INTRODUCTION

Fibroblast growth factors (FGFs)¹ are structurally related proteins encoded by at least 10 distinct genes in mammals (Refs. 1 and 2; reviewed in Refs. 3 and 4). Members of the FGF family demonstrate a variety of properties, depending on cell type and context, such as induction of proliferation, differentiation, stimulation of motility, and enhanced survival (reviewed in Refs. 3 and 4). FGFs induce cell to cell signaling by interacting with high affinity cell surface receptor, although there is also a requirement for ligand presentation by a glucosaminoglycan-containing low affinity receptor (reviewed in Ref. 5). Four receptor genes have been identified that encode high affinity FGF binding receptors, but the number of receptor species is extended by alternative splicing, which changes the ligand binding specificity of some receptors (reviewed in Refs. 6 and 7). Binding of FGF to its receptors initiates signal transduction, ultimately leading to a change in gene expression and cell function. However, there is evidence that FGF1, FGF2, and FGF3 can directly translocate to the nucleus, thereby providing the potential for an alternative intracrine signaling pathway (8-16).

FGF3 (int2) was first identified as a proto-oncogene in virally induced mouse mammary tumors (reviewed in Refs. 17 and 18). Its role in tumorigenesis was strengthened by transgenic mouse studies, which showed that inappropriate ectopic expression in the mammary gland resulted in a partial recapitulation of the disease (19-21). In situ hybridization revealed that normal expression of FGF3 is primarily restricted to prenatal mouse development, where RNA was found at a variety of sites at specific times (22, 23). An unusual feature of FGF3 and FGF2 biosynthesis is the use of alternative CUG initiation codons to generate isoforms extended at the amino terminus that localize to the cell nucleus (24-26). A single CUG start codon is the major start site for FGF3 translation, and the resulting protein is directed in similar proportions to the nucleus and secretory pathway (27). The choice between nuclear import and endoplasmic reticulum entry appears to result from finely balanced opposing signals near the amino terminus. Next to the signal peptide for entry into the endoplasmic reticulum is a classical bipartite nuclear localization signal, but the strength of these signals is modulated by an amino-terminal domain upstream of the signal peptide and the relative position of the NLS to the signal peptide (24). In addition to the bipartite NLS, a carboxyl-terminal sequence has been identified as a motif essential for nucleolar accumulation (28). When expressed in a mammary epithelial cell line, a signal peptide deletion mutant of FGF3 that locates exclusively to the nucleus resulted in inhibition of cell proliferation (28). This finding suggested that FGF3 may have opposing effects, depending on whether it is a secreted paracrine effector or intracrine signaling molecule.

As these findings could have significant implications for the role of FGF3 in mouse development, we have further characterized the signals that target FGF3 to dual subcellular sites and investigated the cooperation between different motifs within the protein. Specifically we demonstrate that in the presence of the competing signal peptide for secretion, a second NLS is necessary to cooperate with the amino-terminal NLS for nuclear uptake. The import of FGF3 into the nucleus is not only signal-mediated but is also shown to be energy-dependent and mediated by karyopherin , the NLS binding subunit of a heterodimeric cytoplasmic receptor for translocating proteins through the nuclear pore. Retention in the nucleus is facilitated by a nucleolar binding motif that is conserved between mouse and human FGF3 despite considerable surrounding differences in primary structure.

MATERIALS AND METHODS

COS-1 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal calf serum and passaged once a week at a ratio of 1 to 10. For transient DNA transfections, 20 µg of purified plasmid DNA was introduced into 5 × 10⁵ COS-1 cells by electroporation (450 V/250 micorfarads) using a Bio-Rad Gene-Pulser. Between 48 and 72 h after transfection, the cells were harvested for immunoblot analysis or processed for immunofluorescence.

The construct pKC4.16, which expresses a mutant mouse FGF3 lacking the signal peptide, has been previously described (28). The plasmid pKC4.28 was created by replacing the 3 sequences of pKC4.16 with those of pKC4.22 (28), which has a deletion of codons 238-247. To construct pKC4.29, the 5 sequences of pKC4.16 were joined to the 3 sequences of a human FGF-3 cDNA via a conserved BamHI site. For pGem4.12, pGem4.12_NUC, and pGem4.12_1,2, the inserts from the corresponding pKC plasmids (28) were transferred to the vector pGem4Z. The cDNAs were oriented for transcription with the T7 RNA polymerase. The -galactosidase-FGF3 fusion proteins were based on the expression plasmids pGAL1, pGAL1.1, pGAL1.2, and pGAL1.4, which have been previously described (28). Using polymerase chain reaction, the carboxyl-terminal 88 codons of FGF3 were amplified using a 5 primer that introduces an EcoRI site. The resulting polymerase chain reaction fragment was fused to the single EcoRI site of LacZ in pGAL1, pGAL1.2, and pGAL1.4 to obtain the plasmids pGAL1.5, pGAL1.7, and pGAL1.8, respectively. The -galactosidase-FGF3 chimeric clone pGAL1.6 was created by replacing a PstI-EcoRI fragment of pGAL1.5 with the corresponding region of pKC4.22. Pyruvate kinase (PK) fusion protein constructs were based on the expression XSRL40-PK (Refs. 29; kindly provided by W. Richardson, University College London). To construct PK-1, a pair of complementary oligonucleotides encoding amino acids 50-77 of FGF3 encompassing the bipartite NLS were used to replace the 8-base pair XhoI-EcoRI fragment of XSRL40-PK. A similar strategy was used to introduce the sequences encoding amino acids 187-205 into the single BstEII site of XSRL40-PK to create PK-2. PK-3 was created by replacing an appropriate restriction enzyme fragment of PK-1 by the corresponding fragment of PK-2.

COS-1 cells grown on glass coverslips were transfected with the appropriate plasmids, and 48 h later the cells were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) for 20 min. The cells were then permeabilized with 0.2% Triton X-100 in PBS for 4 min and treated with 3% bovine serum albumin in PBS to block nonspecific binding of the antibodies. The coverslips were exposed to primary antibodies, and fluorescently tagged secondary antibodies were diluted in 3% bovine serum albumin, PBS. After washing in PBS, the stained cells were mounted in 90% glycerol containing p-phenylenediamine and viewed with a 100 × oil immersion lens on a Zeiss microscope equipped with the appropriate barrier filters for Texas red optics. Mouse and human FGF3 carboxyl-terminal antipeptide rabbit polyclonal serum were diluted 1 in 200 in PBS. -Galactosidase was detected using a monoclonal antibody kindly supplied by H. Durban, Imperial Cancer Research Fund, and the pyruvate kinase polyclonal rabbit antiserum was diluted 1 in 200 (generously provided by W. Richardson, University College London).

The procedures used for preparing cell lysates have been described in detail elsewhere (27). Samples from equivalent numbers of cells were fractionated by SDS-PAGE in 12.5 or 15% gels, transferred to nitrocellulose membranes (Schleicher & Schuell), then probed with rabbit polyclonal antibody to the carboxyl terminus of FGF3. The immunoreactive proteins were detected with ¹²⁵I-labeled protein A (26) or by enhanced chemoluminescence (ECL) using horseradish peroxidase-coupled anti-rabbit immunoglobulin antibodies as described by the manufacturer (Amersham International). Mouse Fgf-3 and NPI-1 cDNAs in pGem4Z and pGem7Z, respectively, were used in a coupled in vitro translation system (TNT; Promega) to generate products for use in binding assays as described in the text (30).

A karyopherin  cDNA encoding NPI-1 and additional six carboxyl-terminal histidine residues were subcloned into a pGem7Z vector. 1 µg of cDNA was added to an in vitro transcription and translation system and used as described by the manufacturers. FGF3 cDNAs as described in the text were also in vitro translated, and 5 µl of karyopherin  and 5 µl of FGF3 in vitro translates were incubated with 20 µl of Co²⁺-nitrilotriacetic acid-agarose beads (CLONTECH) for 1 h at 4 °C in binding buffer (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM MgCl₂) at a final volume of 100 µl. After three washes with binding buffer, the bound proteins were eluted with binding buffer containing 100 mM EDTA, and the eluates were analyzed by immunoblotting using a polyclonal antibody against FGF3. The immune complexes were detected by the enhanced chemoluminescence technique as described under "Experimental Procedures."

COS-1 cells were transfected with the appropriate plasmids by electroporation, and 40 h after transfection, the medium was replaced with glucose minus Dulbecco's modified Eagle's medium supplemented with 6 mM deoxyglucose (Sigma), 10% dialyzed fetal calf serum, and 50 µM oligomycin B (Sigma). To block de novo protein synthesis, cycloheximide (CHX) was added in a concentration of 50 µg/ml. The reversibility of the energy depletion was tested by washing the cultures and incubating them for a further 3 h in Dulbecco's modified Eagle's medium, 10% fetal calf serum medium containing CHX.

RESULTS

In previous studies, we showed that the choice between secretion and nuclear uptake of FGF3 resulted from an interplay between an amino-terminal domain joined to a downstream signal peptide and an adjacent bipartite NLS (Fig. 1). The bipartite NLS was shown to confer nuclear localization to -galactosidase, although it was unable to do so in the presence of the signal peptide. Surprisingly, deletion of the signal peptide and bipartite NLS still resulted in the nuclear uptake of the truncated protein, suggesting that other sequences in FGF3 could also mediate nuclear import and/or retention. To determine whether the carboxyl-terminal sequences of FGF3 could confer nuclear import upon a heterologous cytoplasmic protein, we fused these 3 sequences, encoding 88 amino acids, to the bacterial LacZ gene (pGAL1.5) and examined the subcellular localization of the chimeric protein in COS-1 cells by indirect immunofluorescence (Fig. 2B). The parental plasmid (pGAL1) encoding the LacZ gene showed the expected cytoplasmic staining of -galactosidase (Fig. 2B). In contrast, the LacZ gene with the carboxyl terminus of FGF3 showed exclusive nuclear staining. As previous deletion analyses of FGF3 cDNA had implicated motif 5 in nucleolar retention, we determined whether it was necessary for nuclear uptake. Deletion of motif 5 (pGAL1.6 in Fig. 2A) resulted in expression of a fusion protein that retained its capacity for nuclear localization, although there was some weak cytoplasmic staining, indicating that nuclear accumulation was less efficient. As nuclear localization signals can act additively, we next asked whether the FGF3 carboxyl-terminal sequences could overcome the previously described negative effect of the signal peptide on nuclear localization directed by the amino-terminal bipartite NLS. A LacZ gene construct (pGAL1.7) was generated by adding the amino-terminal region of FGF3 at the amino terminus of pGAL1.5, and a similar construct but without the signal sequence was also made (pGAL1.8). The chimeric protein encoded by pGAL1.7 showed staining in the secretory pathway and nucleus similar to the dual localization of FGF3, whereas, as expected, the protein without the signal peptide was exclusively nuclear. Hence, the combined amino- and carboxyl-terminal domains of FGF3 containing both the NLSs were able to counter the suppressive effect of the signal peptide and direct nuclear uptake.

[View Larger Version of this Image (36K GIF file)]

Properties of FGF3 nuclear targeting signals appended to -galactosidase.

[View Larger Version of this Image (41K GIF file)]

The extended signal peptide and the adjacent bipartite NLS of FGF3 are weak targeting signals, since substitution with stronger signals changes the balance between secretion and nuclear uptake in a predictable manner (27). The weak signals are thought to be mechanistically important to allow competition between the intracellular trafficking pathways; but a weak NLS is a disadvantage for efficient nuclear uptake. This could be overcome by the presence of an additional weak NLS that acts concertedly to enhance nuclear uptake without affecting the balance between trafficking pathways. The additive effect of weak NLSs was demonstrated by Roberts et al. (29) using pyruvate kinase as a cytoplasmic protein in which to test NLSs in combination at different structural locations. As illustrated in Fig. 1, there is a cluster of basic amino acids that resembles a NLS in the carboxyl-terminal sequences (motif 4), which could function as a second nuclear uptake signal. To determine whether the two proposed NLS elements in FGF3 can act together to promote nuclear uptake, we used the same vector to combine these elements. The sequences encoding the bipartite NLS (motif 3) and motif 4 were inserted into the expression plasmid XR30PK separately or together, as illustrated in Fig. 3A. After transfection into COS-1 cells, the subcellular location of the encoded fusion proteins and control construct (PK) were determined by indirect immunofluorescence, using a polyclonal rabbit antiserum against chicken pyruvate kinase (Fig. 3B). Cells expressing pyruvate kinase containing a single NLS (PK-1 and PK-2) showed only cytoplasmic staining, indicating that neither element has sufficient strength to translocate chimeric pyruvate kinase into the nucleus. However, when the bipartite NLS was combined with motif 4 (PK-3), virtually all of the product was found in the nucleus (Fig. 3B). Interestingly, motif 5 could not replace motif 4 in this assay (data not shown). These findings support a model in which motif 4 acts concertedly with the amino-terminal bipartite NLS to increase the efficiency of nuclear uptake.

[View Larger Version of this Image (43K GIF file)]

Although some small proteins can enter the nucleus by passive diffusion through the nuclear pore, most nuclear proteins are actively imported by an energy-dependent process that requires recognition by a cytoplasmic receptor for the NLS. Hence, nuclear import can be abolished in cells by depletion of ATP pools with deoxyglucose and oligomycin (31-33). To determine the importance of the nucleolar retention signal for nuclear accumulation, we examined the distribution of nuclear FGF3 with (pKC4.16) and without (pKC4.28) motif 5. Forty hours after transfection of COS-1 cells, oligomycin was added for 6 h to deplete ATP levels in the absence or presence of CHX to inhibit de novo protein synthesis. Cells containing pKC4.16 in the presence of CHX showed pronounced nuclear and nucleolar staining with virtually no detectable cytoplasmic staining (Fig. 4B). However, in the presence of new protein synthesis (no CHX), there is additional speckled staining for FGF3 in the cytoplasm, suggesting the presence of protein that is not transported to the nucleus. In contrast, cells expressing pKC4.28 showed weak nucleoplasmic staining with no staining of the nucleoli but strong staining of the cytoplasm irrespective of the presence or absence of CHX (Fig. 4B). We interpret these experiments to show that deoxyglucose/oligomycin treatment depletes ATP pools and consequently blocks nuclear uptake of newly synthesized FGF3 (pKC4.16), but protein bound in the nucleoli is retained and does not shuttle back into the cytoplasm at a detectable rate. In the absence of nucleolar retention (pKC4.28), the protein normally shuttles between the nucleus and cytoplasm, consistent with reversible nucleocytoplasmic trafficking, but in the presence of oligomycin nuclear uptake is inhibited, resulting in accumulation of the FGF3 in the cytoplasm. In other experiments (Fig. 4B), similarly transfected COS-1 cells were treated with oligomycin for 3 h and then withdrawn for 3 h. During the whole 6-h period CHX was included to inhibit de novo protein synthesis. FGF3 expressed from pKC4.16 did not change its localization during the course of the experiment, whereas in contrast, after 3 h deoxyglucose/oligomycin treatment, p4.28-encoded products showed a peri-nuclear distribution, but after oligomycin withdrawal, there was re-entry of the mutant protein into the nucleus consistent with nucleo-cytoplasmic shuttling.

[View Larger Version of this Image (48K GIF file)]

Binding of FGF3 to karyopherin .

[View Larger Version of this Image (30K GIF file)]

To examine the stability of protein in the different compartments, COS-1 cells were transiently transfected with the plasmids pKC4.16 and pKC4.28, and 46 hours post-transfection, the cells were treated with cycloheximide to block de novo protein synthesis. Cell extracts were analyzed by immunoblotting at intervals from 0 to 72 h (Fig. 4C). The results show that products derived from pKC4.28 were significantly less stable than those from pKC4.16, suggesting that nucleolar sequestration extends the half-life of the bound protein.

Karyopherin  Interacts with FGF3

The import of nuclear proteins containing a classical NLS involves binding to a heterodimeric cytoplasmic receptor complex (karyopherin -karyopherin /importin -importin ) that translocates to the nuclear pore (reviewed in Ref. 34). A 60-kDa subunit of this complex, karyopherin  (also known as NPI-1) recognizes and binds proteins containing a NLS (30, 35). As it is possible that FGF3 might enter the nucleus by an alternative import pathway or by binding to a shuttle protein, we asked whether FGF3 could bind to the karyopherin/importin nuclear transport machinery. Using a column retention assay, we tested the capacity of full-length FGF3, a similar protein but with the bipartite NLS substituted for the nucleoplasmin NLS and a mutant deleted for the bipartite NLS, to bind karyopherin  attached to an affinity matrix. The protein was generated by in vitro translation of appropriate cDNAs. The FGF3 proteins were then tested for their ability to bind NPI-1 attached via a hexahistidine tag to a metal affinity column. After washing the columns, the bound proteins were eluted, separated by SDS-PAGE, and analyzed by immunoblotting using a FGF3 specific antibody. The results show that all forms of FGF3 were able to bind karyopherin , indicating the presence of a NLS. The affinity matrix without karyopherin  was unable to bind FGF3 proteins. FGF3 deleted for the bipartite NLS was also bound karyopherin , consistent with the presence of a second NLS.

The extreme amino- and carboxyl-terminal regions of human FGF3 show a remarkable lack of sequence conservation to the mouse homolog. Despite the lack of primary sequence conservation, the CUG initiation codon and the functional requirements of the amino terminus necessary for dual localization are present in human FGF3 (27). To determine whether the carboxyl-terminal domain of human FGF3 retains the ability for nucleolar retention, these sequences were exchanged for the equivalent mouse sequences just upstream of motif 4 to generate pKC4.29 (Fig. 6A). COS-1 cells were transiently transfected with the parental plasmid (pKC4.16) and pKC4.29, and the subcellular localization of FGF3 was monitored by immunofluorescence 40-h post-transfection. The results show that the same nuclear-nucleolar distribution was obtained with the chimeric FGF3 as with the mouse protein, indicating that the nucleolar retention function is conserved between these homologs (Fig. 6B).

[View Larger Version of this Image (38K GIF file)]

DISCUSSION

Previous analyses have identified a bipartite NLS located in the amino-terminal region of mouse FGF3, which is sufficient to confer a nuclear localization upon -galactosidase (27). However, the addition of amino-terminal sequences from FGF3 that encode an extended signal peptide were sufficient to prevented nuclear uptake, causing the protein to enter the secretory pathway and cytoplasm. This demonstrates that the competition between the signal peptide and bipartite NLS, which in native FGF3 results in the dual localization to the secretory pathway and nucleus, could not be directly conferred to -galactosidase. Subsequent analysis suggested a second NLS might reside in the carboxyl-terminal region of FGF3 (28), and two candidate sequences (motifs 4 and 5) are present in this region (Fig. 1). A functional NLS in the carboxyl terminus was confirmed by fusing it to -galactosidase and targeting the chimeric protein to the nucleus (Fig. 2). Deletion of motif 5 resulted in a fusion protein that retained the ability for nuclear uptake, although it was clearly less efficient. However, a contribution of motif 5 to nuclear accumulation can be explained by its nucleolar targeting properties, which lead to a retention of FGF3 in the nucleus (see below).

The inhibitory effect of the signal peptide on nuclear localization by the bipartite NLS fused to -galactosidase was overcome by adding the carboxyl-terminal domain of FGF3. Furthermore, the resulting chimera was then shown to recapitulate the dual localization of native FGF3 (Fig. 2). The additive effect of NLSs has been shown before using weak NLSs that were not able to effect nuclear uptake alone but could when present in multiple copies (29). The concerted action of the two FGF3 NLSs was confirmed by their ability to locate pyruvate kinase protein to the nucleus, a process that was not possible with the individual motifs (Fig. 3). It should be noted that in these experiments the nucleolar retention signal (motif 5) could not substitute for motif 4 despite its highly basic nature, underlining a functional difference between these elements. It is interesting that the weak NLSs were able to import -galactosidase in to the nucleus, although as shown in Fig. 2, sequences containing motif 4 in the absence of motif 5 were relatively inefficient at conferring nuclear uptake. The bone fide nature of the FGF3 NLSs is supported by their ability to bind karyopherin , which is the NLS recognition subunit of the cytoplasmic nuclear pore targeting receptor (reviewed in Ref. 34). Interestingly, karyopherin  interacts with mutant FGF3s lacking the bipartite motif, in agreement with the presence of a second independent NLS in the carboxyl terminus.

As signal-mediated nuclear import is energy-dependent, depletion of ATP levels in the cell by the addition of deoxyglucose and oligomycin will inhibit nuclear uptake (31-33). In the presence of cycloheximide (to also inhibit new protein synthesis), previously synthesized FGF3 is retained in the nucleus/nucleolus, indicating nucleoplasmic and nucleolar binding of the protein (Fig. 4). In the absence of cycloheximide, FGF3 protein accumulates in the cytoplasm, consistent with nuclear import of FGF3 being energy-dependent. In contrast, mutant FGF3 protein lacking the retention signal (motif 5) still enters the nucleus, implying that the retention signal is essential for nucleoli association and nuclear retention. Hence, nucleolar targeting of FGF3 is a two-step process: nuclear uptake and nuclear/nucleolar retention.

Motif 5 resembles the nucleolar targeting motifs described for the retroviral proteins Rex of HTLV and Tat and Rev of HIV (36-38). The accumulation of mouse FGF3 in the nucleolus depends on motif 5, but it is a context-dependent signal, since it does not confer this property to -galactosidase but can to another member of the FGF family (Fig. 2) (28). The carboxyl-terminal regions of all sequenced FGF3s are poorly conserved downstream of motif 4, and a comparison between mouse and human sequences demonstrates this point (Fig. 6). Nevertheless, a motif 5 equivalent is present in the human sequence and seems to function as a nucleolar retention signal when substituted for the mouse carboxyl-terminal region (Fig. 6). Although the nuclear/nucleolar localization of mammalian FGF3 is established, there is no evidence that chicken, Xenopus, or zebra fish homologs have CUG-initiated amino-terminal extensions, suggesting that the mechanism for dual localization of FGF3 appears to have evolved after divergence of mammals and birds (39-41).

Two other members of the FGF family have acquired nuclear isoforms, but they utilize different mechanisms for nuclear translocation. The nuclear translocation of FGF1 seems to require an exogenous pathway in which FGF receptor-1 may be involved (42, 43), although it can enter the nucleus from the cytoplasm (10, 16). Once in the nucleus, it may then be retained by binding to some nuclear structures, facilitating its accumulation. Mutation of a putative NLS in FGF1 alters its stability but does not abrogate its mitogenic activity, which is consistent with a retention rather than a classical NLS uptake mechanism (16, 44). Human FGF2 has three CUG-initiated protein isoforms that localize to the cell nucleus, although the mechanism for nuclear uptake again differs from that of FGF3 (11-14). In this case, the amino-terminal sequences appear to undergo methylation, which facilitates nuclear retention of FGF2 (45). One possible reason why FGF3 should have acquired an active nuclear uptake capacity rather than a retention mechanism is the difference in the second location sites; FGF1 and FGF2 are cytoplasmic proteins, whereas FGF3 is secreted. A function for these nuclear FGFs remains elusive, but circumstantial evidence suggests the translocation is functionally important. For example, FGF2 has been shown to move into the cell nucleus concomitantly with the induction of mesoderm in Xenopus laevis, whereas in the chick, FGF2 is expressed in epithelial cells of the developing kidney and translocates into the nucleus in podocytes of meso and metanephric glomeruli as the cells start to differentiate (46, 47). In cell culture, the nuclear forms of FGF2 have been shown to impair growth at low expression levels, whereas high expression levels of FGF2 may reduce serum dependence and cause morphological transformation (48-50).

Extensive mutation analyses have shown that FGF3 contains an array of protein motifs that act in concert to direct the major primary translation product into alternative intracellular trafficking pathways (see Fig. 1). The secreted product can presumably interact with cell surface transmembrane receptors in either an autocrine and/or paracrine context to transduce a signal (51, 52). Nuclear uptake that is mediated by interaction via two NLSs and the heteroduplex karyopherin receptor results in nuclear uptake and subsequent nucleolar association through a retention sequence. As an initial approach to elucidate the properties of nuclear FGF3, we demonstrated that expression of a mutant FGF3 with an exclusively nuclear location in mouse mammary cells resulted in an inhibition of cell growth (28). In cells transfected with cDNAs where the encoded FGF3 lacks the carboxyl-terminal retention signal, the growth rates are similar to the parental line (28).² Taken together, these findings strongly support the idea that FGF3 targeted directly to the nucleus may induce an autocrine signal that relies on efficient nuclear uptake and nucleolar retention.