Improperly folded green fluorescent protein is secreted via a non-classical pathway

Green fluorescent protein (GFP) provides bioluminescence in coelenterates such as the jellyfish Aequorea victoria. GFP folds into a barrel-shaped structure made up of beta-strands, with the chromophore located in the center (Tsien, 1998). Fully folded GFP does not require cofactors or substrate to be functional, and excitation of the chromophore at a specific wavelength results in the emission of fluorescent light (Chalfie,1995). Furthermore, the fluorescence is stable,species-independent and can be observed non-invasively in any cell expressing GFP (Kain et al., 1995). These convenient properties of GFP have been exploited as a tool to study various cellular processes (Cinelli et al.,2000).

Owing to its intrinsic fluorescence, GFP is commonly used as a molecular tag to study intracellular protein trafficking. GFP appears to be an inert and stable molecule localized in the cytosol. Furthermore, it is small enough (29 kDa) to be used as a passenger protein for fusion constructs. In these fusion proteins, it is assumed that GFP does not contain intrinsic targeting information. To date, various proteins have been tagged with GFP to study localization and sorting in compartments such as the mitochondria, nucleus,chloroplasts, endoplasmic reticulum (ER), Golgi apparatus and plasma membrane(Chatterjee and Stochaj, 1996;Choy et al., 1999;Lee et al., 2001;Niwa et al., 1999). GFP-tagged proteins can be observed and their movement tracked in intact cells simply by looking for fluorescence at the targeted location.

Protein secretion in mammalian cells generally occurs via the classical secretory pathway that traverses the ER and Golgi apparatus. Secreted proteins contain a signal sequence with all the necessary information required to target them for secretion. A protein that is not normally secreted can be targeted for secretion by attaching a signal sequence(Simon et al., 1987). The classical secretory pathway is completely inhibited by brefeldin A (BFA),which causes reversible resorption of the Golgi apparatus back into the ER(Doms et al., 1989).

Over the past decade, it has been shown that several proteins are secreted independently of the ER-Golgi pathway. For example, basic fibroblast growth factor (FGF), interleukin (IL)-1β, HIV-tat, galectin-3, and thioredoxin are secreted in a non-classical manner(Chang et al., 1997;Mehul and Hughes, 1997;Mignatti et al., 1992;Rubartelli et al., 1992). These proteins do not display any signal sequence or protein motif known to act as a signal for export (for a review, seeMuesch et al., 1990;Rubartelli and Sitia, 1997). Furthermore, their secretion is not inhibited by the addition of BFA. The pathways used for the export of these proteins are still poorly understood,and multiple pathways for non-classical protein secretion may exist in cells.

In our laboratory, we wished to use GFP as a passenger in a fusion construct to study non-classical secretion by Chinese hamster ovary (CHO)cells. As a control, we expressed GFP alone. To our surprise, most of the GFP expressed in CHO cells is secreted into the medium in a non-classical manner. Many but not all types of cells examined also secrete GFP. We characterized the kinetics, sensitivity to temperature and factors in serum and effects of known specific chemical inhibitors on the export of GFP from transfected CHO cells. In contrast to cytosolic GFP, secreted GFP does not fluoresce,indicating that it is not properly folded.

Trans [³⁵S]-label (>70% L-methionine; >1000 Ci/mmol) was purchased from ICN Biomedicals. Mouse monoclonal antibodies specific for GFP used for immunoprecipitation and western blotting were purchased from QBiogene. Monoclonal antibodies against the myc epitope were purified by protein G affinity chromatography from media of MYC 1-9E10.2 cells(Evan et al., 1984). Polyclonal antibodies against the FLAG epitope (DYKDDDDK) were purchased from ProSci. Protein A-agarose, Lipofectamine 2000, OptiMEM, fetal calf serum and Glutamax I were purchased from Gibco/Life Technologies. The pQBI25-fN3 plasmid, encoding modified GFP with a single emission (474 nm) and excitation peak (509 nm), is from QBiogene. BFA and cycloheximide was purchased from Sigma.

The GFP construct used in this study was created as follows. Using the pQBI25-fN3 plasmid as a template, the forward (encoding a BamHI site,translation initiation consensus sequence and initiation methionine:CGGGATCCGCCACCATGGCTAGCAAAGGAGAAGAACTCTTC) and reverse (encoding a stop codon and XbaI site: CGTCTAGATAGTCAATCGATGTTGTAGAG) primers were used to amplify the GFP-coding region with Platinum pfx DNA polymerase (Gibco/ Life Technologies). The PCR product was digested, isolated and subcloned into pcDNA3 at the BamHI and XbaI site of the multiple cloning site. A similar method was used to create the preprolactin-myc construct using the forward (CGGGATCCGCCACCATGGACAGCAAAGGTTCGTCGC) and reverse(ATAGTTTAGCGGCCGCGCAGTTGTTGTTGTAGATGATTCTGC) primers to amplify the bovine preprolactin gene from the plasmid pT7-Bprl. The PCR product was isolated and cloned between the BamHI and NotI sites of pcDNA3 with a pre-existing myc epitope coding sequence following the NotI site. The murine dihydrofolate reductase (DHFR)-FLAG mammalian expression construct was created as follows. Using pDS5/3 plasmid[(Rassow et al., 1989) a kind gift of Elzbieta Glaser] as a template, the DHFR gene was amplified using forward (CGGGATCCGCCACCATGGAATTCATGGTTCGACCATTG) and reverse(CCTCTAGATTACTTGTCGTCATCGTCCTTGTAGTCGTCTTTCTTCTCGTAGACTTCAAAC) primers. The PCR product was digested with BamHI and NotI and subcloned into pcDNA3 with a pre-existing FLAG-epitope-coding sequence following the NotI site. The same approach was used to create the Schistosoma japonicum glutathione S-transferase mammalian expression construct except that pGEX-3 plasmid (Pharmacia) was used as a template and the different forward (GGAATTCTATGTCCCCTATACTAGGTTATTGG) and reverse primers(CCTCTAGATCACGATAAATTCCGGGGATCCC) were used for amplification. All constructs were verified by DNA sequencing.

CHO, A375, COS, NIH 3T3, HEK-293, MCF-7 and HT-29 cells were maintained at 37°C/5% CO2 in culture medium (DMEM, 10% fetal bovine serum, 1%penicillin/streptomycin and 1% non-essential amino acids). One day prior to transfection, the cells were trypsinized and counted. One million cells were used to seed 60 mm culture dishes and left overnight to form >90% confluent monolayers. Confluent layers of cells were transiently transfected with Lipofectamine 2000 according to the manufacturer's instructions. Briefly, 3μg of plasmid DNA and 15 μl of Lipofectamine 2000 were used for each transfection. Each of the components were resuspended in 450 μl of OPTI-MEM, incubated for 5 minutes at room temperature, mixed and incubated for a further 23 minutes before the addition of 4 ml of OPTI-MEM. The DNA-Lipofectamine 2000 suspension was then added to the cells (prewashed twice in PBS to remove residual serum proteins) and incubated at 37°C overnight. The next day, transfected cells were washed twice in PBS to remove residual DNA-Lipofectamine 2000 complexes prior to metabolic labeling.

The cells were incubated in 1 ml starving medium (DMEM minus methionine and cysteine, 5% fetal bovine serum, 1% Glutamax I, 1% non-essential amino acids,1× penicillin and streptomycin) for 30 minutes at 37°C in 5%CO₂. Metabolic labeling was carried out by the addition of 100μCi of [³⁵S]-methionine and incubating at 37°C for 30 minutes. At the end of the labeling period, the cells were washed in 1 ml of chase medium (DMEM, 1× penicillin and streptomycin, 1% non-essential amino acids) and chased in 800 μl of chase medium for the indicated amount of time. Some cells were treated with 1 μg/ml BFA during the starvation,labeling and chase periods.

For each pulse-chase assay, the medium was collected and cells on the dish were lyzed by the addition of 1 ml of 1X Triton X-100 salt wash buffer (TXSWB;1% Triton X-100, 100 mM Tris-HCl pH 8, 100 mM NaCl, 10 mM EDTA) in the presence of 2 mM PMSF. Both the cell lysate and medium were centrifuged at 16,000 g for 15 minutes to remove cell debris, and the supernatant was transferred to a fresh 1.5 ml tube. At this point, cleared cell lysate and medium were used for spectrofluorometric measurements or LDH assays (see below). For immunoprecipitation, 200 μl of 5× TXSWB (5% Triton X-100,500 mM Tris-HCl pH 8, 500 mM NaCl, 50 mM EDTA) and 2 mM PMSF were added to the cleared medium. Either 1 μl of anti-GFP antibodies or 6 μl of anti-myc antibodies bound to protein G beads was added to each 1 ml of the cell lysate or medium. The samples were mixed by inversion and incubated at 4°C for 1 hour before addition of 10 μl of a suspension of protein G-agarose beads. The samples were rotated at 4°C overnight, washed twice in 1× TXSWB and twice in wash buffer (100 mM Tris-HCl pH8, 100 mM NaCl) and resuspended in 1× SDS-PAGE buffer with 500 mM DTT. The samples were incubated at 37°C for 30 minutes prior to boiling and loaded on a 15% polyacrylamide gel. At the completion of electrophoresis, the gels were destained for 30 minutes, soaked in 1 M sodium salicylate for 30 minutes and dried. The gels were exposed to x-ray film for viewing or a phosphor-imaging screen (Molecular Dynamics) for quantification.

Assays for lactate dehydrogenase (LDH) were carried out on 5 μl samples of cell lysate and medium after the chase period to assess cell lysis. The TOX7 LDH assay kit (Sigma) was used according to the manufacturer's instructions.

200 μl of cleared cell lysate and medium were aliquotted into 96-well plates with blank 1× TXSWB and fresh chase medium as the respective controls. The plates were scanned on a Cytofluorescent 2350™Fluorescence Measurement System (Millipore) with filter sets covering the GFP excitation and emission wavelengths (exciation: 485±20 nm; emission:530±25 nm).

We wished to use GFP as a passenger protein to assay for non-classical secretion. As a control, we expressed GFP to investigate its cellular location in CHO cells. [³⁵S]-methionine-labeled GFP was detected as an immunoreactive 29 kDa protein in cell lysate(Fig. 1, lane 5), which is absent in the lysate or medium of untransfected cells(Fig. 1A, lanes 1 and 2). To our surprise, however, we saw GFP in the medium after 6 hours of chase(Fig. 1A, lane 6). We investigated whether GFP was secreted via the classical secretory pathway. CHO cells were transiently transfected with a plasmid encoding GFP then pulse labeled and chased in the presence of 1 μg/ml BFA, an effective inhibitor of the classical secretory pathway. GFP was detected in similar amounts in both the cell lysate and medium in the absence or presence of BFA(Fig. 1A, lanes 7 and 8 versus lanes 5 and 6). To verify the effectiveness of BFA on the classical ER-Golgi pathway, CHO cells were transfected with bovine preprolactin. This protein was synthesized and secreted (Fig. 1, lanes 9 and 10), and its export was inhibited by 1 μg/ml of BFA (Fig. 1A, lanes 11 and 12). Thus, this concentration of BFA was sufficient to inhibit ER-Golgi dependent protein secretion. We confirmed that the GFP detected in the medium was secreted and not released by cell lysis by carrying out an assay for LDH. In each sample, less than 5% of total cellular LDH was detected after a 6 hour chase period (data not shown). To investigate whether other cytosolic proteins of a similar size are non-specifically exported after transient transfection,we transfected CHO cells with genes coding for murine dihydrofolate reductase(mDHFR) and Schistosoma japonicum glutathione S-transferase. However,overexpression of these proteins in CHO cells did not result in their secretion into the medium (Fig. 1B). Recent reports have shown that the expression of GFP in mammalian cells can be toxic (Liu et al.,1999). It is possible that GFP expression in CHO cells causes these cells to be more susceptible to lysis. Our assay for LDH does not discriminate between lysis in transfected or untransfected cells. Therefore,we co-transfected CHO cells with the plasmid constructs coding for GFP and mDHFR proteins. If GFP expression is toxic and results in cell lysis, then GFP and mDHFR proteins would be expected to be present in the medium in equivalent proportions after 6 hours of chase. When both proteins are co-expressed in CHO cells (Fig. 1C, lane 1),however, only GFP is detected in the medium after 6 hours of chase(Fig. 1C, lane 2). This result indicates that GFP is specifically secreted by CHO cells and is not present in the medium because of lysis of the transfected cells.

We also assessed the ability of other commonly used cell lines to secrete GFP. NIH 3T3 and HEK293 cells transiently transfected with a plasmid encoding GFP secrete the protein non-classically over 6 hours in the presence of BFA(Fig. 2, lanes 1 to 4). By contrast, COS cells secrete GFP poorly over 6 hours(Fig. 2, lane 5 vs 6). Certain cancer cells export thioredoxin via a non-classical pathway(Berggren et al., 1996). Therefore, we investigated whether thioredoxin-secreting cancer cells, such as A375, MCF-7 and HT-29, can also efficiently secrete GFP. We observed that GFP is generally secreted by these cancer cells in the presence of BFA(Fig. 2, lanes 7-10). One exception, however, are HT-29 cells, which do not secrete GFP(Fig. 2A, lanes 11 and 12) yet export thioredoxin well (Berggren et al.,1996). For each cell line tested, no more than 5% of total cellular LDH was detected in the medium, indicating that very little cell lysis had occurred. Overall, compared with CHO cells, less GFP is secreted from the cell lines studied in Fig. 2. Taken together, these results indicate that GFP can be secreted by some cells via a brefeldin A-resistant pathway that is particularly active in CHO cells.

We next investigated the kinetics of GFP secretion. CHO cells transfected with GFP were labeled with [³⁵S]-methionine and chased at 37°C in medium lacking serum for 2, 4, 6 and 8 hours. After separation by SDS-PAGE,the protein bands were quantified using a phosphorimager, and the relative secretion for each time point was calculated(Fig. 3). The secretion of GFP is a slow but steady process with up to 60% being secreted into the medium after 8 hours of chase. The increase in secretion is not caused by cell lysis since an assay for LDH showed that less than 5% of total cellular LDH was detected in the medium after the maximum chase time of 8 hours (data not shown). Since an adequate proportion of GFP is secreted into the medium after 6 hours of chase (Fig. 3, lanes 5 and 6), we chose to use this time point for subsequent assays of GFP secretion.

Facilitated protein secretion is generally a temperature-dependent process. For example, secretion of proteins via the classical ER-Golgi secretory pathway is affected by alterations in temperature(Saraste et al., 1986). On the other hand, passive diffusion through a pore is not affected by a change in temperature (Melchior and Gerace,1995). We investigated the temperature sensitivity of GFP secretion to gain insight into whether it is a facilitated or passive process. CHO cells transfected with GFP were starved, labeled at 37°C and then incubated in serum-free chase medium at 25°C, 37°C and 42°C for 6 hours. As expected, about 45% of GFP synthesized in the cells is secreted after 6 hours at 37°C (Fig. 4, lanes 3 and 4). Lowering the chase temperature down to 25°C drastically inhibits GFP secretion (Fig. 4, lanes 1 and 2). On the other hand, increasing the chase temperature to 42°C increased the secretion of GFP modestly by 10%(Fig. 4, lanes 5 and 6). However, assays of LDH indicated an increase of cell lysis at this temperature(15% of total cellular LDH detected in the medium), which can account entirely for the slight increase in GFP detected in the medium.

Some non-classical secretory pathways are sensitive to factors in serum. For example, the secretion of HIV-tat and IL-1β is inversely proportional to the amount of serum in the medium (Chang et al., 1997; Rubartelli et al., 1990). The secretion of thioredoxin is also reduced with increasing amounts of serum in the medium (M.T., S.H. and S.L.C.,unpublished). We investigated whether factors present in fetal bovine serum also affect the secretion of GFP. CHO cells transfected with the GFP plasmid were starved, labeled with [³⁵S]-methionine and chased in medium containing 0%, 5%, 10% and 20% fetal bovine serum(Fig. 5). Quantification using a phosphorimager revealed no significant differences in the proportion of secreted GFP despite increasing concentrations of serum.

The results above suggest that GFP is secreted by a non-classical secretory pathway in CHO cells. To investigate whether the pathway for GFP secretion might be the same as that used by other non-classically secreted proteins, we tested the effect of various chemical inhibitors on the secretion of GFP(Table 1)(Hughes, 1999;Mignatti et al., 1992;Rubartelli et al., 1990). Although some of these compounds alter the non-classical secretion of other proteins, none appeared to have a significant effect on the secretion of GFP. For example, methylamine, which inhibits the non-classical secretion of thioredoxin by blast cells (Rubartelli et al., 1992), has no effect on the secretion of GFP. Compounds such as monensin and A23187, which enhance the release of thioredoxin, IL-1βand galectin-3 (Mehul and Hughes,1997; Rubartelli and Sitia,1991), also have no effect on GFP secretion. These data suggest that GFP may use a different non-classical secretory pathway.

At least one non-classically secreted protein, galectin-3, is secreted via plasma membrane blebbing (Mehul and Hughes, 1997). We studied whether GFP is also secreted in externalized membrane vesicles. Such a mechanism could enable the post-translational export of fully folded cytosolic proteins from the cell. CHO cells transfected with the GFP plasmid were subjected to a 6 hour pulse-chase assay. At the end of the chase, the medium was clarified by low-speed centrifugation followed by high-speed centrifugation to pellet membrane vesicles in the medium (Mehul and Hughes, 1997). However, after immunoprecipitation, no GFP protein was detected in the pellet recovered after the high-speed centrifugation(Fig. 6, lane 4). Indeed, all of the GFP still remained in the supernatant fraction(Fig. 6, lane 3) and in similar amounts to that detected in the medium after low-speed centrifugation(Fig. 6, lane 2). This result suggests that GFP is not secreted into the medium via membrane blebbing.

GFP is a stable molecule that is resistant to spontaneous unfolding or degradation in the cell (Bokman and Ward,1981). By fluorescence microscopy, GFP-transfected CHO cells are brightly fluorescent (data not shown) indicating that some of the GFP is folded properly in these cells. However, a loosely folded or unfolded conformation is generally required for protein translocation through channels(Schatz and Dobberstein,1996). Since the structure of fluorescent GFP is a bulky beta barrel, it would seem likely that the protein must be at least partly unfolded for export through a channel. We investigated whether GFP is secreted as an unfolded molecule by assessing its fluorescence in the medium. First, we assessed the quantities of unlabeled GFP secreted into the media relative to the cell by carrying out a western blot analysis. Untransfected and transfected CHO cells were placed in serum-free medium for 6 hours. At the end of the incubation, the medium and cell lysate were harvested, aliquots were saved for spectrofluorometric assays and the rest of the samples were subjected to immunoprecipitation, separated by SDS-PAGE and immunoblotted with anti-GFP antibodies (Fig. 7A). As expected, no GFP was detected in untransfected CHO cells and medium(Fig. 7A, lanes 1 and 2). GFP was detected in comparable amounts in the cells and medium of transfected cells by immunoblotting (Fig. 7A, lanes 3 and 4). We then carried out the spectrofluorometric assay on aliquots of the same samples used for the western blot. Very little auto-fluorescence was detected in the cell lysate or medium of untransfected CHO cells (Fig. 7B, CHO). On the other hand, the cell lysate of GFP-transfected CHO cells showed a large increase in fluorescence in agreement with the fluorescence seen by microscopy(Fig. 7B, CHO+GFP). However, no change in fluorescence was seen in the medium despite significant secretion of GFP (Fig. 7A).

We ruled out two possible artifacts that might account for the lack of fluorescence in the media. First, to test whether the lack of GFP fluorescence was caused by quenching by components in the medium, we carried out a 1:1 dilution of cell lysate with fresh medium or phosphate-buffered saline (PBS). The relative fluorescence detected in the sample diluted with fresh medium was half that of the undiluted sample and similar in value to the fluorescence measured when PBS was used as a diluent, indicating that the medium does not quench GFP fluorescence (Fig. 7B, CHO+GFP lysate diluted). Second, we examined whether the lack of fluorescence in the medium results from GFP becoming non-fluorescent after export into the medium. CHO cells expressing GFP were lyzed in fresh medium containing 1% Triton X-100. After clarification by centrifugation, the medium containing released cytosolic GFP was incubated at 37°C for 0, 3 and 6 hours, and the fluorescence was measured at each time point. Fluorescence remained virtually constant over 6 hours of incubation in medium(Fig. 7C). Thus, the lack of fluorescence in the medium is not caused by inactivation or unfolding of functional GFP once it is exported. From these results, we conclude that secreted GFP is not folded properly and therefore not fluorescent.

The secretion of unfolded, non-fluorescent GFP could result from either of two possible models. In the first model, all cytosolic GFP is fluorescent, and to be exported it must first be unfolded prior to or during secretion. This implies that the reduction in intracellular GFP would be reflected by a similar reduction in fluorescence under conditions where no new GFP molecules are synthesized. In the second model, GFP is present in the cell in two different forms as either a properly folded, fluorescent molecule that is not secreted or as an unfolded protein that can be secreted. In this model, the fluorescence from cellular GFP would remain constant despite a decrease in the total cytosolic GFP.

We investigated these two models of GFP secretion. CHO cells transiently transfected with GFP were placed in medium supplemented with cycloheximide to a final concentration of 100 μM, which is sufficient to inhibit protein synthesis (data not shown). The cells were incubated at 37°C in this medium for 0 or 6 hours, and the media and cell lysates were harvested. Aliquots were saved for spectrofluorometric assays, and the rest of the samples were analyzed by western blotting(Fig. 8A). As expected,initially all of the GFP synthesized is present in the cell(Fig. 8A, lane 1) and not in the medium (Fig. 8A, lane 2). This data correlated with the spectrofluorometric result that showed that all of the fluorescence is associated in the cell lysate and none is detected in the medium (Fig. 8B, 0 hours chase, cell lysate versus medium). After 6 hours of chase, the immunoblot showed that approximately one third of the synthesized GFP is present in the medium (Fig. 8A, lanes 4), with a similar reduction of GFP in the cell lysate(Fig. 8A, lane 3). The relative amount of GFP secreted into the medium is similar to the secretion observed in the absence of cycloheximide (Fig. 7A), indicating that the protein synthesis inhibitor did not affect the non-classical export of GFP. Assays for LDH confirmed that the GFP dectected in the medium does not result from lysis caused by cycloheximide(data not shown). However, the fluorescence in the cell lysate after 6 hours is virtually identical to that at the outset(Fig. 8B, 6 hours, cell lysate)whereas no significant fluorescence is detected in the medium despite the significant secretion of GFP protein (Fig. 8B, 6 hours, medium). These data strongly support the second model. Thus, GFP exists in two different pools in transiently transfected CHO cells — a pool of properly folded, fluorescent molecules, and a pool of unfolded, non-fluorescent proteins — and only the second form can be secreted (Fig. 9).

We demonstrate that GFP is secreted by a variety of cells via a non-classical secretory pathway. Significant secretion was observed from transiently transfected CHO cells with over half of the labeled GFP exported to the medium. The secretion of GFP is sensitive to low temperature but not factors in the serum. Commonly used cell lines and several cancer cell lines also demonstrate the ability to secrete GFP albeit with less efficiency. GFP secreted from transiently transfected CHO cells does not fluoresce, indicating that the improperly folded form is secreted via this non-classical pathway.

It is surprising that GFP is secreted from transiently transfected cells. GFP is a cytosolic protein with no known targeting signal. The intrinsic targeting of GFP for non-classical export might not have been observed previously because the molecule is often targeted to a specific location in cells. To our knowledge, this is the first report of GFP secretion by mammalian cells. In yeast, GFP can be localized to the vacuole presumably through a cryptic targeting signal (Kunze et al., 1999). Previously assumed to be an inert molecule, GFP has other unanticipated effects on cells. For example, transgenic mice that express GFP in the heart develop cardiomyopathy(Huang et al., 2000). Furthermore, prolonged overexpression of GFP triggers apoptosis in cells(Liu et al., 1999). Many visual applications, such as fluorescence microscopy, may not be sensitive to minor toxic effects that go unnoticed(Schmitz and Bereiter-Hahn,2001). Taken together, the unanticipated effects and secretion of GFP raise doubts regarding the inertness of this molecule.

Our experiments indicate that the secretion of GFP is not simply due to cell lysis caused by either a cytotoxic effect of GFP or liposome-mediated transfection (Fig. 1C). GFP but not mDHFR is secreted from CHO cells that are co-transfected with cDNA encoding both proteins. If GFP or liposome-mediated transfection caused cell lysis, both expressed proteins would be found in the media. Furthermore, CHO cells, when transfected using calcium phosphate instead of a lipsosome formulation, also secrete GFP (data not shown). Finally, only non-fluorescent GFP is secreted (Fig. 7). If released by cell lysis, both fluorescent and non-fluorescent GFP would be expected in the media. Hence, the export process can discriminate between properly and improperly folded forms. This specificity indicates that GFP is not released by cell lysis.

Although GFP can diffuse through the nuclear pore into the nucleus(Chatterjee and Stochaj, 1996;Chatterjee and Stochaj, 1998),we believe that the export of GFP is not caused by non-selective diffusion for several reasons. First, lowering the chase temperature to 25°C significantly inhibits GFP secretion (Fig. 4). Low temperatures inhibit a variety of facilitated translocation processes but not diffusion(Melchior and Gerace, 1995). Second, GFP is poorly secreted from certain cell lines such as COS and HT-29(Fig. 2, lanes 6 and 12). This result suggests that certain cellular factors or machinery are required for GFP export and that export of GFP is not caused by non-specific diffusion of a molten globule isoform through the plasma membrane as has been suggested for other proteins (Bychkova et al.,1988). Such translocation machinery has been characterized in the ER, mitochondria and chloroplasts (Chen et al., 2000; Johnson and van Waes, 1999; Rehling et al.,2001). Finally, the pathway used by GFP for export appears to be selective since other heterologous, cytosolic proteins overexpressed in CHO cells are not secreted (Fig. 1B,C). This selectivity implies that a specific region of the GFP protein may behave as a cryptic targeting signal as has been demonstrated in yeast (Kunze et al., 1999). Thus, the non-classical secretion of GFP is selective and not universal among mammalian cells.

Little is known about non-classical secretory pathways in eukaryotes. Several proteins such as basic FGF, IL-1β, HIV-Tat, thioredoxin and galectin-3 are secreted in a non-classical manner(Cleves, 1997;Rubartelli and Sitia, 1997). However, only a few of these pathways have been studied, and their exact mechanisms have not been identified. For example, the Na⁺/K⁺ ion channel has been implicated in the secretion of basic FGF (Florkiewicz et al.,1998), whereas an ATP-binding cassette (ABC) transporter appears to be involved in the export of IL-1β(Andrei et al., 1999). Galectin-3 is secreted in vesicles from membrane blebbing; this pathway appears to be capable of post-translational export of fully folded proteins(Hughes, 1999). We failed to detect any GFP associated with vesicles. Furthermore, factors in serum and various chemical inhibitors that affect the secretion of other non-classically secreted proteins have no apparent effect on GFP secretion. Our data suggest the existence of another non-classical export pathway in eukaryotic cells capable of secreting unfolded GFP.

Protein translocation across membranes generally requires proteins to be in a loosely folded or unfolded conformation. One exception is the prokaryotic twin arginine translocase (Tat), which is capable of exporting fully folded proteins — including GFP — into the periplasm(Berks et al., 2000;Thomas et al., 2001). In our studies, the GFP secreted from CHO cells is not fluorescent and therefore not properly folded. Perhaps the GFP is loosely folded or unfolded as a condition for secretion. Once secreted, GFP does not fold into a fluorescent conformation because of the absence of chaperone proteins(Feilmeier et al., 2000;Sacchetti et al., 2001). By contrast, GFP secreted via the ER-Golgi pathway is fluorescent in the medium because proper folding is maintained during secretion(Laukkanen et al., 1996).

GFP is fluorescent in the cytoplasm of transfected cells, indicating that some of the protein is properly folded. The GFP in the medium, however, is not fluorescent (Fig. 7B). Thus, to be secreted by CHO cells, GFP must be in an unfolded conformation. Two models could account for this unfolding. In the first model, fully folded GFP must be unfolded prior to or during its export. In the second model, not all of the GFP synthesized in the cytosol is folded to the fluorescent conformation. This pool of nascent, unfolded GFP might remain associated with chaperones in the cytosol to maintain a loosely folded or unfolded conformation. This scenario is much more likely for post-translationally translocated proteins(Schatz and Dobberstein,1996). The prolonged but efficient secretion of GFP despite treatment with cycloheximide (Fig. 8) suggests that export occurs post-translationally. We discovered that the pool of intracellular, fluorescent GFP did not decrease over 6 hours,whereas a substantial fraction was secreted into the medium(Fig. 8). This result indicates that the second model is correct: two separate pools of nascent GFP —one folded and fluorescent, the other unfolded, non-fluorescent and able to be secreted — exist in cells (see model inFig. 9). Being a protein of jellyfish origin, GFP does not always fold properly at 37°C(Ogawa et al., 1995;Patterson et al., 1997;Siemering et al., 1996;Tsien, 1998). Thus, the two pools of GFP in CHO cells probably arise from an inefficiency in proper folding even though we used an engineered form of GFP. Our data does not indicate the relative size of these two pools, although the high proportion that is secreted over several hours (Fig. 3) suggests that the pool of unfolded, non-fluorescent GFP consists of at least half the newly synthesized GFP. The pool of unfolded GFP does not appear to localize in aggresomes formed by aggregates of misfolded proteins in the cytosol (Garcia-Mata et al., 1999; Garcia-Mata et al.,2002). We observed a diffuse pattern of GFP fluorescence throughout the cytosol as is typically observed(Garcia-Mata et al., 1999;Kain et al., 1995). By immunofluorescence microscopy, an identical pattern of GFP staining is seen in the cytosol of these cells (data not shown). Thus, in cells, two overlapping pools of GFP can be found throughout the cytosol.

Since GFP is not present endogenously in CHO cells, the physiological substrate for this non-classical pathway remains to be identified. To be active extracellularly, proteins secreted via this pathway must be able to attain the proper conformation outside the cell in the absence of chaperone proteins. Alternatively, this non-classical secretory pathway might be a means to dispose of improperly folded proteins from the cytosol.

This work was supported by grant DAMD17-01-1-0150 from the United States Army.