The cysteine residue responsible for the release of fibroblast growth factor-1 residues in a domain independent of the domain for phosphatidylserine binding.

Fibroblast growth factor (FGF)-1 lacks a classical signal sequence to direct its secretion yet utilizes high affinity cell surface receptors to signal its heparin-dependent angiogenic and neurotrophic activities. We have previously reported that FGF-1 is released in response to temperature stress as a latent homodimer through a pathway that is potentiated by the Golgi inhibitor, brefeldin A (Jackson, A., Tarantini, F., Gamble, S., Friedman, S., and Maciag, T. (1995) J. Biol. Chem. 270, 33-36). In an attempt to further characterize this unconventional secretion mechanism, we sought to define the Cys residue(s) critical for FGF-1 dimer formation and release and to determine whether FGF-1 can associate with known phospholipid components of organelle or plasma membranes, which may be disturbed by brefeldin A. Utilizing FGF-1 Cys mutants, we were able to demonstrate that residue Cys30 is critical for FGF-1 release in response to heat shock. In addition, using solid phase phospholipid binding assays we demonstrate that FGF-1 is able to specifically associate with phosphatidylserine (PS). Heparin inhibits the association between FGF-1 and PS, and synthetic peptide competition assays suggest that the PS-binding domain of FGF-1 lies between residues 114 and 137. These observations indicate that FGF-1 may be able to associate with the PS component of organelle and/or plasma membranes and that the domains responsible for FGF-1 homodimer formation and PS binding are structurally distinct.

The fibroblast growth factor (FGF) 1 gene family is currently comprised of nine members (1,2); the two prototypes, FGF-1 (acidic) and FGF-2 (basic), lack a classical signal sequence to direct their secretion through the conventional endoplasmic reticulum (ER)-Golgi apparatus. However, their biological activities are mediated by their interaction with high affinity tyrosine kinase receptors on the cell surface (1,2). Because of the important role played by the FGF prototypes in angiogenic and neurotrophic processes, both during development and in the adult (1,2), it is important to understand the mechanism responsible for their secretion.
We have previously reported that FGF-1 is released from FGF-1-transfected NIH 3T3 cells after heat shock (3). The secreted form of the protein is a biologically inactive homodimer that can be reduced in the extracellular environment to obtain a functional FGF-1 monomer (4). Cysteine residues are necessary for FGF-1 release since a FGF-1 Cys-free mutant is not secreted in response to heat shock (4). Moreover, the FGF-1 release pathway is independent of the ER-Golgi-mediated pathway since pretreatment of cells with brefeldin A, a drug known to block membrane transport from the ER to the cis-medial Golgi (5) prior to heat shock, augmented rather than decreased the level of FGF-1 secreted in response to temperature stress (4). Furthermore, methylamine and verapamil, drugs known to inhibit the exocytotic (6) and multidrug resistance (7) pathways, respectively, also failed to interfere with the release of FGF-1 (4).
In an effort to gain further insight into this unconventional mechanism of secretion, we attempted to determine which cysteine residue(s) is essential for the dimerization of FGF-1 prior to its release. Because intracellular FGF-1 exists primarily as a cytosol-associated protein (8) and a partially unfolded form of FGF-1 is able to interact with acidic phospholipid-containing liposomes (9), we analyzed the ability of FGF-1 to bind acidic and neutral phospholipids. We report that FGF-1 is able to associate specifically with phosphatidylserine and the phospholipid-binding domain is located between FGF-1 residues 114 and 137. We also report that residue Cys 30 is critical for the release of FGF-1 in response to heat shock. These data suggest that the domains responsible for FGF-1 homodimer formation and phospholipid binding are structurally independent. This property may therefore enable FGF-1 to dimerize and utilize the carboxyl-terminal region to interact with membrane/organelle-associated phosphatidylserine and/or other phospholipid-binding proteins during the secretion pathway.

EXPERIMENTAL PROCEDURES
Materials and Reagents-Human recombinant FGF-1 was purified as described previously (10). FGF-1 was radiolabeled with Na 125 I (Amersham Corp.) using the Bolton-Hunter reaction in which 4 g of protein was incubated with 1 g of chloramine T and 0.7 mCi of Na 125 I for 30 s, and the reaction was stopped with 10 g of metabisulfite. The protein was processed over a 0.4-ml heparin-Sepharose column and eluted with Tris-buffered saline (TBS; 24 mM Tris-HCl, pH 7.4, containing 136 mM NaCl and 2 mM KCl) containing 1.5 M NaCl. The specific activity was approximately 250 -300 mCi/mM. Human recombinant (3-[ 125 I]iodotyrosyl) epidermal growth factor (EGF) was purchased from Amersham Corp. L-␣-Phosphatidyl-L-serine (PS), L-␣-phosphatidylcholine (PC), L-␣-phosphatidylinositol (PI), L-␣-phosphatidylethanolamine (PE), and L-␣-phosphatidyl-DL-glycerol (PG) were purchased from Sigma. Synthetic peptides corresponding to FGF-1 residues 114 -137 (KKHAEKNWFVGLKKNGSCKRGPRT) and 15-29 (FNLPPGNYKKP-KLLY) were synthesized using an Applied Biosystems (Foster City, CA) model 431A peptide synthesizer and purified by reversed-phase high pressure liquid chromatography. Amino acid analysis was conducted using a Waters Associates Pico-Tag system and amino acid sequencing was done using an Applied Biosystems model 477A protein sequencer.
Transfection, Heat Shock, and Immunoblot Analysis-NIH 3T3 cells were maintained in DMEM, 10% (v/v) fetal bovine serum (FBS). When the cells were 60 -70% confluent, fresh medium was added, and the transfection was conducted using the calcium phosphate precipitation method with either 5 g of Qiagen-purified FGF-1 Cys 30 , FGF-1 Cys 97 , or FGF-1 Cys 131 DNA as described (4). The medium was changed with fresh DMEM containing 10% (v/v) FBS after 24 h, and 48 h after transfection the selection was started with DMEM, 10% (v/v) FBS containing 800 g/ml geneticin. The medium containing the antibiotic was changed every 2nd day, and geneticin-resistant colonies were isolated and grown separately in DMEM containing 10% (v/v) FBS and 400 g/ml geneticin. The individual NIH 3T3 cell transfectants were grown to confluence and subjected to heat shock (42°C, 2 h) as described previously (4). Conditioned media were collected, and latent FGF-1 was activated with 0.1% (w/v) dithiothreitol and fractionated by heparin-Sepharose chromatography, and the 1.5 M NaCl elution fractions were resolved by 12.5% (w/v) SDS-polyacrylamide gel electrophoresis and subjected to FGF-1 immunoblot analysis as described (4).
Solid Phase Phospholipid Binding Assay-Phospholipid binding assays were performed using polystyrene flat bottom wells (Dynatech Labs) as described (12). Briefly, PS, PC, PI, PE, and PG were dissolved in ice-cold methanol to the desired concentrations (ranging from 1 to 10 g/ml) and sonicated twice at 4°C for 5-10 s, and 100 l of the individual phospholipid was used to coat each well. Wells coated with methanol alone were used as a control, and the wells were kept at 25°C for 18 h to evaporate the methanol. Nonspecific binding was minimized by incubation of the phospholipid-coated wells with TBS, containing 0.5% (w/v) gelatin, for 1 h at 37°C. The wells were washed three times with TBS containing 0.05% (v/v) Tween 20, and the binding assay was performed in TBS. For the heparin competition experiments, TBS containing 4 units/ml heparin (The Upjohn Co.), pH 7.4, was used as binding buffer. The labeled protein, 125 I-FGF-1 or 125 I-EGF, was added to the binding buffer, and the wells were incubated for 1 h at 37°C, aspirated, and washed three times with TBS containing 0.05% (v/v) Tween 20. The amount of radiolabeled protein bound to the phospholipid was quantitated by an automatic ␥ counter. When the phospholipid binding assay was performed under conditions of increased ionic strength, 25-150 mM NaCl was added to the binding buffer prior to the addition of 125 I-FGF-1. Similarly, the synthetic peptides were also added to the well prior to the addition of 125 I-FGF-1 during the synthetic peptide competition experiments.

RESULTS AND DISCUSSION
We have previously reported that FGF-1 is released from FGF-1-transfected NIH 3T3 cells in response to heat shock as a latent homodimer, and we were able to utilize a FGF-1 Cys-free mutant to demonstrate the functional importance of Cys residues (4). Because two of the three Cys residues (Cys 30 and Cys 97 ) present in the human FGF-1 sequence are conserved among all known species of FGF-1 (1), we sought to establish whether a critical Cys residue exists for FGF-1 homodimer formation during heat shock-induced FGF-1 secretion. Using the recombinant circle polymerase chain reaction, we created three FGF-1 mutants in which two of three Cys residues were converted to Ser (Fig. 1A). Stable NIH 3T3 transfectants for each of these mutants were obtained, and their expression levels were examined by immunoblot analysis. As shown in Fig. 1B, the level of FGF-1 mutant expression was comparable with the expression of FGF-1 in the wild-type NIH 3T3 transfectants. These transfected cells were individually examined for their ability to release FGF-1 in response to heat shock. As shown in Fig. 1B, extracellular FGF-1 was observed in the conditioned medium from FGF-1 wild-type and FGF-1 Cys 30 NIH 3T3 transfectants but not in the media conditioned by heat shock from either the FGF-1 Cys-free, FGF-1 Cys 97 , or FGF-1 Cys 131 transfectants. These data suggest that Cys 30 , positioned near the NH 2 -terminal nuclear translocation signal in FGF-1 (8), is critical for the entry of the protein into the heat shock-induced FGF-1 secretion pathway.
Because brefeldin A potentiates rather than inhibits the release of FGF-1 in response to heat shock (4), we considered the possibility that FGF-1 may be associated with Golgi-derived membranes. Phospholipids are known components of plasma and organelle membranes; thus, several phospholipids

FGF-1 Binds Phosphatidylserine 29040
were analyzed for their ability to interact with FGF-1. The acidic phospholipids, PS, PI, and PG, and the neutral phospholipids, PC and PE, were evaluated in solid phase phospholipid binding assays to assess the ability of the protein to bind phospholipid-coated polystyrene wells. In these assays, 125 I-FGF-1 was able to associate with PS but not with either PI, PG, PC, or PE ( Fig. 2A). In contrast, 125 I-EGF, also an acidic protein of similar size (13), was unable to interact with any of the phospholipids examined (data not shown). The association between 125 I-FGF-1 and PS was saturable and dependent on the concentration of PS with half-maximal binding occurring at approximately 4 g/ml (Fig. 2B).
An analysis of the FGF-1 protein sequence for the presence of a consensus sequence for phospholipid binding (14) revealed a putative structural motif in the carboxyl-terminal region of FGF-1 between residues 112 and 132. Because of the presence of several basic amino acids in this sequence (1), we anticipated that the interaction between PS and FGF-1 may involve ionic bonds. Indeed, the interaction between 125 I-FGF-1 and PS was reduced by increasing the ionic strength of the binding buffer (Fig. 3A). We also sought to determine whether this domain was indeed responsible for PS binding. Using a synthetic peptide containing residues 114 -137 from the FGF-1 sequence, it was possible to compete for the PS binding of 125 I-FGF-1 in the solid phase binding assay (Fig. 3B). In contrast, a second peptide corresponding to FGF-1 residues 15-29, which includes the basic residues involved in nuclear translocation (8), was not able to compete with 125 I-FGF-1 for PS binding (Fig. 3B). Interestingly, the crystallographic structure of FGF-1 2 demonstrates that Lys 127 and Lys 132 of the putative phosphatidylserine binding domain are positioned around a sulfate molecule from the freezing medium (15). Since this sulfate may be interchangeable with phosphate, it is possible that Lys 127 and Lys 132 may be responsible for the binding of FGF-1 to the phosphate present in phosphatidylserine. Moreover, because the heparin-binding domain has been localized to a domain within the carboxyl-terminal region of FGF-1 (16), we asked whether heparin could compete with PS for 125 I-FGF-1 binding. Indeed, heparin was able to inhibit the interaction between 125 I-FGF-1 and PS (Fig. 3C). Several proteins have been described that have Ca 2ϩ -dependent and -independent phospholipid-binding domains, and those that bind specifically to acidic phospholipids include members of the annexin family (17). Annexins have been implicated in membrane trafficking, signal transduction, cell-cell and cell-matrix interactions, inhibition of coagulation, and other activities, including Ca 2ϩ -regulated exocytosis and membrane fusion (17). Neuromodulin and neuroregulin (18,19) are among the proteins sharing the putative phospholipid-binding consensus sequence with FGF-1 and are brain-specific protein kinase C substrates. Interestingly, several functions have been proposed for these proteins, including an involvement in neurotransmitter release (19). Synaptotagmin, a component of exocytotic vesicles whose functional domains as a trafficking macromolecule are cytosol-oriented, has been described as a Ca 2ϩdependent PS-binding protein (20) and is also involved in the recycling of clathrin-coated endocytotic vesicles (21). While the ability of FGF-1 to interact with PS appears to be Ca 2ϩ -independent and FGF-1 does not contain a Ca 2ϩ -binding structural motif, it is likely that FGF-1 is not able to bind Ca 2ϩ . However, FGF-1 is well described as a Cu 2ϩ -binding protein (10,22). It is FIG. 2. Solid phase 125 I-FGF-1 phospholipid binding assay. A, phosphatidylserine (PS), phosphatidylcholine (PC), phosphatidylethanolamine (PEA), phosphatidylinositol (PI), and phosphatidylglycerol (PG) diluted in methanol at a final concentration of 8 g/ml were used to coat the wells. Wells coated with methanol alone served as a negative control. Nonspecific binding was blocked with 0.5% (w/v) gelatin, and the wells were washed with 0.05% (v/v) Tween 20. The binding of 125 I-FGF-1 was performed as described under "Experimental Procedures," and the data are reported as counts/min bound per well. C, control. B, PS diluted in methanol at a final concentration of 1, 3, 6, and 9 g/ml was used to coat the wells. Wells coated with methanol alone served as a negative control. Nonspecific binding was blocked with 0.5% (w/v) gelatin, and the wells were washed with 0.05% (v/v) Tween 20. The binding of 125 I-FGF-1 was performed as described under "Experimental Procedures," and the data are reported as counts/min bound per well.

FGF-1 Binds Phosphatidylserine 29041
also interesting that Cu 2ϩ is able to oxidize the formation of FGF-1 homodimer (10), and since the FGF-1 homodimer is released in response to heat shock, Cu 2ϩ may be involved in the regulation of the FGF-1 secretion pathway. This may be particularly appropriate for the FGF-1 release mechanism since the domain responsible for FGF-1 homodimer formation resides near the NH 2 terminus while the domain involved in PS binding resides near the COOH terminus of FGF-1.
Recently, Mach and Middaugh (9) reported that a thermally modified state of the FGF-1 protein is able to interact with negatively charged lipid vesicles at a neutral pH. In this system, a change in protein structure from its native, globular state to a partially unfolded "molten globular" state is required to obtain the binding of FGF-1 to unilamellar vesicles (9). Interestingly, the temperature range that is required to obtain the unfolded state (9) is consistent with the temperature that is able to stimulate the release of FGF-1 (3), and our data are also consistent with the observation that FGF-1 is able to bind to acidic phospholipids (9), with a specific affinity for PS. Further, the crystal structure for FGF-1 also predicts that residue Cys 30 may not be accessible to solvent and may also require a significant structural rearrangement for it to participate in intermolecular disulfide bond formation. Indeed, our data argue that heat shock may enable FGF-1 to unfold and thus not only provide access of residue Cys 30 for intermolecular disulfide bond formation but may enable FGF-1 to disrupt membrane structural integrity attaining characteristics of "molten globules" as described by Bychkova et al. (23) and van der Goot et al. (24) for protein-membrane insertion and/or translocation. Since "molten globule" character has been assigned to FGF-1 as an acidic phospholipid-dependent state (9,25), it is likely that the "molten globule" character of FGF-1 may be specific for PS.
It is intriguing that the PS-binding domain in FGF-1 resides near the carboxyl-terminal region of the protein. It is possible that under conditions of temperature stress, the FGF-1 secretion pathway is activated and thermally modified FGF-1 undergoes dimerization through the Cys residue at position 30. Under these conditions, FGF-1 may still be able to maintain its ability to interact with the PS component of membranes through the carboxyl-terminal region. The interaction of FGF-1 with membranes may stabilize the partially unfolded state of the protein exposing domains that may be involved in the association of FGF-1 with other phospholipid-related molecules and/or other phospholipid-binding proteins. Alternatively, the interaction of the FGF-1 homodimer with PS may be dependent upon a structural rearrangement, which may provide solvent accessibility to PS. Due to the prevalent representation of PS in the inner face of plasma membranes it is likely that this cytosolic surface represents the target of FGF-1-PS binding activity. However, the significance of this protein-phospholipid interaction in relation to the transport of FGF-1 through the plasma membrane during temperature-induced secretion remains to be determined.