Characterization of a novel RP2–OSTF1 interaction and its implication for actin remodelling

Retinitis pigmentosa 2 (RP2) is mutated in 7–18% of patients suffering from X-linked retinitis pigmentosa (XLRP), a severe and untreatable form of progressive retinal degeneration. RP collectively represents a heterogeneous group of retinal degenerative disorders marked by photoreceptor loss and is a leading cause of inherited blindness affecting 1:3000–1:7000 people worldwide (Pelletier et al., 2007; Prokisch et al., 2007). RP2 is a 350-residue protein whose expression at the protein level is not restricted to the retina but is also expressed in a wide range of tissues including the brain and the kidney (Proteomics database, https://www.proteomicsdb.org). Structurally, RP2 encompasses two distinct domains that appear to serve distinct roles (Kühnel et al., 2006). The N-terminal β-helix domain is homologous to tubulin co-factor C (TBCC), and the C-terminal ferredoxin α/β domain displays similarity to nucleoside diphosphate kinases (NDPK) (Kühnel et al., 2006). RP2 is predominantly targeted to the plasma membrane due to dual acylation of its N-terminus (Chapple et al., 2000, 2002); this subcellular localization of RP2 is important for its function and is pathologically relevant (Rosenberg et al., 1999). Notably, RP2 is also associated with primary cilia and the specialized connecting cilia of the photoreceptors, both in terms of localization and function (Hurd et al., 2010; Evans et al., 2010; Gotthardt et al., 2015).

While several proteins have been reported to interact with RP2, the most well-characterized interaction partner is ARL3, a small GTPase with suggested roles in the intracellular trafficking of ciliary proteins (Kim et al., 2014; Wright et al., 2016; Ismail et al., 2011, 2012), loss of which leads to photoreceptor degeneration and renal ciliopathy in mice (Schrick et al., 2006; Hanke-Gogokhia et al., 2016). We previously revealed the crystal structure of RP2 coupled to ARL3, demonstrating that the interaction of the two proteins is mediated exclusively by the N-terminal domain of RP2 and that RP2 displays GTPase-activating protein (GAP) activity against ARL3, stimulating its GTP hydrolysis by several orders of magnitude (Veltel et al., 2008).

The disruption of the RP2–ARL3 complex is clearly a major contributor to the development of retinal degeneration, not only because the ARL3-binding interface on RP2 is a mutational hotspot (Liu et al., 2017), but also because an ARL3 mutation was recently described to cause autosomal dominant RP (Strom et al., 2016). The most well-characterized function of the complex is in intracellular trafficking pathways for lipid-modified retinal and general ciliary proteins. In particular, structural and biochemical studies have proven the ability of GTP-bound ARL3 to allosterically displace protein cargo from chaperones PDE6δ and UNC119, which bind and solubilize lipid-modified membrane-anchored proteins such as NPHP3 and Rheb (Ismail et al., 2011, 2012). Animal model studies have linked this function of ARL3 with the correct targeting of lipid-modified phototransduction proteins, such as the prenylated PDE6 holoenzyme, rhodopsin kinase GRK1 and acylated transducin Tα, to the photoreceptor connecting cilium (Hanke-Gogokhia et al., 2016). It is therefore believed that RP2 has a role in the regulation of intracellular shuttling of those proteins, via its ability to stimulate ARL3 GTP hydrolysis (Zhang et al., 2015). ARL13B, mutated in Joubert's syndrome, was recently identified as a guanine nucleotide exchange factor (GEF) for ARL3, completing the cycle of GTPase regulation by stimulating GDP to GTP exchange (Gotthardt et al., 2015). It is believed that a gradient of ARL3 GTP- and GDP-bound forms is created in ciliated cells due to the combination of the exclusive localization of ARL13B inside the ciliary axoneme and the presence of RP2 at the basal body, which ensures release of lipid-modified cargo within the correct locations (Gotthardt et al., 2015).

However, other studies report different phenotypes associated with RP2 loss that might be independent of its function in this pathway, such as global post-Golgi transport of proteins and targeting of opsins to the connecting cilium (Evans et al., 2010; Li et al., 2013). In addition, ARL3 was recently described as an interaction partner of STAT3, suggesting a new role for RP2 in the regulation of transcription (Togi et al., 2016).

Fundamental questions remain unanswered in the field. For example, given the wide tissue distribution of RP2, it is reasonable to ask whether the function of RP2 is restricted to the photoreceptor connecting cilia and primary cilia or if it has other functions elsewhere. Furthermore, while the function of the N-terminal TBCC-like domain of RP2 has been well-characterized as the binding interface for the interaction with ARL3 and the catalytic centre of the protein, the C-terminus has been poorly characterized with no currently known function. It is known that although it shares weak sequence similarity with NDPK, the C-terminal domain of RP2 does not possess kinase activity, as key catalytic residues are not conserved (Kühnel et al., 2006). Apart from a report attributing DNA binding and exonuclease activity to this domain (Yoon et al., 2006), and another report that it contributes to the overall protein stability (Liu et al., 2017), it has been generally assumed, without any experimental evidence, that the domain participates in protein–protein interactions (Kühnel et al., 2006). In this study, we report a novel interaction between RP2 and OSTF1, a 214-residue protein that was initially identified in a screen for molecules whose overexpression leads to the secretion of factors that indirectly stimulate osteoclast maturation and bone resorption in culture (Reddy et al., 1998). Expression of OSTF1 at mRNA and protein level has been detected in several human tissues including the retina (Reddy et al., 1998) (Proteomics database, https://www.proteomicsdb.org). The exact function of this molecule is as yet unknown, but it has been suggested to be an adaptor for the function of Src in the actin cytoskeleton (Szymkiewicz et al., 2004). It was also recently identified as a negative regulator of cell motility in cancer cell lines, under control of the ERK1 and ERK2 (ERK1/2; also known as MAPK3 and MAPK1, respectively) pathway (Tanimura et al., 2011). OSTF1 exerts this effect by acting as a cytoplasmic anchor for long-tailed type 1 myosin (Myo1E), preventing it from localizing to sites of actin nucleation in lamellipodia (Tanimura et al., 2011, 2016). We conducted structural and biochemical analyses to characterize the RP2–OSTF1 interaction and show that the interaction interface on RP2 does not overlap with the ARL3-binding interface and that the formation of a trimeric RP2–OSTF1–ARL3 complex is possible. In addition, we show that the OSTF1-binding interface spans both the C- and N-terminal domains of the protein and that the residues that participate in the interaction belong to a surface ‘patch’ on RP2 that displays remarkable conservation status among vertebrates, arguing for major functional significance. Finally, we provide evidence that RP2 functions as a regulator of the ability of OSTF1 to influence cell motility, via modulating the availability of OSTF1 to interact with Myo1E.

To look for novel RP2 functions through the identification of novel interaction partners, we conducted a yeast two-hybrid screen by using human full-length RP2 as the bait protein. Of 91 clones, 41 encoded full-length OSTF1 (Fig. 1A). Direct interaction between OSTF1 and RP2 was confirmed in two directions in a targeted yeast two-hybrid assay (Fig. 1B). OSTF1 is a 214-residue protein consisting of an N-terminal proline-rich region, a Src-homology 3 domain (SH3), four ankyrin repeats (ANK) and a well-conserved acidic amino acid cluster at the C-terminus. Co-immunoprecipitation (co-IP) confirmed the interaction between exogenously expressed OSTF1 and RP2 in HEK293T cells (Fig. 1C). In addition, the interaction was confirmed by reciprocal pulldown analyses in bovine retinal extracts using recombinant GST–RP2 or GST–OSTF1 as bait proteins (Fig. 1D). To determine the thermodynamic parameters of the RP2–OSTF1 interaction in solution, we conducted isothermal titration calorimetry (ITC) using full-length RP2 and full-length His-tagged OSTF1 (Fig. 1E). ITC analysis showed robust complex formation with 1:1 stoichiometry; the equilibrium association constant of the interaction was derived as 5.28 µM and the reaction is exothermic with an enthalpic contribution of ∼31.26 kJ/mol. Moreover, indirect fluorescence polarization experiments in which an equimolar complex of fluorescently labelled small GTPase ARL3-mantGppNHp (a non-hydrolysable GTP analog) with RP2 was titrated with increasing amounts of full-length OSTF1 (OSTF1FL) showed an affinity of OSTF1FL to RP2 of 0.672 µM (Table S1).

In order to investigate whether OSTF1 acts as a regulator of RP2 activity, we used fluorescence polarization analysis to assess whether OSTF1 modulates the binding affinity of RP2 to ARL3 or its catalytic activity in vitro. For this purpose, we used full-length RP2 and OSTF1–His as well as the G-domain of ARL3 (residues 18–177) fluorescently labelled with mantGppNHp (a non-hydrolysable GTP analog). RP2 was previously shown by us (Veltel et al., 2008) to bind to ARL3-mantGppNHp and ARL3-mantGDP/AlFx with higher affinity than to ARL3-mantGDP (Table S1). An increase in fluorescence polarization signal of ARL3-mantGppNHp is observed upon addition of RP2 followed by a further increase upon addition of OSTF1 (Fig. 2A). As OSTF1 does not bind directly to either ARL3-mantGDP or ARL3-mantGppNHp (Fig. 2B), we assume that OSTF1 and ARL3 can bind to RP2 simultaneously to form a trimeric complex.

The affinity of RP2 for ARL3-mantGppNHp did not change depending on the absence and presence of saturating concentrations of OSTF1 (Table S1). Therefore, OSTF1 does not inhibit or compromise the RP2–ARL3 complex formation and might also be functionally independent. Moreover, it seems that a stable four-subunit complex seems to be able to be formed with the ARL3 effector, UNC119a (Fig. S1).

We next investigated whether the presence of OSTF1 affects the stimulation of ARL3 GTP hydrolysis upon incubation with catalytic concentrations of RP2 in a charcoal assay. ARL3 was loaded with radioactively labelled [³²P]GTP. The GAP activity was monitored by measuring hydrolysis and release of ³²P leading to an increase in the radioactive signal, as described previously (Veltel et al., 2008). We found that OSTF1 does not influence the intrinsic GTP hydrolysis rate of ARL3 or the GAP activity of RP2 (Fig. 2C).

Since OSTF1 does not compete with ARL3 for binding to RP2, we next asked whether it binds exclusively to the C-terminal NDPK-like domain of RP2. To conduct this domain analysis of the RP2–OSTF1 interaction, we used RP2 and OSTF1 deletion constructs (Fig. 3A) in analytical gel filtration experiments. First, we confirmed that full-length RP2 and OSTF1-His complex formation can be detected with this method, since the complex, being larger, elutes from the stationary phase earlier than the monomeric proteins (Fig. 3B, left). Next, we repeated the assay using an N-terminal deletion construct of RP2 (RP2 ΔN, residues 230–350) and showed that the C-terminal domain alone cannot form a complex with OSTF1 (Fig. 3B, right). However, deletion of only 34 residues from the N-terminus (a flexible region of RP2 that is essential for the interaction with ARL3; Veltel et al., 2008) does not affect complex formation (Fig. S2). Subsequently, we tested the binding of RP2 to OSTF1 deletion constructs (Fig. 3C). Interestingly, deletion of amino acids 1–74, comprising the SH3 domain of OSTF1 (OSTF1 ΔΝ), as well as deletion of amino acids 74–214, the ANK repeats domain (OSTF1 ΔC), both abolish complex formation, showing that both the SH3 and ANK domains of OSTF1 take part in the interaction (Fig. 3C). This was further confirmed by pulldown assays from HEK293T cell lysates using GST-tagged OSTF1 constructs as bait (Fig. 3D). While the ΔC OSTF1 construct cannot form a complex with RP2, it can still pulldown endogenous Cbl, in accordance with previous studies (Szymkiewicz et al., 2004).

Although most RP2 mutations found in patients likely lead to null alleles due to severely compromised or absent RP2 expression, caused by either mRNA degradation or destabilized protein structure, missense mutations that do not affect the expression and overall stability of RP2, according to biochemical studies, in silico predictions or circular dichroism spectroscopy have been reported (Kühnel et al., 2006; Veltel et al., 2008; Liu et al., 2017). Among them, R118H and E138G are known to severely impact on the RP2–ARL3 interaction and thus affect the GAP activity of RP2 (Veltel et al., 2008). G2A and C3S mutations lead to altered RP2 localization by impairing dual acylation at its extreme N-terminus (Chapple et al., 2000). It is apparent that pathogenic missense RP2 mutations that are predicted to be non-destabilizing often occur on important residues that mediate critical interactions or modifications and, therefore, can be particularly informative about the function of the protein. An exception might be R282W, which has been suggested to be a low-frequency polymorphism even though it has been identified in RP patients (Thiselton et al., 2000; Miano et al., 2001).

In order to assess the importance of the RP2–OSTF1 interaction for the function of RP2, we tested the effect of six of these mutations (G2A, C3S, R118H, E138G, R211L, R282W) on the interaction. We conducted GST pulldown analyses using recombinant full-length GST–OSTF1 as bait in lysates of HEK293T cells expressing either wild-type (WT) or mutant RP2 with a V5 tag, and identified R211L as a mutation that completely abolishes the RP2–OSTF1 interaction (Fig. 4A). Interestingly, R211 is not known to participate in ARL3 binding as it resides within a spatially distinct region. Upon multiple sequence alignment of human RP2 with a range of divergent orthologues (from 38% identity with the choanoflagellate Salpingoeca to 94% identity with the mole rat), it is apparent that the R211 position is strictly conserved among vertebrates, but not entirely conserved among non-vertebrate chordates and non-chordates (Fig. 4B). Hence, for the subsequent conservation analyses, only the vertebrate subphylum was used.

Because R211 lies away from the ARL3 interaction interface on the opposite face of the GAP domain, we hypothesized that this area of the protein along the C-terminal NDPK-like domain might form the binding interface for the RP2–OSTF1 interaction. To test this, we mapped the binding interface on the surface of RP2 by using a targeted RP2 structure-guided mutagenesis approach and testing the effect of missense mutations on the OSTF1 binding. Residues to be mutated were based on a comprehensive in silico analysis of the conservation status among vertebrates, the surface properties and the proximity of the residues to R211, as well as the potential for post-translational modifications (Fig. 4C,D). Electrostatic properties and hydrophobicity of residues were taken into account when choosing which residues to mutate, as highly charged and/or exposed hydrophobic amino acids often participate in protein–protein interactions (Fig. S3). In addition, structurally and functionally important regions in proteins typically appear as patches of evolutionarily conserved residues that are spatially close to each other (Ashkenazy et al., 2016). To assess the effect of missense RP2 mutations on the RP2–OSTF1 interaction, we expressed V5-tagged WT and mutant forms of RP2 in HEK293T cells and used the lysates for pulldown assays using GST–OSTF1 as bait. All the mutations generated were not predicted to affect the overall stability of the protein (Table S2) and were, indeed, expressed at WT levels. As an additional control to ensure that the absence of RP2 in the pulldown samples is specifically due to the loss of the RP2–OSTF1 interaction and not an unspecific effect of destabilized secondary structure or folding, we conducted pulldown assays using GST–ARL3 Q71L as bait and confirmed that none of the mutations impaired the RP2–ARL3 interaction, as was predicted. Using this approach, we identified three more residues whose substitution severely impacts on the RP2–OSTF1 interaction: Y245, D255 and K160 (Fig. 4C). While the K160A mutation, which abolishes the positive charge has a mild impact on the RP2–OSTF1 interaction, the K160E mutation, which reverses the charge severely impairs the interaction. This confirms an important role for K160 on the interaction interface. Moreover, we have noted that mutating K252 also affects the RP2–OSTF1 interaction, albeit to a lesser extent.

Sequence conservation and variation was mapped on the surface of RP2 showing two highly conserved clusters of residues on the RP2 surface, one that forms the ARL3-binding site within the N-terminal β-helix domain and a second cluster located on the opposite face of the molecule that spans both the β-helix domain and the C-terminal NDPK-like domain (Fig. 4D). Residues Y245, D255 and K160 were localized to this second conserved cluster. Therefore, it is likely that this area forms the interaction interface with OSTF1. Since all identified residues are strictly conserved among vertebrates (Fig. 4D), we suggest that the interaction occurs widely among vertebrates and thus has an important function that merits further examination.

Based on the characterization of the RP2–OSTF1 interaction presented here, and the known structures of both proteins, we used a macromolecular docking strategy in order to build a 3D model of the RP2–OSTF1 complex. Although our experiments do not tell us the specific interacting residues in OSTF1, they do reveal that both the N- and C-terminal domains participate in the interaction. Thus, we performed a series of docking calculations whereby the interaction was centred around different residues on the surface of OSTF1, and then evaluated models for their compatibility with the mutagenesis data (see Materials and Methods). Although there were multiple models compatible with the experimental data, we observed a single dominant cluster of structural similar models (Fig. 5). Moreover, the binding surface on OSTF1 for this dominant cluster was highly conserved, whereas all other compatible docking models involved poorly conserved regions of the OSTF1 surface (Fig. S4), thus strongly suggesting that this model is reflective of the actual complex.

A function that has been previously attributed to OSTF1 in cancer cell lines is the regulation of cell motility. In particular, OSTF1 negatively regulates cell motility in an ERK1/2 pathway-dependent manner via the regulation of the intracellular localization of Myo1E, with whom it directly interacts (Tanimura et al., 2011, 2016). In order to test whether RP2 also has a role in cell motility, we used the CRISPR-Cas9 genome editing method to generate clonal RP2-null immortalized human retinal pigment epithelial cell (hTERT-RPE1) lines. hTERT-RPE1 cells were chosen because they are motile in cell culture, have a stable diploid karyotype and are of retinal origin. Two clones (denoted G and I) have distinct 2-bp deletions in exon 2 that lead to complete absence of RP2 expression (Fig. 6A). Comparison of the motility of parental RPE1 cells and RP2-null clones demonstrated that cells that lack RP2 expression are less motile in both wound healing and random migration assays (Fig. 6B,C). Live-cell imaging of individual cells in the random migration assay revealed a lack of directionality in their movement and the inability to form proper lamellipodia-like extensions (Fig. 6C). Since OSTF1 was previously identified as regulating cell motility, we asked whether RP2 regulates OSTF1 expression, especially since its target ARL3 was recently identified as an activator of the STAT3 transcription factor (Togi et al., 2016). However, this appears not to be the case as OSTF1 protein expression remained unchanged in all control and RP2-knockout cells (Fig. 6A).

To examine whether RP2 exerts a positive impact on cell motility by modulating OSTF1 localization, biochemical fractionation was performed in HEK293T cells. This showed that overexpression of RP2 causes translocation of a pool of endogenous OSTF1 from the cytoplasm to the membrane fraction (Fig. 7A). Similarly, as determined by immunofluorescence analysis, overexpression of RP2–EmGFP in HeLa cells led to the recruitment of V5–OSTF1 to the cell membrane, where they colocalize (Fig. 7B). By contrast, overexpression of the RP2 G2A pathogenic mutant, which does not display the plasma membrane localization of the WT protein (Chapple et al., 2000), did not recruit OSTF1 to the membrane compartment, as shown by subcellular fractionation and immunofluorescence studies (Fig. S5A,B). We hypothesized that by recruiting OSTF1 to the membrane, RP2 prevents it from interacting with Myo1E, thus promoting cell motility. To test this hypothesis, we used HEK293T cells transiently overexpressing FLAG–OSTF1 and empty vector, or FLAG-OSTF1 and RP2–V5, for anti-V5 immunoprecipitations and observed that less Myo1E associated with FLAG–OSTF1 when RP2 was overexpressed (Fig. 7C; Fig. S6). The interaction of Myo1E and OSTF1 has been shown to be disrupted by ERK1/2-mediated phosphorylation of OSTF1 S202. Serum stimulation and subsequent ERK1/2 activation did not impair the ability of endogenous OSTF1 to co-immunoprecipitate with RP2 (Fig. 7D). This suggests that RP2 interaction with OSTF1 is distinct to the OSTF1-Myo1E interaction.

Studies on the function of RP2 have so far mainly focused on its role in intracellular protein trafficking, coordinated via the GAP activity of its N-terminal domain towards the small GTPase ARL3. However, the existence of non-canonical roles of RP2 in the retina or other tissues has not been fully explored. In addition, the functionality of the C-terminal NDPK-like domain of RP2 has not been elucidated. In order to discover novel RP2 functions, we sought to identify novel interaction partners by conducting a yeast two-hybrid screen. We identified OSTF1 as a novel direct interaction partner and confirmed the interaction by co-IP from HEK293T cells and pulldown assays from bovine retinal extracts. Subsequent to our discovery, the interaction was recently identified in a large-scale study of the human interactome, further validating the interaction (Huttlin et al., 2015). OSTF1 has been associated with osteoclast differentiation (Reddy et al., 1998; Kurihara et al., 2001) and the regulation of cell motility (Tanimura et al., 2011, 2016), but not with retinal processes. We investigated the potential relationship of this interaction with the ARL3 regulation pathway and found that OSTF1 and ARL3 can simultaneously bind on RP2 to create a trimeric complex. These results suggest that OSTF1 does not antagonize the ability of ARL3 to bind to RP2 and it is unlikely that this newly identified partner is a negative or positive regulator of the RP2 GAP activity, at least in vitro. Moreover, we screened several non-destabilizing missense RP2 mutations for their ability to interact with OSTF1 and found that the R211L mutation entirely abolishes the interaction, potentially implicating OSTF1 in the pathogenesis of XLRP2. As R211 is found on a spatially distinct surface of RP2 from the ARL3 interaction interface, and on a previously uncharacterized RP2 region, this prompted us to perform a site-directed mutagenesis approach, producing strictly non-destabilizing missense mutations, to map the interaction interface, which led to the identification of three more residues that are essential for the interaction to occur. The identified residues are good candidates for participating in a protein–protein interaction. For example, although the surface of RP2 is predominantly negatively charged and, like OSTF1, has an acidic theoretical pI, R211 residue belongs to a streak of positively charged residues on the surface of RP2, which suggests the occurrence of oppositely charged interactions (Fig. S3). On the other hand, Y245 has an exposed aromatic side-chain and belongs to a hydrophobic cluster at the inter-domain junction of RP2 (Fig. S3). In addition, this hydrophobic cluster lies within a noticeable surface cavity. Based on our results, we propose that OSTF1 binds to a cluster of residues on the RP2 surface that are strictly conserved among vertebrates and distinct from the ARL3 interaction interface. This conservation analysis hints at an important function of the interaction among vertebrates that is most likely distinct from the ARL3 regulation pathways.

OSTF1 is known to exert a negative effect on cell motility through its direct interaction with Myo1E, which leads to Myo1E being ‘anchored’ to the cytoplasm and prevents it from translocating to sites of actin nucleation in lamellipodia, an event which is essential for their proper formation (Tanimura et al., 2016). The OSTF1–Myo1E interaction is regulated by the phosphorylation of OSTF1 by RSK1 under the direct control of the ERK1/2 pathway; activation of ERK1/2 and PI3K signalling pathways has been long known to be required for actin polymerization and subsequent lamellipodium formation (Tanimura et al., 1998). Here, we report another way of regulating the negative effect that OSTF1 exerts on cell motility: RP2 overexpression results in OSTF1 recruitment to the plasma membrane, leading to less OSTF1 being available for interaction with Myo1E. It will be interesting to study which signalling pathway activates RP2 to act as an OSTF1 regulator, an effect that is simulated by RP2 overexpression in our experiments.

Myo1E is a long-tailed class I myosin that has been suggested to physically connect actin to the plasma membrane and thus generate plasma membrane tension (McConnell and Tyska, 2010). Its deletion in mice leads to defects in kidney filtration (Krendel et al., 2009) and it has also been associated with poor outcome in breast cancer patients and identified as a tumour promoter in a mouse study (Ouderkirk-Pecone et al., 2016). Moreover, Myo1E has a role in coordinating actin assembly and trafficking in clathrin-mediated endocytosis (Cheng et al., 2012). The potential role of RP2 in those processes needs to be clarified further. Interestingly, disruption of actin turnover is observed upon knockout of another XLRP gene, RPGR, suggesting that actin dysregulation may contribute to the pathology on RP2 patients (Megaw et al., 2017). Besides, it is important to note that a dynamic organization of the acto-myosin cytoskeleton is a prerequisite for most retinal pigmented epithelia functions in the retina, and many classes of non-muscle myosins are implicated in them (e.g. myosins VIIa and Va in phagocytosis, and myosin II in melanosome aggregation) (Ruiz-Loredo and López-Colomé, 2012).

An important question that arises from our study is whether the RP2–OSTF1 interaction participates in ciliary traffic and is relevant for RP pathogenesis, or whether it has a non-canonical function in a different context or tissue. One hypothesis is that there are two distinct RP2 functions, the regulation of ciliary traffic via the ARL3 interaction in post-mitotic cells, and the regulation of OSTF1 localization and cell motility in dividing cells, where cilia are retracted. On the other hand, the two functions of RP2 might be related. Interestingly, ARL13B mutant mice show a defect in directional migration of interneurons of the cerebral cortex (Higginbotham et al., 2012), while knockdown of ARL3 impairs the migrational capacity of HeLa cells (Togi et al., 2016). Future studies should also examine whether OSTF1, in turn, affects RP2 localization by preventing its localization to the ciliary axoneme.

Given that R211L is not predicted to destabilize the folding of RP2s in in silico analyses (Liu et al., 2017), our data suggest that loss of the RP2–OSTF1 interaction may lead to retinal degeneration. This could be tested in the future by using CRISPR-mediated genome editing to introduce R211L and other mutations that disrupt the RP2–OSTF1 interaction (such as Y245F) into mice and examining the retinal phenotype. Examination of expression data from the Fantom 5 project (http://fantom.gsc.riken.jp) shows that OSTF1 is expressed in both embryonic and adult eyes in mice, and more specifically in both the retina and retinal pigmented epithelia in humans. More research will be needed to determine the cellular expression pattern of OSTF1 in the retina.

The RP2 protein has come into the spotlight because of the devastating effect of its mutations on the viability of photoreceptors, leading to one of the most early-onset and severe forms of RP. However, a multitude of different and often conflicting studies point to a multifunctional protein with more than one role in cellular biology (Lyraki et al., 2016). RP2-mediated disease also presents with a distinct set of symptoms that often deviate from the classic clinical image that characterises RP patients (Jayasundera et al., 2010). Gene augmentation therapy is perhaps the strategy that is closest to the clinic for RP, having already proven its safety in Leber congenital amaurosis (LCA) patients (Pierce and Bennett, 2015). In fact, gene augmentation therapy has already successfully prevented degeneration in a mouse RP2-knockout model (Mookherjee et al., 2015). However, a thorough understanding of the functions of the protein are essential, and non-canonical functions of RP2 must be taken into account in order to prevent adverse effects and aid the development of complementary therapies before a therapeutic approach can reach the clinic. Our study presents a novel aspect of RP2 function and thus contributes to that direction.

For the yeast two-hybrid screen, we used the ProQuest™ two-hybrid system (Thermo Fisher) according to the manufacturer's instructions. In brief, the full-length human RP2 open-reading frame (ORF) was cloned into the pDEST32 bait plasmid. This was transformed into S. cerevisiae MaV203 yeast cells. These cells were subsequently transformed with the ProQuest™ human kidney 3-frame cDNA library according to the manufacturer's instructions. For co-transformation assays, the RP2 and OSTF1 ORFs were cloned into both the pDEST32 bait and pDEST22 prey plasmids. Interaction of bait and prey molecules were tested by growth of yeast cells on selective medium containing 75 mM 3-amino 1,2,4,-triazole but lacking leucine, tryptophan and histidine and by performing LacZ assays with filter-immobilized cells.

Nickase paired guide sequences were identified in exon 2 of human RP2 using the Zhang laboratory CRISPR design tool (http://crispr.mit.edu/). Two paired guide sequences (guide A, 5′-TCAAATTTGGATGTTTTCAA-3′ and guide B, 5′-TTTGAGGAAGACTCAATGAT-3′) were cloned into pX461 and pX462 CRISPR plasmids, respectively. These plasmids were co-transfected into RPE1-hTERT cells that, after 24 h, were subjected to 24 h selection in 5 μg/ml puromycin-containing medium to select for cells transfected with pX462. After 24 h selection, individual GFP-positive cells were FACS-sorted into individual wells of a 96-well plate to select for cells transfected with pX461. After expansion, 20 clones were Sanger sequenced (sequencing primers: 5′-GGACCCGTGAAAGGCAGCGTG-3′ and 5′-CCTGACACAGGTGTAAAGTCATGAATGTTACTCC-3′) and clones harbouring deleterious mutations kept together with clones harbouring no mutations as controls.

For immunofluorescence experiments, cells were seeded onto acid-etched coverslips and transiently transfected. Cells were fixed in 4% paraformaldehyde (PFA) in PBS for 10 min, washed and then permeabilized with 0.1% Triton X-100 for 5 min. Cells were blocked in 10% goat serum (Sigma) in PBS for 1 h. Primary antibodies (monoclonal anti-V5, 1:5000; Life technologies, R960-25) were applied in blocking buffer for 1 h at room temperature. Cells were washed in PBS prior to incubation with Alexa Fluor-conjugated goat secondary antibodies (Thermo Fisher) for a further hour at room temperature. Finally cells were mounted onto slides using ProLong^® Gold antifade mountant (Thermo Fisher). Slides were imaged on a Leica SP2 inverted confocal microscope.

Point mutations were introduced in the human RP2 coding sequence by using the QuikChange II site-directed mutagenesis kit following the manufacturer's instructions. Briefly, pENTR/D-TOPO entry vector (Invitrogen) containing the WT human RP2 coding sequence was used as a template in PCR reaction using the primers described in Table S3 (for K160A, E205A, E206A, Y245A, K208A, R218S, S213A, E215A, E256A, K160E, K252A and D255A). Primers for G2A, C3S, R118H, E138G, R211L and R282W have been previously published.

To achieve expression of the constructs in mammalian cells, the RP2 coding sequences were transferred into the destination vector pCDNA-DEST40 (Invitrogen), which adds an N-terminal V5 tag to the construct, by recombination reaction using LR Clonase II (part of the Invitrogen Gateway cloning system).

HEK293T, hTERT-RPE1 and HeLa cells were obtained from the American Type Culture Collection (ATCC) and maintained as adherent cultures at 37°C and 5% CO₂. HEK293T and HeLa were grown in Dulbecco's modified Eagle's medium (DMEM) with 4.5 g/l L-/D-glucose and pyruvate (Gibco, Life Technologies), with the addition of 10% (v/v) fetal calf serum and penicillin-streptomycin cocktail at 1 mg/ml final concentration. hTERT-RPE1 cells were grown in DME-F12 medium (Hyclone, Thermo Fisher Scientific) with the addition of 10% v/v) fetal calf serum, penicillin-streptomycin cocktail at 1 mg/ml final concentration and hygromycin (10 μg/ml final concentration). Cells were regularly split using the TrypLE trypsin substitute (ThermoFisher Scientific). All cells were routinely tested for contamination by MRC HGU technical services.

Cells were transiently transfected with 0.5 μg purified plasmid DNA containing the relevant constructs (per 0.2×10⁶ cells, using the nonliposomal reagent Fugene HD (Promega).

Cell motility of hTERT-RPE1 cells was assessed by performing a random migration assay and a scratch wound assay using the Incucyte ZOOM live-cell imaging system (Essen Bioscience). For the random migration assay, cells were sparsely plated (1000 cells per well of a 96-well plate), left overnight to adhere and imaged the next day. The distance covered by individual cells was analysed by a researcher that was blind to the experimental condition with FiJi software. For the scratch wound assay, cells were plated in 96-well plates and were subjected to a scratch wound when completely confluent (using the WoundMaker tool by Essen Bioscience). The density of cells migrating into the wound during the next 24 h was measured by the Incucyte cell migration software.

Cell lysate proteins were separated according to three subcellular fractions (cytoplasmic, membrane/organelle and nuclear/cytoskeletal) using a detergent-based cell fractionation kit (Cell Signaling Technology) according to the manufacturer's instructions.

The following primary antibodies were used for western blotting: mouse monoclonal anti-V5, 1:5000 (Life Technologies, R960-25), rabbit polyclonal anti-OSTF1, 1:10,000 (Bethyl laboratories, A303-004A), mouse monoclonal anti-β actin, 1:10,000 (Sigma-Aldrich, A3853), rabbit polyclonal anti-RP2, 1:2000 (custom-made; Hurd et al., 2010), mouse monoclonal anti-histone H3, 1:1000 (Abcam, ab18521), rabbit polyclonal anti-integrin β1, 1:500 (Millipore, AB1952), rabbit monoclonal anti-FLAG, 1:2000 (Cell Signaling Technology, 8146), mouse monoclonal anti-Cbl, 1:1000 (BD Bioscience, 610441) rabbit polyclonal anti-Myc, 1:1000 (Cell Signaling Technology, 2278) and rabbit monoclonal anti-pERK, 1:1000 (Cell Signaling Technology, 4370).

V5 and FLAG immunoprecipitations (IPs) were conducted in Triton lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl and 1% Triton X-100). For V5 IPs, 10 μl of anti-V5 agarose affinity gel (Sigma-Aldrich) was incubated with 2–3 mg of total cell lysate protein for 1 h at 4°C under continuous rotation. For FLAG IPs, 5 μl of EZ View Red anti-FLAG M2 agarose affinity gel (Sigma-Aldrich) was incubated with 1 mg of total cell lysate protein for 2 h at 4°C under continuous rotation. After incubation, extensive washes followed to minimize non-specific protein binding. Finally, immunocomplexes were eluted off the beads by adding LDS buffer (NuPage, Life Technologies) and dithiothreitol (DTT), and subjected to SDS-PAGE and western blotting.

Native full-length ARL3 or truncated ARL3 [amino acids (aa) 17–177 (ARL3ΔN)], RP2, RP2 aa 250–350 (ΔN), RP2 aa 35–350 (ΔN34) and RP2 mutant R211H were purified as previously described to be completely loaded with GppNHp (Kühnel et al., 2006; Veltel et al., 2008). OSTF1FL, OSTF1 aa 1–74 (OSTFΔC) and OSTF1 aa 74–214 (ΔN) were cloned into pET20 vector and purified in a three-step procedure via affinity chromatography (self-packed NTA material, QIAGEN), a hydrophobic phenyl sepharose column (GE Healthcare) and gel filtration using a Superdex 75 16/60 column (GE Healthcare). Buffers were used as previously described (Chen et al., 2006; Tong et al., 2009). The proteins were stored in buffer M containing 25 mM Hepes pH 7.5, 150 mM NaCl, 5 mM MgCl₂, 1 mM dithiothreitol and 5% glycerol.

GST–ARL3 Q71L, GST–OSTF1 and RP2–GST recombinant proteins used in pulldown experiments were expressed in BL21 DE3 E. coli cells transformed with pGEX4T3 vectors (GE Healthcare) containing the relevant constructs. GST-tagged (at the N-terminal or C-terminal in the case of RP2) recombinant proteins were purified from the cell lysate by immobilization on glutathione–Sepharose bead slurry (GE Healthcare). The protein-bound slurry was stored in PBS containing 10% glycerol and 1 mM DTT.

GST pulldown assays were conducted in mild lysis buffer (25 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 1% NP40 and 5% glycerol) by incubating 0.7–1 mg of total cell lysate or bovine retinal extract protein with ∼10 μg of GST-tagged protein bound on glutathione–Sepharose beads for 2 h at 4°C under continuous rotation. The bead-bound GST protein complexes were washed at least three times in lysis buffer to remove non-specifically bound proteins. Finally, GST complexes were eluted off the beads by the addition of LDS buffer (NuPage, Life Technologies)/DTT and submitted to SDS-PAGE and western blotting.

Interactions of RP2 and RP2 R211H with OSTF FL were investigated by isothermal calorimetry using an ITC₂₀₀ microcalorimeter (MicroCal). Measurements were carried out in buffer M at 25°C. 585 µM OSTF FL protein in the syringe was injected into the cell containing either of 58 µM RP2 or 52 µM RP2 R211H. Analysis of the data was performed by using the Origin 7.0 Software provided by the manufacturer (MicroCal, LLC ITC).

Formation of complexes with RP2, RP2ΔN or RP2ΔN34 with OSTF1 FL, OSTFΔN or OSTFΔC was investigated by analytical size exclusion chromatography using a Superdex 75 10/300 column (GE Healthcare). A total of 0.5 mg RP2 protein was incubated with equal molar amounts of OSTF1 protein for 2 h at room temperature. The mix was then applied to the size exclusion chromatography column and eluted with one column volume of buffer M. The elution profile was recorded and eluted fractions analysed by SDS-PAGE.

ARL3ΔN was loaded with mantGDP or mantGppNHp (Pharma Waldhof) overnight at 12°C by incubation with a 1.5-fold molar excess of nucleotide and purified the following day on a desalting column in buffer M (Veltel et al., 2008). In case of mantGppNHp, alkaline phosphatase was added. Nucleotide loading was determined by high-pressure liquid chromatography (HPLC) measurements on a C18 column. Polarization data was recorded with a Fluoromax-4 spectrophotometer (Jobin Yvon, München, Germany), with excitation and emission wavelengths of mant-nucleotides at 366 and 450 nm, respectively.

Qualitative binding and complex formation of OSTF1 FL with ARL3ΔN was determined by titration of 1 µM ARL3ΔN bound to the respective mant-nucleotide with increasing amounts of OSTF1 FL. For formation of the three-subunit complex, 1 µM ARL3ΔN bound to mantGppNHp was supplemented with 10 µM RP2 followed by 20 µM OSTF1 FL or buffer, respectively.

The binding affinity between RP2 and ARL3ΔN was measured by monitoring the polarization signal during titration of 1 μM ARL3 loaded with mantGppNHp with increasing amounts of RP2 in the presence or absence of 10 µM OSTF1 FL at 20°C in buffer M. Obtained data points were fitted to a first-order reaction using Grafit5 (Erithacus software) to obtain the dissociation constant K_d.

The binding affinity between OSTF1 FL and RP2 was determined indirectly by allowing formation of a complex upon incubating 1 µM ARL3 mantGppNHp with 1 µM RP2, which was titrated with increasing amounts of OSTF1 FL at 20°C in buffer M. Obtained data points were fitted to a first-order reaction by using Grafit5 (Erithacus software) to obtain the dissociation constant K_d.

This was performed as described previously (Veltel et al., 2008). ARL3 was loaded with GppCH₂p with the help of bead-linked alkaline phosphatase, which was removed via centrifugation (12,000 g for 20 min) and another step of gel filtration. Briefly, a mix of 20 µM ARL3-GppCH₂p was mixed with 60 nM [γ-³²P]GTP and 18 µM GTP in buffer M. Upon a short incubation, the reaction was started by addition of 0.2 µM RP2 or 5 µM OSTF1 or a mix of 0.2 µM RP2 and 5 µM OSTF1. Aliquots of 10 µl were taken at 10, 20, 40, 60, 90, 120 180, 240 and 300 s and immediately mixed with 400 µl of charcoal solution (50 g l⁻¹ charcoal in 20 mM phosphoric acid) to stop the reaction. The charcoal was pelleted (12,000 g for 20 min) and the amount of free ³²Pi in the supernatant determined by scintillation counting. Data was plotted by showing ratio of specific counts per minute (cpm) of the supernatant over total counts per minute of the sample at each point.

The empirical forcefield FoldX (Schymkowitz et al., 2005; Guerois et al., 2002) under the YASARA (Van Durme et al., 2011; Krieger et al., 2002) molecular visualization program was used to estimate the free energy difference (stability change) upon mutagenesis from WT, as previously described (Liu et al., 2017).

In order to retrieve and select a divergent set of RP2 orthologue sequences for multiple sequence alignment, the ConSurf server (Glaser et al., 2003; Ashkenazy et al., 2010) was used with the following options (Homologue search algorithm: CSI-BLAST; Number of iterations: 3; E-value cut-off: 0.0001; Protein database: UniRef-90; Maximal % identity between sequences: 95%; Minimal % identity for homologues: 35%). From the returned results, the set of sequences was further manually curated to exclude all ‘fragment’ proteins and those proteins that could not be assigned to any genus after cross-checking their UniRef Accession IDs. Thus, eighteen sequences were finally selected and aligned externally using MUSCLE (Edgar, 2004). The alignment was used as input for the ConSurf web server to map sequence conservation on the protein structure of RP2. The ‘Bayesian’ method was used to calculate the rate of evolution at each site using the ‘JTT’ evolutionary substitution matrix model. The PyMOL Molecular Graphics System (Schrödinger, LLC) was used for visualization, analysis and figure preparation.

The electrostatic surface representation of RP2 was generated by GRASP (Nicholls et al., 1991) and the lipophilic surface rendition was generated using MOLCAD (Heiden et al., 1993).

We used the program HADDOCK (Dominguez et al., 2003) using the standard protein–protein docking protocol in order to generate docking models of RP2–OSTF1 dimers from the known structures of RP2 (PDB ID: 3BH7) and OSTF1 (PDB ID: 3EHR). The interaction surface on RP2 was based on the mutagenesis experiments: the four residues that can disrupt the interaction when mutated (Arg211, Lys160, Tyr245 and Asp255) are defined as ‘active’ and were constrained to be directly involved in the interface, whereas other surface residues within 8 Å of these residues were defined as ‘passive’ and were allowed but not strictly constrained to participate in the interface. However, since we do not know what residues from OSTF1 directly interact, we performed a series of docking runs where the OSTF1 interface was constrained to be centred around each of the 121 surface residues (defined as exposing >20 Å of solvent-accessible surface area). Surface residues within 3 Å of the central residues are defined as core, while other residues within 8 Å are defined as passive.

Initially, we generated 55 water-refined docking models for each OSTF1 surface residue. For each docking model, we used FoldX to predict the effects of each experimentally tested mutation on the RP2–OSTF1 interaction. Next, we identified docking models that meet the following criteria: (1) the interaction involves both the N-domain (residues 13–73) and C-terminal domain (residues 74–190); (2) the OSTF1-binding surface does not overlap with the known SH3-binding pocket (residues 21–24, 26–30, 35, 46–49, 60, 62, 64–65, 68–69); (3) no steric clashes between OSTF1 and ARL3; and (4) P<0.05 for a Wilcoxon rank-sum test comparison between the predicted FoldX effects of the disruptive mutations (R211L, K160A, K160E, Y245A, D255A) versus non-disruptive mutations (R118H, E138G, R282W, E205A, E206A, K208A, S213A, E215A, R218S, K252A, E256A). For the 73 OSTF1 surface residues that had at least one structure compatible with these criteria, we generated 100 additional docking models. We then further identified docking models that meet the above criteria, except they had a P<0.002. For the 53 OSTF1 surface residues that had at least one structure compatible with these criteria, we generated 200 additional docking models.

Out of the 24,555 docking models generated in total, there were 21 that met the above criteria, but also had HADDOCK scores <−50, and in which the predicted effects of all disruptive mutations are stronger than all non-disruptive mutations. These 21 structures were then clustered on the basis of their correlation coefficient root-mean-square deviation (RMSD), with all structures in each cluster having <3 Å RMSD with all others.

To assess the evolutionary conservation of OSTF1 interaction surfaces from each docking model, we used the residue-specific values calculated by ConSurf (Ashkenazy et al., 2016) and then calculated the mean value for interface-forming residues, normalized by the amount of surface area per residue buried in the interface.

We would like to thank the staff of the IGMM Advanced Imaging Resource.