Repetitive Ser-Gly sequences enhance heparan sulfate assembly in proteoglycans.

We showed previously that the synthesis of heparan sulfate on betaglycan occurs at a Ser-Gly dipeptide flanked by a cluster of acidic residues and an adjacent tryptophan (Zhang, L., and Esko, J. D.(1994) J. Biol. Chem. 269, 19295-19299). A survey of the protein data base revealed that most heparan sulfate proteoglycans contain repetitive (Ser-Gly)n segments (n ≥ 2) and a nearby cluster of acidic residues. To study the role of these amino acid sequences in controlling heparan sulfate synthesis, we have examined the assembly of glycosaminoglycans on Chinese hamster ovary (CHO) cell syndecan-1. The glycosylation sites were mapped by making chimeric proteoglycans containing segments of CHO syndecan-1 cDNA fused to Protein A. Two sites near the transmembrane domain (-EGS205GEQ- and -ETS215GEN-) were used solely for chondroitin sulfate synthesis, whereas three sites near the N terminus (-DGS35GDDSDNFS45GS47GTG-) supported both heparan sulfate and chondroitin sulfate synthesis. The strongest sites for heparan sulfate synthesis consisted of the repeat unit, -S45GS47G-. An unusual coupling phenomenon occurred across the adjacent SG dipeptides, leading to a greater proportion of heparan sulfate than predicted by the behavior of each site acting independently. The clusters of acidic residues adjacent to the heparan sulfate sites play important roles as well. These sequence motifs suggest a set of rules for predicting whether heparan sulfate assembles at glycosylation sites in proteoglycan core proteins.

Chondroitin sulfate and heparan sulfate proteoglycans contain glycosaminoglycan (GAG) 1 chains attached to specific serine residues of core proteins. Studies of hybrid proteoglycans 2 such as betaglycan, ryudocan, and syndecan-1 showed that two types of GAG attachment sites exist (1)(2)(3)(4). One type carries heparan sulfate or chondroitin sulfate chains, whereas the other type bears only chondroitin sulfate. In the proteoglycan, betaglycan, a specific amino acid sequence drives heparan sulfate synthesis (2). This site consists of a Ser-Gly dipeptide, a nearby cluster of acidic residues, and an adjacent tryptophan that augments the proportion of heparan sulfate made (2). These structural elements may enhance the interaction of glycosylated core protein intermediates with a key ␣-GlcNAc transferase that initiates the formation of heparan sulfate (5,6).
Almost all cloned heparan sulfate proteoglycans contain a cluster of acidic residues near one or more putative heparan sulfate attachment sites, but only a few contain nearby tryptophan. To study whether other amino acid sequences surrounding GAG attachment sites might enhance heparan sulfate assembly, we have examined the assembly of GAGs on syndecan-1, a hybrid proteoglycan containing both heparan sulfate and chondroitin sulfate (7). Kokenyesi and Bernfield (4) showed recently that the N-terminal half of mouse syndecan-1 contains heparan sulfate and chondroitin sulfate chains, whereas the C-terminal half contains only chondroitin sulfate. Five conserved Ser-Gly dipeptides may act as the sites for GAG addition, but the identity of the sites that prime heparan sulfate was not determined. Our study reveals that most of the heparan sulfate occurs at an attachment site containing (Ser-Gly) 2 through an unusual coupling mechanism. This sequence motif occurs frequently in heparan sulfate proteoglycans and always includes a cluster of acidic amino acids flanking the site. We also show that the biosynthetic capacity of the cell to make heparan sulfate and the amount of core protein expression affect the proportion of heparan sulfate assembled.
Radiolabeling Studies-Na 2 35 SO 4 (25-40 Ci/mg; 1 Ci ϭ 37 GBq) was purchased from Amersham Corp. Syndecan-1 was labeled biosynthetically by incubating cells in sulfate-deficient medium containing 35 SO 4 (100 Ci/ml) for 12 h. The medium was discarded, the radiolabeled monolayer was rinsed five times with cold PBS, and the proteoglycans were extracted with Triton X-100 buffer (1% (w/v) Triton X-100, 150 mM NaCl, 10 mM Na 2 HPO 4 , 2 mM KH 2 PO 4 , pH 7.4) containing protease inhibitors (5 mM N-ethylmaleimide, 5 mM benzamidine-HCl, 1 mM phenylmethylsulfonyl fluoride, 0.5 g/ml leupeptin, and 1 g/ml of pepstatin A). The extract was clarified by centrifugation (10,000 ϫ g, 60 min), and the pellet was re-extracted with 5 ml of Triton X-100 buffer for 1 h at room temperature. The supernatants were combined, and solid urea was added to achieve a final concentration of 6 M. Bovine tracheal chondroitin sulfate A (1 mg, Calbiochem) was added as carrier, and the samples were applied to a 0.5-ml column of DEAE-Sephacel (Pharmacia Biotech Inc.) prepared in a disposable pipette tip and equil-* 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. ‡ Current address: Dept. of Biology, Massachusetts Institute of Technology, 77 Massachusetts Ave., Bldg. 68-480, Cambridge, MA 02139. ¶ To whom correspondence and reprint requests should be addressed. Tel.: 205-934-6034; Fax: 205-975-2547; E-mail: jesko@bmg.bhs.uab.edu. 1 The abbreviations used are: GAG, glycosaminoglycan; CHO, Chinese hamster ovary; PCR, polymerase chain reaction; PBS, phosphatebuffered saline; HPLC, high performance liquid chromatography; PAGE, polyacrylamide gel electrophoresis. 2 A hybrid proteoglycan contains more than one type of GAG chain, such as heparan sulfate and chondroitin sulfate. ibrated in urea buffer (6 M urea, 0.2 M NaCl, 0.5% (w/v) Triton X-100, 25 mM Tris-HCl, pH 7.0, and protease inhibitors). The columns were washed with urea buffer (15 ml), and radiolabeled proteoglycans were eluted with 2.5 ml of urea buffer containing 1 M NaCl. The proteoglycans were desalted over a PD-10 column (Pharmacia) equilibrated in 10% (v/v) ethanol in water containing protease inhibitors. After lyophilizing the samples, they were dissolved in PBS containing protease inhibitors (10 ml).
Syndecan-1 Purification and Characterization-The proteoglycan extract was passed over an affinity column (0.5 ml) of anti-syndecan-1 2E9 monoclonal antibody 3 coupled to Sepharose CL-4B (10). The column was washed with 8 ml of PBS, followed by 2 ml of PBS supplemented with 0.5 M NaCl. Syndecan-1 was eluted from the resin with 5 ml of a solution containing 4 M guanidine hydrochloride, 50 mM sodium acetate (pH 5.8), and protease inhibitors. Salt and guanidine were removed from purified syndecan-1 by chromatography over a PD-10 column before further analysis. More than 95% of purified syndecan-1 rebound to the affinity column, and binding was independent of GAG composition (10,11).
A portion of affinity-purified syndecan was resuspended in 0.5 M NaOH containing 1 M sodium borohydride and incubated at 4°C for 24 h to ␤-eliminate the chains. The base was neutralized by adding 10-l aliquots of 10 M acetic acid until bubble formation ceased. Radiolabeled syndecan-1 and the released GAG chains were analyzed by gel filtration HPLC (TSK G4000SW, 30 cm ϫ 7.5 mm inner diameter, Pharmacia). Samples were eluted with 0.5 M NaCl in 0.1 M KH 2 PO 4 (pH 6.0) containing 0.2% Zwittergent 3-12 at a flow rate of 0.5 ml/min, and radioactivity in the effluent was determined by in-line liquid scintillation spectrometry using Ultima Gold XR scintillant (Packard Instrument Company, Downers Grove, IL). GAGs were digested with 20 milliunits of chondroitinase ABC (Seikagaku) in 50 mM Tris-HCl and 50 mM sodium acetate buffer (pH 8.0) containing protease inhibitors. Complete digestion of chondroitin sulfate by chondroitinase ABC was assured by monitoring the extent of conversion of the carrier to disaccharides (1 mg ϭ 11.4 absorbance units at 232 nm). Nitrous acid-catalyzed deaminative cleavage of heparan sulfate was performed according to the low pH method of Shively and Conrad (12).
cDNA Cloning of CHO Syndecan-1 4 -Sequences coding for syndecan-1 were amplified from a CHO-K1/cDNA quick-clone library (Clonetech, Palo Alto, CA). The reaction mixture contained 2 units of pfu polymerase (Stratagene, La Jolla, CA), 1 ng of cDNA, and 100 pmol of sense and antisense primers. The primers (Oligo etc., Inc., Wilsonville, OR) were identical to the murine syndecan-1 cDNA sequence from nucleotides 240 -264 and 1151-1175 (13). After 30 thermal cycles (1 min of denaturation at 94°C, 2 min of annealing at 55°C, and 3 min of extension at 72°C), the amplification products were analyzed by gel electrophoresis in 1% agarose gels and detected by ethidium bromide staining. The ϳ900-base pair amplification product was excised from the gel and treated with the Gene Clean Kit (BIO 101, La Jolla, CA). The PCR product was phosphorylated, ligated into the SmaI restriction site of plasmid pBluescript SK ϩ (Stratagene), and introduced into competent E. coli DH5␣ cells. Positive clones from two separate PCR reactions were chosen and sequenced by the dideoxy chain termination method (14) using supercoiled plasmid, Sequenase version 2.0 kit (U.S. Biochemical Corp.), T3 and T7 primers, dGTP, and dITP.
The numbers before the colon in the following list of constructs refer to a segment of amino acids from CHO syndecan-1 (mutations in parentheses), and the information after the colon refers to the set of oligonucleotide primers and templates used to amplify the desired sequence. 26 The plasmid, pC17, containing the full-length fibroglycan cDNA sequence (M81687, a gift from Dr. John Gallagher, Manchester, United Kingdom) was used as a template for PCR of the fibroglycan fragment (DMYLDSSSIEEASGLYPIDDDDYSSASGSGAYEDKGSPDLTTSQ). The wild-type betaglycan construct, SPGDSSGWPDGYEDLE, and the mutant, SPGDSSGAPDGYEDLE, were made using primers described previously (2). Overlapping PCR primers were designed to generate the following betaglycan fragments (M77809). Overlapping PCR primers were designed for the following peptide sequences. PCR fragments were extracted with phenol/chloroform, precipitated with ethanol, digested with EcoRI, purified by agarose gel electrophoresis, and transferred to DEAE paper (Schleicher & Schuell). After elution, the fragments were ligated into a derivative of the eukaryotic expression vector pPROTA (16) that had been treated with EcoRI and calf intestinal alkaline phosphatase. All constructs were sequenced to confirm their identity.
Transfection-For transient transfection experiments, each well of a 12-well plate was seeded with 2 ϫ 10 5 cells in F12 medium containing 10% (v/v) NuSerum (Collaborative Research, Inc., Bedford, MA). After one day, the medium was removed, and 0.5 ml of F12 medium containing 0.25 mg/ml of DEAE-Dextran (Sigma D-9885), 50 mM Tris-HCl (pH 7.4), 50 g/ml of chloroquine, and 10 g/ml of plasmid DNA was added to each well. After 2 h at 37°C, the solution was aspirated, and 0.5 ml of 10% (v/v) Me 2 SO in PBS was added. After 2 min, the solution was aspirated and the cells were washed with 2 ml of F12 medium without serum. Cells were then labeled at 37°C for 2 days with 50 Ci/ml of Stable transfectants were created by transfecting a monolayer of wild-type CHO cells (50% confluence) with 16 g of pPROTA containing CHO syndecan-1 cDNA coding for amino acid residues 26 -240. The vector pMAMNEO was included (2 g) along with 20 g of Transfectam reagent using conditions recommended by the manufacturer (Promega Corp., Madison, WI). Two days later, the cells were harvested with trypsin, and a serial dilution (1:3, v/v) series was prepared in a 96-well plate. Strains resistant to G418 (400 g/ml, corrected; Life Sciences Technologies, Inc.) were selected. The medium was changed every 4 days for 2 weeks, and individual colonies were picked, expanded, and evaluated for expression of chimeras as described above.
Fusion Protein Purification-Spent culture medium (1 ml) was mixed in an Eppendorf tube with 10 l of a solution containing 1 M Tris-HCl (pH 7.5), 5% (w/v) Triton X-100, and 0.2% (w/v) sodium azide. Samples were centrifuged at 10,000 ϫ g for 5 min, and the supernatant was decanted into a fresh tube. IgG-Agarose beads (20 l, Sigma) were added, and the sample was mixed end-over-end overnight at 4°C. Samples were centrifuged at 10,000 ϫ g for 2 min. After aspirating the supernatant, the pellets were washed 3 times with 1.4 ml of buffer containing 50 mM Tris-HCl (pH 8), 0.15 M NaCl, and 0.02% sodium azide. Samples were dissolved in 20 mM Tris buffer (pH 7.0) and treated at 37°C for 4 h with 10 milliunits of chondroitinase ABC, 2 milliunits of heparin lyase III (EC 4.2.2.8), or both enzymes. The reactions were stopped by adding SDS-PAGE sample buffer and heating them for 7 min in a boiling water bath. Samples were loaded on a 5-16% linear gradient SDS-PAGE gel (200 ϫ 160 ϫ 1.5 mm) and electrophoresed overnight at constant voltage (80 V). The gel was dried onto a piece of filter paper and visualized by either autoradiography or by imaging (Molecular Dynamics PhosphoImager, Sunnyvale, CA). The distribution of counts was quantitated using ImageQuant software (Molecular Dynamics).
Another set of samples was treated for 24 h at 4°C with 100 l of 0.5 M NaOH containing 1 M NaBH 4 to ␤-eliminate the GAG chains. [ 35 S]GAGs were isolated by anion exchange chromatography as described (17). Samples labeled with TRAN 35 S-LABEL were dissolved in 1 ml of 25 mM Tris-HCl buffer (pH 7.0) containing 6 M urea, 0.5% Triton X-100, and 0.25 M NaCl. An aliquot was counted by liquid scintillation spectrometry.

RESULTS
CHO Syndecan-1-Syndecan-1 represents 15-20% of CHO cell 35 S-proteoglycans based on binding to an affinity column composed of the 2E9 monoclonal antibody against human syndecan-1 described by Lories et al. (10). All of the affinitypurified material behaved like proteoglycan since the 35 Scounts eluted in the V o of a gel filtration column and shifted to included fractions after ␤-elimination ( Fig. 1). Treating purified syndecan-1 with low pH nitrous acid, which depolymerizes heparan sulfate chains by cleaving at N-sulfated glucosamine residues (12), converted 80 -85% of the counts to small 35 Soligosaccharides and free 35 SO 4 . A residual core protein containing heparan sulfate stubs and chondroitin sulfate emerged at a K av of 0.27. Treating a sample with chondroitinase ABC, which depolymerizes chondroitin sulfate chains to 35 S-disaccharides, caused 15-20% of the 35 S-counts to elute in the V t of the column (Fig. 1A). Thus, CHO syndecan-1 consisted of 80 -85% heparan sulfate and 15-20% chondroitin sulfate. The same composition was obtained when the released GAG chains were analyzed by enzymatic digestion (Fig. 1B).
There are five possible GAG attachment sites in mouse, human, and rat syndecan-1 based on the presence of Ser-Gly dipeptides (7). The same five sites were present in CHO syndecan-1, which was cloned by PCR using the mouse cDNA sequence to design appropriate primers ( Fig. 2A; Ref. 13). The CHO syndecan-1 amino acid sequence showed 94, 87, 86, and 77% homology to the Syrian golden hamster, rat, mouse, and human sequences, respectively (Fig. 2B). More importantly, the amino acid sequences near the putative GAG attachment sites are highly conserved among all the different species (7).
fate and chondroitin sulfate chains were present on the Nterminal half of murine syndecan-1, but only chondroitin sulfate was found on the C-terminal half. Of the five potential GAG attachment sites, three reside in the N-terminal half and two in the C-terminal half. We showed previously that the synthesis of heparan sulfate on betaglycan requires a cluster of acidic residues located near a specific Ser-Gly attachment site (2). Two clusters of acidic amino acids border the three Ser-Gly sites on the N-terminal part of the syndecan-1 (-EDQDG-S 37 GDDSDNFS 45 GS 47 GTG-, Fig. 2B), consistent with the idea that one or more of these sites might support heparan sulfate assembly.
To examine how GAGs assemble at the individual sites, we prepared chimeras composed of CHO syndecan-1 segments fused to the IgG binding domain of Staphylococcal Protein A with a signal peptide from the secreted metalloprotease, transin, attached to the N terminus (16). The chimeras were introduced into wild-type CHO cells by transient transfection, and the attachment of GAG chains was assessed by 35 SO 4 incorporation into secreted chimeras ("Experimental Procedures"). The activity of each Ser-Gly site was measured independently by mutating other potential sites in the constructs (Table I). All three Ser-Gly dipeptides toward the N terminus (-EDQDGS 37 -GDDSDNFS 45 GS 47 GTG-) were capable of priming heparan sulfate chains, but their relative ability varied from 9% at Ser 37 to 23% at Ser 47 . In contrast, the two sites near the transmembrane domain (-PTGEGS 205 GEQDFTFETS 215 GENTA-) primed only chondroitin sulfate. A sixth Ser-Gly dipeptide exists in both CHO and Syrian golden hamster syndecan-1 (Ser 181 ), but not in mouse, rat, and human syndecan-1. Expression of a 156-amino acid chimera containing residues 49 -204 did not result in incorporation of 35 SO 4 into GAGs (Table I), suggesting that Ser 181 does not act as a glycosylation site.
The extent of substitution of the chimeras with heparan sulfate was less than expected from the study of endogenous syndecan-1, which contained 80 -85% heparan sulfate (Fig. 1). The difference was not due to variation in the extent of sulfation of heparan sulfate and chondroitin sulfate chains, since comparable results were found when the cells were labeled with [6-3 H]GlcN and chain length did not vary significantly (Fig. 3). The difference also was not due to poor expression of the chimera since only ϳ20% of material labeled with TRAN 35 S-LABEL was converted to high molecular weight proteoglycan.
We also tested the chimeras in CHO ldlD cells, which cannot obtain UDP-GalNAc from UDP-GlcNAc due to a deficiency in the 4Ј-epimerase that interconverts the nucleotide sugars (18). This strain makes less chondroitin sulfate when deprived of exogenous GalNAc (19). When the chimeras were introduced into ldlD cells, the amount of heparan sulfate was enhanced (up to 79%), but the relative amount made at each N-terminal site remained about the same (Table I,  Ser 47 yielded about 48% heparan sulfate (Table I). An identical construct containing all three N-terminal sites gave 60% heparan sulfate. Enhanced synthesis also occurred in constructs containing either Ser 37 and Ser 45 or Ser 37 and Ser 47 (Table I). These findings suggested that some type of coupling occurs across nearby attachment sites that increases the proportion of heparan sulfate.
To test whether coupling was dependent on flanking amino acid sequences, a repetitive SGSG sequence was introduced into betaglycan. In betaglycan, the site that supports heparan sulfate synthesis contains Ser-Gly-Trp flanked by a cluster of acidic residues (Table II). We showed previously that converting the Trp residue to Ala reduced the level of heparan sulfate from 54 to 21% (2). When a second Ser-Gly dipeptide was substituted for the Ala residue, the normal level of heparan sulfate synthesis was restored (51% , Table II). Thus, the coupling across adjacent Ser-Gly sites appeared to be independent of surrounding sequence. Separating the Ser-Gly repeats by one or more residues reduced the proportion of heparan sulfate, suggesting that the coupling depended on proximity (Table II). Inserting a Trp residue next to the repeat Ser-Gly segment gave a slight enhancement of heparan sulfate (58 versus 51%). A stable transfectant expressing a chimera with all five sites (residues 26 -240) was analyzed (Table III). Unlike the behavior of the transient transfectants, the stably transfected cell line converted all of the chimera to high molecular weight proteoglycan (data not shown) and produced about 3-fold more total 35 S-proteoglycan than nontransfected control cells (Table  III). The proportion of heparan sulfate was reduced in the chimera (55%) compared with the proteoglycans found in non-   35 SO 4 for 2 days (see "Experimental Procedures"). The numeric designations in the segments indicate the first and last amino acid residue of CHO syndecan-1 (Fig. 2) and the position of each potential GAG attachment site. The dashes (-) indicate that some of the residues are not shown. The boldface letters indicate mutated residues. The chimeras were affinity-purified (see "Experimental Procedures"), and the 35 S-counts were quantitated. The amount of 35 SO 4 in chimeras in nontransfected cells or in cells transfected with the constructs in the wrong orientation was 200-fold less than the amount incorporated with the correct orientation. The amount of TRAN 35 S-LABEL incorporation into constructs in the wrong orientation was comparable with the values shown since most of the Cys and Met was present in the Protein A portion of the chimera (2). The GAG composition was determined by chondroitinase ABC digestion of a portion of the affinity-purified proteoglycan labeled with 35 SO 4 . The amount of resistant heparan sulfate was quantitated and expressed relative to the total amount of 35 S-proteoglycan treated. The values in parentheses reflect the proportion of heparan sulfate produced when ldl D cells were transfected. transfected parental cells (72%). Interestingly, the endogenous CHO cell proteoglycans also contained less heparan sulfate in the stable transfectant (54 -62%). These findings suggested that the reduced level of heparan sulfate may have been due in part to enhanced core protein expression. Similar effects have been observed when full-length syndecan and betaglycan constructs were expressed in cells (1,4).
To gain insight into the mechanism that results in a higher proportion of heparan sulfate at adjacent attachment sites, we analyzed the GAG substitution pattern of syndecan-1 chimeras containing one site ( 26 -EDQDGAGDDSDNFS 45 GTGTG-64 ) and two sites ( 26 -EDQDGAGDDSDNFS 45 GS 47 GTG-64 ). As expected, the two-site chimera migrated with a greater hydrodynamic volume during gel filtration than the chimera containing one site (Fig. 3A). ␤-elimination shifted the [ 35 S]GAG to a more included position, but the mixture of GAG chains on both chimeras had essentially the same elution position (Fig. 3B). Analysis of the heparan sulfate and chondroitin sulfate chains on the two-site chimeras showed that they were comparable in size as well (Fig. 3C).
The two-site chimera presumably consisted of glycoforms containing one or two GAG chains, with various combinations of heparan sulfate and chondroitin sulfate chains. Analysis of 35 S-labeled material by SDS-PAGE revealed smeared bands characteristic of proteoglycans (Fig. 4, lane 3). The material of faster mobility migrated in the same position as a chimera containing only one attachment site (compare lane 3 with lanes  1 and 2), suggesting that it represented material with only one GAG chain. The smear above this region presumably contained 35 S-chimeras with two chains, although some overlap occurred with molecules containing one chain. Using the divisions shown in Fig. 4, we found that ϳ50% of the total 35 S-counts were in chimeras containing two chains. This material represented about one-third of the chimeras, given that the size of the chondroitin sulfate and heparan sulfate chains did not vary significantly (Fig. 3C). The degree of sulfation of heparan sulfate and chondroitin sulfate differs somewhat (0.8 sulfate groups/disaccharide versus 1 sulfate/disaccharide) (20), but this effect only moderately underestimates the extent of substitution by heparan sulfate.  (42 ϩ 32)), or ϳ9% of the total proteoglycans. These findings are summarized in Table IV.
The extent of substitution by heparan sulfate in the twochain glycoforms (14% of total chimeric proteoglycan) was greater than one would predict based on the independent behavior of each site. Constructs containing only Ser 45 or Ser 47 produced 15 and 23% heparan sulfate proteoglycans, respectively (Table I). If the sites acted independently in the chimera containing both Ser 45 and Ser 47 , then the proportion of chimeras containing two heparan sulfate chains should have been ϳ1.1% based on the probability of producing heparan sulfate at each site (0.15 ϫ 0.23) multiplied by the proportion of chimeras containing two chains (one-third). Therefore, glycoforms with two heparan sulfate chains accumulated more than 10-fold over the predicted value (14 versus 1.1%). The proportion of the other glycoforms was less than predicted (11 versus 22% for proteoglycans containing two chondroitin sulfate chains (onethird of 0.85 ϫ 0.77] and 9 versus 10% for hybrids containing one heparan sulfate and one chondroitin sulfate chain).
Previous studies of betaglycan showed that a cluster of acidic residues downstream from the site that supported heparan sulfate synthesis dramatically affected GAG composition (2). The sites supporting heparan sulfate synthesis in syndecan-1 also have nearby clusters of acidic residues (Fig. 2). To examine if these clusters played a similar role in syndecan-1, a chimera containing residues 32-53 and all three N-terminal sites was

TABLE III GAG composition of proteoglycans in wild-type CHO cells and a stable transfectant
Wild-type CHO cells and a stable transfectant expressing a fusion protein consisting of residues 26 -240 were labeled with 35 SO 4 for 2 days. The proteoglycan fraction from one set of cultures was prepared by Triton X-100 extraction (see "Experimental Procedures"). The medium was removed from another set of cultures, and the proteoglycans in the monolayers were isolated. Chimeras in the spent growth medium were affinity-purified (see "Experimental Procedures"). All samples were treated with protease and ␤-eliminated to liberate the GAG chains (17). The chains were isolated by anion exchange chromatography, and their composition was determined by enzymatic assay (see "Experimental Procedures"). The values in parentheses reflect the relative proportion of heparan sulfate in each sample. prepared. Mutating the Asp and Glu residues in the clusters to Asn and Gln, respectively, reduced the proportion of heparan sulfate from 38% in the control construct to 8 and 13%, depending on the cluster (Table V). Mutating both clusters diminished heparan sulfate synthesis to background (4%). Thus, the clusters of acidic residues are important elements in both syndecan-1 and betaglycan.

DISCUSSION
This report describes the use of protein chimeras to probe the assembly of heparan sulfate chains on proteoglycans. The results suggest a set of rules for predicting whether heparan sulfate will assemble at glycosylation sites. These rules include features of the core protein, the cellular capacity to produce individual GAGs, and the relative abundance of core proteins. Together, they support a model for GAG chain assembly in which specific core protein determinants interact with a key biosynthetic enzyme involved in heparan sulfate biosynthesis.
Clusters of Acidic Residues Are Necessary but Not Sufficient-Studies of betaglycan suggested that one type of heparan sulfate attachment site consists of a Ser-Gly dipeptide located near a cluster of acidic residues (2). Mutation of the acidic residues diminished heparan sulfate synthesis and led to greater substitution of the site with chondroitin sulfate. Studies of syndecan-1 showed that sites supporting heparan sulfate synthesis also contain nearby clusters of acidic residues (-EDQDGS 37 GDDSDNFS 45 GS 47 GTG-). Mutation of the Asp and Glu residues to Asn and Gln in either cluster reduced heparan sulfate synthesis, and altering both clusters caused full inhibition (Table V).
Inspection of various cloned proteoglycans reveals that a cluster of acidic residues always flanks sites thought to contain heparan sulfate chains, although the position and exact composition of the cluster varies (Table VI). These clusters occur less frequently in chondroitin sulfate proteoglycans. ␤-Glycan, decorin, invariant chain, and HI-30 contain clusters of acidic residues, but the natural proteoglycans contain only chondroitin sulfate chains at the indicated sites (1, 21-23). Decorin and the betaglycan chimeras behaved similarly (Table VI). These findings suggest that other elements work along with the cluster of acidic residues to drive heparan sulfate synthesis. Thus, clusters of acidic residues appear to be necessary but not sufficient determinants for heparan sulfate assembly.
Hydrophobic Amino Acids Can Act as Enhancer Elements-In betaglycan, the site that supports heparan sulfate synthesis contains a Trp residue adjacent to the Ser-Gly unit (2). Changing this residue to Ala reduced heparan sulfate synthesis. More importantly, inserting Trp next to a chondroitin sulfate site with a nearby cluster of acidic residues enhanced heparan sulfate formation. Although Trp occurs relatively rarely in proteoglycans (Table VI), aromatic and aliphatic residues are quite common, enhancing the overall hydrophobicity of the immediate region by the GAG attachment site (4). The hydrophobicity of chondroitin sulfate sites tends to be less pronounced (Table VI).
Earlier studies of ␤-D-xylosides in CHO cells showed that priming of chondroitin sulfate most likely occurs by default (24,25). Simple xylosides containing alkyl chains as an aglycone prime chondroitin sulfate but not heparan sulfate. Compounds containing fused aromatic rings, which may simulate the structure of the indole side chain of tryptophan, drive heparan sulfate as well as chondroitin sulfate synthesis (25). Thus, aromatic ␤-D-xylosides may mimic the hydrophobic patch found near heparan sulfate attachment sites. Synthesis of heparan sulfate by aromatic ␤-D-xylosides requires a relatively high concentration of primer (Ͼ30 M) compared with endogenous intermediates, possibly because they lack negative charges imparted by the acidic residues in natural core proteins.
The enhanced synthesis of heparan sulfate across adjacent Ser-Gly sites deserves further study. The upstream site in syndecan-1 (Ser 37 ) couples to the downstream sites at Ser 45 and Ser 47 (Table I) 26 -EDQDGAGDDSDNFS 45 GS 47 GTG-64 and labeled with 35 SO 4 for 2 days. The chimeras were affinity-purified and analyzed by SDS-PAGE before and after treatment with chondroitinase ABC or heparin lyase III (see "Experimental Procedures"). The GAG composition of the twochain glycoforms was determined by imaging the gel (Fig. 4). The experimental values for the various glycoforms were determined by the amount of material migrating in the two-chain region after enzymatic digestion as explained in the text. The theoretical percentage of each two-chain form was calculated by assuming that each Ser-Gly attachment site functioned independently and primed heparan sulfate to the same extent as chimeras containing only one attachment site (Table I). Chimeras containing 26 -EDQDGAGDDSDNFS 45 GTGTG-64 or 26 -ED-QDGAGDDSDNFTGS 47 GTG-64 yielded only 15 and 23% heparan sulfate proteoglycans, respectively. Therefore, if the sites acted independently, then the proportion of chimeras containing two heparan sulfate chains should have been 1.1% (15% ϫ 23% ϫ 1/3, the proportion of chimeras containing two chains). The proportion of chimeras containing two chondroitin sulfate chains should have been 22% (85% ϫ 77% ϫ 1/3  V Acidic residues are important for heparan sulfate assembly Protein A chimeras containing wild-type or mutated syndecan-1 segments (see "Experimental Procedures") were introduced into wild-type CHO cells and labeled biosynthetically with 35 SO 4 for 2 days. The numeric designations in the segments indicate the first and last amino acid residue of CHO syndecan-1 (Fig. 2) and the position of each potential GAG attachment site. The dashes (-) indicate that some of the residues are not shown. The boldface letters indicate mutated residues. The chimeras were affinity-purified, and the GAG composition was determined by chondroitinase ABC digestion. The amount of resistant heparan sulfate was quantitated and expressed relative to the total amount of [ 35 16 4 transfectants expressing ryudocan contain more heparan sulfate when multiple attachment sites were present. These findings may indicate that coupling can occur at a distance, possibly by juxtaposing the sites through secondary and tertiary structure. Interestingly, chimeras containing the sites in syndecan-3 (-SDLEVPTSSGPSGDFEIQEEEETT-) and in N-syndecan (-TTTQDEPEVPVSGGPSGDFELQEE-) do not prime much heparan sulfate (7 and 8%, respectively). Both sequences contain a proline residue between the Ser-Gly units, which may act as an inhibitor. The low degree of coupling between closely spaced Ser-Gly sites in betaglycan also may have been due to an intervening proline residue (2), which may have altered the conformation of the peptide (Table III). If correct, this idea is reminiscent of the inhibitory effect of Pro on the attachment of Glc 3 Man 9 GlcNAc 2 oligosaccharides from dolichyl-P intermediates to Asn-Xaa-Ser/Thr sites in glycoproteins (27). Perhaps negative regulatory sequences/amino acids play a role in proteoglycan assembly as well.
As shown in Table VI, repetitive Ser-Gly dipeptides with a flanking cluster of acidic residues represent a common motif in a variety of heparan sulfate proteoglycans. Thrombomodulin also contains adjacent Ser-Gly dipeptides, but lacks a cluster of acidic residues and therefore does not prime heparan sulfate. Searching the Wisconsin gene bank for sequences consisting of (Ser-Gly) n Ն 1 and a nearby cluster of acidic amino acids ((D/E) n Ն 3) within 6 residues yielded 15 out of 16 heparan sulfate proteoglycans cloned to date. The search also yielded proteins from bacteria, viruses, parasites, and eukaryotic subcellular organelles. Cell surface and secreted proteins, including agrin, a tyrosine kinase receptor, prostatic spermine-binding protein, and sporozoite surface antigen have a high chance of bearing heparan sulfate chains based on their location on the surface or outside the cell. Among them, agrin was shown recently to bear heparan sulfate chains (28). Interestingly, a chimera prepared from the tyrosine kinase receptor sequence (M35196) primed heparan sulfate, suggesting that the native protein might contain a heparan sulfate chain.
Biosynthetic Capacity Affects GAG Composition-A few heparan sulfate proteoglycans lack the repetitive Ser-Gly sequences of syndecan-1 or the aromatic residue found in betaglycan (e.g. proline-rich proteoglycan), suggesting that other enhancing factors may exist. One possibility is that the proportion of chains depends on the relative capacity of cells to produce GAG chains. In ldlD cells, which have reduced ability to make chondroitin sulfate, the relative proportion of heparan sulfate increases at the expense of the chondroitin sulfate chains (Table I). Some sites that do not make much heparan sulfate will even become active when expressed in ldlD cells (Table I). Conversely, the proportion of chondroitin sulfate increases dramatically in pgsD mutants, defective in heparan sulfate synthesis (6,29).
Proteoglycans vary in composition in different tissues and cells. For example, serglycin contains chondroitin sulfate chains when expressed in various myeloid cells (30) and a mixture of heparin and chondroitin sulfate chains when ex- Protein A chimeras containing the following proteoglycan segments were introduced into wild-type CHO cells and analyzed for heparan sulfate as described in Table I: decorin, EDEASGIGPEVPDDRD; thrombomodulin, DSGKVDGGDSGSGEPPPSPTPGST; fibroglycan, DMYLDSSSIEE-ASGLYPIDDDDYSSASGSGAYEDKGSPDLTTSQ; epican, EDERDRHLSFSGSGIDDDEDFIS; glypican, DFQDASDDGSGSGSGDGCLDDLCG;  syndecan-3, DDELDDIYSGSGSGYFEQESGLE; syndecan-3, DLEVPTSSGPSGDFEIQEEEETT; N-syndecan, TTTQDEPEVPVSGGPSGD-FELQEE; perlecan, DDEDLLADDASGDGLGSGDVGSGDF; proline-rich-proteoglycan, DENGDGDDNDDGDDDGSGDDVN; tyrosine kinase receptor, YFIVNVTDALSSGDDEDDNDGSED; betaglycan, SPGDSSGWPDGYEDLE and DSIGWPDGYEDLESGDNGFPGDGDEG. Data for syndecan-1 were taken from Table I  EEEOETSGDFGSGGSVVLLDDLEY ND pressed in connective tissue mast cells. 5 Syndecan-1 also varies in composition in different cells (31) and in response to growth factors (32). No systematic study of the biosynthetic enzymes has been undertaken in these tissues, but it is reasonable to assume that changes in enzyme expression could alter GAG composition. Thus, the actual complement of chains borne by a proteoglycan depends on biosynthetic capacity as well as permissive elements embedded in the core protein sequence.
The Amount of Core Protein Can Affect GAG Composition-The amount of core protein substrates passing through the biosynthetic compartments may also affect GAG composition. Stable transfectants expressing a syndecan chimera overproduce GAG and cause a decline in the proportion of heparan sulfate (Table II). The effect was not limited to the chimera, since endogenous proteoglycans also contained less heparan sulfate chains. Stable transfectants expressing syndecan-1 and betaglycan show similar effects (1,2,4). Recent studies suggest that core protein sequences outside the GAG attachment sites may regulate the fine structure of the chains as well (33).
Although the proportion of heparan sulfate declines, the actual amount of material increases after transfection. The data presented in Table II show that wild-type cells produced about 3.2 ϫ 10 6 cpm of [ 35 S]heparan sulfate (4.4 ϫ 10 6 cpm ϫ 72%). The stable transfectant produced 8.1 ϫ 10 6 cpm of [ 35 S]heparan sulfate (1.5 ϫ 10 7 ϫ 54%). This 2-3-fold increase in heparan sulfate synthesis is offset by a much larger increase in chondroitin sulfate synthesis, causing the relative composition to change. The compositional difference mimics the effect of ␤-D-xylosides, which depress the proportion of heparan sulfate because the primers preferentially stimulate chondroitin sulfate assembly (25).
A Model for Heparan Sulfate Biosynthesis-The above information suggests a model for controlling GAG composition, in which a key enzyme in the biosynthetic pathway senses permissive elements of the core and controls if heparan sulfate assembles. The identity of this enzyme emerged from earlier enzymatic studies showing that a unique ␣-GlcNAc transferase (␣-GlcNAc TI) acts on the tetrasaccharide intermediate, -GlcA␤1-3Gal␤1-3Gal␤1-4Xyl-protein (5,6). This unique transferase may bind to structural elements in the core protein and select specific linkage tetrasaccharides for heparan sulfate assembly. In betaglycan and related proteoglycans, the recognition element consists of the cluster of acidic residues and the adjacent Trp, which together may determine the affinity of the substrate for the transferase. Analogous interactions between distal glycosyltransferases and nascent glycoproteins occur during the addition of GalNAc to GlcNAc on the termini of N-linked oligosaccharides of glycohormones (34) and mannose 6-phosphate to the termini of chains found on lysosomal glycoproteins (35). In syndecan-1, the mechanism involves a cluster of acidic residues and perhaps a hydrophobic pocket defined by aromatic and aliphatic residues. In addition, a coupling phe-nomenon occurs across nearby Ser-Gly attachment sites. The juxtaposition of two or more sites may raise the probability that ␣-GlcNAc TI will act on adjacent linkage fragments. Thus, initiation of the heparan sulfate chains may involve both sequence recognition and a quasi-processive mechanism with regard to the carbohydrate acceptors.