Drosophila homolog of the human nuclear encoded mitochondrial gene NDUFAF6) and marf (Drosophila homolog of the mitochondrial fusion GTPases, Mitofusin 1 and 2) mutants and that ectopic expression of Bmm-lipase rescued photoreceptors in Aats-met mutants [9,27]. One possible explanation as proposed by Liu et al. [27], is that high levels of ROS induce an abnormal initial raise followed by a decrease of LD as the pathology progresses. This perturbed turnover could override LD beneficial role and enhances lipid peroxidation and photoreceptor death. Collectively, our findings and those of others suggest that LDs may play a protective or beneficial role under physiological conditions and be deleterious under pathological conditions.
In conclusion, despite the difference in mouse and fly retinal architecture, RPCs appear to be functionally homologous in both species with respect to providing metabolic support for adjacent photoreceptors. Drosophila RPCs form a tight barrier between the hemolymph and photoreceptors, similar to the mammalian outer retina [44], implying that dRPCs, like mammalian RPCs, may play an active role in metabolite exchanges with photoreceptors [32]. Similar to dRPCs, it was recently shown that Drosophila cone cells, which are also accessory cells of photoreceptors, express glial cell markers and are important to maintain photoreceptor function in the adult fly [45]. Thus, combined with our data, these results support that both types of accessory cells-dRPCs and cone cells-carry glial functions similar to the support functions of Müller glia and retinal pigment epithelium (here mRPC) in the vertebrate retina. We propose that the RPC-photoreceptor interaction is a conserved paradigm between Drosophila and mammalian that will be a useful model to investigate mammalian RPC disorders, such as AMD, particularly given Drosophila genetic tractability.

Ethics statement
Experiments were carried out in accordance with the European Communities Council Directive of 24 November 1986 (86/609/EEC) and the French Ethical Committee (CEEA-LR-12141) guidelines for the care and use of animals for experimental procedures.
For BODIPY 493/503 , dissected retinas were fixed first in 4% PFA for 15 min at RT, washed three times and then incubated with 75 ng/mL BODIPY 493/503 and 25 ng/mL phalloidin-TRITC for 16 hours at 4˚C.
Subsequently, samples were washed three times in wash buffer, mounted on a bridge slide in Vectashield (Vector Laboratories), and stored at −20˚C until analysis. 16-bit image stacks were acquired on a Zeiss LSM800 microscope and processed for quantification using ImageJ2 software [52]. Images were filtered for noise using Gaussian Blur 3D (σ = 1), after which a Zprojection was made. LDs were identified by thresholding the images, and the integrated density of the signal and total retinal area were measured. The integrated density of BODIPY staining divided by the total area of the retina was normalized to the control for each experiment.

Statistical analysis
Statistical analyses were performed using R. Group differences were analyzed by t-test or Tukey's HSD paired sample comparison test, as specified in the figure legends.

Imaging of photoreceptors and RPCs by the cornea neutralization method in living flies
Living flies, maintained for 30 days at 25˚C on a 12h light/dark cycle, were anesthetized using CO 2 , embedded in 1% agarose covered with cold water [53], and imaged using a Leica SP5 upright confocal microscope. RPCs were visualized by pigment autofluorescence (excitation 514 nm, detection 530-630 nm) and photoreceptors were identified by Rh1-GFP expression (excitation 488 nm, detection 500-570 nm) as described [24,47,53]. Ommatidia lacking one or several RPCs lose their lozenge shape and are counted as affected ommatidia. Photoreceptor rhabdomeres were quantified using Fiji cell counter tool.

Transmission electron microscopy of Drosophila eyes
Dissected Drosophila eyes were fixed in 0.1 M cacodylate buffer, 2.5% glutaraldehyde, and 2 mM CaCl 2 for 16 h at 4˚C. After rinsing with 0.1 M cacodylate buffer at RT, the eyes were incubated with 1% OsO 4 in 0.1 M cacodylate buffer for 2 h at RT. Tissues were then progressively dehydrated in acetone at RT and mounted in 100% epoxy resin (Epon 812) in siliconeembedding molds. After resin polymerization for 48 h at 60˚C, samples were sliced into 60 nm sections, which were stained with lead citrate and examined with a Philips CM120 transmission electron microscope (TEM) operating at 80 kV. For quantification, only LDs were considered that were clearly localized within either dRPC or photoreceptors. Vesicles surrounded by double membrane layers were excluded.

Mice
Transgenic mice overexpressing human FATP1 (hFATP1TG) specifically in the RPCs (driven by the RPC-specific VDM2 promoter) were generated on a C57BL/6J genetic background as previously described [31]. The hFATP1TG mice were maintained on a standard 12 h light (90 lux)/12 h dark cycle at~22˚C and were fed ad libitum with a standard rodent diet. Mice were housed in facilities accredited by the French Ministry of Agriculture and Forestry (B-34 172 36 -March 11, 2010). Mice were euthanized by cervical dislocation and the eyes were enucleated and dissected.

Co-staining of lipids, perilipin, and ATP synthase
Nile Red staining of lipids [54] was used alone or in conjunction with fluorescent protein immunostaining. The eyes of 3-month-old mice were enucleated and the retina was separated from the RPC/choroid layer to obtain an empty eyeball [31]. The eyes were fixed in 4% PFA for 1 h at RT, washed with PBS, and permeabilized with 0.1% sodium dodecyl sulfate. They were then incubated for 20 min at RT in blocking buffer (10% fetal calf serum in PBS), and incubated overnight at 4˚C with a mouse anti-mouse ATP synthase (Millipore MAB3494, 1:500) or rabbit anti-mouse perilipin (D1D8, Cell Signaling Technology # 9349, 1: 200) primary antibody. The tissues were then incubated for 4 h at RT with Alexa Fluor 488-conjugated anti-rabbit or anti-mouse secondary antibodies diluted in blocking buffer. The immunostained eyeballs were gently rinsed in PBS, incubated with Nile Red solution (10 μg/ml) for 30 min at RT in the dark, washed twice in PBS for 5 min at RT, and incubated with 4 0 ,6-diamidino-2-phenylindole (DAPI, 1:1000) for 5 min at RT. The eyeballs were finally rinsed 5 times in PBS for 5 min at RT and mounted in Dako mounting medium. Confocal imaging was performed with a Zeiss LSM 5 LIVE DUO Highspeed/Spectral Confocal system. Images were acquired with Zeiss Zen software, and LDs were counted with ImageJ software.

Quantification of neutral lipids in mouse retina
Mice were euthanized by vertebral dislocation, the eyes were enucleated, and the neural retina was removed from the eye cup. Neural retinas were directly frozen in liquid nitrogen. The RPC/choroid was scraped and collected in PBS, and the samples were centrifuged to remove the PBS. Tissues were stored dry at −80˚C.
Lipids were extracted and analyzed as previously described [55]. Total lipids were extracted twice from tissues with ethanol/chloroform (1:2, v/v). Before extraction, internal standards were added. The organic phases were dried under nitrogen and lipid classes were separated by thin-layer chromatography on silica gel G using a mixture of hexane-ethyl ether -acetic acid (80:20:1 v/v/v). Lipids were transmethylated and the fatty acid methylesters were analyzed by gas chromatography. Briefly, samples were treated with toluene-methanol (1:1, v/v) and boron trifluoride in methanol. Transmethylation was carried out at 100˚C for 90 min in screwcapped tubes. After addition of 1.5 mL, 10% K 2 CO 3 in water, the resulting fatty acid methyl esters were extracted with 2 mL of isooctane and analyzed by gas chromatography, using an HP6890 instrument equipped with a fused silica capillary BPX70 SGE column (60 × 0.25 mm). The vector gas was hydrogen. The temperatures of the Ross injector and the flame ionization detector were set at 230˚C and 250˚C, respectively.

Oxygen consumption
Respiration was measured on RPC/choroid and neural retinas permeabilized by incubation for 2 min with 15 μg digitonin per mg and resuspended in a respiratory buffer (pH 7.4, 10 mM KH 2 PO 4 , 300 mM mannitol, 10 mM KCl and 5 mM MgCl 2 ). The respiratory rates were recorded at 37˚C in 2 ml glass chambers using a high-resolution Oxygraph respirometer (Oroboros, Innsbruck, Austria). Assays were initiated in the presence of 5 mM malate/0.2 mM octanoyl carnitine to measure state 2, basal respiration (EIImo basal). Complex I-coupled state 3 respiration was measured by adding 0.5 mM NAD + /1.5 mM ADP (EIIImo β-ox). Then, 5 mM pyruvate and 10 mM succinate were added to reach maximal coupled respiration (EIIImpgso CxI+CxII), and 10 μM rotenone was injected to obtain the CII-coupled state 3 respiration. Oligomycin (8 μg/mL) was added to determine the uncoupled state 4 respiration rate. Finally, carbonyl cyanide-4-(trifluoromethoxy) phenylhydrazone (1 μM) was added to control the permeabilization of the tissues.

Mitochondrial enzymatic activities
The activities of the mitochondrial citrate synthase (CS), oxoglutarate dehydrogenase (OGDH), isocitrate dehydrogenase (IDH) and fumarase were measured in 5 μl RPC/choroid and neural retina homogenates (sonicated in 50 μl PBS) at 37˚C using a Beckman DU-640B spectrophotometer (Beckman Coulter) or a CLARIOstar (BMG LabTech) [56]. Briefly, citrate synthase activity was measured using 0.15 mM DTNB reagent (SIGMA Aldrich) which interacts with CoA-SH to produce TNB. The formation of TNB was followed for 1.5 min at wavelength of 412 nm. IDH and OGDH activities were determined in the presence of respective substrates isocitrate and α-ketoglutarate, by monitoring for 3 min the change in NAD + to NADH which absorbs light at 340 nm. Fumarase activity was determined by measuring the conversion of L-malate to fumarate, monitoring the increase in absorbance at 250 nm. The optical density variation per minute is calculated from the curve and the enzymatic activity expressed as nmol of product formed/minute/mg protein.

Electron microscopy of mouse RPCs
The eyes of 3-6 month-old hFATP1TG mice were rapidly enucleated, the corneas were removed, and the eyeballs were fixed by immersion in 2.5% glutaraldehyde in Sorensen's phosphate buffer (0.1 M, pH 7.4) overnight at 4˚C. The tissues were then rinsed in Sorensen's buffer and post-fixed in 1% OsO 4 for 2 h in the dark at RT. The tissues were rinsed twice, dehydrated in a graded series of ethanol solutions (30-100%), and embedded in EmBed 812 using a Leica EM AMW (Automated Microwave Tissue Processor for Electronic Microscopy). Sections (60 nm thick) were cut near the optic nerve (Leica-Reichert Ultracut E), counterstained with uranyl acetate, and examined using a Hitachi 7100 TEM (Centre de Ressources en Imagerie Cellulaire de Montpellier, France). The thickness of Bruch's membrane was determined by measuring both the thickest and the thinnest parts of 5 fields throughout the retinal section per mouse. Data are presented as the median value per eye. We also enumerated the vacuoles in 5 fields of retinal sections per mouse at 10,000× magnification.  Fig. dFatp is a functional ortholog of human FATP1. (A, B) Confocal fluorescence microscopy of retinas from dFatp -/-(dFatp k10307 ) mutant flies without (A) or with (B) photoreceptorspecific expression of human FATP1 (dFatp -/-;Rh1>hFATP1). hFATP1 rescues the loss of photoreceptors in dFatp -/mutant clones. Retinas were generated using the Tomato/GFP-FLP/FRT technique [47], in which all photoreceptors are marked by GFP, and homozygous mutant mosaic retina is marked by the absence of TdTomato.