ABSTRACT
Many membrane receptors activate phospholipase C (PLC) during signalling, triggering changes in the levels of several plasma membrane lipids including phosphatidylinositol (PtdIns), phosphatidic acid (PtdOH) and phosphatidylinositol 4,5-bisphosphate [PtdIns(4,5)P2]. It is widely believed that exchange of lipids between the plasma membrane and endoplasmic reticulum (ER) is required to restore lipid homeostasis during PLC signalling, yet the mechanism remains unresolved. RDGBα (hereafter RDGB) is a multi-domain protein with a PtdIns transfer protein (PITP) domain (RDGB-PITPd). We find that, in vitro, the RDGB-PITPd binds and transfers both PtdOH and PtdIns. In Drosophila photoreceptors, which experience high rates of PLC activity, RDGB function is essential for phototransduction. We show that binding of PtdIns to RDGB-PITPd is essential for normal phototransduction; however, this property is insufficient to explain the in vivo function because another Drosophila PITP (encoded by vib) that also binds PtdIns cannot rescue the phenotypes of RDGB deletion. In RDGB mutants, PtdIns(4,5)P2 resynthesis at the plasma membrane following PLC activation is delayed and PtdOH levels elevate. Thus RDGB couples the turnover of both PtdIns and PtdOH, key lipid intermediates during G-protein-coupled PtdIns(4,5)P2 turnover.
INTRODUCTION
Eukaryotic cells are composed of membrane bound subcellular compartments each of which has a unique protein and lipid composition that is central to its function. Despite this, when cells respond to external stimuli, the chemical identity of the plasma membrane is altered. Cell surface receptors that transduce ligand binding by activating phospholipase C (PLC) enzymes exemplify this problem. Many clinically important receptors including receptor tyrosine kinases (e.g. EGFR, T-cell receptor) and G-protein-coupled receptors (GPCRs; e.g. muscarinic acetylcholine, metabotropic glutamate receptor) utilise PLC-based signalling pathways. Hence, understanding the regulation of PLC signalling might provide important insights into the biology and treatment of diseases where such receptor signalling is implicated.
When PLC is activated, the plasma membrane experiences changes in lipid composition. PLC hydrolyses phosphatidylinositol 4,5-bisphosphate [PtdIns(4,5)P2], a key plasma membrane lipid, to generate diacylglycerol (DAG), which is then converted into phosphatidic acid (PtdOH); thus the levels of DAG and PtdOH at the plasma membrane rise. Remarkably, even when cells experience high PLC activity, PtdIns(4,5)P2 levels at the plasma membrane remain relatively stable (Balla et al., 2008; Hardie et al., 2001; Willars et al., 1998). This is achieved by restoring its levels by the sequential phosphorylation of phosphatidylinositol (PtdIns) at position 4 and 5 by phosphatidylinositol-4-kinase (PI4K) and phosphatidylinositol-4-phosphate-5-kinase (PIP5K) (Balla et al., 2008; Wang et al., 2008). However, in order to do so the plasma membrane must maintain adequate supply of PtdIns that can be used by PI4K. Thus, during PLC signalling, the plasma membrane faces two challenges, namely to replenish the levels of PtdIns as well as remove PtdOH that accumulates downstream of PLC activation.
PtdOH and PtdIns are both lipids, are thus anchored to membranes and are incapable of diffusing to and from the plasma membrane. How do cells control the levels of these two lipids at the plasma membrane? PtdIns is synthesised at the endoplasmic reticulum (ER) by PtdIns synthase. By contrast, PtdOH is generated by the action of DAG kinase (DGK) but can also be converted back into DAG by a type II PtdOH phosphatase (Garcia-Murillas et al., 2006; Kai et al., 1996). PtdOH must also move from the plasma membrane to the ER where it is converted into cytidine diphosphate diacylglycerol (CDP-DAG) by CDP-DAG synthase (CDS). Thus, during PLC signalling, cells need a strategy by which PtdOH can be transported from the plasma membrane to the ER and, conversely, one that allows PtdIns to be transported in the reverse direction, namely from the ER to the plasma membrane (Michell, 1975).
Lipids can be transported to and from the plasma membrane by multiple mechanisms. Vesicular transport is a possibility, although the time scale of such transport is not compatible with the high rate at which PLC utilises PtdIns(4,5)P2 (Liu et al., 2009; Raghu et al., 2012). An alternative mechanism is the exchange of lipids by lipid transfer proteins, and it has been suggested that this might occur more effectively at membrane contact sites (Chang et al., 2013; Levine, 2004). Several studies have suggested that PtdIns might be transported by PtdIns transfer proteins (PITPs), which are able to transfer PtdIns between membranes in vitro (Cockcroft and Carvou, 2007; Helmkamp, 1990; Phillips et al., 2006; Wirtz, 1991). Using cellular reconstitution assays PITPs have been defined as factors required to support sustained receptor-activated PLC signalling in cells (Kauffmann-Zeh et al., 1995; Thomas et al., 1993). PITPs are conserved in eukaryotes (reviewed in Cockcroft and Garner, 2011; Routt and Bankaitis, 2004) with two classes of metazoan PITPs being identified, class I and II. Although all PITPs contain a PITP domain (PITPd), in class IIA PITPs, several additional domains and motifs are present. Class II PITPs are often referred to as RDGB proteins because the founding member of this class is the retinal degeneration B (RDGBα; CG11111, hereafter RDGB) protein in Drosophila (Harris and Stark, 1977; Hotta and Benzer, 1970). By contrast there is limited information on PtdOH transfer proteins. In yeast, the protein Ups1 in complex with Mdm35 has been shown to function in PtdOH transfer at mitochondrial membranes (Connerth et al., 2012). Additionally, mammalian RDGBβ (also known as PITPNC1), a member of the class II PITP family, has been shown to bind and transfer PtdOH in vitro (Garner et al., 2012).
Although several studies have implicated PITPs in cellular functions including cytokinesis, neurite outgrowth and membrane trafficking (Bankaitis et al., 2005; Carvou et al., 2010; Cosker et al., 2008; Giansanti et al., 2006), the biochemical basis of PITP function in vivo remains unclear. Much attention has focussed on the ability of the PITP domain to bind and transfer PtdIns and the importance of this activity in supporting PLC-based signalling in vivo. Although it has been assumed that these are connected, some studies have refuted this idea. It has been suggested that human RDGBα (Nir2, also known as PITPNM1) is required to maintain DAG levels at the Golgi by modulating the CDP-choline pathway (Litvak et al., 2005) although a more recent study in human cells has shown that Nir2 functions at the plasma membrane (Chang et al., 2013). A previous study of the PITP domain of Drosophila RDGB has also suggested a lack of correlation between PtdIns binding and transfer in vitro and its function in vivo (Milligan et al., 1997). Thus, the biochemical function of the PITP domain and its role in supporting PLC-mediated signalling remains disputed.
Sensory transduction in Drosophila photoreceptors has been a useful model for analysis of PLC signalling in vivo (Hardie and Raghu, 2001). In these cells, photon absorption by the GPCR rhodopsin is transduced into electrical activity by G-protein-coupled PLCβ-mediated PtdIns(4,5)P2 hydrolysis (Fig. 1A) (Raghu and Hardie, 2009; Raghu et al., 2012). Photoreceptors have high basal PLC activity and, during bright light illumination, PLCβ activity and consequent PtdIns(4,5)P2 hydrolysis are substantially enhanced (Hardie and Raghu, 2001). Thus, in order to maintain their ability to respond to continuous illumination, PtdIns(4,5)P2 needs to be resynthesised to match its consumption by PLCβ activity. To date, there is no report identifying the lipid phosphatase activity that acts on rhabdomeral PtdIns(4,5)P2 during PLC signalling. However, consistent with the requirement of PITPs in PLC signalling, Drosophila photoreceptors are enriched for the PITP-domain-containing protein RDGB (Vihtelic et al., 1993).
rdgB mutants exhibit defective light responses and light-dependent retinal degeneration (reviewed in Trivedi and Padinjat, 2007). However, the molecular basis for the requirement of RDGB in these cells remains unresolved; a previous study [using Kir channels as biosensors to report PtdIns(4,5)P2 levels] reported delayed recovery of PtdIns(4,5)P2 levels in rdgB mutant photoreceptors following light stimulation (Hardie et al., 2001). It has also been reported that the PITP domain of rdgB is sufficient to rescue mutant phenotypes, yet the same study reported that a mutant version of this PITP domain (T59E) was able to transfer PtdIns in vitro but was unable to rescue function (Milligan et al., 1997). Here, we have carried out a comprehensive analysis of the lipid binding and transfer properties of the PITP domain of RDGB (RDGB-PITPd) in vitro and compared these with its ability to rescue rdgB mutant phenotypes in vivo. We find that RDGB-PITPd has distinctive lipid binding and transfer properties in vitro and regulates levels of two key lipids, namely PtdOH and PtdIns(4,5)P2, during PLC signalling.
RESULTS
RDGB is required to support normal phototransduction
Although rdgB mutants show both light-dependent retinal degeneration and defective electrical responses to light (Harris and Stark, 1977), the relationship between these two phenotypes is unclear, given that structural abnormalities in a degenerating photoreceptor are likely to result in an abnormal light response. This has raised questions about the independent role of rdgB, if any, in supporting phototransduction. We used optical imaging to visualise rhabdomere integrity in intact retinae (Franceschini and Kirschfeld, 1971). Two alleles of rdgB, namely rdgB2 (a protein null allele) (Vihtelic et al., 1991) and rdgB9 (a severe hypomorph), were tested. As previously reported, peripheral photoreceptors underwent complete loss of rhabdomeres by day 5, whereas the central UV-sensitive photoreceptor was spared. The timecourse of degeneration was similar in both rdgB2 and rdgB9 (Fig. 1B). This degeneration could be rescued by expression of either the full-length wild-type rdgB transgene or the PITP domain of rdgB (rdgB-PITPd) (Fig. 2A,B). These findings were confirmed by examination of photoreceptor ultrastructure in sections of fixed samples (supplementary material Fig. S1A,B). We studied the response of photoreceptors to light using electrical recordings from the eye (by electroretinogram, ERG). There have been several reports (Harris and Stark, 1977; Rubinstein et al., 1989) demonstrating negligible electrical responses to light in rdgB mutants. However, most studies to date have analysed flies older than 2–3 days, by which time substantial retinal degeneration has set in (Fig. 1B). Therefore we tested light responses of young flies reared in the dark prior to the onset of any detectable retinal degeneration (age ≤20 h posteclosion). Under these conditions, rdgB9 showed a reduced ERG amplitude (3±0.5 mV compared to 9±0.5 mV in wild type) (Fig. 1C). When exposed to a light–dark cycle for 1 day posteclosion, light responses from rdgB9 were further reduced (Fig. 1C,D); under equivalent conditions wild-type responses were unaffected. This reduction in ERG amplitude could be rescued by either the full-length wild-type rdgB transgene or the PITP domain of rdgB alone (Fig. 2C,D). We also studied the sensitivity of rdgB9 photoreceptors to light by recording the response to increasing intensities of light stimulus. Compared to wild-type controls of matched eye colour, rdgB9 photoreceptors showed a substantially reduced sensitivity to light (Fig. 1E) and this could be rescued by a wild-type rdgB transgene (Fig. 2E). These findings demonstrate the existence of phototransduction defects in rdgB9 prior to the onset of any detectable structural abnormality.
RDGB function is required to support light-activated PtdIns(4,5)P2 turnover
The protein encoded by rdgB has a PITP domain (Fig. 1A) suggesting that during phototransduction RDGB might regulate PtdIns metabolism including PtdIns(4,5)P2 resynthesis at the plasma membrane. To monitor changes in plasma membrane PtdIns(4,5)P2 levels during PLC stimulation we used a well-established probe (Várnai and Balla, 1998), the PH domain of PLCδ (PH-PLCδ) fused to GFP (PH-PLCδ–GFP) (Chakrabarti et al., 2015; Sengupta et al., 2013). In Drosophila photoreceptors, this probe is uniformly distributed on the plasma membrane (supplementary material Fig. S2F). Given that the phototransduction machinery is limited to the apical plasma membrane (rhabdomeres) and we were interested in monitoring changes in PtdIns(4,5)P2 levels at the rhabdomere, we exploited the technique of pseudopupil imaging, which reports molecules present in the rhabdomere (Franceschini and Kirschfeld, 1971) (see schematic of protocol used in Fig. 1Fi). Under resting conditions, the probe binds to PtdIns(4,5)P2 and hence localises to the microvillar plasma membrane giving rise to a fluorescent pseudopupil (Fig. 1Fii, panel a). Following a flash of blue light that triggers PLC activity, microvillar PtdIns(4,5)P2 is depleted and the probe is displaced from the plasma membrane and diffuses out of the microvilli, resulting in loss of the fluorescent pseudopupil (Fig. 1Fii, panel b). (The central photoreceptor, R7 is not stimulated by blue light and hence retains the probe.) In the dark, presumably following PtdIns(4,5)P2 resynthesis, the probe relocalises to the microvillar plasma membrane resulting in the recovery of the fluorescent pseudopupil. Increasing durations of red light illumination (that converts metarhodopsin into rhodopsin, thus terminating PLC activity) following each image acquisition results in progressively faster recovery of fluorescence (Fig. 1Fii, panels c–i). In 1-day-old rdgB9 flies, the fluorescent pseudopupil is diffuse and lower in intensity to begin with [despite western blotting showing equivalent level of probe expression (supplementary material Fig. S1C)] and following stimulation with blue light recovers with a much slower timecourse (Fig. 1G). These observations suggest that RDGB is required both to maintain basal PtdIns(4,5)P2 levels at the micovillar plasma membrane and to support PtdIns(4,5)P2 resynthesis following light-induced PLC activation.
We tested the ability of RDGB to rescue the PtdIns(4,5)P2 resynthesis defect of rdgB9 photoreceptors. Expression of RDGB in rdgB9 photoreceptors was able to rescue both the basal pseudopupil fluorescence (supplementary material Fig. S1D,E) as well as allowing it to recover back to 100% albeit with delayed kinetics (Fig. 2F,G); this delay was presumably due to a lower level of expression of RDGB from the transgenic construct (supplementary material Fig. S2A).
The PITP domain of RDGB binds and transfers both PtdIns and PtdOH in vitro
We expressed and purified RDGB-PITPd from E. coli and examined its lipid binding and transfer activities in vitro in comparison to the well-characterised human PITPα (hPITPα, also known as PITPNA) protein. Unlike hPITPα, which can bind either a molecule of phosphatidylcholine (PtdCho) or PtdIns, RDGB-PITPd showed limited PtdCho binding (Fig. 3A,B). Further, as recently reported for human RDGBβ (Garner et al., 2012), we found that RDGB-PITPd was able to bind PtdOH (Fig. 3D). These binding studies were conducted in permeabilised HL60 cells, where the recombinant proteins were exposed to the cellular phospholipids present at their physiological levels. PtdOH represents a very small fraction (<1–2%) of the total phospholipids (Fig. 3C). Nonetheless, RDGB-PITPd bound a substantial amount of PtdOH. PtdCho is also bound by RDGB-PITPd to a similar level, but considering that in cells PtdCho represents nearly 50% of the total lipids, the affinity for PtdCho is likely to be significantly lower than that for PtdOH. PtdIns, which represents ∼5–8% of the total cellular lipids, was the dominant lipid bound to RDGB-PITPd (Fig. 3D). In contrast, hPITPα bound PtdIns and PtdCho equivalently, and insignificant binding to PtdOH was observed (Fig. 3D). When the binding of each lipid to the PITP was normalised to the total amount of lipid available, it was clear that both hPITPα and RDGB-PITPd had a higher affinity for PtdIns compared to PtdOH or PtdCho (supplementary material Fig. S2E). The most important difference between the class I PITPα and class II RDGB-PITPd is that the class I protein has a preference for binding PtdCho whereas RDGB-PITPd has a strong preference for PtdOH (supplementary material Fig. S2E).
We also performed in vitro transfer assays and compared the activity of RDGB-PITPd with that of hPITPα (Ségui et al., 2002; Tilley et al., 2004). Under equivalent conditions, both RDGB-PITPd and hPITPα showed PtdIns transfer activity (Fig. 3E); compared to hPITPα, RDGB-PITPd was less active. Whereas hPITPα showed substantial PtdCho transfer activity, RDGB-PITPd showed very little transfer activity towards PtdCho (Fig. 3F). In sharp contrast, RDGB-PITPd showed substantial PtdOH transfer activity, whereas hPITPα was unable to transfer PtdOH (Fig. 3G). Taken together, these data show that, in vitro, RDGB-PITPd can bind and transfer both PtdIns and PtdOH.
Molecular basis of PtdIns binding and transfer by the PITP domain in vitro
A conserved property of the PITP domain (across all classes) is the ability to bind and transfer PtdIns in vitro. The structural basis of PtdIns binding has previously been investigated for hPITPα (Tilley et al., 2004). The crystal structure of PITPα reveals the identity of those amino acid residues in this protein that directly co-ordinate with the inositol ring of PtdIns and appear to be essential for both in vitro and in vivo function, hereafter called PtdIns-binding residues (PIBRs) (Tilley et al., 2004). These PIBRs are conserved in nearly every metazoan PITP domain (that has been sequenced) regardless of the class of PITPs to which the parent protein belongs to, presumably reflecting an evolutionarily conserved role in the function of this domain (Tilley et al., 2004) (Fig. 4A). We studied the requirement of three of these PIBRs, namely T59, K61 and N90 (mouse PITPα numbering used throughout), for the binding of PtdIns to RDGB-PITPd. The single point mutants K61A, N90F and, in the case of T59, two versions T59A and T59E were generated in RDGB-PITPd, then expressed in bacteria and purified (Fig. 4B), and their biochemical properties were studied in vitro. We found that all the four mutant proteins (RDGB-PITPdT59A, RDGB-PITPdT59E, RDGB-PITPdK61A and RDGB-PITPdN90F) showed a dramatically diminished ability to bind PtdIns compared to wild-type RDGB-PITPd (Fig. 4C,D). The T59A, T59E and N90F mutants did not show altered PtdCho binding, whereas in the case of K61A, the lowered PtdIns binding was associated with an increase in PtdCho binding. Notably T59A bound more PtdIns than T59E (Fig. 4C,D). In all mutant versions of RDGB-PITPd, PtdOH binding was reduced (Fig. 4C,D).
We also studied the transfer activity of each PIBR mutant using an in vitro PtdIns transfer assay and found that, with the exception of T59A, all other mutant versions of RDGB-PITPd showed essentially no PtdIns transfer activity (Fig. 4E). T59A retained some residual PtdIns transfer activity, although this was substantially reduced compared to the wild-type protein. These results differ from those reported before for the PITP domain of RDGB. RDGB-PITPdT59A has previously been reported to be null for PtdIns transfer whereas RDGB-PITPdT59E was reported to have activity comparable to wild type (Milligan et al., 1997). Our results presented here are in accordance with what has been reported for PITPα-T59A (Tilley et al., 2004). The PtdOH transfer activity of the mutant proteins was also analysed and was largely unaffected at saturating concentrations of protein (supplementary material Fig. S2B); at very low concentrations, PtdOH transfer by the mutants was marginally reduced. We noted a discrepancy between the lipid binding and transfer activities described here. The biochemical assays for monitoring lipid binding and transfer were undertaken in different membrane environments and therefore represent different properties of the protein. Lipid binding was carried out in permeabilised cells, whereas lipid transfer monitored the movement of radiolabelled lipid from a donor membrane (microsomes or liposomes) to an acceptor membrane (liposomes or mitochondria). Thus, PtdOH binding is modest for RDGB-PITPd, but PtdOH transfer is substantial (Fig. 3). In the PIBR mutants, PtdOH binding is highly reduced although PtdOH transfer is only marginally affected. A small amount of PtdCho is bound by RDGB-PITPd but is unable to transfer PtdCho. This discrepancy has been observed previously for the WW/AA mutants of PITPα (Tilley et al., 2004) and is also evident for the analogous YW/AA mutants described here (see Fig. 8A–F).
PtdIns-binding residues are essential to support RDGB function in vivo
We tested the requirement of PIBRs in supporting the function of RDGB in vivo. Transgenic flies were generated that allowed the expression of either wild-type RDGB-PITPd or each of the PIBR mutants of RDGB-PITPd in photoreceptors. Transgenic lines that showed equivalent level of protein expression (supplementary material Fig. S2C) were used in the following experiments. We studied two phenotypes of rdgB, namely retinal degeneration and the electrical response to light. As previously indicated (Milligan et al., 1997) and in this study (Fig. 2A,D), RDGB-PITPd was able to rescue the retinal degeneration phenotype of rdgB9. By contrast, none of the mutant versions (RDGB-PITPdT59A, RDGB-PITPdT59E, RDGB-PITPdK61A and RDGB-PITPdN90F) were able to rescue retinal degeneration in rdgB9 (Fig. 5A,B). In the case of RDGB-PITPdT59A, the rate of retinal degeneration was somewhat slower compared to rdgB9.
We also tested the requirement of PIBRs to support a normal electrical response to light. ERGs were recorded from rdgB9 reconstituted with either wild-type RDGB-PITPd or a mutant version of each of the PIBRs. We found that although RDGB-PITPd was able to rescue the light response of rdgB9 to near wild-type levels, none of the PIBR mutant versions were able to do so (Fig. 5C). Unlike the case with the retinal degeneration phenotype, RDGB-PITPdT59A was not able to improve the light response in rdgB9. These results suggest that binding of PtdIns to the RDGB-PITPd is essential to support a normal response to light in photoreceptors.
In mammalian cells, PITPα has been shown to be required to support G-protein-coupled PtdIns(4,5)P2 turnover in permeabilised HL60 cells (Thomas et al., 1993). We studied this with respect to the function of RDGB-PITPd. Using the permeabilised cell assay, we found that RDGB-PITPd was able to stimulate the production of inositol 1,4,5-trisphosphate (IP3) in response to GTPγS stimulation just as well as PITPα (Fig. 5D). However, in this assay, RDGB-PITPdT59E and RDGB-PITPdK61A were not able to support the production of IP3 (Fig. 5D). RDGB-PITPdT59A retained partial activity, in keeping with the PtdIns transfer activity observed for this mutant (Fig. 4E). These observations suggest that, similar to PITPα (Tilley et al., 2004), the PtdIns-binding activity of RDGB-PITPd is required to support G-protein-stimulated inositol lipid turnover.
Given that RDGB is a large protein (160 kDa) compared to its PITP domain alone (35 kDa), we also generated one of the PIBR mutant forms (K61A) in the context of full-length rdgB (rdgBK61A) and tested its ability to rescue rdgB9 phenotypes. rdgBK61A showed similar behaviour to rdgB-PITPdK61A and could not rescue the retinal degeneration and ERG phenotypes of rdgB9 (Fig. 6A,B). We also tested the ability of rdgBK61A to rescue the PtdIns(4,5)P2 resynthesis defect of rdgB9. Under conditions where RDGB was able to rescue the PtdIns(4,5)P2 resynthesis defect of rdgB9, reconstitution of rdgB9 with rdgBK61A was unable to reverse the delayed kinetics of PtdIns(4,5)P2 turnover (Fig. 6C,D). Collectively these results imply that PtdIns binding is an essential requirement of for the in vivo function of RDGB.
A Drosophila class I PITP cannot substitute for RDGB function in vivo
Class I and class II PITPs show distinct lipid binding and transfer properties (Fig. 3), but a conserved feature is the ability to bind and transfer PtdIns in vitro. Given that we found that PtdIns binding was essential for RDGB function in vivo, we hypothesised that if this biochemical property was the only determinant of RDGB function, then expression of another PITP should rescue rdgB phenotypes. Remarkably, it has been previously reported that mammalian PITPα or a PITPα-RDGB chimera is unable to restore the light response or rescue retinal degeneration when expressed in rdgB2 flies (Milligan et al., 1997). However, it was not clear whether the lack of rescue by mammalian PITPα in that study was due to evolutionary diversification of PITPα between Drosophila and mammals. To resolve this issue we tested the ability of a Drosophila class I PITP to rescue the phenotypes of rdgB9. The Drosophila genome encodes only one member of class I PITP, Dm-PITPα (i.e. vib, CG5269) (Gatt and Glover, 2006; Giansanti et al., 2006). Endogenous Dm-PITPα is expressed in adult heads at very low levels, so we overexpressed it using the ubiquitin promoter (Fig. 7A). Under these conditions, Dm-PITPα was unable to rescue retinal degeneration and the light response of rdgB9 (Fig. 7B–D). This observation strongly suggests that there are likely to be other biochemical properties in addition to PtdIns binding and transfer that might be unique to RDGB-PITPd function.
An RDGB mutant that binds PtdIns but fails to transfer in vitro cannot support function in vivo
To date there is no agreement on the contribution of PtdIns binding and transfer activity for the in vivo function of PITPs. Although PIBR mutants do not transfer PtdIns in vitro, presumably this reflects their inability to bind PtdIns. Thus, a mutant that retains PtdIns binding but cannot perform PtdIns transfer activity in vitro could help address the importance of transfer activity in vivo. It has been proposed that PITP function requires the essential step of membrane docking, and two residues present within the PITP domain, namely W203 and W204 (mice PITPα numbering), have been implicated in this event (Schouten et al., 2002; Shadan et al., 2008; Tilley et al., 2004; van Tiel et al., 2002). Sequence analysis showed that this ‘tryptophan motif’ is conserved across all classes of PITPs with the minor difference that all class II PITPs have a tyrosine in place of the first tryptophan residue of the motif. In case of RDGB, this YW motif is present at position 210 and 211. We mutated both these residues to alanine (YW/AA) and tested the ability of the recombinant protein to bind and transfer lipid in vitro. We found that RDGB-PITPdYW/AA was able to bind PtdIns at levels comparable to RDGB-PITPd (Fig. 8A). However, the PtdIns transfer activity of RDGB-PITPdYW/AA was profoundly reduced (Fig. 8B). These in vitro properties of the YW/AA mutant are similar to those previously reported for the WW/AA mutants both in case of PITPα and for PITPβ (also known as PITPNB) (Shadan et al., 2008; Tilley et al., 2004). Thus, RDGB-PITPdYW/AA represents a protein that can bind PtdIns but not transfer it in vitro.
To test the importance of PtdIns transfer in vivo we generated transgenic flies and tested the ability of RDGBYW/AA to rescue rdgB9. We found that RDGBYW/AA was neither able to rescue the light response nor was able to suppress retinal degeneration in rdgB9 (Fig. 8C–F). These observations demonstrate that a RDGBYW/AA has an intrinsic ability to bind PtdIns but is unable to transfer in vitro and cannot rescue in vivo function.
RDGB function is required to regulate PtdOH levels during phototransduction
Given that Dm-PITPα, which can also bind and transfer PtdIns, was unable to rescue rdgB9, we hypothesised that the additional and unique biochemical activity of RDGB-PITPd that we found, namely PtdOH binding and transfer, might also be required to support in vivo function. If PtdOH binding and transfer activity of RDGB were important in vivo, levels of PtdOH and its immediate precursor DAG might be altered in rdgB9 photoreceptors, especially in response to enhanced PLC stimulation. To test this we used high-resolution lipid mass-spectrometry to measure PtdOH and DAG levels from head extracts. In dark-adapted photoreceptors, PtdOH levels were almost equal between wild-type and rdgB9 (Fig. 8G, also see supplementary materials Fig. S3). Following illumination with a brief (1 min) flash of bright light, PtdOH and DAG levels rose in controls, reflecting the production of these lipids by the sequential action of PLC and DGK (supplementary material Fig. S4). However the light-induced elevation in PtdOH levels seen in rdgB9 was substantially greater than in wild type (Fig. 8G). These observations suggest that RDGB function is required to regulate the turnover of PtdOH generated during phototransduction.
DISCUSSION
To maintain the sensitivity of receptor-activated PLC signalling cascades, PLC must have adequate levels of its substrate. PtdIns(4,5)P2, the substrate of PLC, is a low abundance lipid (there are ∼4000 PtdIns(4,5)P2 molecules/μm2 on the inner leaflet of the plasma membrane, i.e. 0.1–5% of inner bilayer lipids; Xu et al., 2003), and maintaining stable levels of PtdIns(4,5)P2 is especially challenging in cells that experience high rates of PLC activity; in the absence of tight regulation, membranes are likely to show depletion of PtdIns(4,5)P2. However, multiple studies monitoring PtdIns(4,5)P2 levels have found that the levels of this lipid are quite stable and any drops in its levels during ongoing agonist stimulation are transient (Balla et al., 2008; Hardie et al., 2001; Várnai and Balla, 1998; Willars et al., 1998). PtdIns(4,5)P2 hydrolysis triggers a sequence of five reactions collectively referred to as the PtdIns(4,5)P2 cycle (Fig. 1A) that culminates in the resynthesis of this lipid. In order to keep the PtdIns(4,5)P2 cycle running, two key lipid intermediates of the PtdIns(4,5)P2 cycle, PtdOH and PtdIns need to be exchanged between the plasma membrane and the ER. In the present study, we define a single protein RDGB whose PITP domain possesses the biochemical activity to bind and transfer both PtdOH and PtdIns in vitro. RDGB therefore represents a candidate that might be able to exchange lipid intermediates between the plasma membrane and ER and thus keep the PtdIns(4,5)P2 cycle running. In support of this idea, we find that rdgB mutants show delayed kinetics of PtdIns(4,5)P2 resynthesis following PLC activation and this defect is not rescued by the PtdIns-binding-deficient mutants, thus implying PtdIns binding to PITPd is crucial (Fig. 6C,D). A recent study has shown that in Drosophila photoreceptors, loss of PIP5K activity slows the kinetics of PtdIns(4,5)P2 resynthesis (Chakrabarti et al., 2015). Additionally, loss of Drosophila PIP5K results in a further reduction of the residual light response in rdgB9. These findings are consistent with a model where RDGB function is required to support the activity of PIP5K to generate PtdIns(4,5)P2 during Drosophila phototransduction. We found a clear correlation between the ability of RDGB to bind PtdIns in vitro and its ability to rescue rdgB9, strongly suggesting that PtdIns binding to RDGB-PITPd is relevant for function in vivo. Furthermore, our finding that the RDGBYW/AA protein (which can bind PtdIns but cannot transfer it in vitro) was unable to rescue rdgB9, suggests that PtdIns transfer is very likely an important function of RDGB in vivo.
A second crucial step in the PtdIns(4,5)P2 cycle that requires lipid exchange is the transfer of PtdOH to the ER. To date there are few reports of proteins that can transfer PtdOH between membranes. In yeast, a complex of Ups1 and Mdm35 is able to transfer PtdOH across the mitochondrial space (Connerth et al., 2012) and Garner et al. has reported that PITPNC1, a related protein of the RDGB family, is able to bind and transfer PtdOH in mammalian cells (Garner et al., 2012). In this study, we found that RDGB-PITPd is able to bind and transfer PtdOH in vitro. Importantly, we found that following exposure to light (which activates PLC), rdgB9showed an enhanced elevation in PtdOH levels compared to wild-type flies, suggesting that the RDGB protein is required to facilitate PtdOH metabolism during phototransduction. Live imaging of PtdOH levels during phototransduction might help strengthen this idea but is presently limited by the absence of a suitable PtdOH probe. Hence, the specific mechanism by which RDGB participates in PtdOH metabolism is unclear. Our finding of a PtdOH transfer activity for RDGB-PITPd in vitro suggests that this protein could facilitate transfer of PtdOH from the plasma membrane to the ER where it would be metabolised by CDS into CDP-DAG. In rdgB9 photoreceptors, the absence of this activity could result in a reduced availability of PtdOH for CDS to metabolise into CDP-DAG, thus explaining the enhanced elevation of PtdOH levels following illumination. In a previous study, Drosophila DGK (rdgA), which generates PtdOH from DAG has been shown to be localised to the submicrovillar cisternae (SMC) (Masai et al., 1997) raising the question of whether PtdOH transfer activity is required. RDGA is a large protein (∼1457 amino acids) and it is possible that it can act in trans across the 10-nm cytoplasmic gap between the microvillar plasma membrane and the SMC to access its substrate (DAG) produced at the plasma membrane. Such a mechanism by which an enzyme acts in trans to access a lipid substrate on another membrane has been shown for Sac1 at membrane contact sites in yeast (Stefan et al., 2011).
PITPα is a protein with PtdIns and PtdCho transfer activity. Although PtdIns binding and transfer activity is conserved between PITPα and RDGB-PITPd, PtdOH binding and transfer is unique to RDGB-PITPd. This biochemical difference might represent the mechanistic basis of the inability of Dm-PITPα to rescue rdgB9 phenotypes in vivo. It has been reported that one of the mammalian RDGB homologs, Nir2, can rescue the phenotype of rdgB9 (Chang et al., 1997). This observation is likely to be a reflection of the importance of PtdOH-binding activity of RDGB. Consistent with this model, the PITP domain of the mammalian ortholog of RDGB, Nir2 has been shown to bind PtdOH (Garner et al., 2012) and also transfer PtdOH in vitro (S.C., unpublished data). Furthermore, a distinct member of the mammalian class II family, RDGBβ, has been shown to bind and transfer PtdOH (Garner et al., 2012). Thus, class II PITPs represent PtdIns and PtdOH transfer proteins, in contrast to class I PITPs that are PtdIns and PtdCho transfer proteins.
What might be the role of RDGB in the subcellular context of photoreceptor function? Earlier work has shown that RDGB is localised to a small specialised sub-compartment of the ER, the SMC present at the base of the microvillar plasma membrane (Vihtelic et al., 1993). The SMC is separated from the plasma membrane by a cytoplasmic gap of ∼10 nm. Sequence analysis of RDGB does not reveal any possible transmembrane segments and we found that consistent with what has been reported for its mammalian ortholog Nir2 (Litvak et al., 2002), RDGB is a membrane-associated protein rather than a membrane integral protein (supplementary material Fig. S2D). Given that RDGB has an FFAT motif, it is likely that this motif is bound to VAP proteins [vesicle-associated membrane protein (VAMP)-associated protein] present on the surface of SMC thereby anchoring it to the SMC. Conceptually, the spatial organisation of the SMC and the microvillar plasma membrane represents an example of a plasma-membrane–ER contact site (Levine, 2004) at which exchange of material might occur. The known localisation of RDGB to this site and the observation that RDGB-PITPd binds both PtdIns and PtdOH raises the possibility that RDGB-PITPd might be a key requirement for these lipids to move across the plasma membrane and SMC (Fig. 8H). Such movement of PtdOH and PtdIns would be required to prevent rundown of the PtdIns(4,5)P2 cycle. Alternatively, in this subcellular setting, RDGB-PITPd might sample the plasma membrane and its binding to PtdOH and/or PtdIns might transmit information to the ER machinery to stimulate PtdIns synthesis following PtdIns(4,5)P2 hydrolysis. Given that the enzymes involved in supporting Drosophila phototransduction are present across two distinct compartments (i.e. the plasma membrane and SMC) it is likely that RDGB acts as a sensor communicating the status of PtdIns(4,5)P2 turnover at the microvillar plasma membrane to the ER through its ability to bind PtdIns and/or PtdOH. In summary, our study demonstrates that the PITP domain of RDGB is important to support PtdIns(4,5)P2 resynthesis as well as to regulate PtdOH levels at the photoreceptor plasma membrane during G-protein-coupled PLC activation. The design of a protein such as RDGB that can bind and transfer both PtdOH and PtdIns might represent an elegant solution to monitor and/or exchange two key metabolic intermediates required to keep the PtdIns(4,5)P2 cycle running.
MATERIALS AND METHODS
Fly stocks
Flies (Drosophila melanogaster) were grown on standard corn meal medium with 1.5% yeast and were reared at 25°C in incubators with no internal illumination. Ub-vib-expressing flies were obtained from David Glover, University of Cambridge, UK.
Generation of point mutants and transgenic flies
PtdIns-binding point mutants as well as the YW/AA mutation were generated in rdgB-PITPd by site-directed mutagenesis. The entire open reading frame was sequence-verified post-mutagenesis. Transgenic flies were generated (Rubin and Spradling, 1982) and lines showing equivalent expression levels were used in experiments.
Protein expression for biochemical assays
The Drosophila RDGB-PITPd cDNA coding for the N-terminal PITP domain (1–281 amino acids of GI no. 15291155) was cloned into pRSET-C introducing an N-terminal His tag. Proteins were expressed and purified as described previously (Garner et al., 2012).
Optical neutralisation
Optical neutralisation was performed as described previously (Franceschini and Kirschfeld, 1971). A total of 50 ommatidia were scored across five different flies. For all experiments flies were reared in the dark and transferred to 12-h-light–12-h-dark regime post eclosion.
Electrophysiology
External electrical recordings were performed as described previously (Chakrabarti et al., 2015).
Live pseudopupil imaging
Pseudopupil imaging was performed at 16× magnification on flies expressing a single copy of PH-PLCδ–GFP driven by the trp promoter. The protocol is as described previously (Chakrabarti et al., 2015). For quantification of recovery, the fluorescence value of image ‘b’ (see Fig. 1Fi) was subtracted from all images and the fluorescence value of image ‘a’ was set as 100. The fluorescence value of subsequent images (i.e. b–i) were normalised to the value of ‘a’.
Lipid transfer assays for PtdIns, PtdCho and PtdOH
Transfer assays utilised an appropriately radiolabelled donor compartment and an unlabelled acceptor compartment as described previously (Garner et al., 2012). To monitor PtdOH transfer, liposomes (PtdCho:PtdOH, 98:2) containing radiolabelled PtdOH were used as a donor and mitochondria were used as an acceptor. After incubation at 37°C for 30 min in the presence of the transfer proteins, mitochondria were recovered by centrifugation (10,000 g for 10 min at 4°C). The supernatant was removed and the mitochondrial pellet was resuspended in 500 µl sucrose-EDTA (1 mM EDTA, 0.25 M sucrose). The sample was transferred to a new eppendorf tube and layered on a cushion of 14.3% sucrose solution. After centrifugation of the mitochondria through the sucrose cushion, the pellet was solubilised with 10% SDS and the samples incubated at 95°C for 5 min. The radioactivity in the solubilised samples was monitored by scintillation counting. The PtdOH transfer was expressed as a percentage of input counts in the assay after deducting the background counts.
Binding of cellular lipids by RDGB proteins
Association of cellular lipids with the RDGB proteins was analysed in HL60 cells exactly as described previously (Ségui et al., 2002; Tilley et al., 2004).
Reconstitution of G-protein-stimulated PLC activity with PITPs
G-protein stimulated PLC activity was reconstituted exactly as described previously (Cunningham et al., 1995).
Western blotting
Protein extracts from fly heads were separated by SDS-PAGE and transferred onto nitrocellulose membrane using semi-dry transfer (Chakrabarti et al., 2015). Primary antibodies were against: RDGB (1:4000, polyclonal; made in-house against the PITP domain of RDGB), anti-tubulin (1:4000; E7 DSHB), PITPα (1:1000; raised in-house; Cosker et al., 2008) and TRP (1:5000; made in-house). All secondary antibodies (Jackson Immunochemicals) were used at 1:10,000 dilution.
Lipid mass spectrometry
Shotgun mass-spectrometry was used to estimate PtdOH and DAG levels in Drosophila head extracts. The amount of PtdOH was normalised to total lipid phosphate content. 10 heads per sample (1-day-old flies) were homogenised in 0.1 ml methanol [having 51.98 pmole of (12:0/13:0) PtdOH and 11,140 pmole of (10:0/10:0/0:0) DAG as internal standard] using an automated homogeniser. 0.8 ml chloroform was added and left to stand for 10 min. 0.88% KCl (0.4 ml) was added to split the phases. The organic phase was dried and resuspended in 400 µl of chloroform:methanol (1:2). The total lipid phosphate was quantified from the lipid extract prior to infusion.
Experiments were performed on an LTQ Orbitrap XL (Thermo Fisher Scientific). Stable ESI-based ionisation was achieved using a robotic nanoflow ion source TriVersa NanoMate (Advion BioSciences) using chips with a spraying nozzle diameter of 4.1 μm. Ionisation voltage was 1.2 kV; back pressure was set at 0.95 psi. The temperature of the ion transfer capillary was 180°C. Acquisitions were performed at the mass resolution Rm/z400=100,000. For analysis, 60 μl of samples were loaded onto the 96-well plate (Eppendorf) of the TriVersa NanoMate ion source. Each sample was infused for 20 min. Lipids were identified with the LipidXplorer software (Herzog et al., 2012) by matching the m/z of their monoisotopic peaks to the corresponding elemental composition constraints. A Molecular Fragmentation Query Language (MFQL) file was compiled for PtdOH and DAG. Mass tolerance was 10 ppm and the intensity threshold was set according to the noise level reported by Xcalibur software (Thermo Scientific).
Cell fractionation
Fly heads (from ∼500 flies) were homogenised in lysis buffer (pH 7.4) (20 mM HEPES, 30 mM NaCl, protease inhibitor) and centrifuged at 166,000 g in Beckman TLA55 rotor for 30 min to separate membrane from cytosol. The membrane pellet was washed with lysis buffer (without EDTA) and resuspended in the same. Aliquots of membrane fraction were treated with following reagents on ice for 1 h: lysis buffer (control), 1 M NaCl, 0.2 M Na2CO3 pH 11, 6 M guanidine chloride (denaturant) and 2% Triton X-100. Post treatment, each sample was centrifuged at 172,000 g and supernatant was removed. The pellet was mixed with protein lysis buffer and run on a polyacrylamide gel.
Acknowledgements
We acknowledge the contributions of Alison Skippen, Clive Morgan, Adam Grieve, Tim Ashlin and Hantao Lou in the expression and analysis of the recombinant mutant PITP proteins. We are thankful to the NCBS Drosophila facility for the generation of transgenic flies.
Author contributions
S.Y., K.C., M.L., E.G., A.P., S.W., H.O., S.C. performed experiments and analyzed data. S.Y., P.G. and M.L. generated reagents. S.Y., S.C. and P.R. analyzed data, designed the study and wrote the paper.
Funding
This work was supported by project grants from the Wellcome Trust to P.R. and S.C.; from the Biotechnology and Biological Sciences Research Council to S.C.; and the National Centre for Biological Sciences-TIFR and the Wellcome Trust-DBT India Alliance to P.R. We thank the British Heart Foundation for PhD studentships to K.G. and to E.G.-E. Deposited in PMC for immediate release.
References
Competing interests
The authors declare no competing or financial interests.