Distinct sets of FGF receptors sculpt excitatory and inhibitory synaptogenesis

Neurons of the developing brain must form a network of excitatory and inhibitory synapses in a highly organized fashion, which requires differential molecular control between each synapse type (Chia et al., 2013; Dabrowski and Umemori, 2011; Jin and Garner, 2008; Johnson-Venkatesh and Umemori, 2010; McAllister, 2007; Sanes and Yamagata, 2009; Shen and Scheiffele, 2010; Siddiqui and Craig, 2011; Williams and Umemori, 2014; Williams et al., 2010). An imbalance between excitatory and inhibitory drive in the brain may lead to various neurological and psychiatric diseases, including epilepsy, autism and schizophrenia (Lisman, 2012; Southwell et al., 2014). Thus, the components of molecular signaling mediating synapse formation are key factors in the vulnerability or resilience to such disorders. Hippocampal circuits are crucial for memory formation, emotional processing and social behavior. The hippocampal circuits consist of highly organized synaptic connections, in which excitatory projections relay from the entorhinal cortex (EC) to the dentate gyrus (DG) to CA3 to CA1 to the EC, and are balanced by inhibitory inputs from local interneurons. In the hippocampus, two fibroblast growth factors (FGFs), FGF22 and FGF7, are secreted from dendrites of CA3 pyramidal neurons, and promote differentiation of excitatory and inhibitory presynaptic terminals, respectively (Terauchi et al., 2010, 2015). That this FGF-dependent regulation of excitatory and inhibitory presynaptic differentiation is important for proper brain development is evidenced by altered seizure susceptibility of knockout (KO) animals: Fgf22-KO mice have decreased, and Fgf7-KO mice have increased susceptibility to epileptic seizures (Terauchi et al., 2010; Lee et al., 2012; Lee and Umemori, 2013). However, the signaling mechanisms underlying FGF-dependent synapse formation are currently not understood. Understanding the mechanisms through which FGF22 and FGF7 specifically promote excitatory and inhibitory presynaptic differentiation will give an important insight into how molecular signaling tunes discrete synapse formation.

FGFs regulate a variety of processes, including cell proliferation, migration, differentiation, tissue repair and response to injury in almost all organs (Thisse and Thisse, 2005; Turner and Grose, 2010; Ornitz and Itoh, 2001). In the nervous system, FGF signaling is important in various steps of development, including neural induction, patterning, proliferation, axon guidance and synapse formation (Guillemot and Zimmer, 2011; Jones and Basson, 2010; Mason, 2007; Stevens et al., 2010a; Umemori, 2009). For the 18 secreted FGFs, there are four FGF receptors (FGFRs), which, through b and c splice variants, encode seven distinct receptor tyrosine kinases (Chellaiah et al., 1994; Johnson et al., 1991; Ornitz et al., 1996; Belov and Mohammadi, 2013; Powers et al., 2000). Once FGFRs are activated, they recruit and/or phosphorylate a number of signaling molecules, including fibroblast growth factor receptor substrate 2 (FRS2), phosphoinositide 3-kinase (PI3K) and phospholipase C γ (PLCγ), which result in the activation of downstream pathways (Mohammadi et al., 1996; Furdui et al., 2006; Bae et al., 2009; Lemmon and Schlessinger, 2010). Therefore, the cellular responses to FGFs are regulated at multiple levels by a number of different factors. What is remarkable in FGF-dependent synapse formation is that FGF22 and FGF7 are secreted from the same CA3 pyramidal neurons but specifically organize excitatory or inhibitory presynaptic differentiation. Important questions regarding their specific effects include the identity of the receptors, location of their action and the signaling pathways involved. Understanding the precise FGF signaling mechanisms during excitatory and inhibitory synapse development will not only further our understanding of synaptogenesis, but will also yield more insight into how FGFs dictate cellular outcomes with such remarkable specificity.

Compared with what is known about signaling pathways involved in postsynaptic development (Ebert and Greenberg, 2013; Hagenston and Bading, 2011; Mabb and Ehlers, 2010; Shen and Cowan, 2010; Stamatakou and Salinas, 2014; Tolias et al., 2011), less is known about signaling pathways guiding presynaptic differentiation. The presynaptic contribution of individual signaling pathways has been studied using broad inhibition or overactivation: PI3K/AKT (Martin-Peña et al., 2006; Cuesto et al., 2011), extracellular signal-related kinase (ERK)/mitogen-activated protein kinase (MAPK) (Li et al., 2002; Kushner et al., 2005; Nakata et al., 2005; Wairkar et al., 2009; Giachello et al., 2010) and PLCγ (Yoshida et al., 2009) have been all proposed to be involved in presynaptic development. The contributions of different signaling pathways might reflect the molecular complexity of synapse formation, underscoring how crucial it is to understand the signaling pathways downstream of specific synaptogenic receptors to tease apart this complexity.

In this study, we asked the following important questions regarding the mechanisms of FGF-induced excitatory and inhibitory synapse formation: What are the receptors through which FGF22 and FGF7 mediate their synaptogenic effects? Where do the receptors act? What are the signaling pathways involved? Using multiple null and conditional KO mice and primary cultures prepared from them, we found that distinct but overlapping sets of FGFRs mediate excitatory versus inhibitory synapse differentiation through a presynaptic mechanism that involves PI3K and FRS2 signaling. Our work uncovers mechanisms of FGF-induced specific presynaptic differentiation and contributes to a profound understanding of how appropriate neural networks are established during development in the mammalian brain.

We first sought to identify the FGFRs involved in presynaptic differentiation by investigating whether KO mice of candidate FGFRs show similar synaptic deficits observed in Fgf22-KO or Fgf7-KO mice. We focused on FGFR2b and FGFR1b, because in mitogenesis assays with FGFR-expressing BaF3 cells, FGF22 activates FGFR2b and FGFR1b, and FGF7 activates FGFR2b (Zhang et al., 2006). Fgfr2 and Fgfr1b are expressed in excitatory neurons (Terauchi et al., 2010; Beer et al., 2000) and inhibitory interneurons in the hippocampus (supplementary material Fig. S1). To investigate excitatory presynaptic differentiation in mice lacking Fgfr2b or Fgfr1b, we stained brain sections for vesicular glutamate transporter 1 (VGLUT1), a marker of excitatory synaptic vesicles that cluster at nerve terminals as excitatory synapses develop (Fig. 1). Fgfr2b^−/− mice (Fig. 1A) are not viable postnatally due to lung agenesis, whereas Fgfr2b^+/− animals, although viable, have developmental defects such as stunted seminal vesicle development (Kuslak et al., 2007) and hypoplastic submandibular glands (Jaskoll et al., 2005). Hence, we investigated whether Fgfr2b haploinsufficiency affects excitatory presynaptic differentiation. We focused our analysis in CA3 of the hippocampus, because Fgf22-KO mice show specific defects in excitatory presynaptic differentiation in the CA3 (Terauchi et al., 2010). We found that Fgfr2b^+/− mice have decreased intensity of VGLUT1 immunoreactivity in the stratum radiatum (SR) and stratum lucidum (SL) of hippocampal CA3 at postnatal day 8 (P8), an early stage of synapse formation (Fig. 1B,C). Further analysis revealed that VGLUT1 puncta are smaller and sparser in Fgfr2b^+/− mice compared with wild-type (WT) littermates (Fig. 1D). We next investigated Fgfr1b^−/− mice (Fig. 1E) for excitatory presynaptic differentiation. We found that VGLUT1 intensity is decreased both in the SL and SR layers of the hippocampal CA3 (Fig. 1F,G), and that VGLUT1 puncta are smaller and sparser in Fgfr1b^−/− animals (Fig. 1H). We did not observe appreciable changes in postsynaptic differentiation, assessed by PSD95 immunostaining (an excitatory postsynaptic marker), in either Fgfr2b^+/− or Fgfr1b^−/− mice (supplementary material Fig.  S2). Furthermore, VGLUT1 immunostaining in lateral septal nuclei does not perceptibly change in Fgfr mutant mice, suggesting that loss of Fgfr2b or Fgfr1b does not globally decrease VGLUT1 levels (supplementary material Fig. S3). These data suggest that both FGFR2b and FGFR1b are involved in excitatory presynaptic differentiation in the hippocampal CA3. To determine whether FGFR2b and FGFR1b compensate for the function of each other for excitatory presynaptic differentiation, we investigated the effect of Fgfr1b inactivation on an Fgfr2b^+/− background. We found that Fgfr2b^+/−Fgfr1b^+/− (double heterozygotes) do not augment the loss of VGLUT1 staining in Fgfr2b^+/− mice (Fig. 1I,J), suggesting that there is no apparent redundancy between FGFR2b and FGFR1b on excitatory presynaptic differentiation.

We next investigated inhibitory presynaptic differentiation in Fgfr2b^+/− and Fgfr1b^−/− mice by analyzing vesicular GABA transporter (VGAT) immunostaining to detect the clustering of inhibitory synaptic vesicles (Fig. 2). We found that Fgfr2b^+/− mice have decreased VGAT intensity in SL and SR of hippocampal CA3 at P8 (Fig. 2A,B), resembling inhibitory presynaptic defects in Fgf7-KO mice (Terauchi et al., 2010). Further analysis showed that VGAT puncta in Fgfr2b^+/− mice are smaller and sparser compared with WT littermates in SL (Fig. 2C). By contrast, Fgfr1b^−/− animals do not have defects in inhibitory presynaptic differentiation measured by VGAT intensity, puncta size or density (Fig. 2D-F). Even on an Fgfr2b^+/− background, which has quantitatively lower levels of FGFR-mediated signaling, additional loss of a copy of Fgfr1b (double heterozygotes) does not change the level of inhibitory presynaptic defects relative to Fgfr2b^+/− (Fig. 2G,H). This is consistent with the notion that synapse formation is not simply regulated by the levels of FGFR signaling, but rather that FGFR2b specifically is required for inhibitory presynaptic differentiation. However, it is still possible that loss of both copies of Fgfr1b on an Fgfr2b^+/− background might affect the level of inhibitory presynaptic defects, which might pose an interesting future study. We did not observe appreciable changes in gephyrin immunostaining (an inhibitory postsynaptic marker) in Fgfr2b^+/− mice (supplementary material Fig. S2). VGAT immunostaining in lateral septal nuclei does not perceptibly change in Fgfr2b^+/− mice (supplementary material Fig. S3). These results suggest that FGFR2b, but not FGFR1b, contributes to inhibitory presynaptic differentiation in hippocampal CA3.

FGFR2 and FGFR1 also regulate embryonic neuronal development, including neuronal precursor specification and proliferation (Paek et al., 2009; Gutin et al., 2006; Stevens et al., 2010b). To exclude the influence of embryonic events to presynaptic defects in Fgfr2b^+/− and Fgfr1b^−/− mice, we utilized conditional Fgfr-KO mice and tested whether the receptors are necessary during the postnatal synaptogenic stage. We selectively deleted Fgfr2 or Fgfr1 in a temporally controlled manner by crossing mice possessing tamoxifen-inducible Cre (CreER) under the actin promoter (Guo et al., 2002) with mice with a conditional allele of Fgfr2 (Fgfr2^flox) (Fig. 3A; Yu et al., 2003) or Fgfr1 (Fgfr1^flox) (Fig. 3B; Hoch and Soriano, 2006). Tamoxifen injection (100 µg) efficiently induces CreER-mediated excision of floxed genes (supplementary material Fig. S4). We injected tamoxifen into actin-CreER::Fgfr2^flox/flox or actin-CreER::Fgfr1^flox/flox pups at P0, resulting in Fgfr2-KO or Fgfr1-KO animals, in which both b and c receptor splice forms are deleted postnatally. Both Fgfr2-KO and Fgfr1-KO animals have decreased excitatory presynaptic differentiation in hippocampal CA3 at P8, as measured by VGLUT1 immunoreactivity, compared with their respective controls (Fig. 3C-E). VGLUT1 puncta are smaller in Fgfr2-KO animals, and both are smaller and sparser in Fgfr1-KO animals (Fig. 3E). However, only Fgfr2-KO, but not Fgfr1-KO animals, have decreased inhibitory presynaptic differentiation at P8, measured by VGAT immunoreactivity (Fig. 3F-H). In Fgfr2-KO, but not Fgfr1-KO animals, VGAT puncta are smaller and sparser (Fig. 3H). Both the excitatory and inhibitory presynaptic defects in Fgfr-KO mice persisted at P14 (supplementary material Fig. S5), the end of initial synapse formation in the hippocampus, suggesting that Fgfr inactivation does not simply delay presynaptic differentiation. Thus, during the postnatal synaptogenic stage, excitatory presynaptic differentiation in hippocampal CA3 requires both FGFR2 and FGFR1, and inhibitory presynaptic differentiation requires FGFR2 but not FGFR1.

We have shown that FGFR2b and FGFR1b are necessary for excitatory presynaptic differentiation in vivo (Figs 1 and 3). We next tested whether FGFR2b and FGFR1b are receptors for FGF22 to organize excitatory presynaptic terminals in neurons. We cultured hippocampal neurons from Fgfr2b^−/− embryos, Fgfr1b^−/− P0 pups and their respective WT littermates, treated with FGF22 on day two in vitro (DIV2), and examined response to FGF22 by VGLUT1 staining on DIV14 (Fig. 4A). FGF22 acts on inputs synapsing onto CA3 pyramidal neurons (Terauchi et al., 2010), so we focused our analysis on synapses forming onto CA3 pyramidal neurons. CA3 neurons were identified by staining with the antibody Py (Woodhams et al., 1989), and VGLUT1 puncta contacting Py⁺ dendrites were quantified. WT neurons respond to FGF22 with a twofold increase in VGLUT1 signal along CA3 dendrites; however, Fgfr2b^−/− neurons do not respond to FGF22 (Fig. 4B,C). Closer analysis revealed that the response to FGF22 in WT neurons, but not in Fgfr2b^−/− neurons, comprises increases in VGLUT1 puncta density along CA3 dendrites (Fig. 4D) and VGLUT1 puncta size (Fig. 4E). Fgfr1b^−/− neurons also do not respond to FGF22 (Fig. 4F-I): there was no FGF22-induced increase in density (Fig. 4H) or size (Fig. 4I) of VGLUT1 puncta. In control experiments examining neuronal cell fate, dendritic morphology of CA3 neurons and their excitatory postsynaptic differentiation, we found no apparent differences between WT neurons, Fgfr2b^−/− neurons or Fgfr1b^−/− neurons in the presence or absence of FGF22 (supplementary material Figs S6 and S7), suggesting that FGFR2b and FGFR1b primarily affect presynaptic differentiation. Together, these results indicate that both FGFR2b and FGFR1b are required for FGF22-dependent induction of excitatory presynaptic differentiation.

To determine whether FGFRs are required specifically during the synaptogenic stage in vitro, we temporally restricted Fgfr2 and Fgfr1 deletion. Hippocampal neurons were cultured from actin-CreER::Fgfr2^flox/flox or actin-CreER::Fgfr1^flox/flox mice. On DIV1, 10 nM 4-hydroxytamoxifen (4-OHT) was applied to cultures, generating Fgfr2-KO or Fgfr1-KO neurons. We found that this dose efficiently inactivates Fgfrs (supplementary material Fig. S8). FGF22 was applied on DIV2, and the neurons were assessed on DIV14 for VGLUT1 clustering on Py⁺ dendrites (Fig. 5A). Control neurons respond to FGF22 with a threefold increase in VGLUT1 immunoreactivity along CA3 dendrites, whereas Fgfr2-KO neurons do not respond (Fig. 5B,C). There were no FGF22 responses in VGLUT1 puncta density (Fig. 5D) and size (Fig. 5E) in Fgfr2-KO neurons. Conversely, Fgfr1-KO neurons have an attenuated, but still present response to FGF22 (Fig. 5F,G). Where control neurons respond to FGF22 with a threefold increase in VGLUT1 signal along CA3 dendrites, Fgfr1-KO neurons respond with a twofold increase (Fig. 5G), with significant increases in VGLUT1 density (Fig. 5H) and size (Fig. 5I). Therefore, we conclude that FGFR2 is required during the synaptogenic stage to respond to FGF22 to induce excitatory presynaptic differentiation, whereas FGFR1 is not absolutely required, but is involved during this stage.

Fgfr2b^+/− mice exhibit defects in inhibitory presynaptic differentiation (Figs 2 and 3), similar to Fgf7-KO mice (Terauchi et al., 2010). We next tested whether FGFR2b is a receptor for FGF7 to organize inhibitory presynaptic terminals. Hippocampal neurons from Fgfr2b^−/− embryos, Fgfr1b^−/− P0 pups and WT littermates were treated on DIV2 with FGF7, and assessed at DIV14 for VGAT clustering on Py⁺ dendrites (Fig. 6A). WT neurons respond to FGF7 with a twofold increase in VGAT staining along CA3 dendrites, whereas Fgfr2b^−/− neurons do not respond (Fig. 6B-E). In Fgfr2b^−/− neurons, FGF7 does not increase VGAT puncta density (Fig. 6D) or size (Fig. 6E). Meanwhile, Fgfr1b^−/− neurons respond equally to WT controls with a twofold increase in VGAT intensity along CA3 dendrites (Fig. 6F,G), as well as increases in VGAT density (Fig. 6H) and size (Fig. 6I). We observed no apparent changes in dendritic morphology of CA3 neurons and inhibitory postsynaptic differentiation on CA3 dendrites between WT cultures, Fgfr2b^−/− cultures or Fgfr1b^−/− cultures, with or without FGF7 (supplementary material Fig. S9). Together, these data suggest that FGFR2b, but not FGFR1b, is required for FGF7-induced inhibitory presynaptic differentiation.

To confirm that FGFR2 is required during the synaptogenic stage to respond to FGF7, we temporally restricted Fgfr2 deletion. Actin-CreER::Fgfr2^flox/flox hippocampal cultures were treated with 10 nM 4-OHT on DIV1 to induce receptor deletion, FGF7 on DIV2, and assessed on DIV14 for VGAT clustering along Py⁺ dendrites (Fig. 7A). In response to FGF7, control neurons have a threefold increase in VGAT intensity along CA3 dendrites, whereas Fgfr2-KO neurons do not respond (Fig. 7B,C). Where control neurons respond to FGF7 with increases in VGAT puncta density (Fig. 7D) and size (Fig. 7E), Fgfr2-KO neurons do not. Thus, we conclude that Fgfr2 is required during the synaptogenic stage to respond to FGF7.

As FGF22 and FGF7 are secreted from postsynaptic sites and retrogradely induce differentiation at presynaptic terminals, we hypothesize that FGFR2b and FGFR1b are localized and function presynaptically in the axon. Previous studies reported that FGFR1 and FGFR2 are expressed in axons, dendrites and soma of cultured hippocampal neurons (Li et al., 2002); however, there are no adequate antibodies against FGFR2b or FGFR1b to directly test their endogenous localization. Thus, we used an alternative strategy of overexpressing fluorescently tagged FGFR2b and FGFR1b in cultured neurons to follow their localization (Fig. 8). FGFR2b-GFP and FGFR1b-GFP both localized throughout the cell including the axon, along which FGFRs accumulate in a punctate manner (Fig. 8A,B). In order to demonstrate their localization to presynaptic terminals within the axon, we co-transfected fluorescently tagged synaptophysin, which marks synaptic vesicles (Li and Murthy, 2001; De Paola et al., 2003; Umemori et al., 2004). The majority of FGFR2b-GFP colocalized with synaptophysin-mCherry (Fig. 8C), and the majority of FGFR1b-GFP colocalized with synaptophysin-mCherry (Fig. 8D). This indicates that FGFR2b and FGFR1b can indeed localize to axon terminals, consistent with a presynaptic role.

We then examined whether the receptors function presynaptically. FGF22 acts on excitatory axon terminals contacting CA3 pyramidal neurons, including those from dentate granule cells (DGCs). Therefore, we first focused on FGF22-induced excitatory presynaptic differentiation within the axons of DGCs. Neuronal cultures prepared from DG and CA3 from Fgfr2^flox/flox or Fgfr1^flox/flox animals were sparsely co-transfected at DIV1 with Cre, to induce cell-autonomous receptor deletion, and synaptophysin-YFP, to visualize synaptic vesicles. Co-transfection efficiency was nearly 100%, and efficiency of Cre-mediated excision of floxed genes was estimated to be ∼90% (supplementary material Fig. S4). Thus, the Fgfr was deleted in transfected, synaptophysin-YFP⁺ cells, whereas untransfected cells remained WT. The transfection was very sparse to analyze single cells not contacting other transfected cells (note that synaptophysin-YFP is under the CMV promoter so that we could identify all transfected cells, including astrocytes; see supplementary material Fig. S10), ensuring a cell-autonomous effect. DGCs were identified by staining for the DGC marker Prox1 (Oliver et al., 1993; Williams et al., 2011). Fgfr deletion did not obviously affect DGC axon length (supplementary material Fig. S11). The neurons were treated with FGF22 at DIV2 and fixed for visualization at DIV11 (Fig. 9A,B). Control DGCs (transfected with non-Cre plasmid) responded to FGF22 to induce synaptophysin-YFP clustering, whereas both Fgfr2-deleted and Fgfr1-deleted DGCs failed to respond (Fig. 9C,D), both in terms of synaptophysin-YFP puncta density within the axon (Fig. 9E) or size (Fig. 9F). These data suggest that both FGFR2 and FGFR1 are essential and function in the presynaptic neuron to respond to FGF22.

Next, we studied the functional presynaptic requirement of FGFR2 at inhibitory synapses in response to FGF7. To do this, we performed a similar experiment as described above, but treated with FGF7 and identified transfected interneurons using GABA immunoreactivity (Fig. 9G,H). We found that control interneurons respond to FGF7 to induce synaptophysin-YFP clustering both in terms of density and size, whereas Fgfr2-KO interneurons failed to respond (Fig. 9I-L). These data suggest that Fgfr2 functions in the presynaptic interneuron to respond to FGF7.

Because each coverslip in our experiment contained more than one Cre/synaptophysin-YFP-transfected (i.e. Fgfr-KO) cell, we cannot completely exclude the possibility that other Fgfr-KO cells on the coverslip indirectly affected synaptophysin-YFP clustering in the neuron we examined. However, because only ∼0.1% of cells were transfected (supplementary material Fig. S10), untransfected WT cells would probably mask the indirect effect of sparse receptor deletion. Thus, our data suggest that FGFRs function directly in presynaptic neurons.

Our experiments show that FGFR2 is required in the presynaptic neuron to respond to FGF22. We next asked whether kinase activity is required for and which intracellular signaling pathways are involved in FGF22-dependent presynaptic differentiation. FGFRs belong to the receptor tyrosine kinase family of proteins and have an intracellular kinase domain, which becomes activated after ligand-induced receptor dimerization, triggering a multi-step autophosphorylation cascade (Mohammadi et al., 1996; Furdui et al., 2006; Bae et al., 2009; Lemmon and Schlessinger, 2010). Phosphorylation sites within the intracellular domain also provide docking sites for adaptor proteins, such as FRS2, PI3K and PLCγ, which activate downstream signaling pathways, such as the MAPK/ERK and AKT by FRS2, AKT by PI3K, or IP3/DAG/Ca²⁺ by PLCγ (Mohammadi et al., 1992,, 1996; Burgar et al., 2002; Hart et al., 2001; Hatch et al., 2006; Songyang et al., 1993; Hadari et al., 1998; Ong et al., 2001). In order to study signaling pathways specifically downstream of FGFR2b, we created Fgfr2b mutant constructs to perturb specific downstream signaling (Fig. 10A): a kinase dead (KD) mutant and mutants lacking the FRS2, the PI3K or the PLCγ binding site. We examined whether the mutated FGFR2b constructs can restore FGF22-induced synaptophysin-YFP response in Fgfr2-deleted DGCs. We co-transfected Fgfr2b constructs with Cre and synaptophysin-YFP into DG + CA3 Fgfr2^flox/flox neuronal cultures at DIV1, treated with FGF22 at DIV2, and assessed synaptophysin-YFP clustering at DIV11 (Fig. 10B). As in Fig. 9, DGCs transfected with non-Cre plasmid (control) responded to FGF22, whereas Fgfr2-deleted DGCs did not (Fig. 10C,D). Expression of FGFR2b-WT in Fgfr2-deleted DGCs rescued FGF22-responsiveness, but expression of FGFR2b-KD, FGFR2b-FRS2-deficient or FGFR2b-PI3K-deficient mutants did not. Interestingly, FGFR2b-PLCγ-deficient rescued FGF22-responsiveness (Fig. 10D), although only the rescue of synaptophysin-YFP density (Fig. 10E) and not size (Fig. 10F) was statistically significant. These data indicate that kinase activity of FGFR2b is required to respond to FGF22, and that FGFR2b requires signaling downstream of FRS2 and PI3K to induce excitatory presynaptic differentiation in response to FGF22.

Numerous synaptogenic molecules, such as FGFs, Wnts, neurotrophins, Eph/ephrins, neurexins/neuroligins and leucine-rich repeat transmembrane proteins (LRRTMs), have been identified (Dai and Peng, 1995; Umemori et al., 2004; Terauchi et al., 2010; de Wit et al., 2011; Dickins and Salinas, 2013; Park and Poo, 2013; Siddiqui and Craig, 2011; Xu and Henkemeyer, 2012), but the underlying mechanisms through which they organize presynaptic differentiation are largely unknown. In this study, we demonstrated, using both in vivo and in vitro evidence, that differential use of FGFRs by FGF22 and FGF7 contributes to their presynaptic effects on excitatory and inhibitory synapses in the CA3 of the hippocampus, where the receptors are localized presynaptically in DGCs and interneurons, and that FGFR2b utilizes FRS2 and PI3K signaling to promote synaptic vesicle accumulation. Together, we provide novel insights into mechanisms of excitatory and inhibitory presynaptic differentiation in the mammalian hippocampus that are crucial for proper brain function.

The coordinated development of excitatory and inhibitory synapses in the hippocampal CA3 is guided by FGF22 and FGF7 with remarkable specificity. After the fate of excitatory and inhibitory neurons has been specified, and as their axons reach CA3, FGF22 and FGF7 are secreted from CA3 dendrites at discrete excitatory and inhibitory postsynaptic sites, respectively, to promote presynaptic differentiation of each type of synapses (Terauchi et al., 2010, 2015). We found that, in addition to specific localization of ligand secretion, overlapping but distinct sets of FGFRs are required for the differentiation of excitatory or inhibitory presynaptic terminals, providing robustness to the system. FGF22 and FGF7 both activate FGFR2b, but we propose that FGFR1b plays an important role in coordinating specificity. FGF22, but not FGF7, activates FGFR1b in in vitro suspension cell proliferation assays (Zhang et al., 2006), and we found that, in the hippocampus, FGFR1b is required for excitatory synapse formation in response to FGF22 (Fig. 4), but not for inhibitory synapse formation in response to FGF7 (Fig. 6). Consistent with this specificity, Fgfr2b-lacking mice have excitatory and inhibitory presynaptic deficits, and Fgfr1b-lacking mice have only excitatory presynaptic deficits (Figs 1,Fig. 2-3). The receptor KO mice do not completely lose excitatory or inhibitory presynaptic differentiation, which suggests that presynaptic differentiation is not fully dependent on FGF7 and FGF22. It is likely that other FGFs, FGFRs and/or other synaptogenic molecules contribute to and coordinate presynaptic differentiation. What is the relationship between FGFR2b and FGFR1b in promoting excitatory presynaptic differentiation in response to FGF22? Both FGFR2b and FGFR1b appear to be necessary presynaptically for responsiveness to FGF22 in vitro (Fig. 9); however, they do not appear to be redundant in terms of their effect on excitatory presynaptic differentiation (Fig. 1I,J). One possibility is that FGFR2b and FGFR1b form heterodimers. Under conditions of complete ligand absence, FGFR2 and FGFR1 can form heterodimers (Wang et al., 1997) and the ligand can induce transphosphorylation (Bellot et al., 1991), although it has not yet been demonstrated whether the native forms do so during development. Alternatively, they cooperate as independent homodimers to promote presynaptic differentiation, in which FGFR2b is the dominant receptor required for both excitatory and inhibitory presynaptic differentiation, whereas FGFR1b specifically modulates excitatory presynaptic differentiation in response to FGF22. Interestingly, FGFR1b appears to have a modulatory role in relation to FGFR2b in other systems. For example, FGF10, which, like FGF22, activates both FGFR2b and FGFR1b (Zhang et al., 2006), plays an important role in lung and submandibular gland development. In both systems, FGFR2b promotes glandular proliferation and elongation and induces FGFR1b expression at the end buds, whereas FGFR1b is important specifically for end bud expansion (Steinberg et al., 2005; Patel et al., 2008). Identification of the precise roles of FGFR1b in excitatory presynaptic differentiation is an important next question.

We found that FGFR2b requires kinase activity as well as binding of FRS2 and PI3K for synaptic vesicle recruitment in response to FGF22 (Fig. 10). All three are upstream of AKT signaling, and because the effect of mutating any of these three pathways had a similar effect in our experiments, it is reasonable to speculate that AKT signaling is involved in excitatory presynaptic differentiation downstream of FGFR2b. However, phosphorylation of FRS2 by FGFRs activates not only the PI3K/AKT pathway (through Grb2/Gab1 binding; Ong et al., 2001), but also the MAPK/ERK pathway (through Grb2/Shp2 binding; Hadari et al., 1998). Indeed, deletion of FGFR1 and/or FGFR2 leads to decreases in phospho-MAPK/ERK and phospho-AKT levels (Emmenegger et al., 2013; Zhao et al., 2008; Cai et al., 2013; Hoch and Soriano, 2006; Loilome et al., 2009; Kondo et al., 2007), and overexpression of FGFR2b or FGFR1 increases both phospho-AKT and phospho-ERK levels (Cha et al., 2008; Freeman et al., 2003; Acevedo et al., 2007). It is known that crosstalk can occur between the PI3K/AKT and MAPK/ERK pathways (Moelling et al., 2002), thus, possibly, the MAPK/ERK pathway is being triggered congruently. Indeed, a previous study linked FGF2-induced synaptogenic activity to the MAPK/ERK pathway: application of MAPK/ERK-specific inhibitors blocked the synaptogenic effect of FGF2 (Li et al., 2002). Finally, recent evidence suggests a role for WAVE regulatory complex (WRC) in the regulation of actin cytoskeletal dynamics downstream of synaptic cell adhesion molecules (Chia et al., 2014; Chen et al., 2014). Interestingly, FRS2 contains a putative WRC-binding consensus sequence (Chen et al., 2014), and PI3K can contribute to actin cytoskeletal dynamics (Cain and Ridley, 2009). An interesting possibility is that FGFR2b-FRS2/PI3K signaling might be controlling the local actin environment at the presynaptic terminal for presynaptic differentiation at the site of activation. Finally, two interesting questions remain: Does FGFR1b act through the same or distinct signaling pathways to control excitatory presynaptic differentiation? And do distinct signaling pathways contribute to FGFR-dependent excitatory and inhibitory presynaptic differentiation?

Unbalanced excitation-inhibition has behavioral consequences: Fgf22-KO and Fgf7-KO mice have altered epileptic seizure susceptibility (Terauchi et al., 2010; Lee et al., 2012; Lee and Umemori, 2013). Additionally, other FGFs have also been implicated in epilepsy (Paradiso et al., 2013). What are the behavioral, neurological or psychiatric consequences of losing FGFR2b or FGFR1b signaling in humans? Dominant negative mutations and mutations leading to overactivation in FGFR2 and FGFR1 cause craniosynostosis syndromes, dominated by cranial defects, which also include seizures and intellectual disability (Agochukwu et al., 2012; Melville et al., 2010; Stevens et al., 2010a; Williams and Umemori, 2014). Analysis of single-nucleotide polymorphisms in human patients has linked mutations in FGFR2 to susceptibility to schizophrenia (O'Donovan et al., 2009) and bipolar disorder (Wang et al., 2012), and FGFR1 mutations to susceptibility to depression (Gaughran et al., 2006) and schizophrenia (Shi et al., 2011). These analyses do not differentiate between the isoform-specific contributions of the receptors; however, all of this evidence supports the idea that FGFR signaling is crucial for proper neural development, and FGFRs might prove to be important druggable therapeutic targets in neurological diseases.

Fgfr1b mutant mice were from Juha Partanen (University of Helsinki, Finland) (Partanen et al., 1998). Fgfr2b mutant mice (Fox et al., 2007), Fgfr2^flox/flox mice (Yu et al., 2003; Umemori et al., 2004), Fgfr1^flox/flox mice (Hoch and Soriano, 2006) and actin-CreER mice (Guo et al., 2002; Umemori et al., 2004) were described previously. All mice were on a C57BL/6 background, except Fgfr1b^−/− mice, which were on a mixed 129sv/CD-1 background. WT mice used were either C57BL/6 or ICR/CD-1. All animal care and use was in accordance with the institutional guidelines and approved by the Institutional Animal Care and Use Committees at Boston Children's Hospital and University of Michigan.

For FGF responsiveness assays (Figs 4,Fig. 5,Fig. 6-7), hippocampi were dissected from P0 pups or E18-E19 embryos, dissociated with 0.5% trypsin and grown in culture media [B27 (Gibco), 2 mM L-glutamine, 1× penicillin-streptomycin in neurobasal medium (Gibco)] at a density of 36,000 cells/poly-D-lysine-coated coverslip. For DGC transfections (Figs 8,Fig. 9-10), dissociated DG and CA3 cells were plated at a density of 50,000 cells/coverslip, and culture media was supplemented with 5 mM KCl. 1 μg of plasmids per coverslip were transfected using a CalPhos transfection kit (Clontech). FGF22 (R&D Systems) and FGF7 (PeproTech) were dissolved to 2 nM for responsiveness experiments and 3 nM for synaptophysin-YFP experiments.

To generate Fgfr1b-GFP, the coding region of Fgfr1b cDNA was PCR-amplified from neonatal mouse skin cDNA and fused with GFP in pEGFP-N1 (Clontech). To generate Fgfr2b-EGFP, the coding region of Fgfr2b cDNA from IMAGE clone 5349249 (ATCC) was inserted into pcDNA3.1 with EGFP cDNA.

Fgfr2b mutant constructs were generated using PCR amplification and insertion into the pAP-TAG5 plasmid (GenHunter) using the following combinations of primers: 5′-ATACTAGTCATGGGATTACCGTCC-3′ (primer-N) with 5′-ATGGGCCCTCATGTTTTAACACTGC-CG3′ (primer-C) for WT Fgfr2b; primer-N with 5′-ATGTCGACTTC-CAGTCAAGTGGATGGCTCC-3′ and 5′-ATGTCGACCATTTGTGGTC-TTTTTTTCTTCGTCTATGTTGTTG-3′ with primer-C for Fgfr2b-KD; primer-N with 5′-ATCTGCAGAGTCCAGCTCCTCCA-TGAAC-3′ and 5′-ATCTGCAGAAACCTGTCTCCGCAGGGGG-3′ with primer-C for Fgfr2b-FRS2; primer-N with 5′-ATGAGCTAGCCATGATG-ATGAG-3′ and 5′-TCATGGCTAGCTCATTGGTGC-3′ with primer-C for Fgfr2b-PI3K; and primer-N with 5′-ATAAGCTTTGGATCTCACCC-AGCCTC-3′ and 5′-ATAAAGCTTCC TCATTGGTTGTGAGAG-3′ with primer-C for Fgfr2b-PLCγ. All PCR products were verified by sequencing.

P8 animals were sacrificed by decapitation (Figs 1 and 2) or perfusion with 4% PFA (Fig. 3). Brains were fixed in 4% PFA overnight. Sagittal sections (20 μm) were cryosectioned. Cultured neurons were fixed with 100% methanol, at −20°C for 5 min. The following antibodies and dilutions were used: anti-VGLUT1 (Millipore, AB5905; 1:4000), anti-VGAT (Synaptic Systems, 131003; 1:4000), anti-GFP (Aves, 1020; 1:5000), anti-Prox1 (Millipore, MAB5652; 1:500) and antibody Py [1:25, a gift from M. Webb (Johnson & Johnson) and P. L. Woodhams (GlaxoSmithKline)]. Glycerol with p-phenylenediamine or n-propyl gallate was used as mounting medium.

Twelve-bit images were acquired with epifluorescence microscopes (Olympus) using 20× (Figs 1,Fig. 2-3) and 40× lenses (Figs 4,Fig. 5,Fig. 6,Fig. 7,Fig. 8,Fig. 9-10) with F-View II CCD (Soft Imaging System) and XM10 (Olympus) cameras at 1376×1032 pixels resolution.

Section data were analyzed using MetaMorph software. For each image, a threshold was chosen to exclude signal from background, based on intensity of the fimbria, a myelinated tract of axons exiting CA3 medially.

For analysis of VGLUT1 or VGAT puncta on Py⁺ dendrites (Figs 4,Fig. 5,Fig. 6-7), images were merged using ImageJ (NIH) and Adobe Photoshop in 16-bit, and masks were drawn in Adobe Photoshop over Py⁺ dendrites. Image thresholds were chosen to subtract background staining and separate each punctum. The threshold was then subtracted from the intensity measurement. Dendritic lengths were measured manually by tracing the length in MetaMorph.

Synaptophysin-YFP images (Figs 9 and 10) were analyzed using Fiji software package. The entire axonal arbor was imaged and analyzed. Images were processed to exclude dendrites and converted to 8-bit. For each neuron, a threshold was chosen separately to capture the dimmest punctum and remove diffuse background axonal staining. Masks were applied to images, and puncta size and count were calculated in Fiji. Axon length was traced manually in Fiji.

The statistical tests performed were two-tailed Student's t-tests. All data are expressed as mean±s.e.m.

We thank Masahiro Yasuda, Erin Johnson-Venkatesh and Sivapratha Nagappan Chettiar for critical reading of the manuscript; Mei Zhang, Andrew McNamara, Patricia Yee, Hsin-Lan Wen and Noreen Francis for technical assistance.