The guanine-exchange factor Ric8a binds to the Ca²⁺ sensor NCS-1 to regulate synapse number and neurotransmitter release

Neuronal function is largely dependent on the number of established synapses and the release probability per synapse. In many neuron types – motor neurons in particular – these two features appear to be regulated in an antagonistic manner; cells with more synapses exhibit lower probability of release per synapse, whereas cells with fewer synapses show a higher probability of release. Although Ca²⁺ is the major signal regulating both of these neuronal features (Augustine and Charlton, 1986; Komuro and Rakic, 1996; Zheng, 2000; for reviews see Pang and Südhof, 2010; Meriney and Dittrich, 2013; Frank, 2014), and heterotrimeric G protein complexes have been implicated as mediators (Renden and Broadie, 2003; Wolfgang et al., 2004; Klose et al., 2010; reviewed by Ross, 2008; Schmitz, 2014), a mechanistic link between Ca²⁺ surge and G protein signaling is still unknown. Shedding light on this issue is relevant and urgent, because many cognitive diseases, most of which are without an effective treatment, result from the abnormal regulation of synapses.

Frequenin (Frq) is a high-affinity Ca²⁺-binding protein conserved from yeast to humans [where it is named neuronal calcium sensor 1 (NCS-1)]. Frq was first identified in Drosophila (Pongs et al., 1993), in which the gene is duplicated, frq1 and frq2. In this insect and in the zebrafish Danio rerio, the two proteins, Frq1 and Frq2, are 95% identical, which represents an unusual case of non-divergent duplicates. In addition, neither of the two proteins has acquired any amino acid changes throughout the 12 sequenced Drosophila genomes, which span >40 million years of evolution. Conserved gene/protein duplications are hypothesized to result from the functional subspecialization of one of the components within the context of a more general function for which both duplicates would be required, thus explaining the maintenance of the two virtually identical genes throughout such a long period of time (Rastogi and Liberles, 2005; Sánchez-Gracia et al., 2010). We have tested this hypothesis here.

In Drosophila, over- or under- expression of the two Frq proteins yields similar antagonistic effects on synapse number and neurotransmitter release. Overexpressing either Frq1 or Frq2 increases the release of neurotransmitters and decreases the number of synapses, whereas the loss of function of Frq proteins reduces transmitter release and increases the number of synapses (Romero-Pozuelo et al., 2007; Dason et al., 2009). We previously found that Frq proteins control Ca²⁺ levels through the α1 voltage-gated Ca²⁺-channel subunit encoded by the gene cacophony (cac) (Dason et al., 2009). The mammalian and snail NCS-1 also controls Ca²⁺-channel activity (Weiss et al., 2000; Wang et al., 2001; Tsujimoto et al., 2002). NCS-1 interacts with several targets (reviewed by Dason et al., 2012), including calcineurin (Schaad et al., 1996), PI4Kβ (Hendricks et al., 1999), TRPC5 (Hui et al., 2006), D2R (Kabbani et al., 2002; Saab et al., 2009) and PICK1 (Jo et al., 2008). Human interleukin-1 receptor accessory protein-like (IL1RAPL1) interacts with NCS-1, and both proteins are involved in X-linked mental retardation and autism (Bahi et al., 2003; Pavlowsky et al., 2010). Furthermore, a missense (R102Q) mutation in NCS-1 has been reported in one case of autism (Piton et al., 2008), and schizophrenic and bipolar disorder patients show an excess of NCS-1 in their dorsolateral prefrontal cortex (Koh et al., 2003) and a decrease of the protein in leukocytes (Torres et al., 2009). In view of the large repertoire of functional interactions reported so far, the search for unifying mechanisms that could account for the biology of this Ca²⁺ sensor is justified. A long-debated issue in neurobiology, which could underlie the pathologies mentioned above, is whether the two synaptic features – the probability of release and the number of synapses – are regulated by a single mechanism or result from independent processes that sustain neuron homeostasis. With regard to the Frq proteins, the question is: does Ca²⁺ binding to these proteins trigger a single pathway or, alternatively, two separate pathways? This question has been addressed here, taking advantage of the duplicated frq genes in Drosophila, in contrast to the single homologous mammalian gene, NCS-1.

Following a yeast-two hybrid (YTH) screen performed using a Drosophila embryonic library and full-length Frq1 and Frq2 cDNAs as baits, we isolated Ric8a, a guanine exchange factor (GEF) (i.e. an activator of G protein signaling; AGS), as a binding candidate for Frq2. Screening with Frq1 as bait yielded no candidate for a true positive interaction among 1×10⁶ clones tested. With Frq2, however, out of 3.74×10⁶ clones, a total of 20 verified cases were isolated. All of them contained the same sequence matching CG15797, which corresponds to nucleotides 326–971 of Ric8a cDNA, and to amino acids 73–288 of Ric8a protein (Fig. 1A). Because Frq1 is 95% identical to Frq2, we tested whether Frq1 could bind to Ric8a in a direct YTH assay. The result was negative. The specificity was further validated by in vitro GST pull-down assays. The fusion constructs poly-His–Myc–Frq1 and poly-His–Frq2–HA were expressed in bacteria, purified by affinity chromatography and loaded into GST–Ric8a pre-charged columns, followed by analysis of the corresponding eluates by western blotting. The data showed positive binding for Frq2 but not for Frq1 (Fig. 1B). An in vivo validation was obtained from co-immunoprecipitation (co-IP) assays using HEK293 cell extracts co-transfected with Myc–Ric8a and Frq2–HA (Fig. 1C). Although genomic and cDNA sequences were obtained from the same wild-type strain, Canton-S, we noticed that Frq1 and Frq2 RNAs must undergo adenosine-to-inosine editing to yield an I178M amino acid change in both proteins (Fig. 1D). The change, however, proved to be irrelevant to the Ric8a-binding specificity (Fig. 1E). Also, the binding specificity of Ric8a for Frq2 versus Frq1 was confirmed, this time using V5 tags. Based on these in vitro and in vivo assays, we conclude that the interaction between Frq2 and Ric8a is biologically relevant. However, because Frq1 and Frq2 differ at only 10 amino acids, the issue of binding specificity needs to be addressed.

To understand the molecular basis of the binding, we solved the Frq2 structure by X-ray crystallography. The hydrophobic and ionic character of the Ca²⁺-saturated protein allowed its purification to homogeneity. We obtained two types of crystals that were of space groups P2₁ and P2₁2₁2₁ (supplementary material Table S1). The structure of the unmyristoylated Ca²⁺-bound Frq2 was solved by molecular replacement at 2.2 Å (P2₁ data) and 2.3 Å (P2₁2₁2₁ data) (Fig. 2A,B,D; supplementary material Movie 1). The globular structure of Frq2 displays two faces – one with the three Ca²⁺-binding sites exposed to the solvent (EF face) and, opposite to this, another face with a very hydrophobic and elongated solvent-exposed crevice (crevice face) (Fig. 2B,C; supplementary material Movie 1). The overall Drosophila Frq2 structure is similar to the following four cases: human NCS-1 [PDB codes 1g8i (Bourne et al., 2001) and 2lcp (Heidarsson et al., 2012)] and yeast (Saccharomyces cerevisiae) Frq [PDB codes 1fpw (Ames et al., 2000) and 2ju0 (Strahl et al., 2007], and the root mean square deviation of Ca atoms from the EF-hand regions are 2.1 Å, 3.9 Å (lowest-energy model), 3.2 Å (lowest-energy model) and 2.6 Å, respectively. Among them, the most dissimilar is the human NCS-1 structure 2lcp, solved by nuclear magnetic resonance (NMR; PDB code 1lcp), where helices E from EF-hands 1 and 4 acquire different orientations and the loop between EF-hands 3 and 4 is in a closed conformation, due to interactions with the unstructured C-terminus that is inserted into the hydrophobic crevice.

To understand why Frq2, but not Frq1, interacts with Ric8a, we analyzed the position of the ten differential amino acids to identify candidates for target recognition and binding (Fig. 2A). Amino acids N5, A79, I91, H102, D161 and R162 are solvent exposed and located on the EF face (Fig. 2B, right). By contrast, amino acids D58, T138, G187 and R94 are also solvent exposed but are located on the crevice face (Fig. 2B, left; Fig. 2C). We reasoned that amino acids located near the crevice are more likely to be relevant in target recognition, as is the case with other Ca²⁺-binding proteins (Ames and Lim, 2012). Amino acid D161, although not in the crevice, is located in the Ca²⁺-binding loop of the EF4 motif, between positions Z and –Y. This position is occupied by an invariant G in EF hands (Marsden et al., 1990), and it confers flexibility to the Ca²⁺ loop; thus, it should be included among the candidates. D58, R94 and T138 are placed at the edges of the hydrophobic crevice, whereas G187 is disordered and, consequently, not present in the model. D58 in Frq2 is represented by Q58 in Frq1. Because both amino acids exhibit similar side chains, which contain a carboxyl and an amide group, respectively, it is unlikely that the side chain could account for the binding specificity. T138 is located in a loop between EF-3 and EF-4 that is highly motile, a feature that renders it relevant for target binding.

Thus, based on these structural considerations, out of the ten differential amino acids between Frq2 and Frq1, R94, T138 and D161 seem to be the best candidates to account for the binding specificity. To experimentally probe their relevance, we performed co-IP assays in HEK cells, to investigate the interaction between Ric8a and mutated versions of Frq1 that mimic Frq2 at these candidate amino acids. The data show that K94R, S138T and the double mutant pull down Ric8a as effectively as native Frq2. By contrast, the mutated candidate G161D pulls down Ric8a far less effectively than native Frq2 (Fig. 3A). In addition to these key amino acids, the comparison of the structure of the six independent molecules modeled in the asymmetric units of the two crystal forms indicate that the amphipathic C-terminal helix H10 is highly motile and acquires different positions depending on the physicochemical environment. In molecules B and C of the P2₁ crystal, H10 is inserted into the hydrophobic crevice, where it makes multiple hydrophobic contacts (Fig. 2B–D; supplementary material Movie 2). However, in molecules A and D of this crystal and in the two molecules of the P2₁2₁2₁ crystal, H10 is solvent exposed (Fig. 2D; supplementary material Table S1). This argues that helix H10 could be important for target interaction because it could contact Ric8a. To test this hypothesis, we generated a Frq2ΔH10 mutant by deleting the last 12 amino acids of the C-terminus, and we performed co-IP assays with Ric8a (Fig. 3A). Binding was found to be 30% stronger in the absence of H10 than with the native Frq2, suggesting that the helix does not interact with Ric8a and functions as a built-in competitive inhibitor that blocks the Ric8a binding site. This is consistent with a recent study on Caenorhabditis elegans, which reported that the deletion of the C-terminal tail of NCS-1 did not impair its ability to rescue behavioral defects (Martin et al., 2013). Taken together, R94 and T138 appear to be key amino acids to explain the selective binding, whereas the motile C-terminal H10 helix regulates site accessibility to Ric8a, the targeted substrate.

Although Frq is represented by two genes in Drosophila, its human homolog, NCS-1, is encoded by a single gene. Thus, the Frq2-specific interaction with Ric8a identified in Drosophila might not necessarily be conserved between the human counterparts. We addressed this issue by co-IP assays performed in HEK cells transfected with human NCS-1 and human Ric8a (Fig. 3B). The data show a strong interaction between the two proteins. Human NCS-1 contains an arginine at position 94, as does fly Frq2. Also, the fact that Drosophila S138T Frq1 binds to Ric8a means that hydrophobic contacts might occur between Ric8a and the extra methyl group that is provided by T138, but not S138. Human NCS-1 contains a L at position 138, and thus the necessary hydrophobic contact could be maintained. Finally, R94 is hidden in the two solved human NCS-1 structures (Bourne et al., 2001; Heidarsson et al., 2012). The C-terminal end shows a helical fold in the crystal structure (Bourne et al., 2001), as we observe in our structure, and it is unstructured in the NMR model (Heidarsson et al., 2012). Our data indicate that the C-terminal helix H10 would need to destabilize and leave the crevice in order to interact with Ric8a, as demonstrated here for Drosophila Frq2. Thus, we can conclude that the interaction identified in Drosophila is functionally and structurally conserved in humans.

Because Frq2 is a Ca²⁺-binding protein, and the solved crystal structure corresponds to that observed under Ca²⁺-saturated conditions, we questioned whether the binding to Ric8a is Ca²⁺-dependent. To this end, we carried out co-IP assays in HEK cells co-transfected with Drosophila Ric8a and normal and mutated forms of Frq2 (Fig. 3C). In comparison to the wild-type Frq2, versions with mutations in the functional EF hand motifs showed equivalent binding activity, suggesting that Ca²⁺ is not a requirement for Ric8a–Frq2 binding. Furthermore, we assayed the binding of the human homologs in the presence of EGTA (20 mM) or Ca²⁺ (5 mM) (Fig. 3D) and found that the binding is favored in the absence of Ca²⁺. Thus, we interpret that, under normal cellular physiological conditions and conditions of very low Ca²⁺ concentration, Ric8a and Frq2 will be mostly in the bound state (see Discussion).

Drosophila Frq2 is expressed in the larval and adult central nervous system (CNS) (Pongs et al., 1993; Romero-Pozuelo et al., 2007). Ric8a/synembryn is also expressed in the CNS of C. elegans (Miller et al., 2000) and mammals (Tõnissoo et al., 2003). In Drosophila, Ric8a has been found to localize in mitotic neuroblasts, where it regulates G protein complexes, at early stages of embryonic development (Hampoelz et al., 2005; Wang et al., 2005). However, it is not known whether this is also the case for differentiated neurons, where Frq2 is expressed. To clarify this, we performed in situ hybridization for Ric8a in whole fly embryos. The data show a broad expression during late stages (st.14–16) in the ventral ganglion and the brain (Fig. 4A,B). Similar to Frq proteins (Pongs et al., 1993), Ric8a was also detected in both type Is and Ib boutons, in close proximity to synaptic active zones (Fig. 4C,D). This signal is specific, because it was lost by driving UAS-Ric8a^RNAi expression (Fig. 4E,F). Furthermore, we investigated the subcellular localization of the two proteins by co-transfecting Myc–Ric8a and Frq2–HA into S2 cells. Both proteins were present in the cytoplasm but they only overlapped close to the plasma membrane (Fig. 4G–J). Thus, Ric8a and Frq2 fulfil the localization criteria to undergo a functional interaction at the presynaptic cell membrane.

We previously found that Frq restrains synaptic growth. An increase in the amount of Frq produces a decrease in branching, bouton number and total synapses, whereas a decrease in Frq levels yields the opposite effect (Romero-Pozuelo et al., 2007; Dason et al., 2009). Because Ric8a interacts with Frq2, we applied the same experimental strategy to analyze the possible role of Ric8a in synapses. For this in vivo study, in addition to the standard overexpression and knockdown conditions, we analyzed double combinations of Frq2 and Ric8a constructs in order to identify the functional hierarchy of both proteins.

The data show that RNAi-mediated knockdown of Ric8a decreases the number of presynaptic active zones, whereas the same effect requires the overexpression of Frq2 (Fig. 5A; Table 1). This suggests that Frq2 might negatively modulate Ric8a. Ric8a overexpression does not yield a significant effect. However, in conjunction with Frq2 knockdown, a strong increase in synapse number results. This observation further supports a negative regulation of Ric8a by Frq2. Also, combinations of two conditions that, individually, showed identical phenotypes (e.g. synapse number decrease due to Frq2 overexpression or Ric8a knockdown) yielded the same effect as the two independent manipulations (Fig. 5A). This is indicative that the two proteins work in the same pathway. By contrast, the synapse number decrease that results from Frq2 overexpression is suppressed by the simultaneous overexpression of Ric8a (Fig. 5A). These data further suggest that normalcy results from the balanced activities of Ric8a and Frq2, and that Ric8a is downstream of Frq2.

As a guanine-exchange factor (GEF), Ric8a interacts in vitro with several Gα proteins (Tall et al., 2003), for which it serves as a scaffolding protein, thus facilitating further interactions with specific signaling pathways (Andreeva et al., 2007). However, the in vivo relevance of these Ric8a–G-protein interactions is mostly unknown. We tested the four Drosophila Gα proteins, Gαs, Gαq, Gαi (referred to here as Gs, Gq and Gi) and Concertina (Cta) on synapse number. Gs regulates neuron morphology in Drosophila (Wolfgang et al., 2004) and binds to Ric8a in humans (Klattenhoff et al., 2003). Here, we analyzed the putative involvement of Gs in the Frq2–Ric8a interaction output. In agreement with previous reports, attenuated Gs expression reduced synapse number. However, the same effect was observed when Gs expression was increased (Fig. 5B). This property – the same phenotype under conditions of either overexpression or knockdown – suggests that the protein needs to be activated and performs its normal function within a range of activity levels. Most likely, the overexpression of a non-activated form out-competes the endogenously activated Gs (an excess of the non-activated form of Gs would behave as a dominant negative), whereas the under-expression would have resulted in an activity level below the normal threshold. To test this possibility, we used a constitutively active form of Gs, Gs* (Connolly et al., 1996), which yielded a phenotype opposite to that of the under-expression, resulting in synapse number increase (Fig. 5B). The phenotype observed following overexpression of non-activated Gs – a decrease in synapse number – was transformed into the expected synapse number increase when combined with the overexpression of Ric8a, the Gs activator. In addition, the Gs-knockdown phenotype prevailed over that of Ric8a overexpression, demonstrating that Gs is functionally downstream of Ric8a (Fig. 5B). Taken together, the data indicate that Gs is required in vivo for synaptogenesis, but it requires activation by Ric8a, which, in turn, is downregulated by Frq2.

With respect to other Gα types, Gq appears to regulate synaptogenesis, because the Gq-knockdown condition showed a reduction in synapse number (Fig. 5B). This is consistent with the functional similarity that the two Gα subtypes show in many physiological contexts (Wolfgang et al., 2004). The Gi-knockdown condition showed no effect on synapse number (Fig. 5C) and Cta, whose sequence is rather divergent from the rest of Gα subtypes (Boto et al., 2010), also yielded no effect (Fig. 5C). With respect to non-Gα proteins, we assayed Gβ13F and Gβ76C, and found that these also had no effect on synapse number (Fig. 5C). Thus, we can conclude that both Gs and Gq regulate synapse number.

Although the structural analysis demonstrated the binding specificity of Ric8a for Frq2, rather than Frq1, we tested this specificity in vivo. As shown above (Fig. 5A), Ric8a knockdown elicited a reduction in the number of synapses either by itself or in conjunction with Frq2 overexpression, consistent with Frq2 being a negative regulator of Ric8a. By contrast, the equivalent genetic combination with Frq1 yielded a normal phenotype in terms of synapse number (Fig. 5D). A normal phenotype was also obtained under conditions of Ric8a overexpression (Fig. 5D). Representative examples of neuromuscular junctions (NMJs) are shown in Fig. 5E. These results rule out the possibility that Frq1 could be a negative regulator of Ric8a. Nevertheless, because Frq1 and Frq2 have been shown to regulate synapse number (Dason et al., 2009) (Table 1), it is still plausible that Frq1 could participate in this regulation without directly binding to Ric8a.

Given that the over- or underexpression of Frq increases or decreases quantal content per synapse, respectively (Romero-Pozuelo et al., 2007; Dason et al., 2009), we performed intracellular recordings to determine whether Ric8a also modulates synaptic transmission. Ric8a overexpression increased evoked excitatory junction potentials (EJPs), whereas Ric8a knockdown decreased them (Fig. 6A). In both cases, there was no change in the frequency of spontaneously occurring miniature (m)EJPs, but a mild increase in amplitude (Fig. 6B). We further characterized the effects of Ric8a on synaptic transmission by performing macropatch focal recordings on individual type Ib boutons. Ric8a overexpression increased the amplitude of excitatory junction currents (EJCs) and the number of quanta released per bouton, whereas Ric8a knockdown reduced it (Fig. 6B,C). To further analyze whether the effects on synaptic transmission could result from changes in the number of synapses per bouton, we counted the number of synapses in boutons of the same size (5 µm diameter) and position (the end of a branch) as those used for recordings. The over- or underexpression of Ric8a does not change the average number of synapses per Ib bouton (20.8±1.2 in controls versus 19±1.6 following Ric8a overexpression and 17.7±1 following Ric8a knockdown) (Fig. 6B). Thus, we conclude that Ric8a modulates the quantal content per synapse as we have previously found for Frq2 (Dason et al., 2009).

Intriguingly, the changes elicited by manipulating Ric8a on the probability of release are in the same direction as those elicited by manipulating Frq2, whereas, in terms of the regulation of synapse number, Frq2 and Ric8a yield opposite effects (Fig. 5A; Fig. 6A; Table 1). These observations indicate that the Frq2–Ric8a interaction could trigger differential mechanisms responsible for the regulation of synapse number or neurotransmitter release. To clarify this, we studied how the Frq2–Ric8a interaction affects synaptic release. Because in the independent genotypes (Fig. 6A) both proteins caused the same effects on release, a negative modulation of Ric8a by Frq2 seems unlikely. In the double mutant experiments, we combined two conditions that exhibit opposing phenotypes – Frq2 overexpression, which increases release, and Ric8a knockdown, which reduces it. We reasoned that if Ric8a is negatively regulated by Frq2, this double mutant combination should show the same, or stronger, phenotype than Ric8a knockdown alone. However, the data show that it is the Frq2-overexpression phenotype that dominates, suppressing that of Ric8a knockdown (Fig. 6D). Alternatively, if Ric8a is positively regulated by Frq2, their combination should show mutual suppression of the single phenotypes, and this is not observed. Thus, we must conclude that Ric8a and Frq2 regulate release through different mechanisms.

Attending to the role of Ric8a as a GEF, and given the evidence that Gα proteins are involved in neurotransmitter release, we tested their single and combined effects. In agreement with previous reports (Renden and Broadie, 2003), we found that the knockdown of Gs increased release (Fig. 6D). Thus, the GEF Ric8a might regulate synaptic release through at least the Gs subtype of Gα proteins. Concerning other Gα subtypes, in C. elegans the interaction of Ric8a with Gq has been demonstrated in the context of neurotransmitter release (Miller et al., 1999; Miller et al., 2000). Taken together, these data show that Ric8a is a GEF that functionally interacts with Gs, and likely Gq, in the context of regulating the probability of neurotransmitter release.

Here, we describe a novel interaction between Frq2 and Ric8a. In addition, we show that Ric8a regulates both synapse number and neurotransmitter release. Contrary to most other GEFs, Ric8a is not coupled to a membrane receptor but it traffics from the cytoplasm instead. This strategic positioning and trafficking renders Ric8a as a suitable link between the Ca²⁺ surge and G protein signaling in the context of synapse regulation. The effects on synapse number imply cytoskeletal changes, possibly mediated by tubulin (Roychowdhury and Rasenick, 2008; Davé et al., 2011). The Gs-binding site in tubulin is also used to bind to adenylyl cyclase, eliciting further cyclic nucleotide signaling (Afshar et al., 2004). Consistent with the cytoplasm-to-membrane translocation of Frq/NCS-1, Ric8a and Gα, the cAMP-dependent signaling also requires compartmentalization (Willoughby and Cooper, 2007; Baillie, 2009), suggesting that these proteins assemble in membrane micro-domains at an early step of synaptogenesis. Gs also activates voltage-dependent Ca²⁺ channels (Mattera et al., 1989) in a reinforcing feed-back loop following the initial Ca²⁺ inflow due to membrane depolarization. Concerning Gq, in addition to binding to tubulin, it also activates PLCβ, which mobilizes Ca²⁺ from internal stores (Mizuno and Itoh, 2009), further augmenting the Ca²⁺ surge. These Gs- and Gq-mediated events sustain the Ca²⁺ dynamics in Frq mutants that we have described previously (Dason et al., 2009), and reveal Ric8a as a convergence point of both Ca²⁺ sources.

It is worth noting that, like Frqs but contrary to Ric8a, Gs also causes opposite effects on synapse number and release probability (Table 1). This observation argues in favor of a common mechanism that co-regulates these two neuronal properties in an antagonistic manner. A largely debated issue is whether this coordination results from independent mechanisms of cell homeostasis, or from a single mechanism with two separate outputs. Ric8a and Frq2 regulate both synaptic features. However, they yield similar effects on release while also mediating opposing effects on synapse number (Table 1), which rules out the possibility that release and synapse number could result from one single Frq2–Ric8a-dependent mechanism. Double mutant experiments show that Frq2 acts as a negative regulator of Ric8a in the control of synapse number. By contrast, in the context of transmitter release, the two proteins seem to operate through independent mechanisms. We hypothesize that Ric8a could be part of a switch mechanism for the control of release versus synaptogenesis, which would explain the antagonistic regulation of the two neuronal features (Fig. 7). Based on the in vivo study provided by the genetic analysis, and because Frq2 can bind to Ric8a in the absence of Ca²⁺, we envision that the expected conformational change that Ca²⁺ binding will cause in Frq2 structure will lead to the release of Ric8a (that, as a GEF, will activate G proteins), triggering the signaling cascade to regulate synapse number and the probability of neurotransmitter release. In this context, the motile H10-helix-based mechanism (see below) could titrate the amount of Ric8a that binds to Frq2 and, hence, regulate the levels or type of G-protein signaling. Concerning Frq1, its involvement in synapse regulation is clear (Romero-Pozuelo et al., 2007; Dason et al., 2009) but, in light of the data reported here, its mechanism of activity must be independent of Ric8a and remains to be identified.

The Frq2 versus Frq1 specificity of the Ric8a interaction validates the functional specialization hypothesis for the maintenance of conserved gene duplications (Rastogi and Liberles, 2005; Sánchez-Gracia et al., 2010). Benefiting from the duplicated Frq in Drosophila and through the crystallography-inspired site-directed mutagenesis of candidate amino acids in Frq1, residues R94 and T138 are found to determine the binding specificity. At position 94, the presence of an arginine and not a lysine seems to be crucial. The guanidinium group instead of an amine would permit additional H-bond interactions through nitrogen Nε (Fig. 2B). At position 138, the methyl group of the Thr side chain seems to be interacting with hydrophobic contacts. To our knowledge, there is no precedent of two structurally almost identical Ca²⁺ sensors whose binding specificity is determined by only two amino acid differences.

The dynamics of the binding process deserves some comment. The monoclinic structure shows that molecules B and C display an occluded crevice due to the insertion of two H10 helices – their own helix H10 and a second one from another molecule (Fig. 2C). This situation resembles the interaction between yeast Frq and PICK-1, where PICK-1 uses two α-helical segments, and suggests that, in molecules B and C, these H10 helices mimic the interaction with Ric8a. From the crystal structures presented here, we can infer that in cells, under high Ca²⁺ concentrations, the motile H10 must be inserted into the crevice, occluding it to impede promiscuity of this site. In the presence of Ric8, the target will displace the H10 helix and proper insertion will occur (supplementary material Movie 2). A similar mechanism has been proposed for yeast and human Frq and KChIP1 interacting with their corresponding targets, PI4K3β (Ames et al., 2000; Strahl et al., 2007), dopamine D2/D3 receptors (Lian et al., 2011) and the Kv4.3 channel (Pioletti et al., 2006; Wang et al., 2007). It is plausible that the transient occlusion of the crevice by H10 will serve as a mechanism to quantitatively regulate the probability of binding to Ric8a, as suggested for human Frq (Heidarsson et al., 2012) and similar to the chain-and-ball model of voltage-dependent K⁺ channels for its open-closed status (Fan et al., 2012).

In humans, the interaction is conserved, suggesting a potential role in pathology. IL1RAPL1 and NCS-1 are implicated in X-linked mental retardation and autism (Handley et al., 2010; Pavlowsky et al., 2010). The autism-related missense (R102Q) mutation in NCS-1 abolishes Ca²⁺ dependence, owing to a weakened conformational stability of its C-terminus that affects the cytosolic-to-membrane cycling of NCS-1 (Heidarsson et al., 2012). Because Frq2–NCS-1 is a negative regulator of Ric8, it is plausible that modifying the equivalent cycling of Ric8a and Gs or disrupting the protein–protein interactions with small compounds could modulate the pathological processes.

We used the glutamatergic neuromuscular junction (NMJ) of the female third larval instar as our experimental system. Synapses were visualized using confocal microscopy with the monoclonal antibody nc82, which identifies the Bruchpilot protein, a constituent of the presynaptic active zone (Wagh et al., 2006; Owald et al., 2010), located at the edge of the characteristic T bar specialization of fly synapses (Hamanaka and Meinertzhagen, 2010). Also, presynaptic nc82 spots correlate with postsynaptic GluRII clusters (Wagh et al., 2006; Jordán-Álvarez et al., 2012). Throughout the text, we refer to nc82-positive spots as mature synapses. All counts were obtained from muscle fibre 6/7 of the abdominal segment 3, along with the electrophysiological recordings. One NMJ was studied per female larva and 8–12 larvae were examined per genotype, collected from different crosses and vials to prevent rearing artifacts. Synapse counts were obtained from multiple vials and crosses spanning over 2 years. Recordings were obtained from larvae collected from different vials (3–5) of each cross, and crosses were set at various times throughout a year. In all cases, each cross involved ∼10–15 males and 20–30 females.

The Gal4/UAS system was used to over- or underexpress selected gene constructs (Brand and Perrimon, 1993). RNAi transgenes for Frq2, Ric8a, Gαs60A, Gαq49B, Gαi65A, Gαconcertina, Gβ76C and Gβ13F were from the Vienna Drosophila RNAi center or the Bloomington collection. Lines for the overexpression of normal or constitutively active forms of Gαs, P{UAS-Gαs60A.C}9 and P{UAS-Gαs60A.Q215L}16, respectively, were from the Bloomington stock center. As neural drivers, we used routinely OK6-Gal4, D42-Gal4 and elav-Gal4^C155 http://flybase.org/reports/FBst0008760.htmlhttp://flybase.org/reports/FBst0008760.html. All crosses were raised at 27°C to maximize the Gal4 expression (Matsumoto et al., 1978). Control values from the Gal4 lines correspond to siblings from experimental crosses or to Gal4-x/UAS-LacZ outcrosses.

Frq1 and Frq2 cDNAs, cloned into pAS2-1, were used as baits. Y190 yeast colonies expressing the Gal4 DNA-binding domain fused to Frq1 or Frq2 were identified by immunoblotting with anti-Frq antibody (Gomez et al., 2001). The Gal4 activation domain was supplied as an embryo cDNA library in pGAD10 (Clontech). The subsequent transformation followed standard procedures, using a concentration of 25 mM 3-AT. A total of 1×10⁶ and 3.74×10⁶ library transformants were screened with the Frq1 and Frq2 baits, respectively. Validation of YTH results followed standard procedures.

Six copies of the c-Myc epitope were fused to the N-terminus of Frq1 cDNA and cloned into pRSET(A) (Invitrogen) vector in phase with a poly-His epitope. Similarly, a HA epitope was C-terminally fused to Frq2 by PCR and cloned into the pRSET(C) vector (Invitrogen) in phase with a poly-His epitope. Both constructs were used for standard expression in bacteria and purified under native conditions using His-GraviTrap columns (GE Healthcare). Eluted proteins were desalted in PD-10 columns (GE Healthcare). Ric8a cDNA was PCR amplified from the LD22866 clone (DBGP) and cloned in phase with GST into the pGEX-3X vector (GE Healthcare). The GST–Ric8a fusion protein was expressed in bacteria and attached to GST-Trap columns (GE Healthcare). An aliquot of the flow-through was kept for the subsequent gels as the unbound protein, serving as positive control. The previously purified Myc–Frq1, Frq1–HA or Frq2–HA were rinsed six times with washing buffer (50 mM Tris-HCl pH 7.5, 200 mM NaCl) and passed through one of the GST–Ric8a pre-charged columns. The two columns were rinsed six times with washing buffer and eluted with 200 mM glycine pH 2.5. The eluted volumes were neutralized by the addition of Tris-HCl pH 7.5 and concentrated with Amicom Ultra (10 K) filters (Millipore).

The Myc-Ric8a and the Frq2-HA tagged cDNAs (see above) were cloned into pCDNA3.1 vector (Invitrogen). In the co-IP assays, to test the amino acid relevance in the Ric8a interaction, and due to the possible role of the C-terminus in stabilizing the interaction, a V5 epitope was tagged to the N-terminus of Frq2 and Frq1 by subcloning them into the nV5-pCDNA3.1 plasmid. Point mutations in Frq2 were performed with the Change-IT Directed Mutagenesis Kit (Affymetrix USB) following the manufacturer's instructions. The human Ric8a cDNA (a gift from Gregory G. Tall, University of Rochester Medical Center, Rochester, NY) was subcloned in nV5pCDNA3.1. The NCS-1 construct (provided by Robert D. Burgoyne; Institute of Translational Medicine, Liverpool, UK) was subcloned in pCDNA3.1. Co-transfections into HEK293 cells were performed using Superfect reagent (Qiagen), and the cells were lysed 48 h thereafter. Pre-cleared lysates were incubated overnight at 4°C with or without rabbit anti-HA, anti-V5 (Invitrogen), anti-c-Myc or anti-NCS-1 (Cell Signaling Technology) antibodies as indicated in each experiment. For negative controls, inmunoprecipitations were performed with an unrelated antibody (see figure legends). Samples were incubated with Protein-G–Sepharose (Sigma). Standard procedures for SDS-PAGE and protein transfer onto membranes were used. 10% of the lysate prior to immunoprecipitation was run as the input. The following antibodies and dilutions were used for western blotting; rabbit anti-NCS-1 (1∶1000), mouse anti-V5 (1∶2000), mouse anti-c-Myc (1∶1000; Sigma, M4439), rabbit anti-HA (1∶1000; Sigma, H6908), rabbit anti-Ric8a (1∶1000; provided by Juergen A. Knoblich; Institute of Molecular Biotechnology of the Austrian Academy of Sciences, Vienna, Austria). In order to avoid heavy/light chain antibody interference, blots were incubated with mouse or rabbit TrueBlot (Rockland).

Details on the expression, purification and crystallization of Frq2 have been published previously (Baños-Mateos et al., 2014). To summarize, the Frq2 cDNA was obtained by RT-PCR from whole Drosophila purified RNA extracts and was sequenced. A full-length cDNA was cloned into the pET-Duet plasmid (Novagen) for protein expression in Escherichia coli. The Ca²⁺-loaded protein was purified in two chromatographic steps using hydrophobic and ion-exchange columns (GE Healthcare). The final sample was dialyzed against water and concentrated to 15 mg/ml for protein crystallization. Two different types of plate-like crystals were obtained in the presence of 0.1 M CaCl₂ using PEG4K as the precipitant. Crystal form I (CF-I) was obtained at 4°C and belonged to the monoclinic system. Crystal form II (CF-II) was obtained at 20°C using the detergent Triton X-114 as an additive and corresponded to the orthorhombic system. The crystal structures and structure factors were deposited in the Protein Data Bank [PDB codes: 4by5 (CF-I) and 4by4 (CF-II)].

X-ray diffraction data were collected at 100 K using the ESRF Grenoble Synchrotron radiation source. The diffraction data were processed using Mosflm (Leslie, 2006) and scaled with Scala (Evans, 2006). Data collection statistics are summarized in supplementary material Table S1. Crystals from CF-I and CF-II belonged to space groups P2₁ and P2₁2₁2₁ respectively. The structure of the human Frequenin/NCS-1 (PDB code 1g8i) was used to solve the structure of Drosophila Frq2 by molecular replacement with Phaser (McCoy et al., 2007). The structures were refined at 2.2 Å (CF-I) and 2.3 (CF-II) Å using NCS restraints and jelly body refinement with Refmac (Murshudov et al., 1997). The models were updated with Coot (Emsley and Cowtan, 2004). The stereochemistry of the models was checked with MolProbity (Chen et al., 2010). Analysis of the structure was done with CCP4i programs (Winn et al., 2011), and images were created with CCP4mg (McNicholas et al., 2011). The crystal structures and structure factors were deposited in the Protein Data Bank [PDB codes: 4by5 (CF-I) and 4by4 (CF-II)].

Whole-cell intracellular recordings and extracellular focal recordings were obtained from the ventral longitudinal muscle fiber 6–7 (abdominal segment 3) in HL6 saline (Macleod et al., 2002) as described previously (Romero-Pozuelo et al., 2007). Briefly, sharp glass electrodes filled with 3 M KCl (∼40 MΩ) were used to impale the muscle to measure spontaneously occurring miniature excitatory junction potentials (mEJPs) and stimulus-evoked excitatory junction potentials (EJPs). Extracellular focal recordings on single type Ib boutons (identified using Nomarski optics) were made using focal electrodes with tips having an inner diameter of 5 µm. Quantal content was calculated for individual boutons by measuring the amplitude of stimulus-evoked excitatory junction currents (EJCs) and dividing it by the mean amplitude of the spontaneous miniature excitatory junction currents (mEJCs). Cut segmental nerves were stimulated at 2 Hz using a suction electrode.

All numerical data are presented as the mean±s.e.m. Statistical significance was calculated using Student's two-tailed t-test (unpaired two samples for means) after application of the Kolmogorov-Smirnov method to verify the normality of data distribution. ***P<0.001; **P<0.01; *P<0.05.

We thank the European Synchrotron Radiation Facility (ID14-1 and ID23 beam line) and Javier Garzón and Pilar Sánchez (Cajal Institute, Madrid, Spain) for advice on G-proteins. We are also grateful to Milton P. Charlton (funded by the Canadian Institutes of Health, grant number MOP-82827) and Ernesto Sánchez for use of laboratory equipment and space.