Hydroxylated sphingolipid biosynthesis regulates photoreceptor apical domain morphogenesis

A characteristic hallmark of epithelial cells is the subdivision of their plasma membrane into an apical, a lateral, and a basal domain. The apical plasma membrane domain is often expanded and highly differentiated to execute specific functions, for example to increase absorption or to form the outer segment in vertebrate photoreceptor cells (PRCs), thus enhancing vision. Development of the apical membrane includes its specification, differentiation of cell type–specific specialized structures, and growth to achieve the mature size and shape.

To date, we have a good understanding of the molecular players (genes and protein complexes) and some of the underlying principles that regulate the primary polarization of the plasma membrane in different epithelial tissues (Riga et al., 2020; Rodriguez-Boulan and Macara, 2014). However, once specified, how does the apical membrane achieve its mature size and highly stereotyped shape? The formation of specialized structures, such as microvilli, appears to be largely driven by the underlying cytoskeleton, i.e., by the actin network, including ezrin/radixin/moesin proteins and binding partners (Apodaca, 2018; Fath and Burgess, 1995; Morales et al., 2004; Pelaseyed and Bretscher, 2018; Saotome et al., 2004; Sauvanet et al., 2015). Increase in length of microvilli in the apical stereocilia of sensory cells requires the joint activity of motor protein (myosins) and actin-bundling protein Epsin (Lelli et al., 2016). Recently Crb1, an evolutionary conserved apical determinant, was shown to participate, along with myosin 3b and Epsin, in the extension of actin-based protrusions of spinal neurons in zebrafish (Desban et al., 2019). Overexpression of Drosophila crumbs (crb) can lead to an expansion of apical membranes, whereas lowered levels can cause its reduction (Lattner et al., 2019; Muschalik and Knust, 2011; Myat and Andrew, 2002; Pellikka et al., 2002; Wodarz et al., 1995).

While many proteins are known to contribute to the growth of the apical domain, there is little appreciation of the role of other molecules and cellular processes, for example lipid metabolism or redox reactions, in this process. More specifically, how does the cell use lipids to regulate the development and growth of the apical domain? So far, phosphoinositides and their metabolites, such as phosphatidylinositol 4,5-bisphosphate and phosphatidylinositol (3,4,5)-trisphosphate, are the only well-studied signaling lipids in the context of apicobasal polarity (Gassama-Diagne et al., 2006; Martin-Belmonte et al., 2007; Medelnik et al., 2018; Raghu et al., 2012; Krahn, 2020). But these constitute only a small subset of lipid species in relation to the enormous diversity of lipids in cells. An early lipidomics study on Madin–Darby canine kidney (MDCK) cells in culture revealed that polarization of epithelial cells is coupled with major changes in the lipidome (Sampaio et al., 2011), but the mechanisms required to coordinate/integrate the multiple pathways regulated by an altered lipidome were not further analyzed.

Here, we addressed the role of lipids in apical membrane growth using Drosophila melanogaster PRCs. During pupal development, these cells develop a prominent apical domain, the rhabdomere, formed by ∼40,000 closely packed microvilli (Cagan and Ready, 1989; schematized in Fig. 1 A). In the second half of pupal development, the original rudimentary microvilli increase in number and length and become more ordered (Longley and Ready, 1995). This growth is achieved in part by regulators of F-actin, a major structural component of the rhabdomere (Arikawa et al., 1990). Perturbation of the assembly of F-actin or its interaction with moesin results in early defects in the formation of the rudimentary microvilli (Baumann, 2004; Karagiosis and Ready, 2004; Pham et al., 2008; Zelhof and Hardy, 2004).

In addition, and somewhat unexpected, Rhodopsin, the light-sensing molecule, plays an important structural role in rhabdomere morphogenesis (Kumar and Ready, 1995). Biosynthesis of opsin (the protein moiety of Rhodopsin) is initiated in the last quarter of pupal development, followed by its apically directed transport toward the growing rhabdomere, and coincides with dramatic growth of the rhabdomere. Altered levels of Rhodopsin in the rhabdomere during this phase (Kumar and Ready, 1995), as well as improper trafficking (Beronja et al., 2005; Laffafian and Tepass, 2019; Satoh et al., 2005, 2015), impairs rhabdomere morphogenesis. Apart from Rhodopsin 1 (Rh1) and actin regulators, Drosophila crb also plays a role in rhabdomere growth. Mutants in crb exhibit abnormally shaped rhabdomeres, which fail to extend throughout the depth of the retina (Izaddoost et al., 2002; Johnson et al., 2002; Pellikka et al., 2002). Despite the fact that a huge amount of membrane is required to generate the rhabdomere, information about the role of lipids in this process are lacking. So far, studies on lipids in the fly eye have concentrated on their role in phototransduction mostly in mature PRCs (Muñoz et al., 2013; Raghu et al., 2009; Randall et al., 2015), but how rhabdomere morphogenesis is coupled to lipids and their metabolism has not been addressed.

Here, we used shotgun lipidomics of mutant alleles of crb with altered rhabdomeric shapes to elucidate the link between lipid composition and apical plasma membrane differentiation and growth. This strategy led to the identification of hydroxylated sphingolipids as influential players in rhabdomere morphogenesis. Furthermore, it revealed a novel role of hydroxylated sphingolipids in the regulation of apically directed delivery of Rhodopsin and hence in the growth of the rhabdomeres.

Mutations in crumbs (crb) have been shown to affect rhabdomeric growth, resulting in bulky rhabdomeres (Izaddoost et al., 2002; Johnson et al., 2002; Pellikka et al., 2002). To better understand the basis of this phenotype, we compared the different aspects of rhabdomere growth in crb mutant and wild-type PRCs. The complex geometry of rhabdomeres is the combined consequence of growth in three dimensions that are temporally separated. (a) The base/width of the rhabdomere is established in midpupal phase concomitant with the stalk membrane (Fig. 1, B and C; and Fig. S1, D–F′; marked by green lines). (b) The thickness of the rhabdomere is the result of an increase in microvillar length along the apicobasal axis (Fig. 1, B and C; and Fig. S1, D–F′, yellow lines). (c) The extension of the rhabdomere along the proximodistal axis of the retina is due to an increase in number of microvilli (Figs. 1 A and S1, A–C). Any perturbation in PRC development in the last quarter of pupal development mainly affects extension and thickness of the rhabdomere and can be evaluated upon eclosion by staining with an F-actin marker.

In crb mutants, the width of the rhabdomeres in the distal end was increased already at midpupal stage (not depicted) and clearly visible 70 h after puparium formation (APF; compare Fig. 1, B and D, green lines). As in control PRCs, the width barely changed in crb mutants until adulthood (compare Fig. 1, D and E and B and C, green lines, respectively). During the growth phase, crb mutant rhabdomeres increased in thickness similarly to rhabdomeres of control PRCs (Fig. 1, D and E, yellow lines). However, the extension along the proximodistal axis was strongly affected (Fig. 1, K–M; Izaddoost et al., 2002; Johnson et al., 2002; Pellikka et al., 2002). This defect cannot be attributed to early defects in adherens junctions, since they recover during this phase (Pellikka et al., 2002; Tepass and Harris, 2007). It should be noted that this extension defect affects only the rhabdomere, but not the rest of the cell, which reaches the retinal floor. In summary, crb affects only two aspects of rhabdomeric growth: the width of the rhabdomere in the first half of pupal development, and the extension along the proximodistal axis in the second half of pupal development. The thickness of the rhabdomere was not affected. It should be noted that the width and the thickness of crb mutant rhabdomeres could be assessed only in the distal region due to the extension defects.

To identify lipids involved in rhabdomere morphogenesis, we conducted shotgun lipidomics of the eyes of four different crb alleles (crb^11A22, crb^8F105, crb⁴, and crb^13A9) and compared them to two different genetic controls (w* and wild type). While crb^11A22, crb^8F105 (Jürgens et al., 1984; Wodarz et al., 1993), and crb⁴ (Pichaud and Desplan, 2001; Richardson and Pichaud, 2010) exhibit a reduced extension of the rhabdomeres as described above (compare Fig. 1 J with Fig. 1, K–M), crb^13A9 does not show this developmental defect (Spannl et al., 2017). The inclusion of this allele in the lipidomics analyses provided a comprehensive toolkit of multiple alleles with distinct, characteristic phenotypes.

Lipidomic analyses included the absolute (molar) quantification of 174 lipid species, spanning 14 lipid classes (Data S1). The only clear and consistent change across crb mutants was an increase in the abundance of hydroxylated sphingolipids (Data S1), particularly in ceramide phosphorylethanolamines (CerPE), the most abundant class of membrane sphingolipid in Drosophila (Carvalho et al., 2010, 2012; Guan et al., 2013), and in their precursors, the low-abundant ceramides (Cer). For example, a two-fold increase was observed for hydroxylated CerPE (ox-CerPE, Fig. 2 A). It should be noted that ox-CerPE represents only a minor (<10 mol%) fraction of the total CerPE class. Strikingly, the increase in the abundance of ox-CerPE was observed only in those crb alleles that exhibit improperly extended and bulky rhabdomeres (crb^11A22, crb^8F105, and crb⁴; Fig. 1, G–I). In contrast, crb^13A9 mutant retinas, which have normal-shaped rhabdomeres (Spannl et al., 2017), did not show any significant change in hydroxylated sphingolipid abundance (Fig. 2 A). Further analysis by higher collisional dissociation Fourier transform (HCD FT) tandem mass spectrometry (MS/MS) fragmentation (Fig. S2) revealed that increased hydroxylation of sphingolipid species occurred at the fatty acid moieties. This increase is shown in Fig. 2 B for the most abundant sphingolipids, CerPE 34:1 and CerPE 36:1, bearing hydroxylated fatty acids (20:0:1 and 22:0:1). At the same time, we observed no changes in fatty acid hydroxylation in glycerophospholipid species. Therefore, we conclude that increased hydroxylation of long chain fatty acids in crb mutant retinas feeds into the sphingolipid biosynthetic pathway.

This result raised the question as to the mechanisms by which crb regulates the abundance of hydroxylated sphingolipids. Hydroxylated fatty acids are generated by fatty acid 2 hydroxylase (Fa2h, Fig. 2 C), encoded by fa2h in Drosophila (Carvalho et al., 2010). Therefore, we compared fa2h transcript levels in eyes dissected from mutant and control adult flies by real-time quantitative RT-PCR (real-time qRT-PCR). On average, a two- to fourfold increase in fa2h mRNA levels was observed in eyes of only those crb mutant alleles that develop bulky rhabdomeres (crb^11A22, crb^8F105, and crb⁴), but not in crb^13A9 mutant eyes (Fig. 2 D), which have normally organized rhabdomeres. The magnitude of increase of fa2h was not as high in crb⁴ compared with the other two alleles (crb^11A22 and crb^8F105), which reflects the trend in ox-CerPE levels (Fig. 2 A). Interestingly, we already observed an increase in fa2h transcripts at pupal stages (60–80 h APF) of crb^11A22 and crb^8F105 animals (Fig. 2 E). This period coincides with massive rhabdomere growth.

Hydroxylated fatty acids are lipid peroxidation products and are considered a metabolic outcome of the physiological stress response (Kaikkonen et al., 2004; Kuda et al., 2018). In fact, fa2h expression can be up-regulated under conditions of stress (Nagano et al., 2012). This is especially relevant because loss of crb leads to increased oxidative stress owing to an overactivation of NADPH oxidase, which is followed by increased levels of reactive oxygen species in adult mutant eyes (Chartier et al., 2012). We corroborated and extended these finding by showing that, as a consequence of increased reactive oxygen species levels, oxidative stress signaling was up-regulated in crb mutant eyes. As a marker for oxidative stress signaling, we used the expression of GstD1, a well-conserved oxidative stress response gene encoding a detoxification enzyme (Sawicki et al., 2003). We observed a significant increase in GstD1 mRNA in adult eyes of crb^11A22, crb^8F105, and crb⁴ mutant flies (Fig. 2 F), which have improperly extended rhabdomeres. Increased oxidative stress signaling was further corroborated by the use of a transgenic reporter for oxidative stress signaling (GstD-GFP; Sykiotis and Bohmann, 2008) in adult eyes of crb mutants. In control eyes, GFP expressed from the reporter was observed in the pigment cells that envelop PRCs but was barely detectable in PRCs in the center of the ommatidium (Fig. S3, A and A′). In contrast, crb mutant retinas with improperly formed rhabdomeres exhibited increased oxidative stress signaling in PRCs, as revealed by increased GFP expression in the central part of the ommatidia (Fig. S3, B–C′).

Taken together, our data uncover a correlation between improperly formed rhabdomeres, increased oxidative stress signaling, and increased hydroxylated sphingolipids in crb mutant eyes. Increased levels of hydroxylated sphingolipids are due to an up-regulation of fa2h expression, which was particularly prominent at a stage of pupal development when rhabdomere growth takes place.

Given the coincidence between an altered redox state and impaired rhabdomere extension in crb mutants, we next asked whether altering the redox status of PRCs in an otherwise wild-type background affects rhabdomeric extension. To address this question, we genetically interfered with the quenching of superoxide radicals, known to contribute to an oxidizing environment. This can be done by manipulating the function of cytoplasmic Sod1 (Cu-Zn superoxide dismutase 1), which converts superoxide radicals into the less reactive molecule hydrogen peroxide. We reduced Sod1 levels by RNAi expression (Missirlis et al., 2003) and used a heterozygous mutant allele (Sod1ⁿ¹/+; Phillips et al., 1995), thereby increasing superoxide radicals. The mutant heterozygotes (Sod1ⁿ¹/+) are associated with an increase in GstD1 mRNA (Fig. 3 A), indicative of an altered redox state in these animals. This change was associated with increased levels of fa2h mRNA in Sod1ⁿ¹/+ mutant heads (Fig. 3 B). Strikingly, eyes of heterozygous mutant flies (Sod1ⁿ¹/+) and eyes with PRC-specific RNAi-mediated knockdown of Sod1 were associated with abnormal rhabdomeres; they did not completely reach the base of the retina (Fig. 3, C′–E′). Thus far, we demonstrated that an altered redox status, achieved by reducing either Sod1 or crb, which regulates the plasma membrane redox system (NADPH oxidase; Chartier et al., 2012), results in increased fa2h expression and induces defects in rhabdomere extension.

To further determine whether increased fa2h levels are the cause for defects in rhabdomere growth in crb mutants, we manipulated fa2h expression levels in wild-type and mutant retinas using Rh1-Gal4, which is active in the last quarter of pupal development, during which rhabdomeres grow tremendously. Neither overexpression (Fig. 4 B) nor knockdown of fa2h via RNA interference (Fig. 4 C) in an otherwise wild-type background induced any gross abnormalities in rhabdomere extension compared with control (Fig. 4 A). However, overexpression of fa2h in crb^11A22 and crb^8F105 mutant retinas caused even more bulky and irregular rhabdomeres (compare Fig. 4, D and G, with Fig. 4, E and H, respectively, and Fig. 4 N for penetrance and quantitative comparisons). The lack of further enhancement of the crb^11A22 phenotype of short rhabdomeres upon overexpression of fa2h (Fig. 4 N) is in agreement with earlier data showing that crb^11A22 is a protein null allele, while some protein is still produced in crb^8F105 (Wodarz et al., 1993). In contrast, rhabdomeres appeared less bulky upon lowering fa2h levels by RNAi in a crb mutant retina (compare Fig. 4, D and G, with Fig. 4, F and I, respectively, and Fig. 4 N). The effectiveness of fa2h overexpression and knockdown on fa2h mRNA was confirmed by real-time qRT-PCR (Fig. 4 J). The rescuing effect obtained upon knockdown of fa2h, although not completely penetrant, was more pronounced in the crb allele with the weaker phenotype (crb^8F105). Here, most of the rhabdomeres extended to the base of the retina (compare Fig. 4 I to respective controls in Fig. 4 G). The rescue of the extension defect was also obvious in cross sections (Fig. 4, K′–M′). In crb^8F105, fewer and/or thinner rhabdomeres were visible proximally (Fig. 4 L′). Upon knockdown of fa2h in this background, a full complement of seven rhabdomeres was clearly visible in the proximal region. Note that fa2h knockdown did not affect the enlarged width of the rhabdomeres distally. Overall, these results support the conclusion that defects in rhabdomere extension in crb mutant eyes are caused by increased fa2h expression.

A major structural constituent of the rhabdomeric membrane of the outer PRCs (PRC 1–6) is the visual pigment Rh1. Rh1 is composed of the apoprotein opsin 1 (encoded by the gene ninaE), conjugated to the chromophore 3-hydroxy retinal, which is normally supplied by dietary carotenoids from plants (Isono et al., 1988). Defects in ninaE mutant retinas range from thinner rhabdomeres in hypomorphs to completely irregular microvillar structures in null mutants (Kumar and Ready, 1995). Rh1 was clearly observed in the rhabdomere of flies raised with normal food (Fig. 5, A and E). Reduction of the levels of the apoprotein opsin1, for example in the hypomorphic allele ninaE⁷, resulted in reduced Rh1 in the rhabdomeres (Fig. 5 C). A more detailed examination of the mutant phenotype revealed that these lowered rhabdomeric Rh1 levels coincided with impaired rhabdomere extension along the proximodistal axis (Fig. 5 D, bracket, vs. 5 B). Hence, reduction of Rh1 phenocopied the crb mutant rhabdomeric extension phenotype (Fig. 1, H–J).

Rhodopsin undergoes several steps of maturation, which take place during its trafficking through the secretory pathway. Absence of dietary carotenoid prevents maturation of opsin into Rhodopsin. As a consequence, most opsin accumulates in the perinuclear ER (Fig. 5 G; Ozaki et al., 1993; Satoh et al., 1997). In addition, the overall levels of Rhodopsin as well as rhabdomeric Rh1 (revealed by immuno-EM) are reduced (Huber et al., 1994; Sapp et al., 1991). We confirmed the development of thinner rhabdomeres (Lee et al., 1996) in wild-type PRCs upon carotenoid depletion (Fig. 5 G vs. 5 E). Strikingly, this treatment also prevents proper extension of rhabdomeres along the proximodistal axis, as many of them did not reach the retinal floor (Fig. 5, H and H′). This suggests that proper Rh1 levels in the rhabdomere are crucial for rhabdomeric growth, including its length and thickness. Accordingly, carotenoid depletion in crb mutants exacerbated the extension defects (Fig. 5 L vs. 5 J) and also affected the thickness of rhabdomeres (Fig. 5 K vs. 5 I). However, the increased width of crb mutant rhabdomeres at the distal end remained unchanged. This is not surprising, because rhabdomeric width is established at midpupal stages, before Rh1 expression. Overall, these results confirm that Rh1 in rhabdomeres is critical for rhabdomere proximodistal extension and thickness. They further suggested that impaired extension of rhabdomeres in crb mutants also depended on Rh1 levels. Given the observation that fa2h expression is enhanced in crb mutant retinas, this raises the question whether fa2h has an impact on Rhodopsin levels in the rhabdomere and/or its trafficking.

Besides improperly expanded and bulky rhabdomeres, punctate cytoplasmic accumulation of Rh1 is another feature of the crb mutant phenotype (Pocha et al., 2011), pointing to impaired Rh1 trafficking. As recently shown, absolute levels of opsin do not change in crb mutants with bulky rhabdomeres (Raghuraman et al., 2020). To further understand whether up-regulated fa2h impacts Rh1 distribution as well, we analyzed the localization of Rh1 by immunostaining. At steady state, Rh1 antibody labeled rhabdomeres and sub-rhabdomeric punctae in crb mutant (Fig. S4, A′ and D). Upon fa2h overexpression in crb^8F105 mutant PRCs, decreased rhabdomeric Rh1 staining and more Rh1-positive intracellular punctae (Fig. S4, B′ and E) were observed. The opposite effect on Rh1 distribution and rhabdomere shape was observed upon RNAi-mediated down-regulation of fa2h in mutant PRCs. Under this condition, most Rh1 staining was associated with rhabdomeres (Fig. S4, C′ and F). These images imply that Rh1 distribution is altered with changes in fa2h. Whether changes in fa2h levels impact Rh1 delivery and/or its stability cannot be concluded from these experiments.

Rhabdomere extension in the last quarter of metamorphosis coincides with a massive increase in the production of the apoprotein opsin, its maturation, and its trafficking as Rhodopsin through the secretory pathway to the rhabdomere via apically directed trafficking, mediated by the concerted action of Rab1, Syntaxin 5, Rab6, Rab11, Rip11, the exocyst complex, and MyoV (Beronja et al., 2005; Iwanami et al., 2016; Karagiosis and Ready, 2004; Laffafian and Tepass, 2019; Li et al., 2007; Pocha et al., 2011; Satoh et al., 2005). To follow the conversion of opsin and its trafficking as Rh1 to the rhabdomeres, the BLICS (blue light–induced chromophore supply) assay was used (Iwanami et al., 2016; Pocha et al., 2011; Satoh et al., 2005). Following carotenoid depletion, opsin was trapped primarily in the perinuclear ER of PRCs of adult flies (asterisk in Fig. 6 A) and was barely visible in the rhabdomeres. Only upon supplementation of the precursor (retinal) and its isomerization from all-trans retinal to 11-cis retinal via exposure to blue light, Rh1 was formed and trafficked to the rhabdomere (white arrowheads in Fig. 6, C and C').

Based on BLICS assays, it has previously been shown that loss of crb affects Rh1 trafficking to the rhabdomere (Pocha et al., 2011). To explore whether increased levels of fa2h, associated with loss of crb, were responsible for this defect, we performed the BLICS assay in wild-type animals with increased levels of fa2h. At t = 0, we observed opsin accumulation in the ER similar to that in the control (asterisk in Fig. 6 B). However, after feeding retinal and at t = 180 min of exposure to blue light, very little Rh1 was detected in only few rhabdomeres, compared with control eyes (compare Fig. 6, D and D′, with Fig. 6, C and C′). This indicates that fa2h overexpression affects conversion of opsin to Rh1 and/or trafficking through the secretory pathway.

We next sought to quantify the extent of Rh1 being delivered to the rhabdomeres. Hydroxylated fatty acid metabolism in worms has been shown to influence membrane trafficking mediated by the small GTPases, Rab11 and Rab7 (Li et al., 2018). In Drosophila PRCs, Rab11 colocalizes extensively to newly synthesized Rh1 following its exit from the Golgi and before being deposited in the rhabdomere. Indeed, rab11 mutant rhabdomeres are smaller in size owing to a failure of Rh1 in reaching the rhabdomere (Satoh et al., 2005). On the other hand, Rab7 shows little overlap with Rh1 during the normal delivery of newly synthesized Rh1 to the rhabdomeres (Iwanami et al., 2016). We therefore quantified the overlap of Rh1- and Rab11-positive structures along the length of rhabdomeres from confocal images (Fig. S5, A–D) as a proxy/readout for the extent of Rh1 being delivered to rhabdomeres.

As previously published (Satoh et al., 2005), Rh1 overlapped with Rab11 upon release from ER in control retinas (Fig. 6, E–E"). In line with this, addition of carotenoids and exposure to blue light in BLICs assays resulted in a significant increase in Rh1 localized to Rab11-positive structures in control retinas (Fig. 6 G). In contrast, no increase in Rh1/Rab11 overlap was observed upon fa2h overexpression using the same conditions (Fig. 6 G). It should be noted that there were no significant changes in the overall level of opsin protein upon fa2h overexpression (Fig. S4 G), and the numbers of Rh1-positive and Rab11-positive structures were not altered (Fig. S5, E and F). From these results, it can be concluded that increased fa2h overexpression, also observed in crb mutants, impairs Rh1 trafficking to the rhabdomeres. In crb mutants, Rh1 localization in rhabdomeres decreased (Fig. S5 G; Pocha et al., 2011), and no change in Rh1/Rab11 overlap was revealed by BLICS assays (Fig. S5 H).

To summarize, increased fa2h expression, and thereby increased generation of hydroxylated sphingolipids (as observed, for example, in crb mutant PRCs), resulted in impaired Rh1 trafficking to the rhabdomeres and, as a consequence, to defective morphogenesis of rhabdomeres. These results highlight an as-yet-unknown role of hydroxylated sphingolipids in the growth of the apical domain of Drosophila PRCs during morphogenesis.

Despite the importance of intact microvilli for the homeostasis of many epithelia, regulators controlling microvilli formation are only partially understood (Walton et al., 2018). Here, we used the rhabdomere of Drosophila PRCs, a highly expanded and elaborated apical membrane, as a model to dissect mechanisms controlling the formation of microvilli. We show that increased transcription of fa2h, followed by increased levels of hydroxylated sphingolipids, is associated with defective rhabdomere morphogenesis. This is, at least partially, due to impaired apical trafficking of Rh1, an important structural component of the rhabdomeral membrane. These results also contribute to our understanding of the role of crb in rhabdomere morphogenesis. They imply that increased levels of hydroxylated sphingolipids particularly impact one aspect of rhabdomeral growth, namely the addition of new microvilli, resulting in the extension of the rhabdomere along the proximodistal axis. Rhabdomeral thickness, determined by the length of the microvilli, is not obviously affected by changes in these lipids.

The connection between hydroxylated sphingolipid metabolism and apical domain morphogenesis is not unprecedented. Increased hydroxylation of sphingolipids was observed to be one of the major lipidomic changes during the polarization of MDCK cells (Sampaio et al., 2011), and inhibition of sphingolipid biosynthesis resulted in a reduced number of apical microvilli on their apical surface (Poole et al., 2004). Likewise, the Caenorhabditis elegans orthologue of fa2h, fath-1, was identified in a screen for genes affecting intestinal polarity (Zhang et al., 2011). Until now, however, the link between hydroxylated sphingolipids and their metabolism and polarization of epithelial cells was limited to circumstantial evidence. We now clearly demonstrate that manipulating hydroxylated sphingolipid metabolism, at least in a sensitized genetic background, is sufficient to modulate rhabdomere/apical domain extension.

How do hydroxylated sphingolipids and their metabolism affect rhabdomeric/apical domain growth? Given the low abundance of hydroxylated sphingolipids in the overall fly eye lipidome, we consider it unlikely that they play any major structural role in rhabdomere morphogenesis, e.g., by promoting the stability of Rhodopsin in the rhabdomeric membrane. We rather favor the conclusion that fa2h and thereby hydroxylated sphingolipids play a role as regulators of apical trafficking. This conclusion is based on two findings: (1) less newly synthesized Rh1 is delivered to the rhabdomeres upon overexpression of fa2h, as revealed by BLICS assays, which mimic Rh1 pulse-chase experiments; and (2) a similar defect in Rh1 trafficking has been observed in crb mutant PRCs (Pocha et al., 2011), which have increased levels of fa2h expression. In particular, we identified an effect on post-Golgi trafficking of Rh1 via the Rab11-mediated pathway. Accumulation of cytoplasmic vesicles is observed upon genetic perturbation of Rab11 and/or genes that encode for Rab11 interacting proteins (Li et al., 2007; Otsuka et al., 2019; Satoh et al., 2005). However, no obvious accumulation of similar vesicles was observed in crb mutant PRCs (not depicted). Thus, the trafficking defect observed upon fa2h overexpression cannot be attributed to a general increase/decrease in Rab11 compartments, but rather alludes to a defect in sorting apical cargo. These results are consistent with studies in C. elegans, where apical compartments labeled with Rab11 and Rab7 are affected upon loss of fath-1 (Li et al., 2018).

Hydroxylated sphingolipids, as glycosphingolipids, are proposed to regulate the sorting of apically directed vesicles by combining different polarity cues (Zhang et al., 2011). In line with this, we propose that increased fa2h transcription (and hence an increase in hydroxylated sphingolipids) prevents apical membrane (rhabdomere) growth by inhibiting trafficking of Rh1 via the apical Rab11 compartments. By regulating the amount of Rh1 delivered to the rhabdomere, fa2h modulates the amount of rhabdomeric membrane in Drosophila PRCs, and hence has an impact on rhabdomere growth. Whether fa2h additionally promotes stability of Rhodopsin in the rhabdomeric membrane via the presence of more hydroxylated sphingolipids could not be addressed here. Interestingly, exogenously labeled hydroxylated fatty acids (the precursors of hydroxylated sphingolipids) preferentially distributed to apically localized membrane-bound compartments in the epithelial cells of the C. elegans gut (Li et al., 2018).

Unexpectedly, fa2h overexpression or knockdown in an otherwise wild-type background does not result in an obvious rhabdomeric extension phenotypes. This may be because Rh1 trafficking depends on the concerted action of Rab1, Rab6, Rab11, Rip11, and MyoV as well as the Rab-effector proteins such as Parcas (Rab11GEF) and Rab11 interacting proteins such as dRip11 (Iwanami et al., 2016; Karagiosis and Ready, 2004; Li et al., 2007; Otsuka et al., 2019; Pocha et al., 2011; Satoh et al., 2005), and to the amount of F-actin (Pham et al., 2008). Although sphingolipids can induce changes in the cytoskeleton (Zeidan et al., 2008), we have not observed any overt changes in the staining of F-actin in the rhabdomere. Therefore, we speculate that increased fa2h acts to fine-tune growth by slowing down delivery of Rh1 and hence addition of new rhabdomeric membrane.

Yet, in a sensitized background, i.e., in cells homozygous mutant for crb, which have impaired Rh1 trafficking (Pocha et al., 2011) or upon carotenoid depletion, which also reduces Rh1 levels, fa2h overexpression now decreases Rh1 delivery via Rab11. Crb normally limits oxidative stress (Chartier et al., 2012), and the resulting oxidative status (low oxidative stress) normally limits fa2h expression. In crb mutants, however, an altered oxidative status (increased oxidative stress) causes an up-regulation of fa2h transcription, which, in turn, results in severely reduced Rh1 delivery and improperly extended rhabdomeres. The fa2h dependence of rhabdomeric growth described here is only one aspect of the pleiotropic cellular response to an altered redox status of increased oxidative stress signaling due to loss of crb. Interestingly, loss of crb in epithelial cells of larval salivary glands impairs Rab6-, Rab11-, and Rab30-dependent apical trafficking and hence apical membrane homeostasis (Lattner et al., 2019). Whether this is also caused by increased oxidative stress and/or defects in sphingolipid metabolism remains to be analyzed.

In conclusion, this work elucidates an interplay between oxidative stress, lipid metabolism, and apical domain growth during PRC morphogenesis. It will be interesting to investigate whether this link between cell metabolism and apical membrane growth and morphogenesis is a more widespread phenomenon in epithelial biology.

Fly lines used in this study are documented in Table S1. Genotypes are listed in Table S2 for each figure. All fly crosses were maintained at 25°C under a standard 12-h light-dark cycle with a light intensity ranging from 600 to 1,000 lux. Unless specified, all flies were reared on a standard cornmeal-yeast–based diet.

Pupae were collected at 0 h APF as white prepupae and placed on moistened filter paper discs in a Petri dish. They were allowed to age at 25°C until the desired age, and then the heads were dissected and processed as previously described (Spannl et al., 2017). Adults were collected within 0–6 h of eclosion and immediately processed as described (Spannl et al., 2017). Females were used in all experiments with adult flies except Western blotting.

BLICS assays were done as described (Iwanami et al., 2016; Pocha et al., 2011; Satoh and Ready, 2005). Briefly, flies were reared on a carotenoid-depleted diet (Pocha et al., 2011). Adult flies were collected within 0–6 h after eclosion. Some flies were collected to prepare sections of eyes (carotenoid deficient), while others were transferred to vials containing carotenoid-free food supplemented with all trans-retinal (17.6 mM; Sigma-Aldrich; made in ethanol and allowed to crystalize on the surface). Flies were left in the dark for 48 h on this food. After this preincubation period, flies were exposed to blue light for 3 h. At this time point, heads were dissected and fixed with minimum exposure to white light.

12-µm cryosections were obtained and immunostained as described previously (Spannl et al., 2017). Antibodies and fluorescently tagged dyes used are listed in Table S3. Samples were mounted in Vectashield (Vector Labs), and imaged with an Olympus FV1000 confocal microscope with a 60× Plan Apochromat oil objective (numerical aperture = 1.35) at ambient room temperature. Confocal z-stacks were obtained using an optical sectioning thickness of 0.33 or 1.0 µm.

Images were processed using Fiji (Schindelin et al., 2012) for making projections of the z-stacks and Adobe Photoshop CS5.1 for adjusting brightness and contrast. Adobe Illustrator CS3 was used for image assembly.

The percentage extension of the rhabdomere along the proximodistal axis (Fig. 4 N) was calculated as a ratio of the length of the rhabdomere (measured from projections in Fiji from rhabdomere cap to its most proximal part) to the retinal length (measured from projections in Fiji from rhabdomeric cap to the retinal floor). Sample sizes comprised at least three regions of interest (ROIs) from retinal sections of three to five eyes (biological replicates).

For quantification of Rh1- and Rab11-positive punctae, the entire z-stack from longitudinal sections was used, and an ROI of fixed size that comprised the length of three to four ommatidia was selected using the phalloidin channel as reference. Rh1-positive regions and Rab-11 positive regions were segmented for the whole stack. Using Fiji (Schindelin et al., 2012), a spot colocalization plugin, ComDet, was used as in Freal et al. (2019) for this ROI across the whole stack. This plugin provided the number of Rh1- and Rab11-positive structures and overlap (expressed as percentage overlap) of Rh1-positive and Rab11-positive structures for each slice in the z-stack of the ROI. A geometric mean of the various parameters across the z-stack constituted one biological replicate. Two to three ROIs from at least three eyes were used to compute mean ± SEM and make statistical comparisons.

qRT-PCR was performed as described (Spannl et al., 2017). Briefly, RNA extraction was performed using a standard Trizol-based (Invitrogen) procedure with phase separation using chloroform according to the manufacturer’s instructions. Approximately 10 heads or 40–50 dissected eyes per genotype constituted one biological replicate, and at least two to five biological replicates were used. cDNA generation was performed using SuperScript RT kit (Invitrogen) with a starting amount of 1 µg total RNA. qPCR was performed using primers described in Table S4.

Fly eyes were dissected, pooled in groups of 10 in 200 µl isopropanol in 1.5-ml tubes (Eppendorf), and snap-frozen in liquid nitrogen. Sample size consisted of three biological replicates pooled from three to five independent crosses. Tissues were stored at −80°C. Before lipid extraction, samples were thawed and dried in a vacuum centrifuge for 30 min. 140 µl of water and a spatula of 1 mm zirconium beads (BioSpec) were added to the sample. Samples were then subjected to two cycles of bead beating (Qiagen TissueLyzer II) for 1.5 min at 30 Hz and 4°C.

To the homogenate were added 740 µl of a mixture of internal standards dissolved in methyl t-butyl ether/methanol (10:3; vol:vol). The aliquot of internal standards contained 539 pmol of Zymosterol D5, 73 pmol DG-D5 34:0, 138 pmol PC 25:0, 109 pmol LPC 13:0, 67 pmol PS 25:0, 147 pmol PE 25:0, 53 pmol LPE 13:0, 90 pmol PI 25:0, 68 pmol PG 25:0, 102 pmol Cer 30:1:1, and 74 pmol CerPE 29:1:1 (lipids purchased from Avanti Polar Lipids; see abbreviations in Table S5). The mixture was shaken for 2 h at 4°C and centrifuged for 5 min at 13,400 rpm and 4°C. The upper organic phase was collected into two glass vials and dried under vacuum. Before their analysis by MS, lipid extracts were dissolved in isopropanol/methanol/chloroform mixture (4:2:1; vol:vol:vol) with 7.5 mM ammonium formate (mix1) or methanol/chloroform mixture (5:1; vol:vol) with 0.1% trimethylamine (mix2). Shotgun lipidomics analyses were performed on a Q Exactive mass spectrometer (Thermo Fisher Scientific) equipped with a robotic nanoflow ion source TriVersa NanoMate (Advion BioSciences; Schuhmann et al., 2012) using the following acquisition methods: (a) samples dissolved in mix1: broadband FT MS⁺ (range of m/z 350–1,100, target mass resolution R_m/z=200 = 140,000; automatic gain control (AGC) = 3 × 10⁶) followed by targeted HCD FT MS/MS⁺ (precursor isolation width = 1 Th; R_m/z=200 = 140,000, AGC = 2 × 10⁴) under the normalized collision energy (NCE) of 20% and 35%; or (b) samples dissolved in mix1 and mix2: broadband FT MS^– (range of m/z 400–1200, target mass resolution R_m/z=200 = 140,000, AGC = 3 × 10⁶), followed by targeted HCD FT MS/MS^– (precursor isolation width = 1 Th, R_m/z=200 = 140,000; AGC = 2 × 10⁴) with NCE of 23% and 35%.

Lipid species were quantified by HCD FT MS/MS using LipidXplorer software (Herzog et al., 2011; Schuhmann et al., 2017). CerPE quantification relied on HCD FT MS/MS⁺ at NCE 20% producing a lipid class–specific neutral loss fragment of Δm/z =141.0191 (see Fig. S2).

Five fly heads from each genotype were homogenized in 10 µl of 4× SDS-PAGE sample buffer (200 mM Tris-HCl, pH 6.8, 20% glycerol, 8% SDS, 0.04% bromophenol blue, and 400 mM DTT). After dilution with radioimmunoprecipitation assay buffer (150 mM sodium chloride, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, and 50 mM Tris, pH 8), lysates were heated at 37°C for 30 min. Lysates equivalent to 2.5 heads were loaded and run on a 15% acrylamide gel, and the proteins were transferred onto a membrane (Nitrocellulose Blotting Membrane 10600002; GE Healthcare Life Science). Primary antibodies were incubated overnight at 4°C and included anti-Rh1 (4C5; 1:500) from Developmental Studies Hybridoma Bank and anti-α-Tubulin (1:5,000) from Sigma-Aldrich. As secondary antibody, IRDye 800CW goat anti-mouse IgG (1:15,000; LI-COR Biotechnology) was used. The fluorescent signal from the dry membrane was measured using an Odyssey Sa Infrared Imaging System 9260-11P (LI-COR Biotechnology). The intensity of the bands was analyzed using Image Studio v4.0 software. The reported value was obtained following normalization of the intensity values for Rh1 with the corresponding Tubulin intensity values and the number of heads loaded onto the gel.

Statistical comparisons were performed in PrismPad software. Data distribution was assumed to be normal, but this was not formally tested. Statistical significance was determined by ANOVA followed by Tukey test for all comparisons across a set (e.g., lipidomics) or Bonferroni for specific pairwise comparisons (real-time qRT-PCR) with a confidence interval of 95%. When only two conditions were compared with each other, an unpaired, 2-tailed t test was used to determine significance with a confidence interval of 95%. For comparisons of rhabdomere extension, population medians between genotype pairs were compared using the Kolmogorov–Smirnov test with a confidence interval of 95%.

Abbreviations are listed in Table S5.

Fig. S1 shows that the growth of rhabdomeres in the last quarter of pupal development is manifested by an extension and an increase in thickness. Fig. S2 shows the identification of CerPE species by FT MS and HCD FT MS/MS in positive ion mode. Fig. S3 shows the increased intensity of GstD-GFP reporter in crb mutant PRCs. Fig. S4 shows Rh1 localization at steady state. Fig. S5 shows Rh1-Rab11 overlap following BLICS assays. Table S1 describes fly lines. Table S2 lists genotypes. Table S3 lists chemical reagents and antibodies. Table S4 lists primers. Table S5 lists abbreviations. Data S1 contains the lipid species abundance (as mol%) across different alleles and controls (for each condition, there are at least n = 3 biological replicates) used in this study.

Dedicated to the memory of Suzanne Eaton (1959–2019), a pioneer in lipid research in Drosophila.

The authors thank Ian Henry and Gayathri Nadar (Scientific Computing facility) for advice with statistical testing and image analyses, Catrin Hälsig for help with qRT-PCR, Sylke Winkler for sequencing of mutants, Michaela Yuan (Electron Microscopy facility) for help with ultrathin sectioning, Franziska Friedrich (Media Technologies and Outreach) for help with the cartoon in Figure 1, and the Fly facility, the Microarray facility, and the Electron and Light Microscopy facilities, MPI-CBG, for technical support. We thank Uenal Coskun and Michaela Wilsch-Bräuninger for useful discussions.

Work in the A. Shevchenko laboratory was supported by Deutsche Forschungsgemeinschaft, grants TRR83 (Project A17) and FOR2682. This work was supported by the International Lipid Research Center and the National Centre for Biological Sciences-Max Planck Lipid Center and the Max Planck Society.

The authors declare no competing financial interests.

Author contributions: S. Hebbar: conceptualization, data curation, formal analysis, investigation, project administration, resources, visualization, writing–original draft, and writing–review and editing. K. Schuhmann: data curation, formal analysis, investigation, resources, visualization, and writing– review and editing. A. Shevchenko: funding acquisition, resources, supervision, and writing–review and editing. E. Knust: conceptualization, funding acquisition, resources, supervision, writing–original draft, and writing–review and editing.