Protein-disulfide Isomerase (PDI) in FRTL5 Cells

Thyroglobulin (TG) is secreted by the thyrocytes into the follicular lumen of the thyroid. After maturation and hormone formation, TG is endocytosed and delivered to lysosomes. Quality control mechanisms may occur during this bidirectional traffic since 1) several molecular chaperones are cosecreted with TG in vivo, and 2) lysosomal targeting of immature TG is thought to be prevented via the interaction, in acidic conditions, between the Ser789–Met1172 TG hormonogenic domain (BD) and an unidentified membrane receptor. We investigated the secretion and cell surface expression of PDI and other chaperones in the FRTL5 thyroid cell line, and then studied the characteristics of the interaction between TG and PDI. We demonstrated that PDI, but also other chaperones such as calnexin and KDEL-containing proteins are exposed at the cell surface. We observed on living cells or membrane preparations that PDI specifically binds TG in acidic conditions, and that only BD is involved in binding. Surface plasmon resonance analysis of TG/PDI interactions indicated: 1) that PDI bound TG but only in acidic conditions, and that it preferentially recognized immature molecules, and 2) BD is involved in binding even if cysteine-rich modules are deleted. The notion that PDI acts as an “escort” for immature TG in acidic post-endoplasmic reticulum compartments is discussed.

The functional unit of the thyroid gland is the follicle, which is composed of a monolayer of epithelial cells, the thyrocytes, surrounding a closed space, the follicular lumen. The thyroid prohormone, thyroglobulin (TG), 1 is synthesized by thyrocytes and then secreted into the follicular lumen in which the thyroid hormones themselves are synthesized. Hormonal secretion requires the endocytosis of mature TG (i.e. TG bearing hormone residues), its targeting to lysosomes, and the subsequent release of hormones. Each step of this bidirectional trafficking of TG is subject to regulation, in some cases mediated by molecular chaperones (1)(2)(3).
Thyrocytes synthesize large amounts of TG (up to 50% of protein synthesis in the gland; Ref. 4). Glycosylation and chaperone-assisted folding of TG monomers (M r 330,000) take place in the endoplasmic reticulum (ER), and involve several chaperones including calnexin, BiP, ERp72, grp94, and grp170 (5)(6)(7)(8)(9). The dimerization of monomers, a process thought to abolish the interactions between secreted molecules and chaperones, probably determines the sorting and targeting post-ER of glycoproteins (10). The transit through the Golgi apparatus of newly synthesized TG leads to the modification of some high mannose type N-glycans into N-acetyllactosamine chains. Secreted glycoproteins have an almost unique property, mediated by an unknown regulatory mechanism, in that some of these lactosamine chains of TG are incompletely processed and have accessible GlcNAc moieties (11,12). TG molecules are then targeted to small secretory vesicles and released via the regulated secretory pathway of thyrocytes (13,14).
In the follicular lumen, the prohormone undergoes several posttranslational modifications to produce thyroid hormones. These modifications include the iodination of tyrosyl residues, and the oxidative coupling of some of these residues to form triiodothyronine (T 3 ) and tetraiodothyronine (thyroxine or T 4 ) (1,2). They also include the de novo formation of disulfide bridges, the multimerization of some TG molecules (15), and the progressive completion of their incompletely processed lactosamine N-glycans by the addition of galactose and sialic acid sugar moieties (12).
As TG molecules in the follicular lumen are very heterogeneous in terms of iodine and hormone content, it has been assumed that there is a mechanism that either preferentially targets iodine-rich molecules to lysosomes or prevents the catabolism of iodine-poor molecules. It has recently been reported that megalin, a 330-kDa glycoprotein associated with the cell surface, participates, at least partly, in TG internalization in some some thyroid cell preparations (16,17). However, further studies are required to determine whether this process is selective for a given TG subpopulation, and no receptor specific for mature thyroglobulin has yet been shown to be responsible for the selective internalization of TG. The notion of an intrafollicular retention process for immature molecules arose from three main findings: 1) there exists, in vivo in the rat, a major recycling pathway from internalized TG back to the lumen (18); 2) during TG maturation, galactose and sialic acid are incorporated into some glycans (12), by a process known to occur in the trans compartments of the Golgi complex and the trans-Golgi network (19), and 3) immature TG preferentially binds a membrane receptor in acidic conditions, suggesting that there * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Recent reports have established that in secretory cells some chaperones are secreted or cell surface-associated. For example, calnexin as well as many other "ER resident proteins" are present at the surface of thymocytes (23), and grp94 is secreted by pancreatic cells (24,25). The best documented example is the protein-disulfide isomerase (PDI). PDI (M r 57,000) is an abundant soluble cellular protein of the secretory pathway (0.4% of total liver protein, Refs. 26 -28) found on the cell surface of numerous cell types such as B cells (29,30), platelets (31,32), pancreatic cells (33), and hepatocytes (34). PDI may be secreted despite possessing the KDEL retention signal (34,35). It remains fully active and may be involved in structural modification of (glyco)proteins in the extracellular medium. Indeed, PDI catalyzes thiol-disulfide interchanges that may result in the rearrangement of protein-disulfide bonds. This is why cellsurface PDI is involved in the reduction of the disulfide-linked diphtheria toxin heterodimer (36), cell surface events triggering the entry of the human immunodeficiency virus into lymphoid cells (37), shedding of the human thyrotropin receptor ectodomain (38), cell surface recognition during neuronal differentiation of the retina (39), and, more generally, controls the redox state of existing exofacial protein thiols or reactive disulfide bonds (40). PDI, and some other chaperones such as BiP and grp94, are secreted with TG in in vivo conditions into the follicular lumen (41,42), suggesting that these chaperones may be involved in the structural modification of the prohormone in hormone biogenesis.
We investigated the secretion and cell surface expression of PDI and other chaperones in the FRTL5 (thyroid) cell line, and then studied the characteristics of the interaction between TG or derived fusion proteins and PDI in FRTL5 cells. We report two key observations: 1) PDI is secreted and associated with membranes, which implies that it may be present in vivo in post-ER compartments; and 2) PDI interacts with the membrane-binding domain of TG in acidic compartments. The potential implications of these observations are discussed.

Cell Culture
FRTL5 cells (ATCC RL8305) were cultured in Coon's modified Ham's F-12 medium, pH 7.5, containing bicarbonate (2.67 g/liter) and 5% decomplemented calf serum. A mixture of nonessential amino acids (1 mM) was added, along with six hormones: TSH (10 milliunits/ml), trans-ferrin (5 g/ml), insulin (10 g/ml), somatostatin (10 ng/ml), glycyl-1histidyl-1-lysine (10 ng/ml), and hydrocortisone (3.7 ng/ml). Proteins were resolved by SDS-PAGE and transferred to nitrocellulose membrane using a semidry blotter and 180 mA applied for 30 min.  The blot was incubated for 1 h with Tris-buffered saline, 10 mM Tris, 0. 15 M NaCl, pH 7.5, containing 5% milk powder, and 0. 5% Tween 20 and was then washed twice with Tris-buffered saline containing 0. 5% Tween 20 (TBST). The blot was incubated with TBST containing 2% skim milk powder and primary antibodies (1/1000 dilution). The blot was washed four times with TBST, then incubated in 5 ml of TBST containing 2% skim milk powder and the secondary antibody (1/20,000 dilution). The blot was washed five times with TBST and incubated with a chemiluminescence substrate, according to the manufacturer's instructions. It was then placed against x-ray film for detection of the bound primary antibody. The blot was stripped by incubation with Immunopure elution buffer (Pierce) for 2 h at room temperature. The blot was washed, incubated with SuperSignal working solution (Pierce), and placed against x-ray film, to check that the probes had been adequately removed. The stripped blot was then washed four times and incubated with blocking reagent before reprobing.

Metabolic Labeling and Immunoprecipitation
Cells were washed with 20 mM sodium phosphate, 0.15 M NaCl, pH 7.4 (PBS), and incubated in cysteine-and methionine-free labeling mixture containing 10% dialyzed fetal calf serum, in the presence of [ 35 S]methionine/cysteine (200 Ci/ml, Expre 35 S 35 S labeling mix; NEN Life Science Products). The labeling medium was removed and replaced with Coon's modified Ham's F-12 medium, and the cells were incubated for various lengths of time (the chase period). The media were removed after the chase period, and, protease inhibitors were added (1 mM EDTA, 1 mM phenylmethanesulfonyl fluoride, and 10 g/ml each leupeptin and pepstatin) and the mixture centrifuged (1 h at 100,000 ϫ g). For treatment of the cells with colchicine or brefeldin A, cells were incubated in Coon's medium containing either 40 M colchicine or 20 M brefeldin A for 30 min before labeling and the chase period in medium containing the same concentrations of these drugs. The cells were washed four times with ice-cold PBS and harvested by scraping. They were then centrifuged at 1,500 revolutions/min for 10 min, suspended in 10 ml of ice-cold PBS and recentrifuged. The resulting cell pellets were suspended in 1 ml of lysis buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% Nonidet P-40). The cell lysates were centrifuged at 12,000 ϫ g for 10 min to remove cell debris. The resulting supernatants were cleared by incubation with 20 l of non-immune serum and 20 l of a 50% suspension of Protein A-Sepharose for 1 h at 4°C. The suspension was centrifuged to remove the Protein A-Sepharose, and the cleared cell lysates and media were incubated with 5 l of anti-PDI monoclonal antibody (RL90) and 12 l of anti-BiP polyclonal antibody or 12 l of anti-TG antiserum. Immune complexes were recovered by addition of 30 l of a 50% suspension of Protein A-Sepharose and incubation for 1 h at 4°C. The suspension was centrifuged at 12,000 ϫ g for 3 min, the supernatant discarded, and the pellet washed four times with lysis buffer and once with PBS. The immunoprecipitate was suspended in SDS-PAGE sample buffer containing 2% SDS with or without 1% ␤-mercaptoethanol and subjected to SDS-PAGE. Gels were dried by conventional methods, and radiolabeled bands were detected by phosphoimaging (Molecular Dynamics, Inc., Sunnyvale, CA). The leakage of proteins from injured cells was assayed by determining lactate dehydrogenase activity.

Immunodetection of Cell Surface PDI and Chaperones
FRTL5 cells were grown in Lab Tek I chambers, washed briefly in PBS, and fixed in 2% formaldehyde at room temperature for 15 min. The cells were then washed twice in PBS and incubated with blocking buffer (PBS containing 10% BSA) for 10 min at room temperature in a humidified chamber. The cells were mixed directly, without washing, with the primary antibody in 0.1% BSA in PBS, and the mixture was incubated for 1 h. The cells were then washed five times with 0.1% BSA in PBS, and were incubated for 1 h with the fluorochrome-conjugated secondary antibody in 0.1% BSA in PBS. The cells were thoroughly washed in PBS, mounted on slides in Mowiol, and viewed with a confocal microscope (Leitz DMIR BE, Leica TSC 4D, Heidelberg, Germany).

Quantitation of PDI Secretion
We determined the level of PDI secretion, by subjecting serially diluted (standardized based on protein content) samples of culture medium to slot blot analysis on nitrocellulose membranes. The signal obtained was compared with that given by known amounts of purified PDI. Blots were probed with anti-PDI antibody, treated for chemiluminescence, scanned, and analyzed with a PhosphorImager and the Im-ageQuant software package (Molecular Dynamics).

Radiolabeling
Sulfo-SHPP was iodinated according to the manufacturer's instructions. Briefly, 1 mCi of 125 I-Na was added to 10 l of a freshly prepared solution of sulfo-SHPP in dimethyl sulfoxide (0.1 mg/ml). The following were then added sequentially: 10 l of chloramine T (5 mg/ml), 100 l of hydoxyphenyl acetic acid (1 mg/ml), and 10 l of sodium metabisulfite (12 mg/ml). 125 I-Labeled sulfo-SHPP (200 Ci) was incubated in 25-cm 2 flasks with confluent FRTL5 cells, previously washed three times with ice-cold PBS, for 1 h each at 2-4°C. The labeling medium was removed, and the cells were washed with PBS at 4°C and recovered by scraping into ice-cold PBS. The cells were washed, and cell lysates were obtained for immunoprecipitation of PDI as described above.
PDI (10 g) was iodinated by incubation with IODOGEN and 125 I-Na in 50 l of 50 mM Tris-HCl buffer, pH 8.0, for 15 min at 4°C. The reaction was stopped by adding 1 ml of 10 mM phosphate buffer, pH 7.4, 150 mM NaCl, 20 mM KI. Iodinated molecules were isolated by gel filtration through PD10 columns and extensive dialysis. The same protocol was used for TG iodination, except that gel filtration was performed through Sephadex-200 columns (Amersham Pharmacia Biotech).

Construction of Human Thyroglobulin Fusion Peptides
RNA Isolation-Total RNA was extracted from the thyroid tissues of patients with Graves' disease, as described by Chomczynski and Sacchi (43). Briefly, tissue (10 mg) was crushed in 3 ml of RNABlue solution (Eurobio). Chloroform (300 l) was added, and the mixture was vigorously vortexed and centrifuged at 15,000 ϫ g for 30 min. The aqueous phase was transferred to a microcentrifuge tube containing an equal volume of isopropanol. RNA was recovered by precipitation overnight at Ϫ20°C and centrifugation at 12,000 ϫ g for 20 min. The pellet was washed with 75% ethanol and suspended in diethyl pyrocarbonatetreated water.
Amplification by Reverse Transcriptase-Polymerase Chain Reaction (RT-PCR)-The first strand cDNA was synthesized by incubation of 1-3 g of total RNA with Moloney murine leukemia virus reverse transcriptase (5 units, Life Technologies, Inc.) for 1 h at 42°C, in a volume of 30 l. The resulting first strand cDNA was then used for PCR amplification.
The membrane receptor binding domain of TG (Ser 789 -Met 1178 ) was amplified using the following primers: the forward primer was 5Ј-GGGATCCGCTGAGCAGGTCTTCGAGTTG-3Ј (2374 -2394) and the reverse primer, 5Ј-CTCTAGAGAGCACATTTCAGAGGCTTGG-3Ј (3494 -3474). The cysteine-rich repeat unit 9 (RU9, according to Ref. 44) that forms part of the binding domain, was amplified using the primers 5Ј-GGGATCCCTGCAGATTCCACAGGGCCCG-3Ј (3241-3261; forward) and 5Ј-CTCTAGAGAGCACATTTCAGAGGCTTGG-3Ј (3494 -3474; reverse). Each of these primers contained a unique restriction site, BamHI for the forward primers and XbaI for the reverse primers. The expected protein sequences of the TG-derived fusion proteins included Pro 759 -Leu 1163 for the membrane-binding domain of TG and Leu 1080 -Leu 1163 for the RU9 cysteine-rich motif.
The reaction mixture for PCR (50 l) contained 10 mM amounts of each dNTP, 100 ng of each primer, the buffer supplied by the manufacturer, and 1 unit of Vent polymerase (New England Biolabs). PCR amplification involved denaturation at 94°C for 1 min, primer annealing at 55.5°C (PCR 2) for 90 s, and extension for 90 s, with optimum results obtained using 34 cycles.
Cloning and Sequencing of PCR Products-PCR fragments were purified from agarose gel and cut with BamHI and XbaI endonucleases. They were then ligated into the BamHI/XbaI-cleaved PMal-CR1 expression vector (Biolabs) and used to transform Escherichia coli TB1. The TG fragment inserts of the plasmids of positive clones were checked by sequencing. The DNA was sequenced using M13 (Ϫ47) and reverse M13 (Ϫ48) primers (New England Biolabs). We used the cycle sequencing method and the dRhodamine Terminator Cycle Sequencing Kit (Applied Biosystems/Perkin-Elmer). Sequences were determined with an ABI Prism 310 and were aligned with the Tg cDNA sequence using AutoAssembler software (Applied Biosystems/Perkin-Elmer).
RT-PCR generated two cDNA fragments, one with the expected sequence, based on the wild-type mRNA (43,44), and the other with exons 13, 14 and 15 deleted. Exons 14 and 15 encode the cysteine-rich RU9 motif (His 1025 -Cys 1160 , Ref. 45). This finding is novel, and we therefore tried to determine the incidence of this deletion in various physiopathological conditions. The results of this investigation will be published elsewhere. 2 Three clones derived from the TG cDNAs were selected and prepared. One encoded the full-length binding domain (BD), the second, the alternatively spliced binding site (BD⌬RU9), and the third, RU9 (RU9).

Production and Purification of Fusion Proteins
Bacterial clones containing the various constructs were cultured overnight in 50 ml of LB medium at 37°C with shaking. An aliquot (50 ml) of the overnight culture was used to inoculate 1 liter of LB, which was then incubated at 37°C until the optical density at 600 nm of the culture reached 0.4 (i.e. 2 ϫ 10 8 cells/ml). Synthesis of the fusion proteins was induced by adding isopropyl-1-thio-␤-D-galactopyranoside to a final concentration of 0.1 mM and incubating at 32°C for 90 -120 min. The culture was then centrifuged at 1500 ϫ g for 10 min. The bacterial pellet was suspended in 50 ml of lysis buffer (10 mM phosphate buffer, pH 7.0, 30 mM NaCl, 0.25% Tween 20, 10 mM EDTA, 10 mM EGTA) and placed in a freezer at Ϫ20°C overnight. The lysate was thawed at 4°C, in the presence of 1 mM PMSF and subjected to sonication with 1-min pulses for 20 min. NaCl (final concentration 0.5 M) was added and the mixture centrifuged at 40,000 ϫ g for 30 min at 4°C.
The fusion proteins were purified on amylose resin, according to the manufacturer's instructions. The protein-containing fractions were dialyzed and concentrated (1 mg/ml final concentration) against phosphate-buffered saline (PBS) in a MicroProdicon apparatus (BioMolecular Dynamics). The level of purification of these fusion proteins was checked by SDS-PAGE and immunoblotting with anti-maltose-binding protein (MBP) and anti-TG antibodies.
Membrane Preparation and Binding Assay-FRTL5 plasma membranes were prepared as described previously (5,6). Solid-phase assays of TG and fusion peptide binding to FRTL5 membranes were performed according to established methods (21,22). Membranes were added (20 g of total protein in 40 l of PBS buffer) to the flat-bottomed wells of a microtiter plate (Falcon 3911, Becton Dickinson, Oxnard, CA) and were incubated for 1 h at 37°C and then overnight at 4°C. The wells were washed three times with 150 l of PBS and once with 200 l of PBS containing 2% bovine serum albumin. Thyroglobulin (5 ϫ 10 5 to 10 6 cpm in a binding buffer consisting of 25 mM acetate buffer, pH 5.0, 150 mM NaCl, 5 mM CaCl 2 , and 0.1% BSA) was added and incubated alone or in the presence of competitors at 4°C for 90 min. The wells were washed four times with the binding buffer, and radioactivity was counted. The results are means of at least three separate experiments performed in duplicate or triplicate.
Chaperones detected in membrane preparations were determined by immunoassay using microtiter plates coated with membranes (10 g/ well). The plates were coated by incubation for 2 h at 37°C. The plates were washed with PBS and incubated for 2 h with PBS containing 3% skim milk powder. They were then incubated with monoclonal antibodies directed against chaperones (1:500 dilution). Bound antibodies were detected by probing with horseradish peroxidase-linked anti-mouse IgG (1/1000 dilution), extensive wahshing and enzymatic reaction with OPD (Dako). Absorbance was read at 490 nm on an IEMS Reader MF (Labsystem).

Cell Surface Labeling of PDI and TG-FITC
Human or porcine TG was conjugated with fluorescein isothiocyanate (TG-FITC) as follows. FITC (50 l of a 1 mg/ml solution) was added to TG in 0.1 M carbonate buffer, pH 9.0 (20 mg in a 1.95-ml final volume), and the mixture was incubated in darkness for 1 h at room temperature. TG-FITC was recovered by gel filtration through a PD10 column (Amersham Pharmacia Biotech). It was dialyzed extensively, and aliquots (200 l, 10 Ϫ5 M final concentration) were stored at Ϫ20°C.
FRTL5 cells were grown in Lab Tek I chambers, washed briefly in PBS, and fixed in 2% formaldehyde at room temperature for 15 min. Cells were washed twice in PBS and incubated in blocking buffer (PBS containing 10% BSA) for 10 min at room temperature in a humidified chamber. For PDI or TG labeling, cells were mixed immediately, without washing, with primary antibodies in PBS containing 0.1% BSA, and incubated for 1 h. The cells were washed five times with PBS, 0.1% BSA, and incubated for 1 h with fluorochrome-conjugated secondary antibodies in PBS, 0.1% BSA. They were then washed extensively in PBS. For TG-FITC binding studies, cells were incubated with the flu-orescent probe (0.5 ϫ 10 Ϫ7 M) in 10 mM acetate buffer, pH 5.0, 0.15 M NaCl, for 1 h at 4°C. They were washed with acetate buffer and fixed in 2% formaldehyde. They were then mounted on slides in Mowiol and viewed with a confocal microscope as follows.
Cell samples were examined using a Leica TCS 4D invert confocal microscope with argon/krypton multilasers giving excitation lines at 568 and 488 nm, respectively. The data from the channels were collected simultaneously, using the narrow bandpass filter setting built into the instrument. Double labeling experiments usually employed a combination of FITC and Texas Red dyes (secondary antibodies) to eliminate channel overlap effects. Specimen were tested for overlap by turning off individual laser lines while continuously scanning the two channels. Data were collected with 8-fold averaging at a resolution of 512 ϫ 512 pixels using optical slices of between 0.5 m and 1 m. The microscope was a Leitz DMIR BE utilizing a 100ϫ oil immersion objective lens (numeric aperture 1.4). Data sets were processed using the Scanware software, then exported for preparation for printing using Photoshop. Data were projected onto a single plane, and the contrast was adjusted prior to export as TIFF files to Adobe Photoshop.

Surface Plasmon Resonance Analysis of the TG/PDI Interaction
A BIAcore biosensor system (BIAcore AB, Uppsala, Sweden) was used to study the kinetics of TG binding to immobilized PDI. All experiments were performed at 25°C. The sensor chip was washed with 10 mM Hepes, pH 7.5, 150 mM NaCl, 0.005% BIAcore surfactant P20 (buffer A) (BIAcore AB) between injections. Purified PDI (100 g/ml) was directly bound, via its amino groups, to the sensor surface (CM5 sensor chip), activated by N-hydroxysuccinimide (50 mM final concentration) and N-ethyl-NЈ-(dimethylaminopropyl)carbodiimide (200 mM final concentration) according to the manufacturer's instructions. The remaining N-hydroxysuccinimide groups were inactivated with 1 M (pH 8.5) ethanolamine. Acidic acetate buffers (50 mM, pH from 5.0 -6.0) were used for equilibration and binding analysis, and the surfaces were regenerated by washing with 10 mM NaOH, pH 11.8, and equilibration with acetate buffer or buffer A before reuse. The kinetics of binding to PDI of various TG samples or TG fragments were determined by injecting various concentrations of analyte (5-100 g) in either 10 mM acetate buffer, pH 5.0, or 10 mM Hepes, pH 7.4, at a flow rate of 20 l/min, and then at various flow rates to assess mass transport limitations. Dissociation was observed in running buffer without dissociating agents. Regeneration was obtained with 10 l of 50 mM NaOH. The kinetic parameters were determined using BIAevaluation 3.0 software (BIAcore AB).

Detection and Determination of the PDI Content of FRTL5
Cells-In Western blots of proteins isolated from CEM, BHK21, and FRTL5 cells, probed with the anti-PDI IgG raised against the whole molecule, a band of M r 57,000, and sometimes a faint band of M r 63,000, were detected. In some experiments, an additional band of M r 48,000 to 51,000 was also detected (Fig. 1A). The cellulose membrane was stripped and reprobed with a monoclonal anti-PDI antibody (clone RL90). This resulted in the detection of the M r 57,000 and M r 51,000 proteins bands only (Fig. 1B), demonstrating that: (i) the M r 63,000 band was not related to PDI and (ii) the additional M r 51,000 band was probably a product of PDI degradation. This was confirmed by the subsequent finding that the 51-kDa pro-tein was produced after incubation of a freshly prepared FRTL5 homogenate (Fig. 2, lane 1) for 30 min at 37°C or at room temperature (Fig. 2, lanes 2 and 3, respectively).
Densitometry showed that there was more PDI in FRTL5 cells than in the other cell types studied. The relative amount of PDI present was determined by subjecting serially diluted cell lysate samples to slot blot analysis, probing with monoclonal antibodies, and comparing the signal obtained with that obtained using known amounts of purified PDI. The PDI concentrations in CEM cells has been estimated to be 0.42% (0.33-0.56%, n ϭ 6), that of BHK21 cells, 0.78% (0.68 -0.89%, n ϭ 6) and that of FRTL5 cells to be 1.05% (0.85-1.25%, n ϭ 6) of total protein in the homogenates.
PDI Is Secreted by FRTL5 Cells-We metabolically radiolabeled FRTL5 cells and recovered PDI by immunoprecipitation of cell lysates and media. Cellular PDI was recovered as two major protein bands corresponding to the monomer (57-kDa monomer) and its principal degradation product (51 kDa) (Fig.  3, left). PDI was detected in the medium after a 1-h chase period, and the amount present in the medium increased between 1 and 2 h (Fig. 3, right).
It is also possible to isolate PDI from a culture medium in which the cells have been radiolabeled at equilibrium for 24 h. Apart from the major 57-kDa protein, a 330-kDa protein was also precipitated (Fig. 4, lane 1). This protein was identified as the thyroglobulin monomer (Fig. 4, lanes 5 and 6) and is probably present as a contaminant (Fig. 4, lane 7). Indeed, the same pattern was observed, with similar amounts of PDI recovered, in non-reducing conditions (Fig. 4, lane 2), showing that most of the PDI and TG molecules were not covalently bound. This is further supported by the observation that PDI was not detected in immunoprecipitates of TG, with or without ␤-mercaptoethanol (Fig. 4, lanes 5 and 6). We also isolated BiP from the culture medium. Similar fractions were isolated in the presence and absence of ATP, throughout the immunoprecipitation experiments. This excludes the possibility of ATP-dependent binding between the secreted forms of BiP and TG (Fig. 4, lanes  3 and 4). It is interesting to note that BiP was recovered using a polyclonal antibody directed against the eight amino acid residues at the extreme C terminus of BiP, including the KDEL sequence.
PDI was probably secreted by FRTL5 cells rather than leaked from injured cells, because only 0.8% after 24 h, and 1.8% after 48 h, of total lactate dehydrogenase was released from FRTL5 cells, whereas densitometry showed that 7.5% of total PDI was present in the medium (i.e. the medium ϩ cell extracts) after 24 h. To confirm that PDI recovered from the media was secreted and did not result from cell rupture, we treated cells with either brefeldin A or colchicine prior to and during the labeling and chase periods. Densitometric analysis of fluorographs of PDI immunoprecipitates showed that the amount of newly synthesized PDI in cells was slightly increased after drug treatment (control: 0.048 Ϯ 0.017; brefeldin A: 0.055 Ϯ 0.018; colchicine: 0.054 Ϯ 0.016; n ϭ 3), whereas there was much less PDI in the media, with a reduction of 75-95% (control: 0.128 Ϯ 0.016; brefeldin A: 0.022 Ϯ 0.008; colchicine: 0.032 Ϯ 0.006; n ϭ 3).
PDI Is Present at the Surface of FRTL5 Cells-PDI was recovered by immunoprecipitation following the radioiodination of cell surface proteins. Similar amounts of PDI were recovered in the absence (Fig. 5, lane 2) and presence (Fig. 5,  lane 3) of ␤-mercaptoethanol. This rules out disulfide bridge formation as the mechanism of PDI cell association at the cell surface. If the anti-KDEL antibody was used after the anti-PDI antibody, six proteins were immunoprecipitated from the labeled plasma membrane fraction (Fig. 5, lane 4) as reported previously in a study of exocrine pancreatic cells (33). These proteins, thought to include grp94 and BiP (33), are probably chaperones containing the KDEL sequence.
PDI was readily detected on the cell surface of FRTL5 cells by confocal microscopy (Fig. 6a). All cells were labeled. The labeling was punctate and covered the entire cell surface, although it was more dense in some regions than in others, suggesting the formation of aggregates or clusters. Calnexin was also detected but the labeling pattern differed between cells. The most strongly labeled cells also showed punctate labeling (Fig. 6b). In contrast, despite numerous studies on different cell preparations, we failed to detect specific labeling of BiP at the cell surface (Fig. 6c). A possible explanation is that, whereas the subcellular distribution of BiP did not markedly differ from that of PDI (Fig. 6d) and calnexin (Fig. 6e) (all three chaperones presented the same punctate labeling pattern in vesicular structures located throughout the cytoplasm) the level of fluorescence staining was less intense for BiP (Fig. 6f). We concluded that, if present at the cell surface, BiP was not readily detectable using this approach, and we did not further investigate the colocalization of various chaperones by confocal microscopy.
As stated in the Introduction, TG is thought to interact with various molecules, either constituents of the plasma membrane (megalin, interaction at neutral pH) or molecules thought to pass via the surface of the cell during their intracellular transport (the putative TG receptor, interaction in acidic conditions). The cell membrane chaperones associate with misfolded (3, 5-9) and denatured TG (47). This raises the question as to whether these chaperones also associate with particular subpopulations of TG molecules in preparations that we use. To clarify this issue, we fist tried to identify the locations of the putative receptors for TG and PDI at the cell surface (Fig. 7). In our conditions, endogenously expressed rat TG was not detected on the surface of FRTL5 cells, regardless of the type of anti-TG antibody used (monoclonal or polyclonal, data not shown). As deduced from biochemical studies of the TG binding characteristics on plasma membrane preparations (21,22), cell surface-associated TG could only be detected after addition and incubation of exogenous TG in acidic conditions at 4°C (data not shown). In the present study, we have used fluorescein isothiocyanate-conjugated TG to visualize directly the binding sites of TG on the cell surface. This binding was as specific as that observed on membrane preparations, since in control experiments we noted that it could be specifically prevented by preincubation with non-derivatized TG (10 -5 M concentration, data not shown). Fig. 7 showed that PDI was readily detected on the surface of these FRTL5 control cells (Fig. 7a). TG-FITC did not bind to cells at neutral pH (Fig. 7b), but binding was abundant and clearly detectable in acidic conditions (Fig. 7c).

FIG. 3. PDI synthesis and secretion by FRTL5 cells. Cells were radiolabeled by incubation with [ 35 S]cysteine and [ 35 S]methionine for
1 h and subjected to chase periods of various lengths, as indicated. PDI (57-kDa monomer, presumed degradation products at 51 kDa) recovered by immunoprecipitation from cell lysates (panel A) and media (panel B), was resolved by SDS-PAGE (10% acrylamide gel) and detected by fluorography as described under "Materials and Methods." Chase periods were: 0 min (lanes 1 and 6), 15 min (lanes 2 and 7), 30 min (lanes 3 and 8), 60 min (lanes 4 and 9) and 120 min (lanes 5 and 10). The detection in the same cellular preparation of TG bound to cells in acidic conditions (Fig. 7d) and of PDI (Fig. 7e) demonstrated that this receptor and PDI were located in the same plasma membrane domains (Fig. 7f). Thus, PDI is either located in the vicinity of the receptor or is itself the receptor that binds TG in acidic condition.
PDI Is Involved in the Binding of TG to FRTL5 Membranes-We have reported that the membrane receptor that binds TG in acidic conditions interacts with the various species of TG and preferentially binds molecules with a low iodine content (21). The domain of TG responsible for the binding of the molecule to membranes is located within the Ser 789 -Met 1172 sequence (22). This domain contains two N-linked glycan moieties and a cysteine-rich motif encoded by exons 14 and 15. The cysteine-rich motif is not involved in the interaction (22). The glycan moieties are involved in the interaction because their enzymatic cleavage reduces binding affinity by an order of magnitude (21). Bearing these previous findings in mind, we assessed whether TG fusion proteins encoding all or part of the binding domain were fully active, and investigated whether PDI was involved in the recognition of this binding domain.
The binding activities of TG fusion proteins at pH 5.0 were analyzed using 125 I-TG, TG and TG fusion proteins as competitors, and microwell plates coated with FRTL5 plasma mem-branes (Fig. 8). Unlike the control proteins, MBP and MBPparamyosin, the TG-fusion protein containing the binding domain (Pro 759 -Leu 1163 fragment, BD in Fig. 8) inhibited the interaction between 125 I-TG and membranes. However, as expected for a non-glycosylated binding region (21), the binding affinity of this peptide was lower than that of TG itself (K 0.5 ϭ 0.2 ϫ 10 Ϫ8 M and 0.3 ϫ 10 Ϫ7 M). The peptide fragment lacking RU9, due to the alternative splicing of exons 14 and 15 (BD⌬RU9), had the same degree of inhibitory activity as BD. The lack of involvement of RU9 in binding was also demonstrated by its lack of inhibitory activity, similar to that observed for the nonspecific binding of MBP to membranes (Fig.  8). We thus concluded that the binding domain encoded by TG fusion proteins had the same binding properties as those encoded by native TG molecules.
We investigated whether PDI, alone or with other chaperones, was directly involved in this binding. We first identified the chaperones that were present in our preparations of plasma membranes. We found, as observed previously (Fig. 5, lane 4;  Fig. 6b) that in addition to PDI, other chaperones such as calnexin, grp94, and BiP were present in these preparations (Fig. 9, left). Monoclonal antibodies directed against PDI inhib- Polyclonal anti-PDI antibodies detected PDI (a) on the surface of fixed cells. As described under "Materials and Methods," the incubation of TG-FITC with living cells at 4°C in either PBS or acetate buffer at pH 5.0 indicated that TG did not bind to membranes at neutral pH (b) but did bind to membranes in acidic conditions (c). On the same cell preparation, PDI was detected using the anti-PDI monoclonal antibody (d), the TG binding sites were detected using TG-FITC in acidic binding conditions (e), and both were co-localized at the surface of the cells (f) and analyzed as described under "Materials and Methods." ited TG binding to these membrane preparations, confirming that PDI is either in the vicinity of the receptor or is directly involved in TG binding to membranes (Fig. 9, right). In contrast, monoclonal antibodies directed against grp94, BiP, and calnexin had no effect on the binding of TG (Fig. 9, right), demonstrating that none of these chaperones were involved in binding. TG binding was strongly inhibited (80%) at a concentration of 1 g/ml, by a monoclonal antibody with an epitope close to one of the thioredoxin sites (monoclonal antibody RL90) and was inhibited by about 30% by a monoclonal antibody directed against the C-terminal domain of PDI (Fig. 9, right). The same pattern of selective inhibition by the anti-PDI antibody was observed for two TG-derived fusion proteins, MBP BD and MBP BD⌬RU9 (Fig. 9, right), demonstrating that they interact with the same binding site.
We then used a biochemical approach to demonstrate that PDI is directly involved in the binding of TG to the membrane (Fig. 10). The surface proteins of FRTL5 cells were radiolabeled with 125 I. The cells were lysed, and the cell lysates were incubated, in various conditions, with TG adsorbed onto agarose beads. If the agarose beads were incubated and washed in neutral pH conditions, the material associated with the beads gave four major bands in SDS-PAGE. Three of these bands had molecular masses below 26.6 kDa. The fourth had a molecular mass of about 57 kDa (Fig. 10, lane 1), identical to that of the 125 I-labeled PDI used as a standard (Fig. 10, lane 5). If the agarose beads were incubated with cell lysate in acidic conditions (pH 5.0), even with an aliquot one-fifth the volume of that used in neutral conditions, a larger amount of the 57-kDa protein was recovered and the low molecular weight contaminating bands were not detected (Fig. 10, lane 2). The 57-kDa band was not recovered if the lysate was cleared with an anti-PDI monoclonal antibody prior to the experiment (Fig. 10,  lane 3). Therefore, PDI was the only radioiodinated cell membrane protein that bound TG in acidic conditions. Surface Plasmon Resonance Analysis of the TG:PDI Interaction-Finally, we investigated whether PDI was itself sufficient for both TG binding and the selection of immature mole- Left, membranes (10 g) were used to coat multiwell microtiter plates, and chaperones were detected by probing with antibodies and revealed, the primary antibodies then being detected by enzymatic reaction using horseradish peroxidase-linked anti-mouse IgGs as described under "Materials and Methods." A monoclonal anti-␤-galactosidase was used as control. Right, membranes (20 g) were coated on multiwell microtiter plates and the binding of 125 I-labeled human TG or fusion proteins was challenged using anti-grp94, BiP, calnexin, and PDI monoclonal antibodies (1 g/ml). The results shown are those of three experiments performed in duplicate. cules, via binding to the binding domain. Real time studies of TG binding to immobilized PDI were recorded on a biosensor using PDI as a ligand and various sources of TG as analytes. No binding occurred at neutral pH, but binding did occur in acidic conditions (data not shown). We decided to analyze the interactions at pH 5.0 to enable us to compare the results to those obtained previously. We found that TG from pigs and humans bound PDI (Fig. 11). Similar amounts of mature human and porcine prohormone were bound by PDI (Fig. 11), but the binding affinity was higher for human TG (K D ϭ 1.6 ϫ 10 Ϫ9 M and 2.1 ϫ 10 Ϫ8 M, respectively, see Table I). About three times as much iodine-poor TG (iTG, iodine content ϭ 0.1%) as normal TG (TG, iodine content ϭ 1%) bound to PDI (Fig. 11). The binding affinity for iodine-poor TG was also higher than that for normal TG (K D ϭ 1.0 ϫ 10 Ϫ9 M and 1.6 ϫ 10 Ϫ9 M, respectively, Table I). Thus, these results are consistent with previous finding for membrane preparations indicating: 1) that TG molecules are heterogeneous and only a subset is capable of binding, and 2) that modifications clearly correlated with maturation affect the binding activity. We found that fusion proteins encoding the binding domain also bound to PDI. BD bound to PDI with an affinity one-tenth that of native TGs (K D ϭ 3.8 ϫ 10 Ϫ8 ), as expected for the non glycosylated binding domain ( Fig. 8; see also Fig. 4 in Ref. 21). Deletion of the type I cysteine-rich module did not reduce this binding (it even seemed, for still unknown reasons, to increase binding by a factor of 3: K D ϭ 1.1 ϫ 10 Ϫ8 ). No surface plasmon resonance signal was detected with either RU9 or MBP (data not shown). DISCUSSION We report the following novel findings. 1) PDI is secreted by and associated with the surface of FRTL5 thyroid cells.
2) The binding of TG to FRTL5 cell membranes in acidic conditions is selectively inhibited by anti-PDI antibodies. These antibodies also inhibit the binding of fusion proteins encoding the Ser 789 -Met 1178 membrane receptor binding domain. 3) Surface plasmon resonance analysis of TG/PDI interaction characteristics indicated: (i) that PDI binds TG, but only in acidic conditions, and that it preferentially recognizes immature molecules, and (ii) that the Ser 789 -Met 1178 domain is involved in binding, even if the cysteine-rich module is deleted. These results, and the potential function of PDI as a binder of secreted TG molecules in acidic post-ER compartments, are of particular importance.
FRTL5 thyroid cells are an additional secretory cell type in which some molecular chaperones can escape the ER compartment, even if they possess the C-terminal KDEL retention signal (33,34). The mechanism of secretion by the various cell types is unknown. Secretion may be due to the high level of chaperones in the ER of these cells, or may be an overflow of the KDEL or other ER retention mechanisms (48,49). However, the specific overproduction of PDI in CHO cells (50) or HT 1080 fibrosarcoma cells (40) increases the secretion of PDI but not other resident endoplasmic reticulum peroteins containing the KDEL-sequence, such as grp78 and grp94, suggesting that "PDI secretion and cell-localization is a specific event" (40), probably related to a specific function.
PDI has been implicated in the reduction of proteins present at, or interacting with, the cell surface, including thrombospondin (51), plasmin (52), diphtheria toxin (36), and the human immunodeficiency virus type 1 envelope glycoprotein (37). More recently, it has been reported that secreted PDI manipulates the redox state of exofacial extracellular protein thiols/disulfides, thereby controlling the function of extracellular proteins (40). It is therefore possible that in vivo PDI is involved in the structural modification of TG, particularly the formation of disulfide bridges between and within chains, that takes place in the follicular lumen during the synthesis of thyroid hormones (15).
We quantified PDI secretion by FRTL5 cells and found that it was as high as 15.01 Ϯ 3.71 g of PDI/5 ml of culture medium per 48 h (range: 18.7-13.1, n ϭ 4) for semiconfluent cells. This estimation was made in a cell line known to produce large amount of secreted molecules, at least in the case of TG. Further studies are therefore required to compare the secretion of PDI in thyroid cells, primary cultures, and other cell lines. 2 Nonetheless, the quantity outside the ER appears to be far from negligible, and PDI may therefore be involved in TG metabolism in the follicular lumen in vivo. Studies in vitro have shown that PDI has two distinctive features independent of its function as a catalyst; at high concentration, it acts as a chaperone, inhibiting aggregation, but at low concentration, it facilitates aggregation (anti-chaperone behavior) (28,53). One example of possible anti-chaperone behavior by PDI is the small quantities of PDI molecules that have been found associated with highly cross-linked aggregates in the follicular lumen (41). These aggregates are 50 -500 m in size, may reach concentrations of 590 mg/ml, and consist of TG molecules with high iodine but no hormone (41). The formation of these aggregates, which are also known as "colloid globules" (41), therefore prevents the homing of misfolded and hormone-poor prohormones to the endocytotic secretory pathway of thyrocytes.
However, our results suggested a novel function for PDI, different from those previously described. First, the interaction between TG and PDI occurs in acidic conditions. For reasons explained elsewhere (20,21), we decided to analyze this interaction at pH 5.0. However, it also occurs to a significant extent at pH 6.0, and at pH 5.75 fixation was about 30% the maximum (20,21), which suggests that it may take place in vivo, in any of the acidic compartments of the cellular exocytosis (54 -56) and endocytosis (57-59) pathways. Second, in contrast to the situ-  ation for TG misfolded during synthesis (3,(5)(6)(7)(8)(9) and for denatured TG (47), this interaction involves only one domain of TG and takes place between TG and PDI only, to the exclusion of other chaperones such as grp94 and BiP. Third, the catalytic function of PDI is not involved because carboxymethylated TG molecules bind to membranes (Fig. 4 in Ref. 21), and cysteinerich modules are not involved in the binding (Ref. 22 and this work). Thus, this TG binding property of PDI is activated in acidic post-ER compartments such as those of the exocytotic and endocytotic pathways. There is morphological and functional evidence for the existence of a regulated secretion for TG in thyrocytes (see Ref. 3 for a recent review). It is generally accepted that biogenesis of the vesicles of the regulated secretion pathway begins in the trans compartment of the Golgi apparatus and continues via a gradual process of enrichment and condensation of products destined for secretion (60). The vesicles become more acidic as they cross the Golgi apparatus, and this is known to play an essential role in the secretion pathway (60). It is probable that PDI and TG interact in these compartments. Indirect evidence for this is provided by the presence in the TG-binding domain of N-linked glycans with accessible GlcNAC residues, which may be involved in PDI recognition (Refs. 21 and 22; this report). It may be that the addition of galactose and sialic acid, which is known to occur in the trans-Golgi network (19), is prevented by competition between PDI and glycosyl transferases and steric hindrance on the "receptor-binding domain" of TG. If this were the case: 1) all or some of the secreted PDI would be released into the regulated secretory pathway, and 2) interaction between PDI in this acidic compartment would probably be involved in the efflux of this chaperone from FRTL5 cells.
As previously speculated for the membrane receptor of TG that functions in acidic conditions, the interaction between PDI and TG may occur after their internalization, in the endosomes or prelysosomes. The receptor responsible for the interaction has yet to be identified because attempts at its molecular cloning have not been successful (61,62). PDI possesses several of the biochemical characteristics previously reported for this receptor. It is an acidic non-glycosylated protein (63), abundant in the intracellular compartments (63). The molecular mass of the protein initially isolated by large scale purification from membrane preparations (51 kDa; Ref. 63) corresponds to that of the major degradation product of PDI in thyroid FRTL5 cells (this study). In addition, binding experiments with the biosensor have shown that, like the membrane receptor, PDI recognizes ovomucoid (data not shown), a GlcNAc-bearing glycoprotein that competes with TG for the binding site on the membrane (20). As we exhausted our supply of the antibody directed against the receptor (61), we were unable to test the antigenicity of the receptor (NAGR1, Ref. 63) and PDI and were therefore unable to draw the conclusions about the identity of the two molecules. However, we demonstrate here that PDI is a membrane-associated protein capable of binding preferentially to immature prohormones in acidic conditions as it recognizes the Ser 789 -Met 1172 domain of TG. Thus, membraneassociated PDI is a prime candidate for involvement in the quality control thought to be responsible for the recycling of immature TG from the endosomal compartment back to the lumen.
In conclusion, PDI may bind to certain form of immature TG in the exocytosis and endocytosis pathways of thyrocytes. It is possible that PDI acts as an "escort," preventing structural modification of the binding domain of TG during its cellular transit, and that it coordinates the modifications of TG associated with hormonogenesis. There are several lines of evidence for this. 1) The binding domain of Tg contains two clusters of tyrosine residues involved in iodination and hormone formation (22); 2) the binding domain is located in the N-terminal region of TG, which is highly cross-linked via 11 repeated cysteine-rich motifs and undergoes profound structural changes during its maturation in the follicular lumen (46); and 3) iodination and hormonogenesis are thought to occur after excocytosis, at the cell surface, when TG comes into contact with thyroperoxidase. We are investigating this possibility further in two ways. First, as PDI is not an integral membrane protein, we are trying to identify the partners responsible for its association with the membrane and to determine whether and how PDI is transported with TG in the secretory and endocytotic pathway. Second, we are trying to identify the determinants of the binding domain that acts as sensors, that promote or prevent the interaction with TG during its intracellular transit in post-ER compartments.