Serum GP73 levels are significantly increased in patients with hepatocellular carcinoma (HCC), potentially providing a marker for early detection. However, GP73 is an integral membrane protein localized to the cis Golgi and is not known to be secreted. Based on its presence in sera, we sought to determine whether GP73 might normally be released from cells and to elucidate the mechanism of this release. Indeed, a soluble form of GP73 was released from cultured cells and compared with the Golgi-localized full-length protein, the molecular weight was slightly reduced, suggesting that cleavage releases the GP73 ectodomain. Sequence analysis revealed a proprotein convertase (PC) consensus site, and, indeed, the ubiquitous PC furin was capable of cleaving purified GP73. Further, alanine substitutions in the PC site blocked both the in vitro and the in vivo cleavage of GP73. Using a cleavage-specific antibody, cleaved GP73 was found in the trans Golgi network and endosomes, suggesting that GP73 cleavage occurs as GP73 cycles distal to the early Golgi. We conclude that the endosomal trafficking of GP73 allows for PC-mediated cleavage, resulting in GP73 secretion, and provides a molecular mechanism for its presence as a serum biomarker for HCC.
Hepatocellular carcinoma (HCC) is the fifth leading cause of cancer death worldwide, with an incidence and mortality rate nearly equal (1–3). Several serum biomarkers of HCC have been described, but better markers with improved sensitivity and specificity are being sought (3,4). Recent studies have identified GP73 in the sera of patients with liver disease, particularly HCC (1,2,5) at levels more than 30-fold over that of normal sera, and compared with α-fetoprotein, the most commonly used serum marker, GP73 serum levels appear to be more sensitive for early HCC (2). Further, GP73 is hyperfucosylated in HCC, and its hyperfucosylated fraction in serum appears to be an even better marker of disease (1).
Under steady-state conditions, GP73 is an integral membrane protein of the cis Golgi, making it seem, at first, an unlikely serum marker for disease. However, GP73, similar to the structurally related protein GPP130, cycles out of the cis Golgi to endosomes and the cell surface (6). GP73 localizes to the cis Golgi via a lumenal targeting signal within its coiled-coil stem domain, and its retrieval from post-Golgi structures occurs via the microtubule- and PI3 kinase-independent, late endosome bypass pathway (6), which sorts a subset of proteins, notably the Golgi proteins GP73, GPP130 and TGN46 as well as Shiga toxin, at the early endosome away from those en route to the late endosome/lysosome for direct recycling to the Golgi apparatus (6–8). While endosomal cycling of GP73 serves an as yet undefined purpose, it exposes GP73 to a variety of proteases in the trans Golgi network (TGN) and on the cell surface, providing a means by which this integral membrane protein could be cleaved and released into the extracellular milieu.
One important group of such proteases is the mammalian subtilisin-like serine proteases of the kexin subfamily, also known as proprotein convertases or PCs. This seven-member family, consisting of furin, PACE4, PC1/3, PC2, PC4, PC5/6 and PC7/8/LPC, processes a wide range of proproteins including hormones, metalloproteases, growth factors and signaling peptides (9,10). Numerous studies have shown an upregulation of PCs in cancers: furin in lung, breast, head and neck tumors; PC1/3 and PC2 in neuroendocrine tumors; PACE4 in breast, head and neck tumors (11); PC5 in colon cancer (12) and PC7 in breast carcinoma (13). Increased furin levels specifically have been correlated to a more aggressive and metastatic phenotype of several cancers (14–16).
Based on the known endosomal trafficking of GP73 and the involvement of PC expression in cancer, we sought to determine whether the presence of GP73 in the serum of HCC patients may be explained by GP73 secretion resulting from PC-mediated cleavage.
Immunofluorescence analysis of HeLa cells yielded GP73 staining in a Golgi pattern colocalizing with GPP130 (Figure 1A), and GP73 remained Golgi localized after cycloheximide treatment (unpublished results) indicating that GP73 is a resident, rather than itinerant, Golgi component. Nevertheless, when analyzed by immunoblotting, GP73 was detected in both the cell extracts and the extracellular medium (Figure 1B), while little or no GPP130 was detected in the extracellular medium, indicating that the presence of GP73 was not simply due to cell lysis. Released GP73 migrated slightly faster than cell-associated GP73, suggestive of a cleavage-induced release of a large ectodomain. This ectodomain was also detected in the conditioned medium of foreskin fibroblasts and osteosarcoma cells (Figure 1B). Thus, the release of a fraction of GP73 is a feature of several cell types, suggesting that it may be playing some role in the extracellular environment normally. We also tested whether the mobility of this secreted product matched that of GP73 immunoprecipitated from the sera of two patients with HCC. To reduce the heterogeneity in mobility because of glycosylation, the samples were treated with PNGase F, and strikingly, serum GP73 comigrated with the GP73 released from HeLa cells (Figure 1C).
The GP73 sequence encodes a single pass transmembrane protein that lacks a predicted signal peptidase cleavage site (17). In vitro-translated GP73 is incorporated into microsomes and is not extracted from these membranes upon high pH treatment, indicating that, at least under these in vitro conditions, GP73 is an integral membrane protein (17). Consistent with these findings, carbonate treatment of HeLa cell membranes at pH 11.5 left GP73 in the membrane fraction, while the peripherally associated Golgi protein GM130 was released into the aqueous supernatant (Figure 1D). GP73 in the extracellular medium, however, was entirely soluble under the same salt and pH conditions. Further, in contrast to cellular GP73, released GP73 entirely partitioned into the aqueous phase of Triton-X114 extracts (unpublished results). These results suggest that intracellular, Golgi-localized GP73 is anchored by its transmembrane domain and that this domain is removed by proteolytic cleavage, releasing the GP73 ectodomain.
Analysis of the GP73 protein sequence revealed a consensus cleavage site for the PC family of serine proteases, which cleave carboxy-terminal to RX(K/R)R. The putative site in GP73 (R52VRR55) is located within the predicted coiled-coil stem 20 amino acids away from the membrane-spanning domain, on the luminal side of the protein (Figure 2A–D). A high degree of identity and conservation of the PC site was noted within mammalian species. Probable GP73 homologues were also identified in non-mammalian metazoans, but sequence conservation was less evident and the PC site, if present, begins to diverge from the consensus motif.
Furin is responsible for the majority of proprotein processing in the constitutive secretory pathway and is expressed in a wide variety of tissue types (9,10,18), making it an appealing candidate as the protease responsible for GP73 cleavage. To determine whether furin cleaves GP73, we expressed and purified GP73’s lumenal domain as a chimera fused to glutathione S-transferase (GST). The protein was then incubated with varying amounts of purified furin, showing that cleavage was dependent on furin concentration (Figure 3A). Further, addition of the polyarginine inhibitor of furin, hexa-d-arginine (19,20), blocked this cleavage (Figure 3B).
If R52VRR55 were the site of cleavage, mutations of this sequence at the key arginines should prevent cleavage. As a test of this prediction, we expressed and purified two mutated versions of the GP73 lumenal domain chimera: R52VAA55 and A52VRR55. In contrast to wild-type GP73, furin failed to cleave these proteins, indicating that furin cleaves GP73 at the PC consensus site (Figure 3A).
To test whether GP73 is also cleaved at the PC site in vivo, GP73 expression constructs were generated with C-terminal hexa-myc tags. The C-terminal tag allowed detection of GP73 released into the media using anti-myc antibodies. After transfection of wild-type GP73, immunoblotting revealed the two forms of GP73 as observed above (Figure 4A). In contrast, cells transfected with the R52VAA55 or A52VRR55 constructs yielded exclusively the uncleaved, cell-associated form of GP73, indicating that GP73 release requires cleavage at the PC consensus site. Furthermore, these results suggest that GP73 cleavage in vivo involves furin and/or a closely related PC. While matrix metalloproteases may conceivably come into contact with GP73 as it cycles to the cell surface, they differ from furin in their recognition sequences (21–24). Even N-arginine dibasic convertase, an unusual metalloprotease that can cleave PC-like sites, can be ruled out because it would have been expected to cleave N-terminally of the dibasic in the A52VRR55 variant of GP73 (25,26).
We next tested for furin-dependent cleavage of GP73 in LoVo cells, a colon carcinoma cell line that lacks functional furin (27). Although LoVo cells express other PCs including PACE4, PC5/6 and PC7 (28,29), GP73 recovery in the media of LoVo cells was significantly reduced compared with HeLa cells, and this reduction was reversed by reintroducing wild-type furin via stable transfection (Figure 4B). Thus, while furin clearly participates in cleaving GP73 in vivo, it is not the sole PC responsible for GP73 cleavage, at least in LoVo cells where PC5/6 may be upregulated (12).
Interestingly, however, furin overexpression alone does not increase the release of the GP73 ectodomain in HeLa cells (Figure 5A), suggesting that PC activity may not be the limiting factor in GP73 release and that PC upregulation in HCC, if it occurs, may not by itself be the mechanism underlying the increased GP73 serum levels (see below). However, furin upregulation does contribute to GP73 release in cells overexpressing GP73. As shown, GP73 release was increased in HeLa cells overexpressing GP73 alone, and this effect was significantly augmented by coexpression of furin in these cells (Figure 5A). This was particularly evident after quantifying the percentage of GP73 cleaved (Figure 5B). Because GP73 cycling to the plasma membrane and endosomes and its abundance in these membranes is increased upon overexpression, our results suggest that GP73 overexpression causes greater amounts of GP73 to leave the early Golgi compartment and to come into contact with furin as well as with other PCs, resulting in increased cleavage and ectodomain release.
To investigate the cellular site of GP73 cleavage, the GP73 expression construct was modified by insertion of a FLAG epitope immediately C-terminal to the RVRR site. In the modified construct, cleavage of the resulting protein at this site in a PC-specific manner exposes the FLAG tag at the new N-terminus. This allows for selective antibody binding of cleaved GP73 by the M1 anti-FLAG antibody, which fails to bind an internal FLAG epitope but does bind the FLAG epitope at a free N-terminus. Interestingly, HeLa cells expressing this construct exhibited M1 antibody staining of tubular membranes emanating from the Golgi region and in dispersed, punctate structures. These structures were distinct from the Golgi, as revealed by GPP130 Golgi staining in the same cells (Figure 6A–C). The morphologies of the tubular and punctate structures were characteristic of TGN tubules and endosomal membranes, respectively. As a control demonstrating the specificity of M1 anti-FLAG antibody binding to cleaved GP73, a similar construct containing the mutated PC site R52VAA55 failed to yield staining (Figure 6D–F). Treatment of cells with cycloheximide led to the loss of the M1-positive cleaved GP73 presumably because of its secretion in the absence of replenishment by new protein synthesis (data not shown). Significantly, cleaved GP73 was then generated during a 3-h period of cycloheximide washout, and the M1-positive GP73 repopulated the juxtanuclear and distributed structures (Figure 6G–I). Further, generation of cleaved GP73 required traffic beyond the ER and an intact Golgi because presence of brefeldin A (BFA) during the cycloheximide washout blocked the reappearance of M1-positive GP73 (Figure 6J–L). Quantification of the results indicated that only 1% of the BFA-treated cells exhibited M1 staining, whereas more than 40% of the untreated cells exhibited reappearance of M1-positive GP73.
Colocalization studies were also carried out to confirm the endosomal origin of the punctate structures; however, we were unable to detect coincidence of the M1-positive structures with sorting and recycling endosomes marked by either internalized transferrin or coexpressed GFP-tagged RME1 (unpublished results). A small number of M1 structures became labeled by fluorescein isothiocyanate (FITC)-dextran after 30 min, suggesting partial localization with late endosomes, but when the uptake was extended to 3 h to label lysosomes, the M1 structures were completely distinct (unpublished results). Because the M1 structures might be bypass pathway endosomes that lack conventional markers, we next treated the cells with chloroquine to trap recycling GPP130 in dilated endosomes (30). In chloroquine-treated cells, the cleaved GP73 ectodomain was present in dilated endosomes containing redistributed GPP130 (Figure 7A–C). Further, these structures became labeled by externally added antibodies against GPP130 (Figure 7D–F). Antibody internalization was time and temperature dependent, demonstrating that the M1-positive punctate structures are accessible to an endocytic tracer. Note that in chloroquine-treated cells, the Golgi itself is intact and distinct from endosomes and that externally added anti-GPP130 antibodies are blocked in bypass pathway endosomes (6,30). Thus, although we cannot definitively identify the site of GP73 cleavage, the evidence strongly suggests that cleavage is initiated in the TGN and that, prior to its secretion, the ectodomain product accumulates in membranes that are accessible to an endocytic tracer of the bypass pathway.
Altogether, our data demonstrate that GP73 is released from cells by endoproteolytic cleavage at the R52VRR55 PC consensus site and that this cleavage takes place as GP73 traffics out of the early Golgi into the TGN and possibly endosomes where at least a portion of cleaved GP73 accumulates. Secretion of the cleaved GP73 ectodomain accounts for its presence in the media of cultured cells, providing a mechanistic framework to understand its presence in sera from patients with liver disease.
Following from our results, the upregulation of GP73 expression seen in hepatocytes appears to be a major contributor to its increased serum presence in HCC (Figure 8). The GP73 retrieval mechanism is saturable, and increased GP73 expression results in a greater extent of cycling in the PC-containing compartments that are distal to the early Golgi. Additionally, increased GP73 distal cycling because of upregulated expression could result in exposure of GP73 to PCs that it rarely contacts under normal circumstances. While we have shown that furin is involved in GP73 cleavage in LoVo cells, it was not the sole protease involved. In HCC, there could be increased involvement of cell surface PCs such as PC6 and possibly PACE4 [although it is downregulated in HCC (31)] and even the soluble, cell-surface-associated ‘shed’ furin (32).
Additional factors may also contribute to increased GP73 release in HCC. Preventing endosome acidification blocks GP73 Golgi retrieval and causes its accumulation in distal compartments (6). Thus, organelle alkalinization, which has been described in several breast and colon cancer cell lines (33), could result in increased contact of GP73 with PCs and the subsequent release of its ectodomain.
Upregulation of furin expression is an alternative, and perhaps more likely, contributing factor in GP73 release in HCC as our results indicate that furin overexpression increases GP73 cleavage, provided that GP73 is also overexpressed. Furin is upregulated in many cancers, and furin messenger RNA levels are known to be upregulated in hepatocytes in response to hepatic injury (34). Additionally, transforming growth factor-β (TGF-β), an activator of furin expression through Smads 2–4 (35), is upregulated in HCC (3). Thus, furin expression may be enhanced in HCC and, indeed, furin substrates including glycipan-3, vitronectin, des-gamma carboxyprothrombin and vascular endothelial growth factor C exhibit increased cleavage and/or expression in HCC (36). Interestingly, furin also processes pro-TGF-β to its mature form (33,34), suggesting that a positive feedback loop controls furin expression. As GP73 expression appears to be responsive to some cytokines (37), increased TGF-β might also upregulate GP73 expression. Because TGF-β enhances cell invasiveness (38,39), it is tempting to speculate that cancer cells may somehow benefit from enhanced GP73 secretion. Serious consideration of this possibility, however, will require an understanding of the function(s) of intra- and extracellular GP73.
In summary, we have identified the mechanism by which GP73, a resident cis Golgi protein, reaches the cell surface and is released into the extracellular space via PC-mediated cleavage. Our observations provide a molecular basis for the clinical utility of GP73 as a serum marker of hepatocellular cancer.
Materials and Methods
The GP73 coding region was cloned into the pCS2 vector with an in-frame hexa-myc epitope tag encoded at the GP73 C-terminus. Point mutations were made in this construct according to the Quickchange protocols (Stratagene, La Jolla, CA, USA) by the forward primer 5′-CTGGAAGGCAGGGTCGCCGCGGCGGCTGCAGAGAGA-3′ (R52VAA55) or 5′-GGAGCTGGAGGGCGCCGTCCGAAGGGCGGC-3′ (A52VRR55) (reverse compliment primers not shown). A polymerase chain reaction (PCR)-based loop-in strategy was used to insert the FLAG epitope adjacent to the wild-type PC consensus site using the same PCR conditions above and the following primers encoding the FLAG epitope: 5′-GGAAGGCAGGGTCCGCAGGGACTACAAAGACGATGACGACAAGGCGGCTGCAGAGAGAGG-3′ (forward) and 5′-CTTGTCGTCATCGTCTTTGTAGTC-3′ (reverse). The R52VAA55 point mutant was made in this construct by Quickchange procedures using the forward primer 5′-GAGCTGGAAGGCAGGGTCCGCAGGGACTACAAAGACGATG-3′ (reverse compliment not shown).
Tissue culture and antibodies
Hela and MG63 cells were grown in MEM, and HCA and LoVo cells were grown in DMEM and F12-K, respectively (Sigma-Aldrich). LoVo cells stably transfected with furin (40) were a gift from Claire M. Dubois (Universite de Sherbrooke, Sherbrooke, QC, Canada). Cycloheximide, chloroquine and BFA treatments were carried out in growth medium at 100 μg/mL for 4 h, 0.1 mm for 3 h and 2.5 μg/mL for 3 h, respectively. For immunostaining and immunoblotting, the following primary antibodies were used: GP73 (1:2000), GPP130 (1:1000), GM130 (1:500), FLAG-M1 (1:400 with 1 mm CaCl2; Sigma-Aldrich) and myc (1:200). The secondary antibodies used were horseradish peroxidase-conjugated goat anti-rabbit/mouse (1:2000; Biorad, Hercules, CA, USA) and rhodamine and fluorescein-conjugated goat anti-rabbit/mouse (1:200; Invitrogen).
Cells were passed according to the Transfectol guidelines (GeneChoice, Frederick, MD, USA). The cells were grown on 35-mm plates for experiments involving transfection or on 60-mm plates without transfection. For nontransfected cells, media collection was initiated the following day. After three PBS washes, the cells were cultured 24 h in growth medium lacking serum. When transfection was required, it was carried as outlined in the Transfectol product insert. The following day, the PBS washes and 24-h incubation were initiated as above. After the 24-h incubation, the medium was removed, adjusted with a protease inhibitor cocktail (10 mg/mL leupeptin and pepstatin and 50 mm phenylmethylsulfonyl fluoride) and briefly centrifuged to remove any contaminating cells. The cells were lysed in HKT buffer (10 mm HEPES, pH 7.2, 100 mm KCl, 1% Triton-X-100 and the protease inhibitor cocktail). The lysates were centrifuged to remove insoluble material, and the medium fraction was precipitated with trichloroacetic acid and analyzed by immunoblotting. For transfection experiments, 50% of each sample were analyzed. For nontransfection experiments, 25% of the cell fraction and 100% of the medium fraction were analyzed. For LoVo cells, the cell and medium fractions were precipitated with acetone, followed by treatment with 200 units of PNGase F for 24 h according to product insert recommendations (New England Biolabs), using 2000 units, before a final precipitation with trichloroacetic acid.
Anti-GP73 antibodies were cross-linked to protein A–Sepharose beads (Amersham) with 20 mm dimethylpimelimidate as described (41) and incubated over night with either 30 μL of patient’s serum or the medium collected from a 60-mm plate of HeLa cells. After three washes with HKT buffer, the immunoprecipitated GP73 was released by heating in sodium dodecyl sulfate buffer and incubated with PNGase F as described above and then precipitated with trichloroacetic acid as described above, followed by GP73 immunoblotting.
Carbonate treatment was carried out as described (30). Briefly, medium was collected from cells after a 24-h incubation, and the cells were homogenized in media containing 250 mm sucrose, 10 mm triethanolamine pH 7.4, 1 mm ethylenediaminetetraacetic acid and protease cocktail using a 25-gauge needle. The sucrose concentration was then adjusted to 50% (w/v). Membranes were collected by differential centrifugation using membrane flotation at the interface of 10 and 45% sucrose/TEA overlays. The pH of the resulting fraction and of the medium was adjusted to pH 11.5 with 100 mm carbonate, and the samples were kept on ice for 30 min. After centrifugation at 100 000 × gfor 60 min, the resulting pellet and supernatant fractions were analyzed by immunoblotting.
Furin in vitro assay
The lumenal domain of GP73 (aa 36–aa 195) was cloned in-frame after the N-terminal GST in pGEX-2T (Amersham), expressed using isopropyl-beta-D-thiogalactopyranoside induction and isolated on glutathione agarose beads (Sigma-Aldrich). The fusion protein was eluted and dialyzed into PBS to remove DTT and glutathione. Digestion was performed in furin reaction buffer (100 mm Tris pH 7.5, 1 mm CaCl2, 0.01% Triton-X-100) using 20 μg of protein and 0, 1, 2 or 4 units of furin (New England Biolabs), for 3 h at 37°C. Hexa-d-arginine (Calbiochem) treatment was performed under the above conditions using 4 units of furin and 10 mm hexa-d-arginine.
Immunostaining was carried out as described (30). Microscopy was performed using a spinning-disk confocal system with a 100× oil-immersion objective (42). z-Axis sectioning was at 0.3 μm. Images are shown as maximum value projections.
We thank members of the Linstedt Lab and Dr Tina Lee for critical advice. Funding was provided by National Institutes of Health grant GM-56779 to A. D. L.