A homeo-interaction sequence in the ectodomain of the fibroblast growth factor receptor.

Interaction of fibroblast growth factor receptors (FGFR) sufficient for a trans-phosphorylation event in which one intracellular domain is substrate for the other is essential for signal transduction. By analysis of the direct interaction of recombinant constructions co-expressed in baculoviral-infected insect cells, we identified a 17-amino acid sequence that is required for the stable interaction between ectodomains of FGFR. The sequence 160ERSPHRPILQAGLPANK176 (Glu160-Lys176) connects immunoglobulin modules II and III. In insect cells, the interaction between Glu160-Lys176 domains occurs independently of intact heparin or FGF binding domains. The sequence is not required for the binding of heparin or FGF-1, but is essential for mitogenic activity of the FGFR kinase in mammalian cells. The results support a model in which the homeo-interaction between Glu160-Lys176 in the ectodomain contributes to the interaction between intracellular domains in mammalian cell membranes (Kan, M., Wang, F., Kan, M., To, B., Gabriel, J. L., and McKeehan, W. L. (1996) J. Biol. Chem. 271, 26143-26148). We propose that the Glu160-Lys176 domain plays a pivotal role in restriction of the interaction between kinases by pericellular matrix heparan sulfate proteoglycan and divalent cations. Restrictions are overcome by FGF or constitutively by diverse gain of function mutations which cause skeletal and craniofacial abnormalities.

trans-Phosphorylation of tyrosines between fibroblast growth factor receptor (FGFR) 1 kinase domains is thought to be necessary for activation of FGFR. It derepresses the access of substrates to the catalytic site of the tyrosine kinase, and creates binding sites containing phosphotyrosine for recruitment of substrates that relay signals to the intracellular compartment (1)(2)(3). Recombinant FGFR kinases without the ectodomain exhibit autophosphorylation activity of which a part is assumed to occur by an intermolecular mechanism. However, the intracellular kinase domain with or without the COOH-terminal tail sequence does not exhibit sufficient mutual affinity to demonstrate oligomers in solution (2)(3)(4). 2 The FGFR kinase with an intact ectodomain exhibits an intrinsic ability to self-associate (1). In cell membranes where concentration and mobility are restricted, primary control of the proximity and the activity of the intracellular FGFR kinase domains occurs by the interaction of the extracellular domains with heparan sulfate and FGF.
Recently, we proposed a model in which interaction between kinase domains is restricted by conformation due to interaction of Ig loop II with divalent cations and heparan sulfate (5). This model predicts an interface in the FGFR ectodomain whose interaction is modified by occupation of each FGFR monomer by an FGF and contributes to the relationship between intracellular kinase domains in a physiological context. In this report, using deletion and site-directed mutagenesis and analysis of the self-association of recombinant FGFR ectodomains coexpressed in insect cells, we localize the interaction domain of FGFR to the sequence 160 ERSPHRPILQAGLPANK 176 (Glu 160 -Lys 176 ). This domain lies between Ig loop II and loop III and by homology to other Ig motifs extends into the NH 2 terminus of loop III. In addition to constituting a homeo-interaction domain between FGFR monomers, the sequence likely plays a role in maintenance of the structural relationship between Ig loops II and III and may contribute to the high affinity binding of some members of the FGF family.
Heterozygous mutations of the counterpart residues to Arg 161 or Ser 162 to cysteines in FGFR3 result in disulfide links between receptors and cause the neonatal lethality and profound dwarfism associated with thanatophoric dysplasia (6,7). Mutations at the counterpart of Ser 162 to tryptophan in the FGFR2 gene and Pro 163 to arginine in the FGFR1, FGFR2, and FGFR3 genes cause less severe abnormalities (8). Both types of mutations are thought to induce FGF-independent FGFR activity presumably by subversion of restrictions on dimerization and trans-activation of FGFR kinases. We propose that the Glu 160 -Lys 176 sequence is the primary sequence within FGFR ectodomains whose interaction between FGFR ectodomains is modified by the presence of FGF in the context of normal cell membranes.
Construction of Recombinant FGFR-FGFR1 recombinant constructs are shown in Fig. 1A. cDNAs coding for the full-length FGFR1␣1, FGFR1␤1, and FGFR2␤1 tyrosine kinases have been described (1,9,10). Murine FGFR3 was a gift from Dr. David Ornitz (11). The 5Ј-noncoding sequence was removed from the FGFR3 cDNA by treatment with BssHI and Klenow enzyme followed by treatment with BclI. The cDNA fragment was ligated at the BclI site with a polymerase chain reaction (PCR) fragment coding for the COOH-terminal of FGFR3 generated with sense primer R31 and antisense primer R32 using the FGFR3 cDNA as the template. The resultant full-length coding sequence of FGFR3 containing short non-coding sequences was cloned into pBlueScript SK vector at SmaI and EcoRI sites. The human FGFR4 cDNA beginning 42 bp upstream of the translation initiation site and ending 123 bp downstream of the end of coding sequence of the transmembrane domain was generated by reverse transcriptase-PCR from human hepatoma HepG2 cells using sense primer R41 and antisense primer R42. Coding sequence for the intracellular domain of FGFR4 was generated from the same cDNA pool by the sense primer R43 beginning 15 bp upstream of the coding sequence of the transmembrane domain and an antisense primer R44 ending 18 bp downstream of the translation stop codon. The PCR fragment coding for the ectodomain was treated with EcoRI and cloned into pBlueScript SK at an EcoRI site. The PCR fragment coding for the intracellular domain was digested with EcoRI and XhoI. The two cDNAs were fragmented and ligated at the AccI site to yield the full-length coding sequence.
For construction of FGFR1␤ variants fused to GST, the coding sequence for GST was first generated using sense G1 and antisense G2 primers, respectively, and Alk5GST-pVL 1392 (12) as the template in a PCR. After treatment with KpnI and BamHI, the GST coding sequence was ligated in frame with partial cDNAs coding for wild type or mutant FGFR1␤ cDNAs at the KpnI site and the constructs were cloned into pBlueScript SK vector at XbaI and BamHI sites. The cDNAs coding for mutants or fragments of the FGFR1 ectodomain were constructed by PCR-mediated mutagenesis as described previously (13)(14)(15). Mutant cDNA products were then ligated with the flanking coding sequences for either the full-length FGFR1 kinase, the extracellular domain of FGFR1 extending 26 amino acid residues past the transmembrane domain (13,14) or the FGFR1␤GST construct described above.
FGFR1␣⌬II was generated by sense and antisense mutant primers D1 and D2 with primers A1 and P2d (10) as the upstream and the downstream flanking primers, respectively, with human FGFR1␣1 as the template. After treatment with BglI and TaqI, the PCR fragment was ligated with flanking FGFR1 sequences at BglI and TaqI sites and cloned into pBlueScript SK at the EcoRI site. FGFR1␣⌬IIAH was generated with sense and antisense primers D3 and D4 and primers P1a and P2d (10) as the upstream and the downstream flanking primers. After treatment with EcoRI and BglI, the PCR fragment was ligated with a cDNA fragments coding for the COOH-terminal portion of FGFR1␣⌬II at its BglI site and then the cDNA was cloned into pBlue-Script vector at the EcoRI site. The FGFR1␤⌬III cDNA was generated using sense and antisense primers ADP1 (14) and D6 with FGFR1␤ as the cDNA template. After treatment with BstXI and HincII, the PCR fragment was ligated with FGFR1␤ flanking sequences at BstXI and HincII sites and the cDNA was cloned into pBlueScript at BstXI and EcoRI sites. For FGFR1␤⌬II, sense and antisense primers P1a (10) and D7 were used to generate the 5Ј-end segment and primer D8 together with primer P6 (10) were used to generate the 3Ј-end using FGFR1␤1 cDNA as the template. After digestion of the 5Ј-fragment with BstXI and HindIII and the 3Ј-end piece with HindIII and HincII, the two cDNA segments were ligated with the rest of the COOH-terminal coding sequence of FGFR1 at a HincII site and cloned into pBlueScript SK at BstXI and EcoRI sites.
Constructs SFR1K 189 and SFR1K 176 have been described elsewhere (14). Scramble mutants FGFR1␤S1 and FGFR1␤S2 were generated with primers S1 and S2, S3 and S4, respectively, together with primers ADP1 and P1b (10) using the FGFR1␤ cDNA template. The cDNA fragments were digested with BstXI and BglII and ligated with flanking sequences coding for full length FGFR1␤ or the GST fusion protein and then cloned into pBlueScript at BstXI and EcoRI sites.
Primer Th1 together with SF-Lys 189 (14) and primer Th2 was used to generate the cDNAs coding for the FGFR1-Gly 148 -Lys 189 C and FGFR1Gly 148 -Lys 189 S fusion fragments used in the yeast two-hybrid analysis. After digestion with EcoRI and PstI, the above cDNAs were cloned into yeast two-hybrid expression vectors pGBT9 and pGAD424, respectively, at EcoRI and PstI sites to result in the Gal4BD-IL and Gal4AD-IL constructions (Fig. 1).
The complete sequence of mutant cDNA fragments generated by the PCR and the sequence across ligation sites with flanking wild type FGFR cDNA sequence was determined. After verifying the sequence, all the cDNA constructions were excised by treatment with BamHI and cloned into baculoviral transfer vector pVL1393 or pcDNAzeo 3.1 (Invitrogen, San Diego, CA). Recombinant baculoviruses were prepared for expression in Spodoptera frugiperda (Sf9) insect cells as described (1,12). Each recombinant viral stock was standardized by analysis of level of expression of antigen determined by immunoblot.
Binding and Covalent Affinity Cross-linking of 125 I-FGF-FGF-1, FGF-2, and FGF-7 were iodinated, reactivated by reduction and purified to a specific activity of 2-5 ϫ 10 5 cpm/ng and analysis of binding and covalent affinity cross-linking of 125 I-FGF was performed as described (5,14,15,18,19 ). Binding mixtures contained 2 ng/ml radiolabeled FGF and 2 g/ml heparin. Prior to analysis of ligand-labeled receptors by SDS-polyacrylamide gel electrophoresis and autoradiography (14,15), radioactivity of extracts was counted by ␥-counter. Levels of receptor antigen present in binding assays were standardized by the immunoassays described below.
Assay for Interaction between FGFR-Sf9 cells in six-well plates (9.6 cm 2 ) were infected or co-infected with the combinations of recombinant virus bearing FGFR cDNAs indicated in the text. After three days at 27°C and removal of the culture medium, the infected Sf9 cells were incubated in the indicated conditions for 30 min at room temperature. Cells were transferred to microcentrifuge tubes, collected by centrifugation and then lysed with 0.5 ml of lysis buffer containing 0.25% Triton X-100 in phosphate-buffered saline (PBS) and 1 mM p-methylsulfonylfluoride. Samples (250 l) of lysate that were immobilized and purified through a GST component were incubated with 15 l of GSH conjugated to Sepharose beads (Pharmacia Biotech, Uppsala, Sweden) for 30 min at room temperature. After four washes with 0.1% Triton X-100 in PBS followed by a wash with 5 mM HEPES-OH buffer (pH 8.0), the specifically bound FGFR complexes were eluted from the beads with 10 mM of GSH in 5 mM HEPES-OH buffer and the eluates were analyzed by immunoblotting.
For samples immobilized and purified by antibodies, 250 l of each lysate was incubated with 2 g of purified monoclonal antibody or 5 l of anti-FGFR serum and 15 l of protein A-conjugated Sepharose beads for 2 h at 4°C. After the beads were washed with 0.1% Triton X-100 in PBS four times, the bound FGFR complexes were extracted with SDSpolyacrylamide gel electrophoresis sample buffer and analyzed by immunoblot with appropriate antibodies as described above. Antibodies and immunoblot procedures have been described in detail elsewhere (20). The immunoblots were scanned and quantified with ImageQuant software (Molecular Dynamics, Sunnyvale, CA).
Yeast Two-hybrid Analysis-HF7c strain yeast cells were co-transfected with expression vectors indicated in the text according to the manufacturer's instructions (CLONTECH, Palo Alto, CA). The yeast cells were then cultured in agar plates containing SD synthetic medium (ϪTrp, ϪLeu) for 4 -6 days. Positive colonies were transferred to filter paper and frozen in liquid nitrogen. The filter was then placed on another paper soaked with Z buffer (0.15 M phosphate buffer, pH 7.0, 10 mM KCl, 10 mM MgSO 4 ) containing 0.3 mM X-Gal substrate (5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside) for ␤-galactosidase for detection of positive blue colonies that exhibited ␤-galactosidase activity.
Transfection and Analysis of BaF3 cells-BaF3 cells were cultured in RPMI 1640 medium containing 5% fetal bovine serum (FBS), 10% WEHI-3 cell-conditioned medium and L-glutamine (24). Cells (1 ϫ 10 6 ) were transfected with 10 g of pcDNA-Zeo3.1 plasmid bearing an FGFR1 cDNA indicated in the text by electroporation (960 F, 220 V) as described (24). The cells were then distributed to 96-well plates and transfectants selected in the presence of 100 g/ml Zeocin. Transfected cells exhibiting 5000 -20000 receptors/cell were selected by screening in binding assays for radiolabeled FGF-1 and analyzed in DNA synthesis and cell proliferation assays.
BaF3 and transfected stock cultures to be used in proliferation assays were washed three times with RPMI 1640 containing 5% FBS. Cells (2 ϫ 10 4 /well) were plated in 96-well plates in 100 l of RPMI 1640, 5% FBS, and 25 g/ml heparin. The indicated levels of FGF-1 were added and the total volume adjusted to 200 l. After 36 h, 1 Ci of [ 3 H]thymidine was added in a volume of 50 l of RPMI 1640 medium. After 6 h at 37°C, cells were harvested with a Packard cell harvester and incorporated thymidine was determined in a TopCount microscintillation plate counter (Packard, Meriden, CT). Specific binding of 125 I-FGF-1 (2 ng/ml) was performed on 2 ϫ 10 6 cells in 0.5 ml of binding buffer containing 2 g/ml heparin as described above.

A Heparin/Heparan Sulfate-and FGF-independent Interaction Domain Resides in the FGFR␤ Ectodomain-Previously
we demonstrated that FGFR1␤1 3 interacts with FGFR1␣2 in insect cells co-infected with the two constructs with sufficient affinity to survive detergent-extraction and co-precipitation with an FGFR1␣-specific antibody (1). FGFR does not interact under the same conditions with a variety of components at high concentrations in the analytical system including serum proteins, immunoglobulins, recombinant fusion partners, host cell proteins and co-expressed members of other transmembrane receptor families (1,12). 2 This suggested the presence of an interaction domain in FGFR1␣2 which is missing most of the catalytic domain of the tyrosine kinase and the COOH terminus. The co-precipitation of FGFR1␣1 or FGFR1␤1 with a chimeric construct comprised of the ectodomain, the transmembrane domain and six amino acid residues of the intracel- FGFR is numbered throughout this study beginning with the initiator methionine of the FGFR1␤ isoform as residue 1. B, conservation of the inter-loop II and III sequence among FGFR. The junction of loop II, the inter-loop sequence, and loop III indicated by molecular models is indicated (5,21). The single residue mutations associated with craniofacial and skeletal syndromes are indicated (22,23). C, binding of FGFR constructions to FGF. Binding and covalent cross-linking of 125 I-FGF-1 or FGF-2 to the indicated recombinant products expressed in Sf9 cells were performed as described under "Experimental Procedures." Each analysis is from 1 ϫ 10 4 cells. Radiolabeled bands at 150, 120, and 85 kDa for R1␣1, R1␤1, and R1␤GST, respectively, correlated with molecular masses for 1 FGF:1 receptor. Upper radiolabeled bands are cross-linked aggregates and uninformative. The lower 60 kDa FGFR1␣1 band in the antigen panel is due to proteolysis (20), but is not radiolabeled relative to the FGFR1␤1 truncate because of lower affinity for FGF (15). A similar truncate of FGFR1␤1 appears in the more sensitive radiolabeled panels even though the band is below the threshold in the antigen panel. Total antigen from 1 ϫ 10 6 of the same infected cells was analyzed by immunoblot with antiserum A50 (bottom panel). lular juxtamembrane sequence fused to GST (FGFR1␤GST) showed that an interactive domain was upstream of most of the FGFR intracellular domain (Figs. 1 and 2A).
Heparin and FGF increased the interaction of FGFR1␤1 with FGFR1␣2 in the membranes of insect cells when both were expressed at low levels at a ratio of about 1 to 1 (1). However, a significant portion of the two isoforms of FGFR interacted independent of heparin and FGF at the lowest level of expression that could still be accurately measured (1). When either FGFR1␣1 or FGFR1␤1 were co-expressed at a high level and at a ratio of approximately 2 to 1 to FGFR1␤GST ( Fig. 2A), no effect of added heparin and FGF on the amount of either isoform that co-precipitated with FGFR1␤GST could be demonstrated (Fig. 2B). Scanning densitometry indicated that less than 10% of FGFR1␣1 and 35% of FGFR1␤1 complexed with FGFR1␤GST under the conditions. FGFR2␣1, FGFR3␣1, and FGFR4␣1 also complexed with the FGFR1␤GST product under the same conditions (Fig. 3).
These results show that in the absence of restrictions imposed by concentration or other factors in mammalian cells, a domain resides in the FGFR1␤GST product that interacts with another FGFR independent of heparin and FGF. The domain required for the interaction is potentially conserved among the four FGFR genes (Fig. 1B).
The Acidic Box Sequence and Ig Loop II Are Not Involved in the Interaction of FGFR Ectodomains-The ectodomain of FGFR exhibits a characteristic sequence domain called the acidic box. It is located at the NH 2 terminus of FGFR␤ and constitutes the inter-loop sequence between loops I and II in FGFR␣. It is not essential for the binding of heparin/heparan sulfate or FGF (10,15). Although the acidic box is alternately spliced in FGFR2␤ (15,25), it is always present in the FGFR␣ isoform (15). Loop II of FGFR contains an essential heparan sulfate-binding domain in its NH 2 terminus. The domain together with the connecting sequence between Ig loops II and III and a segment of the NH 2 terminus of loop III constitutes a common high affinity binding site for FGF-1, FGF-2, FGF-7, and possibly all members of the FGF family of polypeptides (13,14). To determine whether the acidic box or loop II were involved in the interaction between FGFR ectodomains, we deleted the acidic box and loop II from the FGFR1␣ ectodomain. This resulted in a construct composed of loop I, the inter-loop sequence that normally connects loops II and III and loop III (FGFR1␣⌬II) plus the transmembrane domain and 17 residues of the intracellular juxtamembrane (Fig. 1A). A similar construction (FGFR1␣⌬IIAH) was also built in which the acidic box and the heparin-binding domain were present at the NH 2 terminus of loop I (Fig. 1A). Both constructs missing the entire Ig loop II module were unable to bind FGF (Fig. 1B), but retained the ability to interact with intact FGFR1␣1 (Fig. 4) and FGFR1␤1 (data not shown). Unless a separate interaction domain lies within Ig loop I of the FGFR␣ isoform, these results suggested that an FGFR interaction domain resides between the COOH terminus of loop II and the COOH terminus of the Lysates from 2 ϫ 10 6 Sf9 cells infected with the recombinant baculovirus bearing the constructs indicated at the top were divided into two portions. The first portion was used to determine how much FGFR1␣1 (125 kDa) and FGFR1␤1 (95 kDa) reacted with FGFR1␤GST (68 kDa) by purification on GSH-Sepharose beads. Lower bands of doublets in the FGFR1␤ products are due to proteolytic cleavage of the NH 2 terminus at the COOH terminus of the acidic box sequence (13). The immobilized products were analyzed by immunoblot with anti-FGFR1 antibody M5G10 (top panel). The second portion was used to determine the total amount of each product that was present in the extract by purification on protein A-Sepharose beads bearing anti-FGFR1 monoclonal antibodies M2F12 and M17A3. The immobilized products were analyzed by immunoblot with anti-FGFR1 monoclonal antibody M5G10 (bottom panel). Immobilized products are indicated on the right. Ab denotes the immobilizing mouse antibody detected by the goat anti-mouse Ig conjugated to alkaline phosphatase. The presence (ϩ) or absence (Ϫ) of 200 ng/ml FGF-1 and 25 g/ml heparin during the analysis is indicated at the bottom. Densitometric scanning indicated that 10 and 33% of total FGFR1␣1 and total FGFR1␤1, respectively, were complexed to FGFR1␤GST (top panel). The ratios of total FGFR1␣1:FGFR1␤GST and FGFR1␤1:FGFR1GST ϭ 2.0 (lower panel). B, ligand-independent interaction of the FGFR1␤ ectodomain. Sf9 cells that were co-infected with FGFR1␤GST (68 kDa) and FGFR1␤1 (95 kDa) constructs were incubated with the indicated components for 30 min prior to cell lysis. The FGFR complexes were purified on GSH-Sepharose beads and analyzed by immunoblot as described in A. Total antigen indicates the amount of both expression products in the cell lysate which was purified on protein A-beads bearing the A50 anti-FGFR1 serum. FGFR was detected with anti-FGFR1 monoclonal antibody M17A3. Scanning densitometry indicated that 35% of total FGFR1␤1 was complexed to FGFR1␤GST, the ratio of total FGFR1␤1: FGFR1␤GST ϭ 2.0 and the mean ratio of FGFR1␤1:FGFR1␤GST in the purified complexes ϭ 0.4 (range 0.35-0.54) .   FIG. 3. FGFR1 interacts directly with FGFR2, FGFR3, and  FGFR4. The FGFR isoforms indicated at the top (95, 110, 115-120, and 115 kDa, respectively) were co-expressed with FGFR1␤GST and the complexes purified on GSH-Sepharose beads. After release from the beads by incubation with GSH, the bound FGFR antigens were detected by anti-FGFR1 serum (A50), rabbit anti-FGFR2 serum, rabbit anti-FGFR3 serum, or anti-FGFR4 serum. The right lane indicates an immunoblot of an Sf9 extract after infection with all four FGFR and analysis with a mixture of the four antisera. transmembrane domain.
To further confirm the absence of an interaction domain within the acidic box and loop II, we prepared a construct of FGFR1␤GST from which loop III and the inter-loop II-III sequence between Val 159 and Thr 268 was deleted (FGFR1␤⌬IIIGST). The association of FGFR1␤1 with the construct after purification on GSH-beads was markedly reduced relative to FGFR1␤GST (Fig. 5, top panel). However, the FGFR1␤⌬IIIGST product composed of the acidic box, loop II, the extracellular juxtamembrane and transmembrane domains of FGFR retained high affinity for heparin (Fig. 6A) and FGF-1 (Fig. 6B). The binding of FGF-2 to the construction was reduced, but still detectable under the same conditions. No binding of FGF-7 could be detected under the same conditions. However, specific binding of FGF-7 to the FGFR1␤⌬IIIGST product was observed when radiolabeled FGF-7 was increased 10-fold in binding assays and exposure of autoradiographs was prolonged (Fig. 6B). These results suggest that the loop IIheparin complex exhibits a binding domain for all three FGF polypeptides, but that the highest affinity binding of FGF-2 and FGF-7 requires the inter-loop II/III sequence and a segment of the NH 2 terminus of loop III as reported previously (14). These results show that the acidic box and loop II do not support the direct interaction between FGFR. The heparinand FGF-binding domains associated with loop II are neither required nor are they sufficient for the interaction. These data further suggest that Ig module II independently composes a heparin and FGF-1 binding site and contributes to the binding site for FGF-2 and FGF-7. This suggests that a global disruption of the Ig loop II structure when divorced from downstream domains unlikely underlies its failure to associate with other FGFR. More likely an FGFR interaction domain lies downstream of the COOH terminus of loop II.
The FGFR Interaction Domain Resides within the Inter-loop II/III Sequence 160 ERSPHRPILQAGLPANK 176 -To determine whether loop III contributed to the FGFR interaction domain, we constructed a variant of FGFR1␤GST in which the sequence between Glu 29 and Leu 168 was deleted (FGFR1␤⌬IIGST). This results in a construction comprised of loop III, the extracellular juxtamembrane sequence, the transmembrane sequence and six residues of the intracellular juxtamembrane domain. The mutant FGFR1␤⌬IIGST exhibited no detectable heparin or FGF-1 binding (Fig. 6A). However, the construction complexed with FGFR1␤1, but to less extent than FGFR1␤GST (Fig. 5,  middle and lower panels). To determine whether the interac-tion domain overlapped or resided downstream of the main FGF binding domain (14), Sf9 cells were infected with virus bearing the secreted FGFR1 fragments SFR1K 176 or SFR1K 189 together with FGFR1␤GST. As described previously (14), a complex of heparin and SFR1K 189 exhibits an equal affinity for FGF-1, FGF-2 and FGF-7, while the heparin-SFR1K 176 complex exhibits a reduction in affinity for all three FGFs. Both secreted FGFR1 fragments interacted with FGFR1␤GST (Fig.  7). Taken together with the fact that structural domains upstream of the COOH terminus of loop II do not participate in the interaction between FGFR, these results suggest that the interaction domain lies in the sequence 160 ERSPHRPILQAGL-PANK 176 . This sequence connects Ig loops II and III and, by homology to characterized Ig structures, extends into the NH 2 terminus of loop III. The results also suggest that sequence domains downstream of Lys 176 are not required for the stable homeo-interaction defined by these analyses.
To confirm that the inter-loop sequence composed a sequence-specific interactive interface when in the full-length FGFR ectodomain, we constructed mutants FGFR1␤S1 bearing the scrambled sequence 160 IPHPRESR 167 instead of the wild-type sequence 160 ERSPHRPI 167 and FGFR1␤1S2 bearing the scrambled sequence 168 NPQALLGA 175 instead of the wildtype sequence 168 LQAGLPAN 175 (Fig. 1). The FGFR1␤S1 construct lost ability to bind both FGF-1 and FGF-2. However, The infected cells were preincubated with the indicated components and the formation of FGFR complexes were analyzed as described in Fig. 2B. The first nine lanes are FGFR complexes purified from cell lysates on GSH-Sepharose beads. Total antigen refers to the amount of both expression products purified from cell lysates by the A50 anti-FGFR1 serum immobilized on protein A-beads. Visibility of the FGFR1␤⌬IIIGST band is affected by the large amount of Ig at the same molecular mass. The top two panels were analyzed with FGFR1 monoclonal antibody M17A3. The bottom panel shows the lower part of the same blot from the middle panel after analysis with anti-GST antibody (A19). Scanning densitometric analysis indicated that less than 2% of total FGFR1␤1 was complexed to FGFR1␤⌬IIIGST and the ratio of FGFR1␤1/FGFR1␤⌬IIIGST ϭ 0.04 in the top panel. In the middle panel, 23% of total FGFR1␤1 was complexed to FGFR1␤⌬IIGST. Similarities in apparent molecular masses of the GST products and precipitating antibody and lack of high affinity antibodies for FGFR1 loop III made quantification of the ratio of GST expression product to FGFR1␤1 unreliable (far right lane) as well as the ratio of FGFR␤1 to GST products in the immobilized complexes.
FGFR1␤1S2 bound FGF-1 and FGF-2 (Fig. 6B). Both constructions exhibited a marked decrease in the homeo-interaction between mutant monomers (Fig. 8A). Scanning densitometry indicated that less than 2% of FGFR1␤1S1 was complexed with FGFR1␤S1GST and less than 5% of FGFR1␤1S2 associated with FGFR1␤1S2GST when the pairs were expressed at a ratio of 1 to 1.5 (Fig. 8A). The heterotypic interaction of both mutants with the wild type ectodomain FGFR1␤ was considerably increased over the homeotypic interaction between mutants (Fig.  8B). Moreover, the heterotypic interaction between mutant FGR1␤1S1 and FGFR1␤1S2 was also greater than the homeotypic interaction between mutants. These results suggest that (i) both segments, ERSPHRPI and LQAGLPAN of the Glu 160 -Lys 176 sequence domain contribute to the optimum interaction between the FGFR ectodomains, and (ii) heterotypic interactions in trans between the Glu 160 -Lys 176 domain in two differ-ent FGFR may occur.
A Sequence Containing the Inter-loop II-III Domain Glu 160 -Lys 176 Interacts in the Yeast Two-hybrid Expression System-To determine whether the Glu 160 -Lys 176 structural domain would interact when present in an unrelated protein, the amino acid sequence between Gly 148 and Lys 189 from FGFR1 was fused to the Gal4 promoter activation domain and DNA binding domain. The two constructs, Gal4AD-IL and Gal4BD-IL (Fig. 1), were tested for ability to support growth and activate the Gal4 promoter indicated by expression of ␤-galactosidase in HF7C yeast cells. Only cells co-infected with the two constructs supported growth of HF7C cells in Trp-and Leu-deficient medium and yielded the characteristic blue colonies indicative of the presence of ␤-galactosidase (Fig. 9). Individual constructs expressed alone or with unrelated products fused to the activating or DNA binding partner yielded negative results.
Requirement for the Intact Glu 160 -Lys 176 Sequence for Mitogenic Activity of FGFR1-To determine impact of the Glu 160 -Lys 176 sequence on mitogenic activity of FGFR, recombinant FGFR1␤1, FGFR1␤1S2, and FGFR1␤1⌬III with full-length kinase domains were expressed in BaF3 cells which exhibit little or no endogenous FGFR and do not respond to FGF (24) (Fig.  10). Clonal cultures derived from the transfected BaF3 population that displayed about 10000 specific FGF-1 binding sites per cell were selected in binding assays using radiolabeled FGF-1 (Fig. 10, A, inset, and B). Only cells expressing the wild type FGFR1␤1 exhibited a response to FGF-1 (Fig. 10A). A second set of clonal cultures from an independently transfected BaF3 population expressing about 20000 sites per cell were examined at twice the cell density as the experiment in Fig. 10,  A and B (Fig. 10C). Although the base level of FGF-independent DNA synthesis increases under these conditions, the cells expressing intact FGFR1␤1 still exhibited the largest response to FGF-1. These results confirm that the intact Glu 160 -Lys 176 sequence is not required for binding of FGF-1, but is required for mitogenic activity. The results further suggest that the interaction of ectodomains mediated by the Glu 160 -Lys 176 sequence is essential for mitogenic activity. DISCUSSION Auto-phosphorylation of tyrosines in FGFR tyrosine kinases is an obligatory event for FGF signal transduction (1, 3, 26). A recent structural analysis of the FGFR kinase (4) suggests that, similar to the insulin receptor kinase (27), activation of FIG. 6. Heparin and FGF binding activity of interaction-competent and interaction-defective constructs of the FGFR1 ectodomain. A, heparin binding. The indicated constructions fused to GST were expressed in Sf9 cells, extracted and immobilized on GSH-Sepharose beads, and then incubated with 2 g/ml 3 H-heparin for 1 h at room temperature. After washing with PBS, the beads were eluted with GSH and the 3 H-heparin in the eluate was counted by liquid scintillation. Data are the mean of triplicate samples with the indicated standard error. Inset, a portion of the same samples of FGFR1␤⌬III-GST (46 kDa), FGFR1␤⌬II-GST (48 kDa), FGFR1␤-GST (68 kDa), and noninfected Sf9 cells (left to right) used in the heparin-binding assays was subjected to immunoblot analyses with anti-GST serum. B, FGF binding. The indicated constructions of the FGFR1 ectodomain fused to GST (see Fig. 1A) were expressed in Sf9 cells and immobilized from cell extracts on GSH-Sepharose beads. Binding and covalent affinity crosslinking of the indicated radiolabeled FGF to immobilized products containing equivalent amounts of GST antigen was performed as described under "Experimental Procedures." Assays for FGF-1 and FGF-2 contained 2 ng/ml radiolabeled FGF and were exposed for 24 h. Assays for FGF-7 contained 20 ng/ml FGF-7 and were exposed for 10 days. Radiolabeled bands, besides the indicated 1 FGF:1 FGFR complex, are discussed in Fig. 1A. The band at the bottom is uncross-linked FGF that was presumably specifically bound to FGFR.

FIG. 7. Intact Ig Loop III is not required for the interaction between FGFR ectodomains.
Sf9 cells were co-infected with recombinant virus bearing FGFR1␤GST (68 kDa) and different amounts of virus bearing cDNA for FGFR1 fragments, SFR1K 189 (32 kDa) and SFR1K 176 (27 kDa) as indicated at the top. Lower bands of doublets are proteolytic modification as described in Fig. 2A. Cell lysates were incubated with GSH-Sepharose beads and the FGFR1 antigens indicated at the right were analyzed on immunoblots with monoclonal antibody M17A3 after elution with GSH.
the FGFR kinase requires trans-phosphorylation of tyrosine in a cis-acting autoinhibitory domain within the kinase domain. Therefore, the level of activity of FGFR simply depends on concentration or proximity of FGFR molecules sufficient to sustain an enzyme-substrate relationship between intracellu-lar domains. When free of restraints imposed by concentration, position and co-factors in mammalian cell membranes, the FGFR tyrosine kinase self-activates through auto-phosphorylation independent of FGF and pericellular heparan sulfate (1,26). This activity is also independent of other structural domains of the transmembrane receptor including the ectodomain (2, 3) which contains the heparan sulfate and FGF binding sites (13,14). The activation and presence of FGF is normally required for activity of the FGFR signal transduction complex in mammalian cell membranes. Therefore, it is important to understand the role of the ectodomain and its ligands which include FGF and heparan sulfate in control of the enzyme-substrate relationship between the intracellular FGFR kinases. Here we show that the FGFR ectodomain exhibits a homeo-interaction domain that is within the 17-amino acid sequence 160 ERSPHRPILQAGLPANK 176 (Glu 160 -Lys 176 ). The Glu 160 -Lys 176 sequence is 100% conserved (except for Lys 176 ) in all FGFR isoforms in higher organisms and composes the sequence which connects Ig loops II and III. By homology to structurally resolved Ig domains in other proteins, the sequence ERSPHRPI of the Glu 160 -Lys 176 domain begins at the COOH terminus of loop II and comprises the connection between loops II and III. Residues LQAGLPANK extend into the NH 2 terminus of loop III (Fig. 1). Since the FGFR intracellular domain does not exhibit sufficient affinity to co-precipitate under conditions similar to those described here (4), 2 we propose that the Glu 160 -Lys 176 domain in adjacent FGFR monomers plays a major role in dimerization of FGFR and the enzyme-substrate relationship between adjacent kinases.
Recently we reported that divalent cations cooperate with heparin/heparan sulfate to maintain the FGFR kinase in an FGF-dependent state (5). From the results, we proposed a conformationally based model in which heparan sulfate chains from pericellular proteoglycans rigidly anchor adjacent Ig loops II of the FGFR ectodomain (5). When the binding site is not occupied by FGF, the relationship (inactive state I) between loops II and III within an FGFR is such that a homeo-interaction between ectodomains of adjacent molecules is unstable. FIG. 8. Requirement of the intact Glu 160 -Lys 176 sequence for optimum interaction between FGFR. A, FGFR1␤1S1 (95 kDa) and FGFR1␤1S2 (95 kDa) constructions were co-expressed with the same respective ectodomain through Thr 314 fused to GST (FGFR1␤S1GST and FGFR1␤GST at 68 kDa). After incubation of the infected Sf9 cells with the indicated factors and conditions, complexes were purified on GSH-Sepharose beads and the associated antigens were analyzed with monoclonal antibody M17A3 after release with GSH. Total antigen is the same amount of antigen purified from cell lysate by the A50 anti-FGFR1 antiserum immobilized on protein A beads. Quantification by scanning densitometry revealed that total FGFR1␤1S1: FGFR1␤S1GST ϭ 1.0, GSH-bound FGFR1␤1S1:FGFR1␤S1GST ϭ Ͻ0.01 and Ͻ1% of total FGFR1␤1S1 was complexed to the immobilized FGFR1␤S1GST (top panel). In the bottom panel, total FGFR1␤1S2: FGFR1␤S2GST ϭ 1.5, GSH-bound FGFR1␤1S2:FGFR1␤S2GST ϭ Ͻ0.05 and less than 5% of total FGFR1␤1S2 was complexed to the immobilized GST-tagged product. B, the constructions indicated at the top were co-expressed with the GST-tagged constructions indicated at the right in an approximate 2 to 1 ratio. The complexes were purified from cell lysates on GSH-Sepharose, released with GSH and analyzed as described in A. Homeo-interaction between the IL sequences results in blue colonies equal to the p53-LT interaction, while all other combinations resulted in the white refractory colonies. The coding sequence for cysteine was replaced with that for serine in the FGFR1 cDNA fused to that for the activating domain to avoid potential disulfides between the two products.
Consequently the kinase domains are unable to sustain a trans-activating enzyme-substrate relationship (activated state II). FGF stabilizes the state II conformation or increases the time that FGFR complexes reside in the state II conformation. This model which involves a 2 FGF:2 FGFR complex requires that both adjacent FGFR be occupied by FGF even for activation of one partner. We propose that the Glu 160 -Lys 176 domain in the ectodomain may play a pivotal role, possibly as a contact interface, in control of the enzyme-substrate relationship between FGFR kinases by heparan sulfate and FGF in the context of cell membranes.
A variety of autosomal dominant point mutations occurs in FGFR1, FGFR2, and FGFR3 that results in craniofacial and skeletal abnormalities of graded severity. These include achondroplastic, hypochondroplastic, Apert, Crouzon, Jackson-Weiss, and Pfeiffer syndromes in adults and lethal thanatophoric dysplasia (22). The diverse mutations appear to be gainof-function mutations (7, 28 -31). None to date occur within loop II and most occur between the COOH terminus of loop II and the COOH terminus of the transmembrane domain (22). Some mutations disrupt and some have no effect on binding of FGF (10,28,30,31). 2 Several mutations in the ectodomain result in new cysteines capable of forming disulfides between adjacent FGFR (7,22,28). The most severe syndrome is thanatophoric dysplasia which is characterized by neonatal death and profound dwarfism. The phenotype is caused by (i) a mutation of lysine to glutamate in the kinase repressor loop within the kinase domain of FGFR3 (31); (ii) cysteine substitutions in the FGFR3 counterparts of Arg 161 and Ser 162 in the Glu 160 -Lys 176 domain of FGFR1 (7, 23); (iii) cysteine substitutions just NH 2 -terminal to the transmembrane domain (23); or (iii) noncysteine mutations which cause an extension of the distal COOH-terminal sequence (23). The severity of the cysteine substitutions in the Glu 160 -Lys 176 sequence suggests the sequence may comprise a contact domain in which disulfides form more efficiently between FGFR dimers than in other non-contact domains. This suggests a model for how autosomal dominant mutations at different sites within FGFR can cause diverse phenotypes of graded severity from achondroplastic dwarfism to thanatophoric dysplasia. In cells in which 50% of FGFR bear mutations between the COOH terminus of loop II and the intracellular domain, a maximum of 25% of total heparan sulfate-FGFR complexes that are present as mutant-mutant complexes can be FGF-independent and constitutively active. Since occupancy of both FGFR is required for transphosphorylation in the model, the 50% mutant-wild type and 25% wild type-wild type complexes remain FGF-dependent. The constitutively active fraction of FGFR and thus severity of the phenotype will depend on the degree that an individual mutation can subvert the forces which prevent a sustained enzyme-substrate relationship between intracellular kinases within the mutant-mutant complexes. In the case of non-cysteine mutations, the impact will depend on the extent that the mutation disrupts the relationship between loops II and III that prevents sustained interaction of FGFR contact domains, particularly Glu 160 -Lys 176 . The impact of cysteine substitutions will depend on location of the cysteine and efficiency of disulfide bond formation between mutant FGFR. This is consistent with the differences in severity of phenotype caused by cysteine substitutions at residues Arg 161 and Ser 162 within the Glu 160 -Lys 176 contact domain relative to the mutation of Ser 162 to tryptophan and adjacent Pro 163 to arginine. A large percentage of the mutant-mutant complexes may be locked in an enzyme-substrate relationship through stable disulfide bonds that form efficiently during the normally transient contact between Glu 160 -Lys 176 domains. This is in contrast to cysteine residues in non-contact domains or indirect structural changes caused by the Ser 162 to tryptophan and Pro 163 to arginine in the Glu 160 -Lys 176 contact domain. The fact that thanatophoric dysplasia also results from cysteine substitutions at the NH 2 terminus of the transmembrane domain of FGFR3 also suggests contact between FGFR at that domain. If the domain is a contact point in FGFR dimers, its intrinsic and independent FIG. 10. Requirement of the intact Glu 160 -Lys 176 domain for mitogenic activity of FGFR. A, mitogenic response of cells expressing different FGFR1␤1 constructs to FGF-1. BaF3 cells were transfected with wild type FGFR1␤1 and mutants FGFR1␤1⌬III and FGFR1␤1S2, selected and analyzed as described under "Experimental Procedures." Assays contained 2 ϫ 10 4 cells/well displaying about 10000 FGF-1 sites/cell. A representative experiment of three is shown. Data points are mean of duplicates with the deviations indicated. Solid square, FGFR1␤1; diamond, FGFR1␤1S2; triangle, FGFR1␤1⌬III; open square, untransfected BaF3 cells. Inset, binding and covalent crosslinking of 125 I-FGF-1 to transfected BaF3 cells. Each analysis is from 2 ϫ 10 7 cells. B, specific FGF-1 binding to FGFR1␤1 constructs. A representative experiment of three is indicated. Separate saturation binding analysis suggested that the lower binding of the FGFR1␤1⌬II reflects affinity rather than number of binding sites. C, mitogenic response of independent clones of transfected BaF3 cells. Assays contained 2 ϫ 10 5 cells per well displaying about 20000 sites/cell and 10 ng/ml FGF-1 where indicated. Data are the mean of duplicates with the indicated deviation. affinity for self-association is too low for detection by the methods used in this report. In contrast to the cysteine mutations in the two subdomains in the ectodomain above, an activating K650E mutation in the repressor loop of the kinase domain of FGFR3 results in a still more profound level of constitutive FGFR activity and associated phenotype (31). If the sole role of trans-phosphorylation is to derepress the active site of the FGFR kinase, then this model predicts that in contrast to the ectodomain mutations which are capped at 25%, a larger percentage (potentially the 25% mutant-mutant plus the 50% mutant-wild type) of FGFR complexes could be FGF-independent. Last, this model explains the tissue specificity of diverse heterozygous FGFR mutations and lack of effect on other tissues where the same FGFR isoform is believed to play a regulatory role. For example, the FGFR2 gene exhibiting mutations in the Glu 160 -Lys 176 domain associated with Apert syndrome is widely expressed in the epithelium of adult parenchymal organs and plays an important role in mediation of the communication between stromal and epithelial compartments (16,32,33). Mutations in the FGFR3 gene are associated with multiple craniofacial and skeletal syndromes including achondroplastic dwarfism. Yet there appears to be little or no effect in the adult central nervous system (33) as well as other adult tissues and cells 2 where FGFR3 is expressed and thought to play a role. FGFR functions in these tissues likely require a signal intensity that exceeds the level of signal present as a result of a particular gain-of-function heterozygous mutation. Therefore, FGFR functions remain FGF-dependent in these tissues.