A 35-kDa Protein Is the Basic Unit of the Core from the 2 (cid:51) 10 4 -kDa Aggregation Factor Responsible for Species-specific Cell Adhesion in the Marine Sponge Microciona prolifera*

Dissociated sponge cells quickly reaggregate in a spe- cies-specific manner, differentiate, and reconstruct tissue, providing a very handy system to investigate the molecular basis of more complex intercellular recognition processes. Species-specific cell adhesion in the ma- rine sponge Microciona prolifera is mediated by a supramolecular complex with a M r (cid:53) 2 (cid:51) 10 7 , termed aggregation factor. Guanidinium hydrochloride/cesium chloride dissociative gradients and rhodamine B iso-thiocyanate staining indicated the presence of several proteins with different degrees of glycosylation. Hyaluronate has been found to be associated with the aggre- gation factor. Chemical deglycosylation revealed a main component accounting for nearly 90% of the total pro- tein. The cDNA-deduced amino acid sequence predicts a 35-kDa protein (MAFp3), the first sponge aggregation factor core protein ever described. The open reading frame is uninterrupted upstream from the amino termi- nus of the mature protein, and the deduced amino acid sequence for this region has been found to contain a long stretch sharing homology with the Na (cid:49) -Ca 2 (cid:49) exchanger protein. A putative hyaluronic acid binding do- main and several putative N - and O -glycosylation sig-nals are present in MAFp3, as well as eight cysteines, some of them involved in intermolecular disulfide bridges. Northern blot suggest variable expression, and Southern blot analysis reveals the presence of other related gene sequences. According to the respective lecular one aggregation factor contain suggesting that sponge cell adhesion might be based on the assembly of multiple small glycosylated protein subunits. Bio-Rad.

Sponges have been traditionally used as models to study cell adhesion, since their rather loose and porous extracellular matrix allows a mild cell dissociation and the recovery of intercellular structures in virtually native state. Three basic components were shown to be necessary to aggregate sponge cells (Humphreys, 1963), the cell surface, calcium ions, and an extracellular complex termed aggregation factor. Sponge aggre-gation factors are proteoglycan-like molecules showing either a linear appearance (Halichondria, Haliclona, Terpios) or a closed, sunburst-like morphology (Microciona, Geodia). Their active participation in species-specific cell-cell and cell-matrix interactions (Gramzow et al., 1988a) is in contrast to the rather passive mechanical functions generally ascribed to proteoglycans, although growing evidence is accumulating in favor of their relevant role in differentiation and proliferation of cells (Ruoslahti, 1988). Variability in the glycosaminoglycan moiety of proteoglycans seems to correlate with changes in adhesiveness of the cells to other cells or to the substrate (Sanderson et al., 1994;Roughley et al., 1993), much in the same way as it has been described for the neural cell-adhesion molecule, where electrostatic interactions between charged glycosaminoglycan chains are thought to modulate the strength of cell-cell interaction (Rothbard et al., 1982). Variability in the sizes of proteoglycan core proteins has been described as well (Marynen et al., 1989;Doege et al., 1991), although the physiological relevance of such changes has not yet been explained.
The Microciona prolifera aggregation factor (MAF) 1 is a M r ϭ 2 ϫ 10 7 complex containing roughly equal amounts of protein and carbohydrate Cauldwell et al., 1973). MAF-promoted, species-specific sponge cell adhesion involves 1) a Ca 2ϩ -dependent MAF self-interaction site, and 2) a Ca 2ϩindependent MAF-cell binding (Jumblatt et al., 1980). Highly polyvalent structures have been shown to be involved in both functional domains (Misevic and Burger, 1986;Misevic et al., 1987), and carbohydrate-carbohydrate interactions play a central role in MAF self-association activity (Misevic et al., 1987;Misevic and Burger, 1993). Carbohydrate groups are involved in specificity (Turner and Burger, 1973;Misevic and Burger, 1990), at least in the cell binding site, although whether such specificity lies in the glycan structure itself, in the distribution pattern of glycans on the protein backbone, or both, remains unknown. A biological activity for the protein component of MAF, other than being a mere scaffold for the attachment of glycans, is also possible. Specific matrix interactions mediated by proteoglycan core proteins have been demonstrated (Schmidt et al., 1987;Heremans et al., 1990;Yamaguchi et al., 1990), and convincing arguments that biological activity resides in certain proteoglycan core proteins are also appearing (Kjellén and Lindahl, 1991;Templeton, 1992). Although progress has been made in unraveling the structure of the glycan moiety of MAF (Misevic andBurger, 1990, 1993;Spillmann et al., 1993), very limited information is available concerning the * 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.
The nucleotide sequence (s)  protein components. The complexity of sponge cell aggregation is suggested by the finding of many proteins associated with the aggregation factor, among them sialyltransferases (Mü ller et al., 1977), protein kinase C (Gramzow et al., 1988b), calpactin (Robitzki et al., 1990), and other bridge proteins that seem to link the factor to the cell membrane receptors (Gramzow et al., 1986;Varner, 1995). Nevertheless, the nature of the main protein component of the aggregation factor core structure has remained elusive.

EXPERIMENTAL PROCEDURES
Isolation of MAF, Total RNA, and Genomic DNA from Sponges-Live specimens of the red sponge, Microciona prolifera, were collected in the Woods Hole area (MA). Aggregation factor was isolated as described by Misevic et al. (1987). Total RNA from freshly dissociated sponge cells was prepared according to Chomczynski and Sacchi (1987). A modification of this protocol was used to obtain genomic DNA; 0.1 ml from a pellet of freshly dissociated sponge cells was added in a 2-ml tube to 0.9 ml of solution D (4 M guanidinium thiocyanate, 25 mM sodium citrate, pH 7, 0.5% sarcosyl, 0.1 M 2-mercaptoethanol) and gently mixed by inversion until the viscous solution became homogeneous. The proteins were then extracted with a 1:1 mixture of phenol/chloroform equilibrated with TE (10 mM Tris-HCl, 1 mM EDTA), pH 8. After a second extraction with chloroform/isoamyl alcohol 24:1, the DNA contained in the water phase was ethanol-precipitated and stored at 4°C in TE, pH 7.5.
Dissociative Gradient Fractionation of MAF-In general, the protocol described by Sajdera and Hascall (1969) was followed. To a MAF solution in H 2 O, guanidinium hydrochloride and CsCl were added to final concentrations of 4 M and 50% (w/v), respectively. The resulting volume was centrifuged at 10°C in Beckman quick-seal tubes for 12 h at 48,000 rpm in a VTi 65 vertical rotor (Beckman). Fractions were collected, dialyzed against H 2 O before electrophoresis, and finally stored at Ϫ20°C.
Analytical Methods and Sequencing of Protein and DNA-Protein was determined with the DC Protein Assay from Bio-Rad. Carbohydrate determination was performed as described by Trevelyan and Harrison (1952). Alkaline ␤-elimination data were obtained following the protocol of Antonsson et al. (1989). Amino acid determination was carried out according to Knecht and Chang (1986). Peptides released after cyanogen bromide or trypsin treatments were separated by reverse phase-high performance liquid chromatography in a C-18 column (Vydac), with a 3-h gradient between 0.1% trifluoroacetic acid in water and 0.08% trifluoroacetic acid in water/CH 3 CN, 30:70. For amino-terminal sequencing, Edman degradation was performed in a 477A Protein Sequencer (Applied Biosystems), and the products of each cycle were analyzed in a 120A PTH Analyzer (Applied Biosystems). DNA inserts cloned in pCR TM II or pBluescript SK(Ϫ) were sequenced with an A. L. F. Automatic Sequencer (Pharmacia Biotech, Inc.), using fluorescein-labeled primers.
Enzyme Treatments and Chemical Deglycosylation-PNGase F digestions were performed in 25 mM potassium phosphate, pH 7.5, 12.5 mM EDTA, 1% 2-mercaptoethanol. To denature the molecules prior to deglycosylation, the samples were heated (100°C, 3 min) in the presence of 0.1% SDS. Nonidet was then added to a final concentration of 1% to avoid denaturation of the enzyme by SDS. In general, 0.2 units of PNGase F (Boehringer Mannheim, recombinant N-glycosidase F from E. coli) were used to digest samples containing between 1 and 10 g of protein. Incubation times ranged from 1 to 12 h at 37°C. 0.4 g of hyaluronidase preparation (Sigma, type VI-S from bovine testes) were typically used to digest samples containing between 0.1 and 1 g of carbohydrate in 10 mM citrate, pH 5, with overnight incubation at 37°C. Treatments with cyanogen bromide (Merck) were used according to Hanson and Bentley (1983). Trypsin (Promega, sequencing grade) was used to digest between 1 and 10 g of protein in 0.1 M NH 4 HCO 3 , pH 7.7. The enzyme concentration was about 10 g/ml, and incubation was carried out at 37°C for 3 h. TFMS deglycosylation was performed according to Edge et al. (1981). HF treatments were carried out in a closed system (Peptide Inst. Inc., Minoh-Shi, Osaka, Japan), as described by Mort and Lamport (1977). In both procedures, 1-10 mg of MAF previously dialyzed against H 2 O and lyophilized were incubated at 0°C for the times indicated. TFMS and HF were purchased from Fluka. Anisole (Fluka) was added as scavenger. TFMS reactions were performed under N 2 in 10-ml glass Pierce Reactivials. The proteins deglycosylated with either method were freed of reagents and low molecular weight sugars by addition of a 2-fold excess of diethyl ether (Merck) cooled to Ϫ80°C. To the clear solution an equal volume of ice-cold 50% (v/v) aqueous pyridine (Merck) was added, and the ether phase was discarded. After a second ether extraction, the aqueous phase was extensively dialyzed against water, speed-vac concentrated, and finally stored at Ϫ20°C.
Construction of a Sponge cDNA Library and Selection of Positive Clones-Poly(A) ϩ RNA selection from sponges collected during July was done according to Sambrook et al. (1989). A Lambda ZAP II cDNA library was made using the ZAP-cDNA Synthesis Kit and Gigapack II Packaging Extract from Stratagene (1993). To generate a probe suitable for screening the library, the information contained in the aminoterminal sequence of MAFp3 was used to design two non-overlapping, completely degenerated nested primers: 50.1 (5ЈTT(TC)AC(GATC)GT-(GATC)CC(GATC)AT(ATC)TA(TC)GT3Ј) and 50.2 (5ЈGA(GA)GA(TC)-CA(GA)(TC)T(GATC)GA(TC)GC(GATC)ATG3Ј) derived from the peptide sequences FTVPIYV, and EDQLDAM, respectively. Using 1 l straight from the cDNA library as template, a first PCR round was performed with primers 50.1 and T7. One l of this PCR reaction was used as template for a second PCR round using primers 50.2 (internal to 50.1) and T7. The only major broad band, of about 1.2 kb, visualized in 2% agarose gels after the second round, was excised, and the DNA was extracted (QIAEX II Gel Extraction Kit, QIAGEN) and subcloned in pCR TM II (TA Cloning Kit, Invitrogen). One of the resulting inserts was sequenced, and after confirming that it contained the right sequence, a third PCR reaction was performed using sponge cDNA as template, with primers 50.1 and MAFp3-A (5ЈCGGTATTGAGGTGGT-TGGTT3Ј), derived from the sequenced clone, and binding near the 3Ј end of the putative open reading frame. Again, the resulting single band was excised, extracted, subcloned, and sequenced as described. The deduced amino acid sequence was in agreement with the residues from the amino-terminal end of the sought-after protein. To generate a probe, nondegenerated oligomers from each end of the predicted coding sequence were chosen, MAFp3-A and MAFp3-S (5ЈTTTACTGTTC-CGATTTACGT3Ј), and used as primers for a PCR reaction with sponge cDNA as template. DIG-dUTP (Boehringer Mannheim) was included in the reaction mixture with a ratio DIG-dUTP/dTTP 1:3. The probe obtained (MAFp3-DIG) was analyzed in a 1% agarose gel, and after confirming the presence of a labeled band with the expected size, the PCR reaction volume was stored at Ϫ20°C and used to screen the sponge cDNA library at a working dilution 1:2000 in hybridization buffer. The library was screened according to the instructions in the ZAP-cDNA Synthesis Kit from Stratagene (1993). Plaques (10,000 per 14-cm plate) were transferred to Hybond-Nϩ membranes (Amersham Corp.). Hybridization, stringency washes, and detection were performed following the instructions in the DIG DNA Labeling Kit from Boehringer Mannheim. From the positive plaques, pBluescript SK(Ϫ) was excised from Lambda ZAP II using the In Vivo Excision Protocol (ExAssist/SOLR System, Stratagene, 1993).
PCR Amplifications, Northern and Southern Blot Analysis-In general, the procedures described by Sambrook et al. (1989) were followed. Taq DNA polymerase, dNTPs, and PCR buffer were purchased from Boehringer Mannheim. PCR reaction conditions were those specified by the supplier. As a rule, 30 cycles were programmed in a DNA Thermal Cycler 480 (Perkin-Elmer): 94°C, 1 min/58°C, 2 min/72°C, 2 min. For Northern blots, 30 g of sponge total RNA were loaded per lane in denaturing 1% agarose gels containing formaldehyde. The amount of RNA was calculated by measuring absorbance at 260 nm. To confirm that equal amounts were being compared, the membranes were stained with methylene blue (0.04% solution in 0.5 M sodium acetate, pH 5.6) and destained in water, and the lanes exhibiting similar rRNA amounts were finally densitometered, only considering those differing less than 5%. For Southern blots, about 10 g of sponge genomic DNA were digested with DraI (Boehringer Mannheim), under the conditions spec-ified by the supplier, during 4 h with constant gentle shaking. After electrophoresis, the nucleic acids were transferred to a positively charged nylon membrane (Boehringer Mannheim) by capillary transfer and hybridized to MAFp3-DIG (Southern) or, for Northern blots, to an RNA probe labeled with the DIG RNA Labeling Kit (Boehringer Mannheim), using T7 RNA polymerase and pBluescript SK(Ϫ) containing MAFp3 cDNA as template. This construct was excised from a positive clone having a 1.2-kb insert that spanned the region between the phenylalanine in position ϩ3 and the poly(A). Previous to in vitro transcription, the vector was digested with EcoRI, analyzed in an agarose gel, and the linearized molecule extracted as described. Hybridization, stringency washes, and detection were performed as described by Boehringer Mannheim. Molecular sizes were determined relative to molecular mass markers (Boehringer Mannheim).

Characterization of MAF Subunits-SDS-PAGE of native
MAF gave rise to a series of carbohydrate-containing bands, visualized after combined Alcian blue/silver staining. With 1% SDS in the sample loading buffer, the main components entering the gel were ϳ210-kDa (S1) and ϳ2 ϫ 10 3 -kDa (S2) subunits (based on protein and hyaluronate standards, respectively), although a lot of material was retained by the 3% stacking gel (Fig. 1A). This pattern was not changed in the presence of reducing agent (Fig. 1B). Using the fluorescent dye RBITC, both bands and the excluded fraction have been shown to contain protein. Amino-terminal sequencing of the bands excised from Western blots, without previous deglycosylation, yielded the sequence NELIDYETFSDGRVL for S1, and the less defined -I-NLL-AL for S2. Protein data base searches did not reveal any significant similarities. Attempts to identify larger structures in 1.3-10% polyacrylamide gradient gels (Vilim and Krajickova, 1991) or in agarose/polyacrylamide hybrid gels (McDevitt and Muir, 1971) failed in our hands. In nondenaturing 1 ϫ TB gels, native MAF did not generate any subunits entering the gel, and only after dilution with H 2 O, heating (100°C, 3 min), or overnight incubation in the presence of 5 mM EDTA, a pattern similar to that of SDS gels was observed (not shown). These data confirmed the existence of noncovalent bonds and the importance of Ca 2ϩ ions in maintaining the native structure.
Cesium chloride density gradients in the presence of 4 M guanidinium hydrochloride were able to dissociate the MAF complex (Fig. 1B). The bottom of the gradient (fraction 1) contained S1, which might correspond, according to its density, to a proteoglycan monomer. Direct sequencing of fraction 1 yielded the same amino-terminal sequence already obtained for S1 from Western blots. The center of the gradient (fraction 2) represented 90% of the total protein and carbohydrate (Table I) and contained the excluded material from polyacrylamide gels. Re-loading of fraction 2 in a second dissociative gradient did not further release S1. The upper part of the gradient (fraction 3) contained several minor bands visible in 15% polyacrylamide gels after silver staining with a prominent 18-kDa protein (Fig.  1C). S2 was distributed between fractions 2 and 3, depending on each preparation. Alcian blue staining revealed a hyaluronidase-sensitive component seen as a diffuse band of low electrophoretic mobility in 5% SDS-PAGE (Fig. 1D), which copurified with fraction 2.
However, S1, S2, and fraction 3 accounted for no more than 15% of the protein in MAF, with the remaining 85% found in the aggregates not entering the gel. This excluded material being heavily stained with Alcian blue; chemical deglycosylation was chosen as an attempt to remove glycan chains that might be preventing the analysis of a major undetected protein, since enzymatic deglycosylation was never successful in quantitatively cleaving glycosaminoglycan structures from MAF (Spillmann et al., 1993;Misevic andBurger, 1990, 1993).
Chemical Deglycosylation of MAF-HF and TFMS deglycosylation products of fraction 2 from the dissociative gradient, containing 90% of the total protein, were analyzed in reducing SDS-PAGE, yielding a series of bands between 30 and 60 kDa ( Fig. 2A), visible after Coomassie Blue, Amido Black, or silver protein staining. When deglycosylated samples were run under nonreducing conditions, none of the aforementioned bands was observed (not shown), and aggregated material appeared retained by the stacking gel, suggesting the presence of a large structure stabilized by disulfide bonds. As a control, loading in the same gel of identical protein amounts of nondeglycosylated fraction 2 did not reveal any protein bands, either under reducing or nonreducing conditions. Longer deglycosylations, up to 3 h, enriched the smaller fractions, mainly consisting of a broad band around 33 kDa. The upper bands were slightly stained by Alcian blue, but those under ϳ40 kDa were not, suggesting differences in carbohydrate content originating from incomplete deglycosylation. To confirm that the observed bands represented an integral part of the MAF complex, the reaction products of a brief HF deglycosylation were immobilized on a membrane and decorated with the monoclonal antibody Block 1, which recognizes a carbohydrate epitope involved in sponge cell aggregation (Misevic et al., 1987;Spillmann et   All bands under 60 kDa resulting from both mild and extensive deglycosylations were transferred to a membrane, excised separately, and their amino termini determined, obtaining always the same sequence, shown in Fig. 2C. Sequencing of the smeared band seen under 29 kDa yielded many secondary amino acid peaks in each cycle, suggesting that this region contained peptides hydrolyzed from the main 33-kDa protein, as a possible consequence of the 3-h-long TFMS deglycosylation reaction. The reported amino terminus is different from those contained in S1 and S2 subunits, indicating the presence of a third, major protein component in MAF. The deglycosylation products, after dialysis against H 2 O, formed an insoluble precipitate containing virtually all the starting protein and very little carbohydrate, with a protein to carbohydrate ratio of 55:1. The supernatant was entirely devoid of protein and contained only traces of carbohydrate. The protein precipitate was partially dissolved by SDS and almost completely dissolved after reduction with 2-mercaptoethanol. After deglycosylation, all the material loaded entered the gel and separated when analyzed in reducing conditions, showing that there was no other major unresolved protein. Combined Alcian blue/silver stain did not reveal any carbohydrate-containing bands. The amino acid composition of fraction 2 before and after deglycosylation was very similar to that of native MAF (Table II), confirming that significant damage to the protein backbone did not occur. It was interesting to note the detection of hydroxyproline (0.3% of total amino acid), an anchoring point for O-linked glycosaminoglycans in plants (Showalter and Rumeau, 1990).
The data presented above indicate that the peptide sequence from Fig. 2C represents the amino terminus of the main protein from Microciona aggregation factor, which we will name MAFp3.
Deduced Amino Acid Sequence of MAFp3-Completely degenerated oligonucleotides were designed according to the amino terminus of MAFp3 and used as primers for PCR amplifications intended to generate a probe for library screening. The largest clone obtained is shown in Fig. 3. A second, shorter positive clone, confirmed the sequence from residue Ϫ180. As in other sponge cDNAs (Pfeifer et al., 1993), the typical polyadenylation signal AATAAA is missing. PCR reactions using primers MAFp3-A and MAFp3-S revealed products of identical length when using either cDNA or genomic sponge DNA as template (not shown), indicating that the region covered by the probe is intronless. The proline corresponding to the first residue from the amino-terminal peptide reported in Fig. 2C is designed as position ϩ1. The predicted sequence of 315 amino acids would have a molecular mass of 35.2 kDa, in good agreement with the size range of the bands observed after MAF deglycosylation.
Since one of the peptide bonds more sensitive to acidic conditions is Asp-Pro (Landon, 1977), and precisely those are the residues Ϫ1/ϩ1 from the sequence reported in Fig. 3, one might expect that cleavage of a longer native protein occurs at that position as a consequence of HF and TFMS treatments. In such a case, the 33-kDa main protein band shown in Fig. 2A would not contain the real amino terminus. Nevertheless, deglycosylations in this work have been carried out at 0°C, during, at most, 3 h, while aspartyl-prolyl bonds begin to be extensively cleaved at considerably higher temperatures (37-40°C), after extended incubation times (over 24 h), and preferably with the presence of a denaturing agent in the acid solution (Landon, 1977). Under conditions similar to those used by us, TFMS and HF deglycosylations of fetuin, a glycoprotein containing several Asp-Pro bonds, did not affect the peptide backbone (Mort and Lamport, 1977;Edge et al., 1981). Moreover, taking into account that MAFp3 sequence contains two more such bonds (amino acid residues 206/207, and 245/246), those should have been cleaved as well, yielding a main product well under 33 kDa, which was not observed. Therefore, we conclude that significant Asp-Pro cleavage has not occurred.
The sequence ANVLVGALNVTMVNII (positions Ϫ31/Ϫ16), could be a membrane spanning region, since it contains 12 nonpolar and 4 polar uncharged residues (3 Asn, 1 Thr). According to Engelman and Steitz (1981), asparagine and, especially, threonine are not unfrequently found buried in membranes. This sequence is followed by a more polar stretch, as in  the generic pattern for signal peptides, although these are highly variable and rapidly evolving structures (von Heijne, 1990). Two internal peptide sequences from MAFp3 were identified after trypsin digestion of deglycosylated protein excised from Western blots (FVVMR, residues 152-156) and after cyanogen bromide treatment plus trypsin digestion of nondeglycosylated MAF (ISVPN-NDLTLV, residues 115-126). Data base searches did not find any significant similarities with other described protein or DNA sequences. Several putative Nand O-glycosylation sites are present. In particular, the first target sequence for N-linked glycans in positions 18 -20 might be glycosylated in vivo, since a blank was obtained during the amino-terminal sequencing in the Asn ϩ18 position (Fig. 2C). Alkaline ␤-elimination data reveal that 63% of Thr and 25% of Ser might be glycosylated in MAF. The open reading frame is not interrupted in the 5Ј direction, indicating that MAFp3 could be synthesized, at least in some cases, as part of a longer preprotein. This hypothesis is supported by Northern blot analysis (Fig. 4A), which revealed a dominant ϳ1.2-kb transcript in 14 out of 20 individuals col- lected during the year (lane 2), although in four cases abundant RNAs up to 8.5 kb were also recognized by the MAFp3-specific probe (lane 3). In the remaining two specimens no bands were detected (lane 1). A weak background presence of long transcripts was observed throughout the year, but the high transcription shown in lane 3 was only found in specimens collected in spring and early summer.
The ubiquitous ϳ1.2-kb band seen in Northern blots is probably the lower limit for the transcript size. The fact that the RNA probe strongly binds the 1.2-kb transcript suggests that this contains the coding sequence for MAFp3. Indeed, some of the positive clones isolated from the cDNA library (e.g. the one used to generate the RNA probe) had about that length and were identical to the 3Ј moiety of the sequence from Fig. 3, although we ignore whether they represent original short mRNAs or fragments of longer forms. Unfortunately, all short clones sequenced so far lack the 5Ј region upstream from Pro ϩ1, which might be different from that found in long messengers. Several methionines upstream from the amino terminus of the mature protein (Met Ϫ20 , Met Ϫ83 , Met Ϫ137 , and Met Ϫ180 ) are in a good position to be chosen as translation start (Kozak, 1995), assuming that the same requirements described for vertebrate ribosomes are also valid in sponges. This reinforces the idea that proteins larger than MAFp3 might be translated from longer mRNAs. The variety of RNA lengths observed might be explained by alternative splicing, although Southern blot results (Fig. 4C) indicate the presence of related DNA sequences that can represent the existence of several MAFp3related genes.

DISCUSSION
The aggregation factor from the sponge Microciona prolifera is a supramolecular structure containing hyaluronate and several glycosylated proteins. Since major impurities are discarded in the MAF preparation , those components must be held together in the native complex. The ϳ210-kDa subunit observed in SDS-PAGE is likely to be the same entity described by Varner et al. (1988) as a MAF-binding protein. The MAF complex then could be composed of several interacting units, which might make it difficult to draw the limits between what belongs to the factor and what associates with it. The presence of noncovalent bonds has been confirmed by PAGE and dissociative gradient results. Two proteins have been found to be associated with glycosaminoglycan-containing bands, but chemical deglycosylation revealed the presence of a distinct major protein component.
Electron microscopy (Humphreys et al., 1977) and atomic force microscope studies (Dammer et al., 1995) have shown that MAF has a sunburst-like structure, with a central ring of about 200 nm across and radiating arms, each 180 nm long. MAF-MAF self-binding has been found to be the result of carbohydrate-carbohydrate interactions . Therefore, since two MAF molecules seem to bind through their arms (Dammer et al., 1995), these might be enriched in carbohydrate, leaving for the ring most of the protein. Urea, SDS, or mercaptoethanol have little effect on MAF as long as Ca 2ϩ is present (Cauldwell et al., 1973). Humphreys et al. (1977) described that the aggregation factor was stable in 0.5% SDS, and only after 4 weeks of incubation in the presence of EDTA were all the arms dissociated, although the central ring, containing protein and polysaccharide, remained unaltered even after further treatment with SDS, dithiothreitol, and heating for 5 min at 100°C. Only a drastic 4-h treatment at 80°C in 40 mM EDTA and 5 M urea was able to fragment about 60 -80% of MAF in fractions entering a separating gel, with still as much as 20 -40% retained by the stacking gel (Misevic et al., 1982). We have found that chemical deglycosylation disrupted the aggregated material observed in SDS-PAGE, suggesting that glycosaminoglycan chains play a crucial role in maintaining MAF structure. This idea is strongly supported by the result of the brief HF deglycosylation shown in Fig. 2B, which demonstrates that MAFp3 units, otherwise so difficult to isolate, begin to be released immediately after glycan chopping off starts. Covalent link between proteins through glycosaminoglycan chains has already been described (Enghild et al., 1993), but we have no proof of such a bond existing in sponges. On the other hand, the high negative charge of MAF  might hamper the interaction with other negatively charged molecules like EDTA and SDS, thus limiting their disruptive effects. In such situations, Ca 2ϩ ions buried within MAF could be protected enough to form relatively stable ionic bonds with the abundant acid residues from MAF glycosaminoglycans and/or proteins, helping to stabilize the overall structure. In agreement with this hypothesis, two distinct groups of Ca 2ϩ -binding sites have been found in MAF, which might be related to the requirement of low Ca 2ϩ concentrations to maintain the subunit organization of the aggregation factor complex and of higher Ca 2ϩ concentrations for the factor to keep cells together (Cauldwell et al., 1973).
The presence of putative N-and O-glycosylation sites, PNGase F digestion data, and alkaline ␤-elimination results suggest that MAFp3 contains both N-and O-linked glycans, although its localization in an intermediate fraction of dissociative guanidinium hydrochloride/cesium chloride density gradients indicates a moderate degree of glycosylation. The carboxyl-terminal GSGLGSGIG (positions 293-301) closely resembles sequences found in rat serglycin (Bourdon et al., 1986), syndecan (Saunders et al., 1989), human fibroglycan (Marynen et al., 1989), and in glypican Fig. 5A). Two glutamate residues in positions Ϫ5 and Ϫ6 complete a good consensus sequence for glycosaminoglycan attachment (Bourdon et al., 1987). MAFp3 does not contain sequences, like RGD, often found in adhesive proteins (Kreis and Vale, 1993), thus leaving open the question whether the protein moiety of MAF participates actively in the binding of the factor or only carbohydrate structures are involved in the recognition process.
We have found 30% identity between the region enclosed by residues Ϫ149 and Ϫ16 from the sequence shown in Fig. 3 and the cytoplasmic domain of the Na ϩ -Ca 2ϩ exchanger (Fig. 5B), a membrane protein responsible for the maintenance of low intracellular Ca 2ϩ levels. Canine and rat isoforms of the Na ϩ -Ca 2ϩ exchanger exhibit for the same region a mere 51% identity (Nicoll et al., 1990;Li et al., 1994), which makes the similitude with the sponge sequence even more striking. We regard this homology as most suggestive since, besides the classical secretory vesicle pathway, an alternative route to export MAFp3 to the extracellular matrix could be its synthesis as a larger polypeptide together with a protein targeted to the membrane, to be cleaved afterward. The deglycosylation results shown in Fig. 2A have been regularly reproduced with aggregation factor isolated in all seasons from pools of several individuals, without ever detecting any other amino-terminal or deglycosylation products different from those presented under "Results," thus ruling out the existence of a longer MAFp3 variety that could be incorporated in the factor under certain conditions. Although the structure-function relationship of MAFp3 for sponge cell adhesion remains to be investigated, regulation of its expression might be related to the yearly life cycle (sexual and asexual reproduction, budding, repair after injury), environmental changes (temperature, light, salinity), or both. During winter, the sponge body degenerates and in spring it will grow again from groups of dormant cells. Since most of the physiological uproar takes place in spring and summer, a correlation between phenomena such as gemmulation, release of larvae or sperm cells, and the adhesiveness of the sponge tissue is to be expected. Our results showing clear differences in MAFp3 expression during this physiologically active period point in that direction.
Whether the several bands identified in Southern blots represent different forms of MAFp3 or other related proteins is currently under investigation in our laboratory. The variety of products observed after chemical deglycosylation of MAF is most likely the result of an incomplete removal of glycosaminoglycan chains, although the possibility of the existence of several MAFp3 forms sharing the same amino terminus should not be ruled out. Considering the protein content of MAF, between 250 and 300 MAFp3 units are expected to be found in a single molecule of aggregation factor. This is the number of pyruvate epitopes suggested to be present in each molecule (Spillmann et al., 1993), although this might be an underestimation due to the acid lability of the pyruvate group, which makes it sensitive to the hydrolysis conditions used for its preparation. If the three putative N-glycosylation sites in each MAFp3 molecule were substituted, that would give about 900 pyruvate epitopes per MAF molecule, a value very close to the 1100 sites suggested by Block 1 antibody binding assays (Misevic et al., 1987). PNGase F digestion releases a repetitive glycan from MAF, termed G-6, which should be represented about 950 times, and contains the cell binding site of MAF (Misevic and Burger, 1990). Since Block 1 antibody binds to G-6 (Misevic, 1989;Misevic and Burger, 1993), it seems reasonable to suggest that MAFp3 carries G-6 and, therefore, the pyruvate epitope that would be involved in the cell binding.
Highly polyvalent carbohydrate-carbohydrate interactions have been shown to mediate sponge cell aggregation. Our finding that the main protein of MAF is a relatively small 35-kDa molecule suggests that polyvalency at the protein level might also exist. Highly glycosylated proteoglycans consist of repeated sites in a large protein chain to which glycosaminoglycan substituents are attached. In the sponge aggregation factor, the whole protein itself seems to be the repeating unit, and its polymerization can bring along the carbohydrate polyvalency required for sponge cell aggregation. Yeast agglutinins are cell adhesion molecules sharing many characteristics with the sponge system; the high binding affinity of the sexual agglutination factor from Hansenula wingei was found to be the result of the additive effect of several of the individual binding sites, located on protein subunits interconnected through disulfide bonds (Taylor and Orton, 1971;Yen and Ballou, 1974). Saccharomyces cerevisiae a-agglutinin analogs are highly glycosylated disulfide-linked oligomers, containing a large core subunit mediating cell surface attachment and a small self-binding subunit, the total molecular weight being about 10 6 (Lipke and Kurjan, 1992). In reducing conditions, we have not detected any measurable decrease of MAF-mediated aggregation efficiency of sponge cells, although monomerization of the protein might be overcome by the existence of several active carbohydrate binding units on each single MAFp3, thus suggesting again the existence of strong interactions between glycosaminoglycan chains. Moreover, the presence of hyaluronate in the aggregation factor preparation suggests that binding of MAFp3 to hyaluronic acid might also occur. MAFp3 contains a cluster of basic amino acids (RRYRNRVR, residues 107-114), which could determine the attachment to hyaluronate (Hardingham et al., 1976;Lyon, 1986). This sequence resembles the hyaluronan binding motif B(X 7 )B, where B is either arginine or lysine and X is any nonacidic amino acid (Yang et al., 1994). We have observed that the addition of exogenous hyaluronate to aggregation factor preparations with low activity raised the aggregation efficiency of the factor to equal that found in the most active preparations. Our data indicating (i) that the main protein from the 2 ϫ 10 4 -kDa sponge aggregation factor is a relatively small 35-kDa molecule and (ii) the presence on MAFp3 of carbohydrate structures involved in the aggregation activity, suggest that the cooperative effect of multiple low affinity interactions required for MAF-mediated sponge cell aggregation might be based on the cross-linking of a small glycosylated protein into a large polymer, where the active carbohydrate sites would be correctly exposed.