Extended Synaptotagmin Interaction with the Fibroblast Growth Factor Receptor Depends on Receptor Conformation, Not Catalytic Activity*

Background: Extended Synaptotagmins selectively bind the active FGF receptor but the mechanism is unknown. Results: ESyt2 and 3 but not ESyt1 recognize the active conformation of the FGF receptor independently of catalytic activity. Conclusion: ESyt2 and probably ESyt3 access the active cleft of the FGFR1 kinase domain. Significance: How ESyts recognize FGFRs is essential to understanding how they modulate receptor signaling. We previously demonstrated that ESyt2 interacts specifically with the activated FGF receptor and is required for a rapid phase of receptor internalization and for functional signaling via the ERK pathway in early Xenopus embryos. ESyt2 is one of the three-member family of Extended Synaptotagmins that were recently shown to be implicated in the formation of endoplasmic reticulum (ER)-plasma membrane (PM) junctions and in the Ca2+ dependent regulation of these junctions. Here we show that ESyt2 is directed to the ER by its putative transmembrane domain, that the ESyts hetero- and homodimerize, and that ESyt2 homodimerization in vivo requires a TM adjacent sequence but not the SMP domain. ESyt2 and ESyt3, but not ESyt1, selectively interact in vivo with activated FGFR1. In the case of ESyt2, this interaction requires a short TM adjacent sequence and is independent of receptor autophosphorylation, but dependent on receptor conformation. The data show that ESyt2 recognizes a site in the upper kinase lobe of FGFR1 that is revealed by displacement of the kinase domain activation loop during receptor activation.

We previously demonstrated that ESyt2 interacts specifically with the activated FGF receptor and is required for a rapid phase of receptor internalization and for functional signaling via the ERK pathway in early Xenopus embryos. ESyt2 is one of the three-member family of Extended Synaptotagmins that were recently shown to be implicated in the formation of endoplasmic reticulum (ER)-plasma membrane (PM) junctions and in the Ca 2؉ dependent regulation of these junctions. Here we show that ESyt2 is directed to the ER by its putative transmembrane domain, that the ESyts hetero-and homodimerize, and that ESyt2 homodimerization in vivo requires a TM adjacent sequence but not the SMP domain. ESyt2 and ESyt3, but not ESyt1, selectively interact in vivo with activated FGFR1. In the case of ESyt2, this interaction requires a short TM adjacent sequence and is independent of receptor autophosphorylation, but dependent on receptor conformation. The data show that ESyt2 recognizes a site in the upper kinase lobe of FGFR1 that is revealed by displacement of the kinase domain activation loop during receptor activation.
The Extended Synaptotagmin-like proteins (ESyts) 2 are similar in general structure to the Synaptotagmins, a C2 domaincontaining family of proteins involved in calcium-mediated secretion and endocytosis (1). So far three proteins have been discovered that belong to the Extended Synaptotagmin family, ESyt1, ESyt2, and ESyt3. ESyt1 was originally discovered in vesicle preparations from rat adipocytes and named vp115 for 115kDa vesicular protein (2). However, it was not until 2007 that all three family members were initially studied and the name Extended Synaptotagmin first coined (3). The domain architecture of the human ESyts revealed a putative N-terminal transmembrane domain (TM), an SMP (Synaptotagmin-like Mitochondrial lipid-Binding Protein) domain (4) followed C-terminally by multiple C2 domains. Human ESyt2 and ESyt3 each contain three C2 domains (C2A, C2B, and C2C) while ESyt1 has five (C2A to C2E) (Fig. 1A), and this organization is conserved in mouse (5), Xenopus (6) and to a surprising extent in the yeast Tricalbins (7).
Jean et al. (6) provided the first potential function for ESyt2 when they showed that it acted as an endocytic adapter specific for the activated FGF receptor and was required for functional signaling via the ERK MAP-kinase pathway during early Xenopus development. Xenopus ESyt2 was also later shown to recruit the p21-GTPase Activated Kinase PAK1 and to regulate the dynamics of the actin cytoskeleton (8). More recently, the yeast Tricalbins were shown to be endoplasmic reticulum (ER)-resident proteins that aid in the formation of ER to plasma membrane junction sites or bridges (9), and the human ESyts were shown to play a similar role (10,11). The ESyts were also shown to associate with the ER membrane, probably via a TM hairpin, and to help tether the ER to the PM. Further, ESyt1 was shown to respond to cytosolic Ca 2ϩ by translocating to the sites of ER-PM junctioning and to promote the replenishmentofPM-associatedphosphatidylinositol4,5-bisphosphate (PIP2) (10).
Here we have investigated the molecular basis for the specificity of the ESyt-FGFR interaction. We have characterized the homologous and heterologous interactions between the human ESyts, the interactions of each with FGFR1, defined the homologous interaction and ER targeting domains of ESyt2 and used extensive deletion and point mutations to investigate the molecular specificity of the ESyt2-FGFR1 interaction. The data surprisingly reveal a mode of interaction that is independent of receptor autophosphorylation, or indeed catalytic activity, and is solely dependent on the active receptor conformation. The data show that ESyt2 recognizes a site in the upper kinase lobe of FGFR1 that is revealed when the activation loop is displaced into the active configuration.

Experimental Procedures
Plasmid Constructs-Full-length human ESyt1, ESyt3, and Syt1 cDNAs were amplified from total MCF-7 cDNA and corresponded in coding sequence to FAM62A (NM_015292), FAM2C (NM_031913), and SYT1 (NM_005639). Construction of the human ESyt2b splice variant cDNA was previously described (6) and was equivalent in reading frame sequence to FAM62B (NM_020728). The open reading frame for the human ESyt2a splice variant was created from the ESyt2b cDNA by replacing the sequences 5Ј of the unique SacII site with a synthetic custom gene fragment (Integrated DNA Technologies, IDT) corresponding to the equivalent region of sequence DQ993201. All the ESyt mutants and epitope-tagged constructs were created in these original cDNAs, subcloned in PCDNA3, and the full coding sequence of each was determined. The human FGFR1 was obtained from J. Wesche and E. M. Haugsten and was subcloned along with a FLAG N-or HA C-terminal epitope tag in the pCS2ϩ vector. FGFR1 mutants were created directly in these constructs using the QuickChange strategy (Agilent Technologies) or by direct PCR amplification and the full coding sequence of each final mutant was determined.
Immunofluorescence Imaging-Cells were washed with PBS, fixed in 4% PFA for 15 min, and permeabilized with 0.5% Triton in PBS for 5 min. Incubation with the appropriate primary antibodies was performed for 1 h in PBS, 5% BSA, or goat serum, and cells were then stained with AlexaFluor 488, 568, or 647-conjugated anti-rabbit or mouse secondary antibodies (Molecular Probes) and counterstained with DAPI. After mounting in 50% glycerol, 50% glycine buffer (0.2 M Na-glycine, 0.3 M NaCl), three-dimensional epifluorescent image stacks were acquired using a Leica SP5 II confocal microscope, equipped with a 63ϫ immersion objective, running in standard scanning.
For FGFR1 uptake assays the above protocol was modified as follows; cells expressing N-terminally FLAG-tagged FGFR1 and HA-tagged ESyt2b were rinsed in Opti-MEM (Invitrogen) and then incubated for 1 h at 4°C with rabbit anti-FLAG antibody diluted 1/500 in Opti-MEM. Subsequently cells were rinsed twice in Opti-MEM at 4°C, incubated for 20 min in Opti-MEM-containing bFGF (Sigma), 20 ng/ml, and heparin (Sigma), 5 g/ml at either 37°C or 4°C (control), rinsed twice, and stained "live" with AlexaFluor 568-conjugated anti-rabbit antibody diluted 1/250 in Opti-MEM for 1 h at 4°C. Cells were then fixed and permeabilized before incubation with mouse anti-HA antibody (12CA5) and staining with AlexaFluor 488conjugated anti-rabbit and AlexaFluor 647-conjugated antimouse antibodies, and counterstaining with DAPI.
The use of SNAP-tags (New England Bioloabs) followed the manufacturer's recommendations. Briefly, cells were incubated for 30 min in cell impermeable SNAP-Surface AlexaFluor 488 in culture medium. After three rinses in culture medium, cells were further incubated for 30 min in cell permeable SNAP-Cell TMR-Star in culture medium. Cells were rinsed three times over 30 min to permit unreacted SNAP-Cell ligand to diffuse out of cells, then fixed, and observed as above.
In Vitro Pull-down Assay-Combined in vitro transcription/ translation of WT and mutant FLAG-hESyt2b proteins (cloned in pCDNA3) were carried out using the T7-TNT transcriptiontranslation kit (Promega) using unlabeled and 35 S-labeled methionine. The reaction mixture was diluted to 1 ml with Nonidet P-40 lysis buffer (25 mM Tris-HCl, pH 8, 50 mM NaCl, 1% Nonidet P40). A 250-l aliquot was used for each pull-down assay. Equal amounts of bacterially expressed GST, GST-FGFR1 aa361-752, GST-FGFR1 aa361-562, and GST-FGFR1 aa361-550 were immobilized on G-Sepharose (GE Healthcare) and incubated with in vitro-translated E-Syt2 WT or ⌬TM for 2 h at 4°C. The G-Sepharose was then washed five times with Nonidet P-40 lysis buffer and eluted proteins resolved on SDS-PAGE. After electrophoretic transfer the proteins were revealed by Ponceau Red staining followed by Western blot using FLAG-HRP antibody (Sigma) or, in the case of 35 S labeling, by phosphoimaging on a FLA-5100 (FUJIFILM Life Science).
we were unable to detect these proteins on the PM via N-terminal FLAG epitope-tags (Fig. 1B), or indeed HA-tags (data not shown) despite the same tags being fully available after cell permeabilization. A SNAP-tag™ fused to the N terminus of ESyt2b was also not available before cell permeabilization, despite an N-terminal SNAP-tag fused to the adrenergic receptor b2 (ADRb2) being easily detected with both the cell-impermeable SNAP-surface and cell permeable SNAP-Cell ligands (NEB) (Fig. 1C). Recently it was found that rather than being inserted into the PM the ESyts are probably inserted into the membrane of the endoplasmic reticulum (ER) (10,11). These data exclude the possibility that the ESyts traverse the PM and are consistent with an association with the ER, but also with a non-penetrating mode of membrane association such as recently proposed (10,11).
ESyt2b Is Misdirected to the PM by Fusion to the Syt1 Transmembrane (TM) Domain-Exactly what directs the ESyts to insert exclusively into the ER membrane rather than into the PM as does Syt1 is presently not known. To resolve this question we created a fusion between Syt1 and ESyt2b, such that the potential transmembrane/membrane-insertion (TM) domain and N-terminal sequences of ESyt2b were replaced by those of Syt1. This resulted in an ESyt that associated with and penetrated the PM much as did Syt1 (Fig. 1D). Given that the ESyts and their splice variants display little if any homology preceding the potential membrane-associated domain (Fig. 1E), this strongly suggests that the determinants for association with the ER membrane lie within the membrane-associated domain or the ϳ20 aa preceding it.
The ESyts Homo-and Hetero-dimerize-Given their localization in the ER, to better understand the function of the ESyts in FGF signaling, we first wished to establish if they function as monomers or as hetero-or homodimers. Recent data demonstrated that the ESyts can probably heterodimerize and homodimerize (6,11). Consistent with this, when differentially tagged versions of the three ESyts were expressed in homologous and heterologous pairs, it was evident that not only did all three heterodimerize, but also homodimerize (Fig. 2). ESyt1 interacted with itself, with the two N-terminal splice forms of

Extended Synaptotagmins Target the Active FGFR Conformation
ESyt2, (2a and 2b, Ref. 6) ( Fig. 2A), and with ESyt3. ESyt2a interacted both with itself and its N-terminal splice variant ESyt2b, as well as with ESyt1 and 3 (Fig. 2B), and conversely ESyt3 interacted with both ESyt2a and 2b (Fig. 2C). ESyt3 coexpression with ESyt1 was consistently poor but clearly showed an interaction (compare Fig. 2, C and A). Since none of the three ESyts interacted with Syt1, clearly these interactions were highly specific.
ESyt2 Dimerization Maps to Its N-terminal Sequences and Does Not Require the SMP Domain-We further investigated the domain of ESyt2 responsible for its dimerization. The N-terminal regions preceding the membrane domain of ESyt2a and 2b bear little homology, suggesting that these regions were probably not involved. However, when increasingly extensive C-terminal deletion mutants of FLAG-tagged ESyt2b were coexpressed with full-length HA-tagged ESyt2b, interactions were observed with deletions mutants aa 1 to 785, lacking C2C domain, aa1 to 510 lacking C2B and C2C, aa 1 to 359 lacking all three C2 domains, and even aa 1 to 139, lacking the SMP domain (Fig. 3). Thus, the minimal homo-dimerization/oligomerization domain mapped between aa 1 and 139. This was somewhat surprising in the context of the recent crystal structure of two SMP domains that showed ␤-barrel structures contacting end-to-end to form a dimer (15). The data then suggest that either ESyt2 contains two or more redundant dimerization domains or that it predominantly dimerizes via sequences close to or within its putative transmembrane domain that were not present in the ESyt2 structure determination.
ESyt2 and -3, but Not ESyt1 Interact Selectively with the Activated FGF Receptor-We had previously shown that both Xenopus and human ESyt2 interact in a highly selective manner with the activated forms of the FGF receptor family (FGFR1-4) (6). We also showed that endogenous human ESyt2 interacts with endogenous FGFR1 in HEK293T cells, but due to a lack of adequate antibodies to either ESyt2 or FGFR1 we were unable to determine whether this endogenous interaction responded to FGF stimulation (6). We now show that this is indeed the case (Fig. 4A). Endogenous ESyt2 from HEK293T cells coimmunoprecipitated with exogenous FLAG-FGFR1 after FGF stimulation, and this interaction was significantly suppressed when cells were treated with the FGFR1 inhibitor SU5402. Extending these observations we found that this was a common property of the shorter two ESyts, the splice variants ESyt2a and -b and ESyt3 all displaying a strong selectivity for activated FGFR1 and little or no interaction when the receptor was specifically inhibited using SU5402 (Fig. 4, B and C). By contrast, ESyt1 repeatedly displayed little or no interaction with FGFR1 in co-transfection assays as compared with ESyt2a ( Fig. 4D) or indeed ESyt2b or -3 (data not shown). At first sight this suggests a functional difference to the shorter ESyts. However, ESyt1 is predominantly associated with the cytosolic ER membrane and not with ER-PM junctions ( Fig. 1B) (10,11). Hence, the lack of interaction with FGFR may in part be a function of its different subcellular distribution.
ESyt2 May Transiently Accompany Activated FGFR during Early Endocytosis-The lack of an interaction of ESyt1 with the FGF receptor suggested that the interactions of the ESyts were at least in part determined by their subcellular distribution. We had previously shown that ESyt2 was implicated in determining receptor endocytosis and that it associates with Adaptin2 (AP-2) (6). This suggested that, consistent with its PM proximal distribution, its interaction with the activated FGF receptor initially occurs on the PM during the formation of clathrin-coated pits. We therefore asked if ESyt2b was also internalized along with activated FGFR or if its interaction was limited to the PMassociated receptor fraction. When cells expressing N-terminally FLAG-tagged FGFR1 were subjected to FGF stimulation for 20 min at 37°C, as expected a significant level of initially PM-associated FGFR1 was observed to move into endocytic vesicles, while this internalization was prevented at 4°C (Fig.  4E). In contrast, the distribution of ESyt2b (magenta) appeared to remain predominantly proximal to the PM and was not detected in FGFR1-positive endocytic vesicles (green). Further, ESyt2b did not significantly associate with early EEA1-or Rab5-

FIGURE 2. All three ESyt proteins and splice variants homo-and heterodimerize in vivo.
A, N-terminally FLAG-tagged human ESyt1 was co-expressed in HEK293T cells with each of the other N-terminally HA-tagged human ESyt forms and human Syt1. Subsequently, co-immunoprecipitation of FLAG-tagged ESyt1 was determined by Western blot. B and C, same analysis was performed respectively for N-terminally FLAG-tagged ESyt2a and ESyt3.
positive endosomes or indeed with late Rab7-positive endosomes during FGFR1 endocyosis (Fig. 4F). This suggested that ESyt2b interacted exclusively with the activated PM-associated FGFR1. However, it was possible that ESyt2b was internalized with FGFR1 but rapidly stripped from early vesicles, as is the case for AP-2 and clathrin. In fact, this appeared to be the case since ESyt2b immunoprecipitated with endosomes tagged with FYVE domain of the early endosomal marker EEA1 (Fig. 4G). This association was, however, only significant 5 min after FGF stimulation and was lost 15 min later. Thus, it is likely that ESyt2b is initially internalized with activated FGFR1, but is rapidly stripped from the early endosomes.
Interaction of ESyt2 with FGFR1 Is Mediated by a TM Adjacent Domain-To determine the structural determinants of the ESyt-FGFR interaction we used ESyt2b as a canonical model and investigated the interaction of truncation mutants with FGFR1. Co-transfection of FGFR1 with the series of ESyt2b C-terminal deletion mutants showed that loss of one, two or all of the C2 domains (⌬C2ABC) had no inhibitory effect on the interaction with FGFR1 (Fig. 5, A and B). However, deletion of the N-terminal, TM and adjacent sequences to aa136 (⌬TM) very strongly suppressed or eliminated the interaction (Fig. 5, B and C). Given that the sequences N-terminal of the TM show little or no homology between ESyt2a, 2b and 3 (Fig. 1E), it was not surprising that their deletion from ESyt2b (aa 1 to 87, ⌬Nterm) did not eliminate the interaction with FGFR1. This interaction was clearly much weaker than for the WT, though significantly greater than for the ⌬⌻⌴ mutation, at least in part because the ⌬Nterm mutant repeatedly expressed poorly (Fig.  5B). Specific deletion of the SMP domain had no discernable effect on the interaction with FGFR1, and the interaction was suppressed by receptor inactivation (Fig. 5C). Thus, the data show that the C2 and SMP domains of ESyt2b are not required for the interaction of ESyt2b with FGFR1. By contrast, this interaction does require the TM and sequences immediately flanking it (aa 88 to 138), though probably not the unconserved sequences further N-terminal (Fig. 5D). Hence, the domain required for ESyt2 dimerization (Fig. 3) and its interaction with FGFR1 in greater part overlap.
Interaction of ESyt2 with Activated FGFR1 Is Independent of Receptor Phosphorylation-Interactions of signaling modules with activated tyrosine kinase receptors are often mediated by receptor autophosphorylation (16). To determine if this was the case for the ESyt-FGFR interaction we generated mutations of all seven phosphotyrosine sites on FGFR1 and determined if these affected the interaction with ESyt2b. Phosphorylation of the activation loop sites Tyr-653 and -654 (Fig. 6A) is known to be required for FGFR1 activation (17). Consistent with this, Y to F mutation of these sites reduced receptor autophosphorylation and the interaction with ESyt2b. Further, their combined mutation (Y653,654F) suppressed both receptor autophosphorylation and the ESyt2b interaction to the same degree as the kinase dead (KD) ATP-binding site mutation K514A (Fig. 6, A and B). Thus, either ESyt2b recognized receptor phosphorylation or was able to detect the structural changes that accompany FGFR1 activation.
Individual mutation of the five non-regulatory phosphotyrosine sites (Y463, 583, 584, 730, and 766F) on FGFR1 had little if any effect on its interaction with ESyt2b (data not shown). Combined mutation of all five sites (Y5F) somewhat reduced interaction with ESyt2b, but to a much smaller extent than the activation loop mutations (Y653,654F) or the combined mutation of all 7 phosphotyrosines (Y7F) (Fig. 6C). Thus, interaction of ESyt2b with FGFR1 was not mediated by any single phosphotyrosine and even combined mutation of the 5 non-regulatory sites had only moderate inhibitory effects on the interaction. This left the possibility that a specific interaction with the activation loop phosphotyrosines was involved. To test this we argued that replacement of the activation loop phospho-sites of FGFR1 by phosphomimics (Y653,654E) would prevent their phosphorylation and potentially generate a constitutively activated receptor. The Y653,654E mutant displayed catalytic activity, as demonstrated by receptor autophosphoryla-  tion (pY) and also interacted with ESyt2b (Fig. 6D). Further, the interactions of wild type and Y653,654E FGFR1 with ESyt2b were roughly proportional to their relative autophosphorylation levels and hence their relative catalytic activities, and both autophosphorylation and the ESyt interaction could be suppressed by the FGFR inhibitor SU5402 (Fig. 6D). We also generated an FGFR1 in which the Y653,654E mutation was combined with Y to F mutation of the other 5 phosphotyrosine sites (Y5F,653,654E) (Fig. 6E). This mutant receptor was able to recruit ESyt2b, though less efficiently than the wild type. However, in this case the SU5402 inhibitor unexpectedly caused an enhancement of the ESyt2b interaction. The Y653,654E phosphomimetic mutation of the activation loop clearly induces a structural change resembling that of the activated receptor (Fig. 6D). But, when combined with the five Y to F phosphosite mutations (Y5F,653,654E in Fig. 6E), it was quite possible that this structural change was enhanced by SU5402 binding to the ATP fold.
The data to this point strongly argued that no single phosphotyrosine or phosphotyrosine combination was required for the selective interaction of ESyt2b with the activated FGF receptor, and that even the activation loop tyrosines could be replaced by phosphomimetics. We concluded that either the activation loop phosphomimetics themselves were recognized by ESyt2b or the interaction did not require receptor phosphorylation at all and depended on a structural change.
ESyt2 Interaction Depends on an Active Receptor Conformation, not Its Catalytic Activity-The data from the phospho-site mutants suggested that phosphorylation did not play a direct role in the specificity of ESyt2 for the activated receptor, but that receptor activation was required. This suggested that ESyt2 might recognize a specific receptor conformation occurring on activation. FGFR1 activation is brought about by a displacement of the activation loop that allows access to the active site of the kinase domain (18). Further, FGFR1 autophosphorylation was shown to occur via an asymmetric inter-  action between the kinase domains of adjacent receptors, one acting as enzyme and the other as substrate. Thus, receptor tyrosine phosphorylation actually occurs by a transphosphorylation event between the catalytic domains of a receptor dimer. Mutation of arginine aa 577 to glutamic acid (R577E) within the FGFR1 kinase domain was shown to prevent the asymmetric interaction required for this transphosphorylation event and, unexpectedly, to lock the activation loop in the open "active" conformation (19) (Fig. 7A). Thus, this mutation prevents the in cell interaction between catalytic domains of receptor dimers that permits their autophosphorylation, while at the same time locking the activation loop in the activated configuration. Indeed, the isolated FGFR1-R577E is catalytically active in vitro and able to phosphorylate a PLC␥ substrate (19). We argued that if ESyt2b specifically recognized the conformation of the active FGFR1 it should interact with the R577E FGFR1 mutant regardless of its inability to autophosphorylate. ESyt2b did indeed interact with FGFR1-R577E, despite this mutant being clearly unable to autophosphorylate (Fig. 7B) and hence unable to recruit phospho-PLC␥ (pY-PLC␥). As control we used the Y766F mutant that eliminates the phospho-site bound by PLC␥, but has no effect on receptor activation (20,21). FRS2 is constitutively recruited to the receptor and hence for this reason was unlikely to be implicated in the ESyt2 interaction (22). However, to directly test this we also mutated the essential leucine 422 (L422A) within the FRS2 binding site on FGFR1, but found it had no effect on the ESyt2b interaction (Fig. 7B).
We further investigated the interaction of ESyt2b with the FGFR1-R577E mutant in comparison with the inhibited and activated states of the wild type receptor and the kinase dead ATP binding site mutant K514A (23). Both the wild type FGFR1 and the R577E inactive mutant interacted strongly with ESyt2b, while as should be expected, the K514A (KD) mutant did not (Fig. 7C). Interestingly, the inhibitor SU5402 strongly suppressed the interaction of ESyt2b with wild type FGFR1, but had no effect on the interaction of ESyt2b with the R577E mutant. Since this mutant binds ATP and SU5402, this further confirmed that the active receptor conformation and not its catalytic activity were sufficient for the ESyt2b interaction (24). Together these data show that ESyt2 specifically recognizes the open active conformation of FGFR1 independently of either catalytic activity or receptor autophosphorylation.

FGFR Truncation Reveals an Interaction with ESyt2b That Is
Direct and Independent of Catalytic Activity-Because the data to this point indicated an ESyt2-FGFR1 interaction based solely on the recognition of the open receptor conformation, this suggested that displacement of the activation loop revealed an ESyt2 binding site that was otherwise hidden in the inactive receptor. Thus, we decided to ask if this surface would also be revealed in C-terminal receptor deletion mutants (Fig. 8A). Deletion of the C-terminal tail, leaving the kinase domain intact, as expected, had no effect on the ESyt2b interaction, as did deletion of the transmembrane domain proximal Nedd4-1 ubiquitinylation site (⌬6) shown to be required for receptor internalization (25) (Fig. 8B). Surprisingly, deletion of the lower C-terminal kinase lobe and the activation loop also had no effect on the ESyt2b interaction, but deletion to aa 475, to remove most of the N-terminal kinase lobe, eliminated the interaction. This supported the idea that ESyt2b recognized a binding site on FGFR1 contained within the N-terminal or upper C-terminal kinase lobes that was revealed on displacement of the activation loop during receptor activation. To further delineate the ESyt2b binding site, we created three more C-terminal deletion mutants of FGFR1. Partial deletion of the     upper part of the C-terminal kinase lobe to aa 600, or indeed its full deletion to aa 550, had no effect on the interaction of ESyt2b with the receptor (Fig. 8C). However, deletion to aa 500, also removing half the N-terminal kinase lobe, did eliminate the ESyt2b binding site.
To this point the data, though very suggestive of a direct ESyt2/FGFR interaction, did not rule out the possibility that it required a bridging protein. To test this, we asked if the FGFR1 catalytic domain, expressed in Escherichia coli as a GST fusion protein, would bind ESyt2b translated in vitro. Multiple previous studies showed that the FGFR1 catalytic domain expressed in bacteria is already in the activated tyrosine-phosphorylated conformation, e.g. (26). G-Sepharose-immobilized wild type GST-FGFR1 (aa361-765) and the C-terminal truncation mutants (aa361-562 and -550) retained full-length ESyt2b, but did not retain the N-terminal truncation mutant ⌬TM shown to strongly suppress FGFR1 binding in coimmunoprecipitation assays (see Fig. 5, A and C). Thus, the ESyt2b-FGFR1 interaction involves direct contacts between the N-terminal lobe of the FGFR1 catalytic domain and the N-terminal region of ESyt2b.
Together the data strongly suggest that the selectivity of ESyt2b for the active FGFR1 receptor depends on a binding site within the N-terminal kinase lobe of the receptor that is revealed when the activation loop is displaced to its position in the active receptor conformation. The probable structure of the N-terminal kinase lobe in the aa 550 deletion mutant is shown in Fig. 9A. Since we can fully delete the activation loop without affecting the ESyt2b-FGFR1 interaction, contact with the loop itself is clearly unnecessary for the interaction. However, within the context of the wild type FGFR1 the activation loop would prevent ESyt2 access to the underside of the N-terminal kinase lobe in its inactive configuration, but allow access in its active configuration (Fig. 9B). This in turn suggests that the site of the ESyt2 interaction lies proximal to or corresponds with the ATP-binding pocket.

Discussion
The interaction of ESyt2 with FGFR1 was previously shown to be dependent on receptor activation and to be required for a rapid phase of receptor internalization necessary for functional signaling via the ERK pathway during early Xenopus development and shown to be conserved in human (6). Here we have used the human system to investigate the molecular parameters of the ESyt-FGFR1 interaction in order to better understand its underlying specificity. However, ESyt2 is just one of the three member family of human Extended Synaptotagmins. These proteins were recently shown to be implicated in the formation of ER-PM junctions and in the Ca 2ϩ -dependent regulation of these junctions (9 -11). We provide evidence supporting the   observations (10,11) that the ESyts are not integral PM proteins, but rather are inserted into the ER membrane. We also show that, in the case of ESyt2, this function is an intimate property of the putative transmembrane domain, since its replacement by the transmembrane domain of Syt1 redirects ESyt2 to the PM. We further show that all three ESyts heteroand homodimerize to some degree, though ESyt1 may prefer to heterodimerize with ESyt2 rather than with ESyt3. Analysis of ESyt2b deletion mutants showed that neither the C2 domains nor the SMP domain are essential for its homodimerization in vivo. This suggests that the dimerization via the SMP domain observed in the ESyt2b crystal structure is not essential for its dimerization in vivo (15).
We show that ESyt2 (both a and b splice variants) and ESyt3, but not ESyt1, selectively interact in vivo with the activated form of FGFR1. Interaction of ESyt2b with FGFR1 depends strongly, if not exclusively, on the TM and immediately adjacent sequences of ESyt2 (aa 88 to 138). This region is common to ESyt2a and 2b and shows 45% identity/71% similarity with the equivalent region of ESyt3 and 41% identity/61% similarity with the equivalent of ESyt1. Thus, the lack of an interaction of ESyt1 with FGFR may in part be due to the lower homology with ESyt2/3 or to its predominant localization to cytoplasmic ER membranes (10, 11), or both.
We further investigated the factors that determine the selective interaction of ESyt2b with activated FGFR1. We found that while receptor activation is a prerequisite for the interaction, none of the receptor autophosphorylation events were required. Indeed, even the activation loop phospho-sites Tyr-653 and Tyr-654 of FGFR1 could be replaced by phospho-mimics (Y653/ 654E) without affecting either the interaction with ESyt2 or its specificity. Combining these mutations with Tyr to Phe mutation of the five other phosphotyrosine sites (Y5F,653,654F, Fig. 6E) confirmed that none of the sites were directly involved in the interaction. However, it was clear that such extensive mutation of the receptor could have important structural effects, potentially explaining why ESyt2 binding was now enhanced rather than suppressed by the ATP analog inhibitor SU5402 (Fig. 6E). Though this was an unexpected result, it did serve to finally confirm that the catalytic activity of the receptor was not required for interaction with ESyt2, and suggested a role for receptor conformation. Later data in fact showed that ESyt2 binds a surface in the upper kinase lobe of the receptor close to the ATP fold, providing a potential explanation for the contradictory effect of SU5402 in the context of the Y5F,653,654F FGFR1 mutant. However, the interaction of ESyt2 with the R577E FGFR1 mutant confirmed that the ESyt2-FGFR1 interaction was indeed dependent on precise receptor conforma- tion, not activity. The R577E mutation renders FGFR1 catalytically inactive in vivo. In contrast, atomic structure determination showed that this mutation locked the activation loop in the open, active configuration, but otherwise has little effect on kinase domain structure (Fig. 7A) (19). Thus, the ability of ESyt2 to efficiently interact with FGFR1-R577E strongly argued that the interaction was based solely on activation loop conformation and not catalytic activity. This finding suggested that ESyt2 recognized a surface on the FGFR1 catalytic domain that was revealed by displacement of the activation loop. Receptor deletion mapping confirmed this was the case and showed that the interaction site between ESyt2b and FGFR1 lay within the N-terminal kinase lobe of the receptor, probably close to the ATP binding fold. We further demonstrated in vitro that this interaction was direct. Together, the data strongly suggest that the conformation of the receptor activation loop defines access of ESyt2 to the lower surface of the N-terminal FGFR1 kinase lobe including the ATP-binding pocket (Fig. 9). This then explains the high degree of selectivity of ESyt2 and probably ESyt3 for the active form of the FGF receptor.
Our previous data showed that ESyt2 was required in very early Xenopus embryos for functional FGF signaling via the ERK but not the PI3-kinase pathways. The data further demonstrated a function of ESyt2 in a rapid phase of receptor endocytosis in these embryos. Recent data showing that ESyt2 and 3 localize to ER-PM junctions and may act in concert with other junctioning proteins (9) suggests that this function may be only one part of a more complex pathway regulating FGF signaling that involves the regulation of Ca 2ϩ and phosphatidylinositol 4,5-bisphosphate (PIP2) levels (10). FGF signaling directly activates PLC␥, e.g. see Fig. 7B, and hence stimulates the cleavage of PIP2 and the release of Ca 2ϩ into the cytosol (27). This release of Ca 2ϩ was shown to stimulate ESyt1 recruitment to, or the tightening of, ER-PM junctions and the replenishment of PIP2 on the PM (10). Such a feedback mechanism would effectively prolong or enhance signaling via the PLC␥ pathway and hence have an important modulating influence on intracellular signaling. Consistent with this, parallel mass spectrometric studies have shown that the interaction of ESyt2 with ESyt1 is dependent on FGFR activation. 3 Ca 2ϩ has also been shown in other systems to modulate clathrin-dependent endocytosis (28,29). Thus, the response of the ESyts to Ca 2ϩ release may explain their ability to modulate growth factor signaling in Xenopus by controlling the rate of endocytosis (6). Further studies of the ESyts is certain to generate insight into the mechanisms that underly the cell-specific outcomes of growth factor signaling.