Thyroglobulin Transport along the Secretory Pathway

GRP94 serves as a molecular chaperone in the endoplasmic reticulum (ER). In normal thyrocytes, GRP94 interacts transiently with thyroglobulin (Tg), and in thyrocytes of animals suffering from congenital hypothyroid goiter with defective thyroglobulin, GRP94 and thyroglobulin associate in a protracted fashion. In order explore possible consequences of GRP94 binding, we have studied recombinant nonmutant thyroglobulin expressed in control Chinese hamster ovary (CHO) cells in comparison to that produced in CHO cells genetically manipulated for selectively increased GRP94 expression. Levels of ER chaperones other than GRP94 did not detectably differ, and thyroglobulin achieved transport competence in both kinds of CHO cells. However, increased availability of GRP94 caused the residence time of Tg in the ER to be remarkably prolonged. This was accompanied by a major increase in Tg directly associated with GRP94 and an increase in the ER pool size of Tg. Importantly, co-immunoprecipitation analysis revealed disulfide-linked Tg complexes (previously reported as an early Tg-folding intermediate) especially associated with GRP94. Indeed, non-native Tg, GRP94, and a 78-kDa protein likely to be BiP, appeared in ternary complexes. Under these conditions, GRP94 association appears directly involved in prolongation of Tg folding and export, consistent with a role in quality control in the ER.

group of hereditary illnesses originating from mutations in secretory proteins or other exportable proteins that interfere with protein folding and exit from the ER (7) .
In both normal secretory protein export and in ER storage diseases, the functions of ER chaperones remain poorly understood. BiP, perhaps the most studied ER chaperone, interacts transiently with certain wild-type exportable proteins and exhibits more prolonged interactions with certain mutant or unassembled exportable proteins (8 -18). Nevertheless, it has been difficult to clearly define a single post-translational role of BiP in the export of secretory proteins (see Introduction of Hendershot et al. (19)). Even less is known about the role played by GRP94 in protein export, although its demonstrated ability to bind certain secretory proteins is also well established (20,21). For instance, although it has been concluded that, in conjunction with its chaperone function in the ER, GRP94 is a molecule capable of ATP binding (22), hydrolysis (23), and autophorphorylation activity (24), these conclusions have recently been called into question by new experiments which suggest that peptide interaction with GPR94 is ATPindependent (25).
We have been interested to know more about the effects of binding of GRP94 on the folding and export of thyroglobulin (Tg), the secretory protein precursor for thyroid hormone synthesis, and one of the reported "substrates" for GRP94 association in thyroid epithelial cells (18,26,27). Tg is a large (ϳ330 kDa) glycoprotein that undergoes substantial initial folding, including the formation of nearly 60 intrachain disulfide bonds, before homodimerization, which normally occurs in the ER. Remarkably, even when great precaution is taken to prevent formation of artifactual disulfide bonds at the time of cell lysis, immediately after translation in primary thyrocytes, newly synthesized nonmutant Tg is found in high molecular mass complexes, many containing improper interchain disulfide bonds (27)(28)(29). Subsequently the disulfide-linked Tg complexes become undetectable while Tg is observed to advance toward more folded forms of individual monomers (14). During early folding, nonmutant Tg dissociates from GRP94 with kinetics superimposable upon those of BiP (18). These kinetics precede Tg homodimerization and its vesicular egress from the ER (14).
In thyrocytes as in other cell types, GRP94 levels are physiologically regulated (30). Indeed, in humans suffering from congenital goiter with mutant Tg, thyrocytes can achieve levels of GRP94 that are elevated by Ն1 order of magnitude, which represents a greater increase than that observed for BiP (31). Interestingly, in the cog/cog mouse (32)(33)(34), an animal model of this illness in which thyroid tissue is available for pulsechase analysis, the fraction of Tg molecules that exits the ER is much lower than normal, while an increased fraction of Tg interacts with GPR94 in a prolonged manner (18).
Unfortunately, congenital hypothyroid goiter is not a suitable model in which to independently examine the role of GRP94 binding in Tg maturation, because it is difficult to resolve the consequences of increased GRP94 binding from those related to intrinsic alterations in the biophysical properties of mutant Tg. For this reason, in the present report we have examined the conformational maturation and export of nonmutant Tg as a consequence of manipulating the probability of GRP94 binding to recombinant Tg in Chinese hamster ovary (CHO) cells engineered for wild-type or increased levels of GRP94 expression.
Our results indicate that a major increase in the ER residence time and expansion of the ER pool of Tg occurs when the availability of GRP94 is increased, and this retention occurs as a direct consequence of complex formation between an apparent Tg folding intermediate and this ER chaperone.

EXPERIMENTAL PROCEDURES
Materials-The following vectors were employed: pCB6, a mammalian expression vector carrying a neomycin resistance cassette and cytomegalovirus promoter-driven insert (originally from Dr. M. Stinsky, University of Iowa); pBAT14, used as a shuttle vector, was from Dr. M. German (University of California, San Francisco); pBR322 was purchased from Upstate Biotechnology Inc. Co-vidarabine (pentostatin, inhibitor of adenosine deaminase) was from Parke-Davis. L-Alanosine was obtained from the Drug Synthesis and Chemistry Branch, Developmental Therapeutics Program, Division of Cancer Treatment, NCI) as distributed by Ogden BioServices Corporation. Dithiobis(succinimdyl propionate) (DSP) was purchased from Pierce. Restriction enzymes, T4 DNA ligase, and recombinant endoglycosidase H were from New England Biolabs (Beverly, MA). A polyclonal rabbit antiserum was raised against denatured Tg as described previously (35). Antibody to ribophorin II was the kind gift of Dr. D. Meyer (University of California, Los Angeles). An immunoprecipitating polyclonal antiserum to GRP94 (36) was kindly provided by Dr. P. Srivastava (Fordham University, Bronx, NY). A polyclonal antiserum to BiP was purchased from Stress-Gen (Victoria, Canada). Polyclonal antibodies to protein disulfide isomerase and ER60 were from Dr. T. Wileman (Pirbright Laboratories, Surrey, UK), and polyclonal antibodies to GRP94, calnexin, ERp72, and calreticulin (18) were also provided by Dr. P. Kim (Beth Israel Hospital, Boston, MA). A rhodamine-conjugated, affinity isolated, goat anti-rabbit IgG was purchased from Tago, Inc. (Burlingame, CA), and an alkaline phosphatase-conjugated goat anti-rabbit IgG was from Life Technologies, Inc. The serine protease inhibitor 4-(2-aminoethyl)benzenesulfonyl fluoride was from ICN (Costa Mesa, CA). 125  Construction of Full-length Tg cDNA-Until now, no full-length Tg cDNA has ever been successfully prepared. For this reason, our strategy involved subcloning four contiguous partial cDNAs (37), kindly provided by Dr. G. Vassart (University Libre, Brussels, Belgium), to create a cDNA corresponding to base pairs 519 -8430 (i.e. from a conserved NcoI site to the 3Ј-end) of the bovine Tg coding sequence (Fig. 1). Using the rat Tg-2 cDNA (38) obtained from Dr. P. Graves (Mt. Sinai School of Medicine, New York), the remaining Tg coding sequence (i.e. from the extreme 5Ј-end to the conserved NcoI site) was provided. Thus, the final full-length cDNA we employed encodes for a polypeptide that exhibits overall 99.1% identity to normal bovine Tg, with the remaining 0.9% being conservative bovine 3 rat Tg substitutions contained within the first 155 amino acids.
For stepwise construction, a fragment of the rat Tg-2 cDNA extending from an EcoRI site (17 bases upstream from the translational start site) to NcoI (Fig. 1) was ligated into the shuttle vector pBAT14 along with a partial bovine Tg cDNA extending from the conserved NcoI site to a HindIII site at position 2826. The ligated insert was excised from pBAT14 using BamHI from the polylinker and HindIII, and then subcloned into the BglII and HindIII sites in the polylinker of pCB6, destroying the former site. The resulting plasmid was then digested with HindIII and the sole downstream BamHI site, and a partial Tg cDNA encoding the 3Ј-end (position 7446 -8430) was directionally ligated (Fig. 1). Independently, HindIII-HindIII cDNA fragments extending from positions 2826 -4764 and 4764 -7446, respectively, were ligated together at the HindIII site of pBR322, and appropriately oriented subclones were selected from DNA minipreps. From this, the correctly ligated 4.6-kilobase pair insert was excised from pBR322 by partial digestion with HindIII, and this gel-purified insert was ligated into the HindIII-digested pCB6 which contained the rest of the Tg cDNA. The size and orientation of the final full-length clone ( Fig. 1) was confirmed by identity to the known bovine Tg restriction map (37).
Cell Culture, Tg Transfection, and Selection of Stable Tg-expressing Clones-Two lines of CHO cells were graciously provided by Dr. A. Dorner (Genetics Institute, Cambridge, MA) for use in the present studies. The parental ("CHO-P" cell) line, containing endogenous levels of ER chaperones, is a dihydrofolate reductase-deficient line, previously called DUKX-B11 (39). The GRP94-overexpressing ("CHO-G" cell) line was prepared as pooled transfectants overexpressing murine GRP94 (2) as a consequence of selection by co-amplification of adenosine deaminase (40).
CHO cells were maintained in media based on ␣-minimum essential medium containing 1% penicillin-streptomycin. The medium for CHO-P cells also contained 10% fetal bovine serum and 10 mg/liter each of adenosine, deoxyadenosine, and thymidine, whereas CHO-G cell medium was supplemented with 10% heat-inactivated fetal bovine serum, 1 mM uridine, 1.1 mM adenosine, 10 mg/liter each of deoxyadenosine and thymidine, 1 M pentostatin, and 0.05 mM L-alanosine. Depending upon confluence, cells were fed every 2nd or 3rd day.
For Tg transfection, subconfluent cell cultures were rinsed with PBS and detached with trypsin-EDTA. A 0.25-ml suspension of 2 ϫ 10 6 cells was transfected with the full-length Tg cDNA in pCB6 (20 g) in a 0.4-cm pass electroporation cuvette at 330 V and 250 F (time constant ϳ14 ms); cells were then diluted 400-fold and plated. After 2 days in culture, selection was started by addition of 0.8 mg/ml geneticin. Colonies were picked and screened for Tg expression by immunofluorescence; media and cell lysates from positive clones were then analyzed by immunoblotting. At least two Tg-expressing clones of each type were further studied; by initial characterization, the results obtained with replicate clones were essentially identical (e.g. see Fig. 4), thus the data presented in this report all derive from representative clones.
Immunofluorescence-Cells grown on coverslips were rinsed with PBS, fixed for 15 min at room temperature with 4% formaldehyde, and permeabilized in PBS containing 0.2% (v/v) Triton X-100 and 1 mg/ml bovine serum albumin. Incubation with primary antibody was carried out for 1 h at room temperature in the permeabilization buffer. Cells were then rinsed with this buffer and incubated for another hour with a 1:400 dilution of a rhodamine-conjugated, affinity-isolated goat antirabbit IgG. The rinsed and mounted specimens were examined with a Zeiss microscope equipped with epifluorescence optics.
Metabolic Labeling-CHO cells were washed twice with Cys-free, Met-free medium, prior to pulse-labeling for either 10 or 20 min at 37°C in the same medium containing 0.5 mCi/ml [ 35 S]cysteine and methionine. At the conclusion of the pulse labeling, the cells were washed and chased in normal growth medium. For long term labeling to approach steady state, the medium was not deficient (i.e. the medium contained unlabeled cysteine and methionine, plus serum, at half the usual concentration of that in complete growth medium), and labeling was for 20 -24 h. In preliminary experiments (not shown), results from experiments analyzed by two-dimensional SDS-PAGE (like those shown in Figs. 7 and 8) were not significantly affected by continuous labeling times ranging from 8 to 24 h, although labeled amino acid incorporation was proportional to the length of the labeling period. For the experiment shown in during lysis (indeed, previous detailed investigations of Tg folding have established that detection of high molecular mass disulfide-bonded Tg complexes, as shown in this report, is not due to the presence or absence of alkylating agent during cell lysis) (28,29). The cell lysate was boiled for 3 min, passed 15 times through a 25-gauge needle, and then spun in a Microfuge for 10 min at 4°C to remove debris. An aliquot from this supernatant was diluted Ն20-fold in immunoprecipitation buffer (25 mM Tris buffer, pH 7.5, containing 1% Triton X-100, 0.1% SDS, 0.2% deoxycholic acid, 10 mM EDTA, 100 mM NaCl). 1 ml of the diluted cell lysate was preabsorbed for 30 min at room temperature with 50 l of a 10% suspension of fixed Staphylococcus aureus (Zysorbin). The suspension was pelleted in a Microfuge for 2 min at 3,000 rpm, and the supernatant was incubated for 16 h at 4°C with 10 l of polyclonal rabbit anti-Tg. 50 l of a 10% suspension of Zysorbin was then added, and immune complexes were allowed to adsorb for 1 h at 4°C. Pellets of this suspension were washed once in immunoprecipitation buffer, once in 0.5% Tween-20 in TBS (25 mM Tris, 150 mM NaCl, pH 7.4), once in TBS, and finally in water, before boiling for 5 min in 20 -40 l of 2 ϫ sample buffer in the presence or absence of 10% ␤-mercaptoethanol.
For quantitative determination of Tg by immunoprecipitation, the amount of cell lysate obtained from different clones was quantitated by DNA assay using bisbenzimide fluorescence with a Hoefer (San Francisco, CA) Mini-Fluorometer, according to the protocol provided by the manufacturer.
For sequential immunoprecipitation, cell lines labeled to approach steady state and cross-linked with DSP, were first immunoprecipitated with anti-Tg as described above. These immunoprecipitates were eluted from Zysorbin for 1 h at 60°C in 50 l of 1% SDS plus 62.5 mM Tris, pH 6.8. The supernatant was then diluted to 1 ml in immunoprecipitation buffer containing 1 mg/ml unlabeled bovine Tg (Sigma) and mixed with 2 l of anti-GRP94. Final immunoprecipitates adsorbed to Zysorbin were eluted by boiling for 4 min in 30 l of sample buffer containing ␤-mercaptoethanol.
Digestion with Endoglycosidase H-Cell lysates prepared in 0.5% SDS, 1% ␤-mercaptoethanol in 50 mM sodium citrate, pH 5.5, plus protease inhibitors were boiled for 5 min and then either digested or mock digested for 1 h at 37°C as described previously (35). The samples were then diluted to 1 ml and immunoprecipitated with anti-Tg.
Immunoblotting-For immunoblot analysis, samples resolved on SDS-gels were electrophoretically transferred to nitrocellulose. The membrane was blocked for 1 h with 3% gelatin in TBS plus 0.5% Tween-20, incubated for 1 h with primary antibody in the same solution, and then washed three times. The blot was then incubated for 1 h with a 1:3,000 dilution of alkaline phosphatase-conjugated goat antirabbit IgG, rinsed several times, and then reacted with 4-nitro blue tetrazolium chloride and 5-bromo-4-chloro-3-indolyl phosphate. In the immunoblots of Fig. 2, the lanes were loaded based on total cellular protein, and the secondary reagent employed was radioiodinated protein A; in this case, bands were quantitated by phosphorimaging. It is theoretically possible that normalizing to total cellular protein could unintentionally introduce small quantitative differences from that which might be obtained upon normalizing samples to cellular DNA, but we can estimate that any such error, if it existed, would be Յ10% and would affect neither the outcome nor the conclusions obtained from these experiments.
Two-dimensional SDS-PAGE-Tg immunoprecipitates were separated in the first dimension under nonreducing conditions by 4% SDS-PAGE. For the second dimension, the samples were reduced with 10% ␤-mercaptoethanol in sample buffer and resolved by 6.5% SDS-PAGE. Single dimensional gels used acrylamide concentrations that varied depending upon the particular experiment.

Characterization of CHO Cell Lines with Normal and
Increased Expression of GRP94 -To evaluate potential effects of GRP94 on Tg maturation and intracellular transport, we set out to examine CHO cells that maintain either wild-type levels of ER chaperones (parental CHO-P cells), or pooled transformants achieving elevated levels of GRP94 (CHO-G). Using the adenosine deaminase co-amplification system (40), GRP94 expression in CHO-G cells detected by immunoblotting ( Fig. 2A) was increased ϳ13-fold, on average (see Fig. 2B).
Since increased abundance of GRP94 might indirectly affect Tg trafficking by displacing other ER resident proteins (56, 66, 67), we tested the levels of several other ER proteins by specific immunoblotting (Fig. 2). Importantly, increased expression of GRP94 had no significant effect on the steady state level of BiP (Fig. 2, A and B). Further, compared with CHO-P cells, there was no appreciable change in the total amount of rough ER in CHO-G cells, based on the levels of ribophorin II and calnexin (Fig. 2C), which are among the membrane proteins that appear to reflect the cellular quantity of rough ER (41,42). Moreover, increased expression of GRP94 did not alter the concentrations of any of the other ER lumenal resident proteins that were examined, calreticulin, PDI, ERp72, and ER60 (Fig. 2C). Thus, the levels of GRP94 appear selectively increased in CHO-G cells.
Expression of Recombinant Tg in CHO Cell Lines-We next proceeded to constitutively express an 8.4-kilobase pair cDNA encoding full-length, nonmutant Tg (Fig. 1, see "Experimental Procedures"). CHO-P and CHO-G cells were each stably transfected, and clones were selected initially for neomycin resistance and then screened for Tg expression by immunofluorescence and immunoblotting. As shown in Fig. 3A, in comparison to mock-transfected cells, CHO-P and CHO-G cell lines each stably expressed the Tg polypeptide, although the relative Tg pool sizes in these lines appeared to differ (see below).
In thyroid epithelial cells, most newly synthesized Tg follows a high probability pathway that concludes with homodimerization, arrival in the medial Golgi compartment, and secretion from cells (28), while mutant Tg that exhibits defective homodimer formation is quantitatively deficient in ER exit and arrival in the extracellular space (18). This provides a useful correlation between proper Tg folding/assembly and its subsequent secretion. With this in mind, when screening conditioned media from Tg-transfected CHO-P and CHO-G cells, all clones were found to secrete a full-length polypeptide that could be detected by immunoprecipitation (Fig. 3B) and immunoblotting with anti-Tg (Fig. 3C, lanes 1-3). Even without immunoprecipitation, Tg was readily observed as an autoradiographic band from the SDS-PAGE analysis of secretion from labeled, transfected cells (Fig. 3C, lanes 4 and 5). These results indicate that at least a fraction of recombinant Tg attains a conformation allowing for intracellular transport.
Protracted Residence of Tg in the ER as a Consequence of Increased Availability of GRP94 -To determine the extent to which increased availability of GRP94 might influence the intracellular residence of Tg, we compared Tg synthesis in the FIG. 2. Levels of ER resident proteins in CHO cells genetically engineered for selectively increased expression of GRP94. A, equal amounts of total cellular protein from parental CHO cells (P) and CHO cells selected for amplified expression of GRP94 (G), were immunoblotted on nitrocellulose with antibodies either to GRP94 (M r ϳ94,000) or BiP (M r ϳ78,000), followed by 125 I-labeled protein A. B, quantitation of phosphorimages shows the amount of GRP94 or BiP expressed relative to that in parental cells (error bars ϭ S.D.). C, immunoblots of cell extracts similarly analyzed for lumenal ER resident proteins calreticulin (M r ϳ58,000), protein disulfide isomerase (PDI, M r ϳ59,000), ERp72 (M r ϳ72,000), ER60 (M r ϳ60,000), and for ER membrane proteins ribophorin II (M r ϳ65,000) and calnexin (M r ϳ90,000). The levels of these proteins appear similar in CHO-P and CHO-G cells. different clones (as reflected by 35 S-labeled amino acid incorporation into full-length Tg during a 20-min pulse) to the intracellular Tg accumulated at steady-state (as measured by 35 S-Tg content of cells continuously radiolabeled for 20 h). Two independent lines of CHO-P and CHO-G cells were analyzed. In each case, labeled intracellular Tg was recovered by immunoprecipitation from cell extracts, normalized for DNA content, and analyzed by SDS-PAGE and phosphorimaging (Fig. 4A). From this analysis, the Tg synthetic rate in CHO-G transfectants, on average, was found to be 1.7-fold compared with that of parental cells (Fig. 4B, open bars). Remarkably, however, in CHO-G clones, intracellular Tg at steady state was elevated out of proportion to the synthetic rate, i.e. compared with parental cells, CHO-G cells maintained 11.6-fold more intracellular Tg (Fig. 4B, closed bars).
The increase in intracellular Tg pool size as a consequence of increased availability of GRP94 suggested the possibility of slowed Tg egress from these cells. To test this point, we examined the rate of newly synthesized Tg secretion by noncumulative hourly collections of chase medium after pulse labeling. In CHO-G cells the maximum rate of labeled Tg release was delayed Ն2 h over that seen in CHO-P cells (Fig. 5). Thus, increased availability of GRP94 markedly prolonged the intracellular residence of Tg.
To identify the compartment containing the greatest intra-cellular accumulation of Tg, immunofluorescence with anti-Tg was performed in stably transfected CHO cells. Unlike in untransfected CHO cells, a fine reticular fluorescence throughout the cytoplasm, characteristic of the ER, was observed both in Tg-transfected CHO-P and CHO-G cells (Fig. 6A). This fluorescence pattern appeared essentially superimposable with that observed using antibodies to the ER chaperones, BiP and GRP94 (not shown). To confirm the Tg distribution, lysates of  1 and 4) or after Tg transfection into CHO-P cells (lanes 2 and 5) were labeled, and the media were collected as in B. The media were subjected to SDS-PAGE and electrophoretic transfer to nitrocellulose. The same transfer membrane was first autoradiographed (lanes 4 and 5) and then directly immunoblotted with anti-Tg and an alkaline phosphatase-coupled secondary antibody (lanes 1 and 2). In lane 3, identically analyzed medium from a separate immunoblotting experiment using CHO-G cells is aligned for comparison. The mobilities of full-length bovine Tg and myosin molecular mass standards are shown at right. In each case, the labeled, immunoprecipitated Tg from CHO-P cells was arbitrarily assigned one unit (error bars ϭ S.D.). As the specific radioactivity in the two different types of labeling experiments are markedly different, note that this graph cannot be used to compare the absolute amount of newly made Tg protein to the amount of Tg protein contained within CHO cells at steady state.

FIG. 5. Tg progression through the secretory pathway is retarded in cells with increased available GRP94.
Clones of CHO-P and CHO-G cells at roughly comparable cell density were metabolically labeled with 35 S-labeled amino acids for 10 min. After labeling, noncumulative hourly collections of media were analyzed by SDS-PAGE and phosphorimaging. One clone from each duplicate is shown in the halftone panels (at left). Identity of the Tg band was confirmed by immunoprecipitation with anti-Tg (not shown). Relative band intensity of labeled Tg secreted by a given clone type at each chase time is shown as the mean of duplicates (at right). No attempt was made to normalize the data for cell number between different sets of clones; but within each duplicate, variation in the total amount of labeled Tg protein secreted over 10 h averaged ϳ9% for P cells and ϳ2% for G cells. Importantly, the maximum rate of newly synthesized Tg secretion by CHO-G cells (dashed line) appeared to be delayed ϳ2 h over that seen in CHO-P cells.
steady-state labeled cells were digested with endoglycosidase H, immunoprecipitated for Tg, and analyzed by 4% SDS-PAGE (Fig. 6B). By this assay, the fraction of intracellular Tg residing in Golgi/post-Golgi compartments was undetectable, since virtually all was sensitive to digestion, although recombinant Tg secreted from CHO cells was entirely resistant to endoglycosidase H (Fig. 6B). Evidently, the transit time for recombinant Tg through Golgi/post-Golgi compartments is sufficiently fast such that endoglycosidase H-resistant Tg does not accumulate within the CHO cells. Thus, even though Tg spends most of its intracellular residence time in the ER of both CHO-P and CHO-G cells, the data in Figs. 4 -6, taken together, indicate that retention of Tg in the ER is substantially increased with increased availability of GRP94.
Tg Retained in the ER Is Bound to GRP94 -One simple way to explain the prolonged Tg residence and expanded Tg pool in CHO-G cells is chaperone-mediated retention of Tg in the ER. Support for such a model would require direct proof of trapping Tg in complexes that include the chaperone whose availability is increased. Transient disulfide-linked Tg complexes associated with ER chaperones have been found early in the maturation of wild-type Tg; however, at a moment in time in the steady state in normal thyrocytes, the quantity of these complexes is at nearly undetectable levels (14,28,29,43). We therefore sought to determine whether such complexes might accumulate when recombinant nonmutant Tg was expressed in CHO cells with increased availability of GRP94. After labeling overnight to approach steady state, the cells were extracted under denaturing and nonreducing conditions, immunoprecipitated for Tg, and then analyzed in a first dimension by nonreducing 4% SDS-PAGE, followed by a second dimensional 6.5% SDS-PAGE under reducing conditions. The intention of this two-dimensional gel system was to segregate disulfide-linked Tg complexes from uncomplexed Tg in the first dimension, and then to dissociate disulfide-linked complexes in the second dimension.
From this analysis (Fig. 7), Tg molecules were specifically obtained from both CHO-P and CHO-G cells, although the forms of Tg recovered in the two types of CHO cells were not entirely identical. Radioactive background bands could be detected in control CHO cells that did not express Tg (Fig. 7,  upper left panel) while the expected mobility of labeled Tg from CHO-G cells upon reducing 6.5% SDS-PAGE is indicated by a sample run only in this dimension (Fig. 7, lower left panel). Interestingly, in the two-dimensional gels (Fig. 7, right panels), some of the Tg molecules were spread along the horizontal, indicating separation in the first, nonreducing dimension. The majority of recovered Tg appeared as the uncomplexed species (left of dashed line), and "trailing" of this form (just left of the dashed line) suggested a range of Tg folding states with different numbers of intrachain disulfide bonds. In addition, however, in CHO-G cells, a fraction of immunoprecipitable Tg was broadly smeared in the high molecular mass region (right of dashed line). The apparent molecular mass of this fraction of Tg is incompatible with uncomplexed Tg monomers or native homodimers (which run as monomers upon SDS-PAGE). Persistent band broadening of the high molecular mass fraction of Tg in the second (vertical) dimension suggests incomplete disulfide reduction after the first-dimensional gel. These results indicate that a meaningful fraction of immunoprecipitable Tg had accumulated in disulfide-linked complexes as a consequence of increased abundance of GRP94. Interestingly, when looking directly beneath this high molecular mass fraction of Tg on the two-dimensional gels of CHO-G cells, there was no apparent detection of associated chaperones (Fig. 7). However, such a result was not unexpected, since ER chaperones do not remain associated with exportable proteins under SDS-PAGE conditions.
For this reason, the same protocol was followed in cells exposed to the membrane-permeant, thiol-cleavable cross-linking reagent, DSP, which has been found to be required for preserving protein associations with GRP94 (18,21,26,27,44). Once again, radioactive background areas were detected in control CHO cells treated with the cross-linker that had not been transfected to express Tg (Fig. 8, upper left panel). When primary antibody was omitted, no immunofluorescence signal was detected in these cells (not shown). To compensate for the increase in intracellular Tg (Fig. 4) in CHO-G cells for which film exposure was 30 s, the exposures shown for CHO-P cells not transfected (NT) or transfected with Tg (P) were lengthened (60 s) to increase signal in these cells. Bar ϭ 10 m. B, CHO cell clones were labeled to approach steady state with 35 S-labeled amino acids, extracted, and either mockdigested (Ϫ) or digested with endoglycosidase H (ϩ), followed by immunoprecipitation with anti-Tg and analysis by SDS-PAGE and autoradiography. After digestion, essentially all intracellular Tg shifted down upon SDS-PAGE, indicating that it is contained in a pre-Golgi compartment. By contrast, Tg molecules that passed through the Golgi complex and were released into the medium bathing CHO-G cells (Med) or CHO-P cells (not shown), were resistant to endoglycosidase H digestion. right panel) a significant fraction of intracellularly accumulated Tg again appeared broadly in the first (horizontal) dimension, including high molecular mass material, suggesting the presence of multiple different Tg-containing complexes. Thus, the specific forms of Tg recovered after cross-linking (Fig. 8) were not fundamentally different from those detected in the absence of cross-linking (Fig. 7). Importantly in Fig. 8, however, the second dimensional reducing gel allowed for release of co-precipitated, Tg-associated proteins (described below), upon hydrolysis of the thiol-cleavable cross-linker.
In a single dimensional analysis, specific immunoprecipitation from CHO-G cells with a polyclonal anti-GRP94 (Fig. 8, lower left panel) directly demonstrates the position (arrow) of this band (as well as a band of molecular mass ϳ78 kDa, arrowhead). Interestingly, when examining two-dimensional gels from CHO-G cells (Fig. 8, lower right panel), a labeled band of the identical mobility (arrow) was specifically co-precipitated with Tg and could be observed to run directly beneath Tg in high molecular mass Tg complexes, suggesting the presence of GRP94 in these Tg complexes. Unequivocal proof of the identity of GRP94 in these complexes was obtained by denaturation of these complexes with SDS (and subsequent detergent dilution), followed by a secondary direct immunoprecipitation with anti-GRP94 (see below). In parallel experiments, the 78-kDa band, which also co-precipitates with Tg (Fig. 9, lane 1, marked by arrowhead) was found to co-migrate with BiP immunoprecipitated from the same cells without cross-linker (Fig. 9, lane 2), strongly suggesting the presence of BiP in Tg complexes. Finally, it must be emphasized that, in Fig. 8, the uncomplexed Tg molecules (recovered to the left of the dashed line in the two-dimensional gels of CHO-P and CHO-G cells) were associated with GRP94 and the 78-kDa band (likely to be BiP) to a much lesser degree than were the high molecular mass Tg complexes (lower right panel, right of dashed line).
These data clearly establish the formation of complexes be-tween recombinant nonmutant Tg and GRP94 in CHO-G cells. However, only in highly overexposed autoradiograms did the parental CHO-P sample reveal a small fraction of high molecular mass Tg in association with these bands, indicating that this particular (un)folded form of Tg was specifically accumulated as a consequence of increased abundance of GRP94 in the ER. Stoichiometry of Immunoprecipitable of Tg⅐GRP94 Complexes-If export of recombinant Tg from CHO cells requires escape from GRP94-mediated binding in the ER, then it would be predicted that the accumulation of Tg complexes in CHO-G cells would be accompanied by an increase in GRP94⅐Tg stoichiometry as a consequence of increased binding of GRP94. To test this prediction, cross-linked or uncross-linked cells that had been labeled with pure [ 35 S]methionine to approach steady state, were extracted under nonreducing conditions and then sequentially immunoprecipitated, first with anti-Tg, and then with anti-GRP94, before analysis by reducing SDS-PAGE. Using this protocol, GRP94 could be recovered only from crosslinked samples (Fig. 10).
Using the known numbers of methionine residues in both GRP94 and Tg, the molar ratio of the two proteins associated in these sequentially immunoprecipitated complexes was calculated. Importantly, this stoichiometry doubled (from 3.1 to 6.0) in cells in which GRP94 abundance was increased. These data strongly support the idea that increased GRP94 availability FIG. 10. Composition of Tg⅐GRP94 complexes. Tg-expressing CHO-P and CHO-G cells were metabolically labeled for 22 h with [ 35 S]methionine. Cells were either cross-linked with DSP or incubated with carrier and lysed under denaturing, nonreducing conditions. Cell lysates were immunoprecipitated first with anti-Tg and then eluted and reimmunoprecipitated with anti-GRP94. A phosphorimage is shown of the labeled Tg⅐GRP94 complexes analyzed by SDS-PAGE under reducing conditions. Note that Tg is very poor (Ͻ1% abundance) in methionine but contains 122 cysteine residues, while GRP94 has only 3 Cys residues but contains 18 methionines. Thus, unlike Fig. 8 in which most of the Tg signal is derived from labeling with [ 35 S]cysteine, the ratio of GRP94 to Tg in Tg⅐GRP94 complexes appears much higher after labeling with pure [ 35 S]methionine, which is used to precisely calculate a GRP94⅐Tg stoichiometry in the isolated complexes (see text).

FIG. 8. In CHO-G cells, disulfide-linked Tg complexes include bound GRP94.
After labeling with radioactive [ 35 S]cysteine and methionine as in Fig. 7, Tg-expressing CHO cells (right-sided panels) and control cells not expressing Tg (upper left panel) were exposed to the membrane-permeant cross-linker, DSP, as described under "Experimental Procedures." Tg immunoprecipitates were then analyzed by two-dimensional SDS-PAGE as in Fig. 7. In the second dimension, discrete bands of ϳ94 and ϳ78 kDa can be seen to underlie disulfidelinked Tg complexes. In the single dimension lane (bottom left panel), the migration of immunoprecipitable GRP94 by reducing 6.5% SDS-PAGE is shown, as well as a second band of 78 kDa, which could either represent a cross-reaction with BiP or recovery of protein complexes that include both proteins. Secondary immunodetection, described in Fig. 10, confirmed the identity of the 94-kDa protein as GRP94, while the 78-kDa protein co-migrates with authentic immunoprecipitable BiP from CHO-G cells (see Fig. 9).
FIG. 9. A 78-kDa band cross-linked to immunoprecipitable Tg co-migrates with authentic BiP. CHO-G cells, metabolically labeled as in Fig. 7, were lysed and immunoprecipitated with anti-BiP (lane 2). Alternatively, the cells were cross-linked and immunoprecipitated with anti-Tg (lane 1), in which bands of 94 kDa (identified as GRP94, see Fig.  10) and 78 kDa were co-precipitated as in Fig. 8. In both cases, the samples were then analyzed by reducing 5% SDS-PAGE and phosphorimaging.
leads to increased GRP94 interaction with Tg, a potential "substrate." DISCUSSION It has been known for some years that each secretory protein species exits the ER with its own distinct rate (45,46). The time spent preparing newly synthesized secretory proteins for export from the ER is thought to be comprised of two kinds of activities. First, individual secretory proteins enter the ER lumen in an unfolded state and spend a characteristic period of time converting their raw primary structure into a native or near-native conformation. Second, differences in rates of entry into anterograde transport vesicles may contribute to distinct exit rates for different secretory proteins, possibly as a consequence of their biophysical properties (47) or possibly based on presentation or association with putative cargo receptors (48 -51). Interaction with and dissociation from resident ER chaperones are likely to strongly influence both kinds of activities, and therefore are likely to be important factors influencing the ER exit rates of exportable proteins.
For nonmutant secretory proteins, ER chaperone associations are likely to be near-maximal during the earliest stages of folding when certain features, such as exposed hydrophobic patches, are not yet buried in the tertiary/quaternary structure. A positive role for chaperone binding in the exit pathway of exportable proteins has been described as largely circumstantial (see Introduction), but has been suggested based on pharmacological manipulations that influence either chaperone availability (14) or post-translationally modified chaperone binding sites (52), as well as by experiments expressing ER chaperones in certain model systems (53). Nevertheless, the hypothesis of ER chaperones as quality controllers (54,55) suggests the possibility that the extent of chaperone involvement may be an independent factor influencing the ER residence time of many normal secretory proteins.
It is now established that GRP94 molecules interact with newly synthesized thyroidal Tg transiently after normal Tg synthesis, and for a protracted period in thyrocytes of animals suffering from congenital goiter with deficient Tg (18). The thyrocytes of animals and humans with this illness exhibit the unfolded protein response (7), which causes total thyroidal levels of GRP94 to be elevated (31). Of course, in those thyrocytes, much of the GRP94 binds to accumulated mutant Tg, such that the level of free GRP94 in the ER, if it rises at all, almost certainly does not exhibit the same fold increase as the total elevation of this chaperone. By contrast, to examine the consequences of increased GRP94 binding in a system uncomplicated by instrinsic Tg defects, in this report we have examined the folding and export of nonmutant Tg after genetic manipulation to increase levels of free GRP94 in the ER of CHO cells. The results of this study indicate that increased availability of GRP94 causes major prolongation of Tg residence in the ER, and this results in a significant increase in the ER pool size of Tg in the steady state that is disproportionate to the rate of Tg synthesis (Figs. 4 -6).
How are these results to be interpreted in terms of mechanism of action of GRP94 with respect to Tg? The possibility that the observed phenotype occurs by meaningfully diminishing the levels of BiP or other chaperones with pro-folding activity, would appear to be excluded by our measurements establishing that intracellular levels of these ER resident proteins did not change (Fig. 2). 2 Instead, three new pieces of evidence, taken together, strongly support the notion that direct interaction with GRP94 mediates delayed Tg folding and exit from the ER. First, we report that disulfide-linked Tg complexes, previously found in thyrocytes at an early stage in the Tg folding pathway (28,29), are specifically accumulated in CHO-G cells (Fig. 7). More importantly, co-immunoprecipitation studies establish that this particular form of Tg is directly associated with GRP94 (Fig. 8). Finally, the stoichiometry of GRP94⅐Tg in these complexes is increased as a consequence of increased chaperone availability (Fig. 10). In this context, we conclude that the binding of GRP94 to Tg slows its advance through folding and ER export pathways.
There is reason to believe that GRP94 itself is largely retained in the ER by mechanisms sensitive to lumenal calcium levels (56). A natural consequence of this is the tendency of GRP94-associated proteins to be co-retained in the ER compartment. Second, persistent association of GRP94 is likely to interfere with delivery of secretory proteins to ER export vesicles which (by poorly understood mechanisms) exclude the entry of ER lumenal resident proteins (57,58).
Although the secretory phenotype consequent to manipulation of GRP94 expression has never been previously examined, such studies have been performed with BiP. Interestingly, it has been established that, in cells with diminished levels of BiP, increased secretion of mutant plasminogen activator is observed (59). Moreover, in cells with increased BiP levels, diminished secretion of factor VIII or von Willebrand factor has been observed (39,60). Importantly, selectively increased ER chaperone expression does not produce a generalized effect on anterograde protein traffic, as export is slowed only for those secretory proteins to which chaperone binding can be detected (39,61). In the present study, Tg, a protein shown to bind GRP94, exhibits similarly delayed export from cells with increased levels of GRP94. Thus the data suggest that enhanced binding of ER chaperones of the hsp90 class (GRP94), like that of the hsp70 class (BiP), can influence export dynamics by retarding the progression of suitable secretory protein "substrates" through folding and ER exit pathways.
These findings appear consistent with our view (14) that during conformational maturation of Tg (as well as other secretory proteins), ER chaperones go through cycles of dissociation-reassociation; we have proposed that this cyclical association is the very means by which ER chaperones ordinarily monitor the progress of secretory protein folding. At appropriate moments in the normal Tg folding pathway, different domains and subdomains within nascent Tg take advantage of the limited period of dissociation from GRP94, or BiP, to bury individual binding sites. During successful folding, chaperone rebinding decreases progressively to zero as conformational maturation proceeds to completion. Of course in mutant Tg molecules, structural defect(s) may drastically extend the period needed to bury certain hydrophobic patches, thus, chaperones perpetually rebind to these regions, resulting in sustained chaperone association and retention in the ER (18). Furthermore, based on the evidence presented herein, we propose that the permissible length of time for burying features characteristic of GRP94 binding is inversely related to the availability of this chaperone in the ER. Under conditions where chaperone availability is increased, chaperone "on" time increases and "off" time is diminished to the point where one or more Tg domains do not have enough time to complete the folding step. This leads to a bottleneck in the conformational maturation pathway (Fig. 7), an increased concentration of complexes at steady state (Figs. 8 -10), prolonged residence in the ER (Figs. 5 and 6) and an increase in ER pool size (Fig. 4).
There are certainly differences in the peptide binding specificity of BiP (62-64) from that of GRP94; this is suggested by sequential binding of these chaperones during immunoglobulin chain maturation (44). However, in the immediate post-translational period, the large Tg protein must fold 20 or more different domains (65). Thus, it is perhaps not surprising that, in thyrocytes, Tg may form complexes simultaneously with both GRP94 and BiP (26,27), while dissociation of these two ER chaperones from Tg also appears kinetically indistinguishable (18). Our present data in CHO cells suggest that high molecular mass complexes of recombinant Tg include both GRP94 and a 78-kDa protein (Fig. 8) likely to be BiP (Fig. 9); while sequential immunoprecipitation, first with anti-Tg, and then with anti-GRP94, still co-precipitates the 78-kDa protein, pointing to the likelihood of ternary complexes that include BiP (Fig. 10). Moreover, after steady state labeling of CHO cells genetically manipulated for selectively increased expression of BiP (39) rather than GRP94, recombinant Tg⅐GRP94 complexes analyzed precisely the same way as that shown in Figs. 8 and 10 appear identical except for a selective increase in the 78-kDa band. 3 Thus, our results suggest that GRP94 binds preferentially to non-native forms of Tg in the ER that are also suitable "substrates" for BiP. These findings are consistent with the possibility that GRP94 and BiP are two chaperones with nonidentical binding specificity that can work as a team in the binding of folding intermediates in the Tg export pathway.
In conclusion, we propose that, along with BiP (59), GRP94 binding can contribute to an ER retention function for Tg, as part of the ER quality control system that helps to prevent export of conformationally immature proteins.