The Serum Protein α2-HS Glycoprotein/Fetuin Inhibits Apatite Formationin Vitroand in Mineralizing Calvaria Cells

We present data suggesting a function of α2-HS glycoproteins/fetuins in serum and in mineralization, namely interference with calcium salt precipitation. Fetuins occur in high serum concentration during fetal life. They accumulate in bones and teeth as a major fraction of noncollagenous bone proteins. The expression pattern in fetal mice confirms that fetuin is predominantly made in the liver and is accumulated in the mineralized matrix of bones. We arrived at a hypothesis on the molecular basis of fetuin function in bones using primary rat calvaria osteoblast cultures and salt precipitation assays. Our results indicate that fetuins inhibit apatite formation both in cell culture and in the test tube. This inhibitory effect is mediated by acidic amino acids clustering in cystatin-like domain D1. Fetuins account for roughly half of the capacity of serum to inhibit salt precipitation. We propose that fetuins inhibit phase separation in serum and modulate apatite formation during mineralization.

We present data suggesting a function of ␣ 2 -HS glycoproteins/fetuins in serum and in mineralization, namely interference with calcium salt precipitation. Fetuins occur in high serum concentration during fetal life. They accumulate in bones and teeth as a major fraction of noncollagenous bone proteins. The expression pattern in fetal mice confirms that fetuin is predominantly made in the liver and is accumulated in the mineralized matrix of bones. We arrived at a hypothesis on the molecular basis of fetuin function in bones using primary rat calvaria osteoblast cultures and salt precipitation assays. Our results indicate that fetuins inhibit apatite formation both in cell culture and in the test tube. This inhibitory effect is mediated by acidic amino acids clustering in cystatin-like domain D1. Fetuins account for roughly half of the capacity of serum to inhibit salt precipitation. We propose that fetuins inhibit phase separation in serum and modulate apatite formation during mineralization.
Considerable progress has been made toward a molecular understanding of skeletal development through the identification of local factors that control skeletal patterning and bone shape (1). Stages beyond the condensation of mesothelial cells and the developing chondrogenic phenotype are envisaged as a chain of timed events that lead to a progressive development of the osteoblast phenotype (2). Following that is the assembly of collagen, noncollagenous proteins, and mineral into functional bone tissue with acidic proteins controlling crystal growth at the interphase of the mineral and organic components (3). Most protein components of this assembly are expressed bone tissuespecific. However, serum proteins made in the liver are also known to contribute to the organic phase of bone and dentine (4). The function of serum proteins in biomineralization is poorly understood despite the fact that cell culture systems mimicking bone formation generally include serum.
One particular family of serum proteins, synonymously called ␣ 2 -HS 1 glycoproteins or fetuins, are concentrated in the mineral phase of bone (5) and teeth (6). They are structurally well studied. To date complete primary structures are published for human (7), bovine (8), sheep, pig (9), rat (10), and mouse (11) fetuins and the intron-exon structure has been published for the rat gene (12). From N terminus to C terminus two conserved cystatin domains functionally silent in human fetuin (13) are followed by a third, unrelated and more divergent domain rich in proline. Due to the cystatin domains fetuins are grouped within the cystatin superfamily of proteins (14). Fetuins are both N-and O-glycosylated (15,16) and Serphosphorylated (17). Human plasma ␣ 2 -HS is proteolytically processed (17).
Throughout all species studied high amounts of fetuins consistently occur in mineralized bone (5, 18 -21). Based on this observation a role for fetuins in bone formation and resorption has been suggested (22). Fetuins are abundant in fetal blood and tissues indicating that they may play a more general role during organ development. More specific functions of fetuin like lipid transport or inhibition of insulin receptor tyrosine kinase have been suggested; they are critically reviewed in a recent monograph (23).
Fetuins bind strongly to apatite and can thus be selectively enriched from serum (18). Bovine fetuin has several low affinity calcium binding sites (24) that might mediate apatite binding. However, the molecular basis of apatite binding and the significance of fetuin accumulation in bone have not been elucidated. In this paper we demonstrate that fetuins inhibit apatite precipitation. We propose a biological function for fetuins in the maintenance of high blood calcium levels and in the inhibition of unwanted mineralization.
Recombinant Expression and Purification of Fetuins-The complete cDNAs encoding human and mouse fetuins, respectively, were subcloned into the baculovirus transfer vector pVL1393 (Pharmingen) as EcoRI fragments. Rat fetuin was subcloned from pBluescript vector as an EcoRV/NotI fragment into pVL1393 digested with SmaI and NotI. Sf9 cells were co-transfected with pVL1393 vector and Baculogold TM DNA (Pharmingen). Virus expressing fetuins were cloned by limited dilution of virus mixtures. Recombinant fetuin proteins were purified from protein-free medium (Sf900II, Life Technologies, Inc.) of virusinfected Sf9 cells by a four-step procedure. Protein was concentrated by precipitation with 2 M (NH 4 ) 2 SO 4 . The precipitate was dissolved in buffer (50 mM NaH 2 PO 4 , pH 7, 0.75 M (NH 4 ) 2 SO 4 ), and applied to a butyl-Sepharose (Pharmacia Biotech Inc.) column. Bound protein was eluted at a flow of 2 ml/min with a 30-min linear gradient of 0.75 M ammonium sulfate, 50 mM NaH 2 PO 4 , pH 7, to 50 mM NaH 2 PO 4 , pH 7. Fractions containing recombinant protein (as determined by ELISA) were pooled and concentrated by precipitation with ammonium sulfate. The pellets were redissolved in 50 mM Tris-HCl, pH 7.4, and fractionated on a Superose 200 (Pharmacia) column. The purified fetuins migrated as single bands upon SDS-PAGE and Coomassie staining. Purification of recombinant rat and mouse fetuin was monitored by ELISA and Western blot analysis using antibody against rat serum fetuin (Dr. Nawratil, Munich).
Bacterial Expression of Mouse Fetuin Deletion Mutants-C-terminally truncated variants of mouse fetuin were expressed in Escherichia coli (XL1, Stratagene) as fusion proteins of the maltose-binding protein (MBP) using the expression vector pMal TM -c2 (New England Biolabs). The mouse fetuin cDNA was shortened from the 3Ј-end by digestion with exonuclease III. The expression vector pMal TM -c2 was modified by inserting into XbaI/PstI-digested vector an oligonucleotide hybrid carrying a KpnI restriction site and three TAA stop codons. The hybrid forming oligonucleotides used were 5Ј-CTAGAGGTACCTAACTAAC-TAACTGCA-3Ј and 5Ј-GTTAGTTAGTTAGGTACCT-3Ј. The modified pMal TM -c2 vector was digested with BamHI and KpnI, and a mouse fetuin cDNA fragment was inserted. This fragment was generated by PCR from the complete cDNA (11) subcloned into the vector pBluescript. The primers used were the M13-20 primer as the 3Ј-primer and the oligonucleotide 5Ј-GGCTGCGGATCCGCTCCACAAGGT-3Ј as the 5Ј-primer containing a BamHI restriction site followed by the sequence encoding the first amino acids of the processed mouse fetuin protein (without signal sequence). The PCR amplicon was digested with BamHI and KpnI and ligated into the modified pMal TM -c2 vector. For deletion mutagenesis the plasmid was digested with KpnI (3Ј-overhang) and XhoI (5Ј-overhang), and the mouse fetuin cDNA was truncated from the 3Ј-end with exonuclease III following established protocols (25). Deletion mutants were analyzed by DNA sequencing.
Bacteria were lysed in buffer (20 mM Tris, pH 7.4, 200 mM NaCl, 1 mM EDTA, 10 g/ml benzamidine, 1 mM phenylmethylsulfonyl fluoride, 1 mM DTT). DNase and RNase (final concentration 10 g/ml) were included in the lysis buffer, as contaminating E. coli DNA inhibited apatite formation. MBP fusion proteins were purified by affinity chromatography using amylose resin (New England Biolabs). MBP expressed from the modified vector pMal TM -c2 (lacking the LacZ␣ sequences fused to MBP) and MBP fusion proteins with deleted mouse fetuin proteins smaller than 100 amino acids for reasons unknown bound to amylose affinity matrix very inefficiently. We therefore purified the deleted MBP-mouse fetuin fusion proteins by a three-step procedure. After cell lysis and centrifugation protein was precipitated with 3.2 M ammonium sulfate, dissolved in buffer (20 mM Tris, pH 8, 1 mM DTT, RNase and DNase, 10 g/ml), and fractionated on a Superose 200-gel filtration column equilibrated in buffer (20 mM Tris, pH 8, 6 M urea). DTT (final concentration 100 mM) was added to the eluant and reduction proceeded for 2 h at room temperature. Reoxidation proceeded overnight at room temperature and was achieved by diluting the proteins to a final concentration of less than 50 g/ml (about 150-fold dilution) into renaturation buffer (50 mM Tris, pH 8, including 2 mM reduced, 0.2 mM oxidized glutathione, and 0.01% NaN 3 ). The solution was applied onto a Resource Q column (Pharmacia), and bound protein was eluted with a 10-min linear gradient from buffer A (20 mM Tris, pH 8) to buffer B (20 mM Tris, pH 8, 0.5 M NaCl). Fractions containing fusion proteins were desalted by gel filtration. Protein concentration was determined using a dye reagent (Bio-Rad) and purified MBP (New England Biolabs) as a standard, by Western blotting and by ELISA using antibodies against MBP. Inhibition of apatite formation was assayed as described below, and the protein specificity of inhibition was routinely verified by digestion of test proteins with proteinase K.
Generation of Antibodies-Antisera were generated by injecting rabbits with three 50-g doses of recombinant Sf9 mouse fetuin. The specificity of the antiserum was determined by immunoblotting of serum and tissue extracts from brain, liver, skin, skeletal muscle, spleen, and kidney. In serum and in liver tissue a single protein band of 50-kDa molecular mass (reducing SDS-PAGE) was detected (not shown).
Electrophoresis and Blotting Procedures-SDS-PAGE and chemiluminescence immunoblots were done as described (17). Lectin blots were performed using the DIG-glycan differentiation kit according the manufacturer's protocol (Boehringer Mannheim).
Immunocytochemistry and in Situ Hybridization-Mice were snapfrozen in isopentane cooled by liquid nitrogen. Cryostat sections were cut nominally 8-m thick and mounted on silanized glass slides. Sections were fixed 10 min at room temperature in Histochoice fixative (AMRESCO) for immunocytochemistry and in p-formaldehyde fixative (4% in PBS) for in situ hybridizations and washed in PBS for 5 min.
Co-precipitation of Fetuins with Salts-40 g of human serum ␣ 2 -HS dissolved in 90 l of phosphate-buffered saline were incubated with 1 mCi (10 l) of carrier-free Na[ 125 I] and Iodogen (100 g/tube) for 10 min. Unreacted iodine was separated by spin column chromatography on 1 ml of Sephadex LH20 gel (Pharmacia) equilibrated in PBS. Coprecipitation of ␣ 2 HS with salts was determined by incubating for 90 min at 37°C 125 I-labeled ␣ 2 -HS (5 nM, 200,000 cpm) in buffer (20 mM Hepes-NaOH, pH 7.4) with the salt mixtures indicated in Fig. 3. Precipitate was collected by centrifugation (15,000 ϫ g, 5 min), and ␣ 2 -HS coprecipitated was quantified using a gamma counter.
Inhibition of Apatite Formation-To determine the effect of fetuins on the formation of apatite, we adapted the assays described (27,28). Inhibition of calcium salt precipitation was determined by incubating at 37°C for 90 min a buffered salt solution (50 mM Tris-HCl or 20 mM Hepes-NaOH, pH 7.4, 4.8 mM CaCl 2 , 2 ϫ 10 6 cpm [ 45 Ca]Cl 2, 1.6 mM Na 2 HPO 4 ) containing test proteins as indicated in the figure legends. Precipitates were collected by centrifugation (15000 ϫ g, 5 min at room temperature), dissolved in 0.5% acetic acid, and quantified by liquid scintillation counting. All incubations were done in triplicate and independently repeated at least two times. To estimate the inhibition of salt precipitation by serum and ␣ 2 -HS-depleted serum, assays were done in buffer (20 mM Hepes, pH 7.4) containing salts at their serum concentrations (25 mM NaHCO 3 , 4 mM KCl, 0.75 mM MgSO 4 , 100 mM NaCl, 0.7 mM Na 2 HPO 4 , 2.5 mM CaCl 2 ) and a spike of [ 45 Ca]Cl 2 . Calcium salt precipitation was performed and quantified as above.
Proteolytic Digests-Human ␣ 2 -HS glycoprotein was digested either with chymotrypsin, elastase, proteinase K, trypsin, or endoproteinase Glu-C at a molar ratio of 100:1. All digests proceeded 2 h at 37°C in buffer (50 mM Tris-HCl, pH 7.4, 1 mM DTT). The endoproteinase Glu-C digest was fractionated by HPLC gel filtration. 0.5-ml fractions were collected and assayed for inhibitory potency toward apatite formation. Blot sequencing of peptides separated by nonreducing SDS-PAGE was performed by Dr. R. Kellner at the Central Protein Chemistry Facility, University of Mainz.
Chemical Modification of Fetuin-5 mg of bovine fetuin was chemically reduced by incubation for 3 h at 45°C in 1 ml of buffer (0.2 M Tris-HCl, pH 8.3, 1.5 mM DTT, 6 M guanidinium hydrochloride). The reduced cysteine residues were carboxymethylated by iodoacetamide (final concentration 6 mM) for 30 min at 25°C in the dark. ␤-Mercaptoethanol was added to a final concentration of 0.5% to quench residual iodoacetamide, and the reduced/carboxymethylated fetuin was desalted by gel filtration. Reduction was assayed by SDS-PAGE, as reduced single chain fetuins display higher apparent molecular weights than nonreduced forms (17). Modification of carboxylate groups was performed as published (29). 5 mg of bovine fetuin was reacted for 4 h at 37°C in 1 ml of buffer (12.5 mM glycine ethyl ester (GEE), 50 mM 1-ethyl-3-(3-dimethylaminopropyl)-carbodiimide (EDC), and 5 mM Nhydroxysulfosuccinimide, pH 4.75). The modified fetuin was dialyzed against water. Modification of carboxylate groups was determined by pI shift in isoelectric focusing gel electrophoresis (Phastsystem, Pharmacia).
Primary Osteoblast Cell Cultures-Calvaria from newborn rats were dissected and digested with collagenase (type II from C. histolyticum, 2 mg/ml, 20 min, 37°C). The digestion procedure was repeated five more times. Cells of the sixth digest were harvested by straining through a nylon mesh and seeded into 24-well plates at 20,000 cells/cm 2 . Initial cell culture was in Dulbecco's modified Eagle's medium (DMEM; calcium, 1.8 mM; PO 4 , 1 mM; 5% CO 2 ) containing 10% fetal calf serum (FCS) until the cultures became confluent, typically at day 4 in culture. Mineralization was induced by changing the culture medium to DMEM/ FCS including 10 mM ␤-glycerophosphate and 50 g/ml ascorbate. In serum-free conditions, FCS was substituted by 0.5% bovine serum albumin (dialyzed against distilled water). The influence of fetuins on mineralization was determined by adding purified proteins from stock solutions (in DMEM) to the serum-free medium. Fresh culture medium was replaced every 3 days, and the extent of mineral deposition was scored at day 12 by von Kossa staining (30). For immunostaining, mineralized primary osteoblasts were fixed with cold methanol. Human ␣ 2 -HS was detected using polyclonal antiserum against serum ␣ 2 -HS (diluted 100-fold in PBS) and FITC-conjugated secondary antibody (diluted 100-fold in PBS). Rabbit preimmune serum served as a negative control.

Fetuins Are Made in the Liver and Concentrate in Bones-
We expressed mouse fetuin cDNA in Sf9 insect cells and purified recombinant mouse fetuin protein to homogeneity to raise specific antibodies in rabbits. We studied the tissue distribution of fetuin in fetal and newborn mice by immunostaining, and on consecutive serial sections we determined the expression of fetuin mRNA by in situ hybridization. Fig. 1 shows an autoradiograph of longitudinal sections of a mouse at day 2 after birth (P2) probed with 35 S-labeled riboprobes transcribed in antisense (Fig. 1A) and sense (Fig. 1C) orientation. In this section like in all developmental stages analyzed (E12 through P7, not shown), strong hybridization signals for fetuin mRNA were detected in liver. No other organ expressed fetuin mRNA at levels comparable with liver. Similar hybridization patterns were obtained with 35 S-UTP-labeled riboprobes transcribed from the full-length cDNA (1250 base pairs) and from PCR fragments comprising the 3Ј-terminal half (650 base pairs) of the cDNA (not shown). The 5Ј-terminal part of fetuins contains two cystatin domains, and riboprobes representing this part of the molecule could possibly cross-hybridize with mRNA for other cystatin-like molecules. The 3Ј-terminal part of fetuins is, however, specific for this molecule and should only detect fetuin mRNA. The fact that both riboprobe types yielded similar hybridization patterns therefore indicates that our staining was strictly fetuin-specific.
To study the tissue distribution of fetuin protein, we stained sections consecutive to the ones used for in situ hybridization with antibody against mouse fetuin protein. Antibody staining of sections at low concentration of antibody (antiserum diluted 1 in 2000) outlined the skeleton, most notably the skull, the calcified part of the ribs, long bones, and teeth (Fig. 1B), indicating that these tissues contained the highest amounts of fetuin protein. Earlier stages of development showed no detectable signal at this low concentration of antiserum. At higher concentration of antibodies (antiserum diluted 1 in 500), many more tissues stained positive for fetuin, among them liver, kidney, hind gut, and skeletal muscle (not shown). Sections stained with preimmune serum diluted 1 in 200 remained unstained (Fig. 1D).
The combined results of in situ hybridization and antibody staining indicate that fetuins are primarily made in the liver and are concentrated in bone and teeth. We sought to determine what function in bone fetuins might serve.
Fetuins Inhibit Apatite Precipitation in Primary Osteoblast Culture-To test the influence of fetuins on actively mineralizing bone cells, we established primary rat calvaria osteoblast cultures. These cultures grow confluent until day 4 in DMEM containing 10% FCS. After this initial culture period the cells were further grown in DMEM supplemented with ascorbate and ␤-glycerophosphate to induce mineralization. During this second culture period, the cells deposited calcium mineral as revealed by von Kossa staining at day 12. Fig. 2A represents a typical view of von Kossa stained primary osteoblasts at day 12 in culture. When the DMEM culture medium contained 10% FCS and hence large amounts of bovine fetuin, calcium salt deposits were inconspicuous and only microscopically visible ( Fig. 2A, top left well). Bone nodule formation commenced and the cells stayed alive in long term culture for up to 3 weeks (longer periods were not tested). When FCS was omitted from the culture medium and instead 0.5% BSA was included (serum-free medium), upon addition of ␤-glycerophosphate and ascorbate abundant precipitation of calcium salts occurred starting at day 10, and the cells died until day 12 ( Fig. 2A,  bottom row). When, however, bovine fetuin ( Fig. 2A, left panel) or human ␣ 2 -HS ( Fig. 2A, center panel) were included in the serum-free medium, the initial pattern of FCS-containing cell culture was restored in that precipitation of calcium salts was reduced, mineralized bone nodules were formed, and cells were visually indistinguishable from FCS-containing cultures. This inhibition of salt precipitation was also observed with asialofetuin and recombinant human, rat, and mouse fetuin expressed by baculovirus-infected insect cells (not shown). The inhibition was dose-dependent in that higher concentrations of fetuin more consistently inhibited precipitation than lower concentrations. For example, human ␣ 2 -HS at 2 M (100 g/ml) completely inhibited salt precipitation in 6 out of 6 replicates within 3 independent experiments. At 0.02 M (1 g/ml), complete inhibition was observed in 4 out of 6 replicates. Bovine fetuin, asialofetuin, and recombinant human, rat, and mouse fetuin expressed by baculovirus-infected insect cells effected similar inhibition at 0.2 M like human serum ␣ 2 -HS. Ovalbumin in combination with BSA ( Fig. 2A, right panel) or BSA alone ( Fig. 2A, bottom) could not restore the initial pattern obtained with serum-containing medium.
Immunofluorescence analysis of the calcium salt deposits formed in the presence of human ␣ 2 -HS showed that they contained this protein (Fig. 2C), whereas no such ␣ 2 -HS-containing deposits could be observed in cultures grown without ␤-glycerophosphate and ascorbate (Fig. 2D). We concluded that fetuins modulate the formation of calcium mineral and sought to determine the molecular basis for this activity. We deliberately switched between forms of fetuins from different species to establish a broad base for the interpretation of fetuin biological activity. We used commercial preparations of bovine fetuin and human ␣ 2 -HS for chemical modification studies where mg quantities of proteins were required, and we included recombinant forms of fetuins derived from several species to confirm activity detected in bulk purified protein. Results were essentially congruent whenever fetuins from different species were tested in parallel.
Fetuins Bind to and Inhibit the Formation of Apatite-The accumulation of fetuins in bone matrix has previously been ascribed to its high affinity for hydroxyapatite. An elegant purification procedure for ␣ 2 -HS has been devised exploiting differential precipitation with solutions containing precipitates of calcium and phosphate (19). Elaborating on this observation, we devised a precipitation assay to determine the ion specificity of this precipitation reaction. In a precipitation assay with combinations of calcium, magnesium, carbonate, and phosphate ions, 125 I-labeled fetuins co-precipitated in mixtures of calcium and phosphate and in mixtures of calcium and carbonate but not in mixtures of magnesium and phosphate (Fig. 3). These data indicate that the binding of fetuins to apatite is due to interactions with calcium ions rather than with phosphate ions. The mineral formed in mixtures of calcium and phosphate under the assay conditions was poorly crystalline apatite as judged by x-ray diffraction of spun and dried precipitate powder (data not shown). Henceforth, we refer to the mineral formed as apatite.
We noticed that considerably less apatite but not calcium carbonate was formed in the presence of fetuins than in the protein-free control incubations. Quantification of the salt precipitates under modified assay conditions (Tris-buffered precipitation mixture with added [ 45 Ca]Cl 2 ) revealed that, for example, human ␣ 2 -HS inhibited the precipitation of apatite in a dose-dependent manner. Fig. 4A shows that the transition between no inhibition and maximum inhibition of apatite formation was sudden, resulting in a steep slope of the sigmoid dose-response curve and concomitant large error. The dose response for precipitation inhibition was similar for all forms of ␣ 2 -HS glycoproteins/fetuins tested, in that half-maximum inhibition occurred around 0.5 M (Table I). Fig. 4B illustrates the time course of a precipitation reaction. In protein-free control incubations the precipitation reached its maximum after 2 h (Fig. 4B, f). When ␣ 2 -HS was included in the precipitation mixture a dose-dependent delay of precipitation occurred, and the maximum of precipitation was achieved after 4 h (1 M ␣ 2 -HS, Fig. 4B, ࡗ) and 10 h, respectively (5 M ␣ 2 -HS, Fig. 4B, q). Adding ␣ 2 -HS to the precipitation mixture delayed but did not completely stop the formation of apatite. Accordingly, ␣ 2 -HS added to apatite preformed in the absence of protein did not dissolve the mineral (data not shown).
Fetuins Inhibit Apatite Formation in Serum-It is known that serum contains inhibitors of salt precipitation (31). We therefore sought to determine the contribution of fetuins to the overall inhibitory effect of human serum. The results obtained with human serum containing 10 M ␣ 2 -HS (500 g/ml, measured by ELISA) depleted (to less than 0.15 M) and reconsti- tuted serum are shown in Fig. 5. These experiments were performed in buffer containing the major inorganic components at their respective serum concentrations. Due to restraints of the experimental setup, the maximum serum content of the precipitation mixture was 30%. This amount of serum corresponded to 3 M ␣ 2 -HS and inhibited the formation of apatite by 95% when compared with the protein-free control incubation (Fig. 5). Isolated ␣ 2 -HS at 3 M alone inhibited by 75%, indicating that non-dialyzable serum components other than ␣ 2 -HS also inhibit the formation of apatite. Serum depletion of ␣ 2 -HS by affinity absorption caused a marked reduction in serum's inhibitory activity that could be restored by adding back the original amounts of ␣ 2 -HS. This was also observed when serum was added to the precipitation mixture at 15 and 5% final concentration. In summary, the results illustrated in Fig. 5 demonstrate that (i) the capacity of serum to inhibit apatite precipitation can be partially accounted for by the amount of ␣ 2 -HS present in this serum; (ii) when ␣ 2 -HS is removed from the serum, the capacity of the remainder to inhibit apatite precipitation drops by 50% demonstrating that ␣ 2 -HS significantly contributes to the inhibition of calcium salt precipitation in serum.

Inhibition of Apatite Formation by Fetuins Is Largely Independent of Post-translational Modification-Next we deter-
mined if forms of fetuins from different species also differing with respect to post-translational modification inhibited the precipitation of apatite better than others. To this end we included commercial preparations of bovine fetuin (complex glycosylation) and asialofetuin as well as human ␣ 2 -HS isolated from serum (dephosphorylated (13)) and from HepG2 cell culture medium (phosphorylated (17)). We expressed human, rat, and mouse fetuins in baculovirus-infected insect cells, and we expressed mouse fetuin in E. coli cells that do not glycosylate eukaryotic proteins. We determined the N-linked and Olinked carbohydrate structure of several forms of fetuins by lectin blotting and tested the forms of fetuin thus characterized for their ability to inhibit the formation of apatite. Table I shows that bovine fetuin, asialofetuin, and forms of human ␣ 2 -HS purified from serum, HepG2 hepatoma cell cul- IC 50 values were estimated by curve fitting. Assays were done in triplicate, and the standard error of the mean was calculated (S.E.). Fetuins analyzed were bovine fetuin and asialofetoin, human ␣ 2 -HS isolated from serum and from culture supernatant of HepG2 hepatoma cells, and recombinant forms of human, rat, and mouse fetuins expressed by baculovirus-infected Sf9 insect cells and a fusion protein of MBP and mouse fetuin (MBP-mf327). Note that crude preparations of mouse fetuin fused to maltose-binding protein (MBP-mf327) did not inhibit the formation of hydroxyapatite unless refolded in the presence of glutathione. The glycosylation pattern of forms of fetuins was examined by lectin blotting as described under "Experimental Procedures."  As the forms of fetuins tested differed with respect to glycosylation and phosphorylation, we conclude that these post-translational modifications do not greatly affect the inhibitory activity of fetuins. A fusion protein between maltose-binding protein and mouse fetuin (MBP mouse fetuin, crude) expressed in the cytoplasm of E. coli did not inhibit the formation of apatite unless the protein was refolded in the presence of glutathione (Table I, MBP mouse fetuin, refolded).
Several control proteins were also tested. Bovine serum albumin, a protein similarly acidic like fetuins, did not inhibit the formation of apatite below 10 M. Lysozyme, a protein with high affinity for hydroxyapatite (32), and calmodulin, a well established Ca 2ϩ binding protein, did not inhibit the formation of apatite. This result demonstrates that the inhibition of apatite formation is not simply due to Ca 2ϩ or apatite binding but rather reflects interference with crystal lattice formation. Ovalbumin, a control protein chosen because of its complex disulfide structure, also did not inhibit the precipitation of apatite.
Amino Acids in Cystatin Domain 1 Mediate the Inhibition of Apatite Formation by Fetuin- Fig. 6A demonstrates that blocking of Glu and Asp residues by glycine ethyl ester (GEE) and carbodiimide (EDC) abolished the inhibitory activity indicating that acidic amino acid residues are required for the inhibition of apatite precipitation. Likewise, the reduction and alkylation of disulfide bonds by dithiothreitol (DTT) and iodoacetamide in the presence of guanidinium hydrochloride (GdnHCl) completely abolished the inhibitor activity (Fig. 6B, q). Denaturation of bovine fetuin with GdnHCl alone and subsequent removal of GdnHCl by dialysis did not destroy the protein's ability to inhibit apatite precipitation (Fig. 6B, ࡗ).
To identify minimum primary sequences of the fetuin molecules responsible for the inhibition of apatite precipitation, we generated proteolytic fragments. Limited proteolysis of human ␣ 2 -HS and bovine fetuin with trypsin, chymotrypsin, elastase, and proteinase K all destroyed the ability to inhibit the precipitation of apatite. The results for human ␣ 2 -HS are shown in Fig. 7A. Limited proteolysis of human ␣ 2 -HS with Staphylococcus V8 protease (Endo Glu-C), however, generated a mixture of fragments of roughly 10, 20, and 30 kDa molecular mass that could still inhibit salt precipitation (Fig. 7B). When we isolated the fragments by HPLC gel filtration and tested them individually in the precipitation assay, the inhibitory activity co-purified with fragments of roughly 30 kDa (Fig. 7B). Blot sequencing of the fragments separated on nonreducing SDS-PAGE yielded the N-terminal sequences of human ␣ 2 -HS heavy and light chain in each case, indicating that the 30-kDa fragments comprised microheterogeneous complexes of the N-terminal half of the molecule and the light chain attached through a disulfide bond.
To further map the amino acids required for inhibition of apatite formation, we generated a series of C-terminal deletion mutants of mouse fetuin fused to maltose-binding protein (Fig.  8A). The fusion proteins were expressed in E. coli and required refolding in the presence of glutathione to gain inhibitory activity. Fig. 8B illustrates that the full sequence mouse fetuin FIG. 6. Chemical modification abolishes inhibition of apatite formation by bovine fetuin. A, the carboxylate groups of bovine fetuin were modified by glycine ethyl ester (GEE) alone or by GEE in combination with the activating agent, ethylcarbodiimide (EDC). B, bovine fetuin was denatured using guanidinium hydrochloride (GdnHCl), reduced by dithiothreitol (DTT) and carboxymethylated by iodoacetic acid (IAA) as described under "Experimental Procedures." The influence of these modifications was assayed in the apatite precipitation assay as described in Fig. 4. Assays were done in triplicate; bars indicate the standard error of the mean.  Table I. MBP-mf81 (lacking all naturally occurring disulfide bonds) still inhibited the formation of apatite, whereas MBP-mf52 did not, indicating that acidic amino acids in the N-terminal half of domain D1 are required for the inhibition of apatite formation. In view of these results intact disulfide bonds do not seem to be required for the inhibition of apatite formation. The lack of inhibitory activity in reduced/alkylated bovine fetuin (Fig. 6B) and in incorrectly folded MBP-mf327 after expression in bacteria (Table I) therefore suggests that the inhibitory sequences in domain D1 are inaccessible in those forms. This interpretation gains further support by the fact that 1 M MBP-mf81 inhibited apatite formation equally without refolding in the presence of glutathione. DISCUSSION We developed antibodies against mouse fetuin and studied its tissue distribution on longitudinal whole body sections by immunocytochemistry and in situ hybridization. Our results demonstrate that fetuin mRNA and protein are made in the liver and that fetuin protein is transported to the bone to be sequestered in the mineralized matrix of the developing bone. These findings corroborate earlier physiological experiments in rabbits (5) and immunodetection of fetuin in human (33) and rat tissues (34). The accumulation of fetuin in bone suggests a relatively high affinity of these serum proteins for calcium which should represent an aspect of a more basic involvement in bone formation or remodelling.
Rat calvaria cells are an established model system of osteogenesis and mineralization in vitro (35). In this system mineralization commences after the addition of ␤-glycerophosphate or phosphate to the culture medium, whereas the formation of nodules and the expression of bone-specific proteins are independent of an external phosphate source (36). Serum-free culture conditions demonstrate that conditions of ion supersaturation causing spontaneous salt precipitation can exist in this system ( Fig. 2A). It is known that phase separation in supersaturated solutions of calcium and phosphate is slowed down in the presence of serum proteins (31). Serum albumin has been described as an inhibitor of crystal formation in serum, but serum albumin alone could not account for whole serum's ability to inhibit salt precipitation (37). Our data indicate that ␣ 2 -HS glycoproteins/fetuins are efficient inhibitors on a molar basis and could account for the remainder of serum's inhibitory activity. Fetuins can prevent phase separation in serum efficiently as the concentration causing half-maximum inhibition of apatite formation in vitro (0.5 M) is 20-fold below the observed serum concentration (10 M). Specifically, fetuins might be the major inhibitors of salt precipitation during fetal life, when serum albumin concentration is low. This conclusion gains support from the observation that serum Ca 2ϩ concentration and fetuin concentration closely correlate during fetal development in any single one species and also when several species are compared (24).
Common structural features of protein inhibitors of salt precipitation can be stretches of poly(Asp) and extensive serine/ threonine phosphorylation (27), ␥-carboxyglutamic acid (28), and sequences that contain no specific motifs but are generally acidic (3) or highly charged like the N-terminal fragment of salivary statherin (38). Fetuins contain clusters of Asp residues in both cystatin domains 1 and 2 (9) and are also Ser-phosphorylated (17). The inhibition of apatite formation by fetuins was abolished after chemical blocking of carboxylate groups with glycine ethyl ester, indicating that aspartic and glutamic acid residues might confer calcium binding. The combined results from chemical modification of fetuin (Fig. 6), limited proteolysis (Fig. 7), and C-terminal deletion mutagenesis (Fig. 8) restrict the amino acid clusters to the N-terminal half of domain D1. Acidic amino acids conserved in all known fetuin domains D1 are Asp 15 -Asp-X-(Asp/Glu)-X-Glu 20 , Asp 46 and (Asp/Glu 61 )-X-Glu-X-Asp-X-X-Glu 68 ). Neither Asp 15 -Glu 20 alone (see MBP-mf52) nor a synthetic peptide comprising Ile 62 -Arg 81 of human ␣ 2 -HS alone (data not shown) caused inhibition; therefore, these amino acids are likely to inhibit the formation of apatite in a concerted fashion. FIG. 8. Inhibition of apatite formation by truncated forms of mouse fetuin. Mouse fetuin cDNA was progressively shortened at the 3Ј-end by exonuclease III digestion as described under "Experimental Procedures." The resulting C-terminal deletion products (A) were expressed in E. coli as fusion proteins with maltose-binding protein (MBP) and purified to homogeneity. Proteins were refolded in the presence of glutathione as described under "Experimental Procedures," and inhibition of apatite formation was estimated as described in Fig. 4. Assays were done in triplicate; bars indicate the standard error of the mean (B).