Accumulation in Marine Sponge Grafts of the mRNA Encoding the Main Proteins of the Cell Adhesion System*

Specific cell adhesion in the marine spongeMicrociona prolifera is mediated by an extracellular aggregation factor complex, whose main protein component, termed MAFp3, is highly polymorphic. We have now identified MAFp4, an ∼400-kDa protein, from the aggregation factor that is translated from the same mRNA as MAFp3. The existence of multiple potential sites forN-glycosylation and calcium binding suggests a direct involvement of MAFp4 in the species-specific aggregation of sponge cells. The deduced partial polypeptide consists of a 16-fold reiterated motif that shows significant similarity to a repeat in an endoglucanase from the symbiontic bacterium Azorhizobium caulinodans and to the intracellular loop of mammalian Na+-Ca2+exchangers. Restriction fragment length polymorphism analysis indicated that the genomic variability of MAFp4 is high and comparable to that of MAFp3. Their combined polymorphism correlates with allogeneic responses studied in a population of 23 sponge individuals. Peptide mass fingerprinting of tryptic digests of the polymorphic MAFp3 bands observed on polyacrylamide gels after chemical deglycosylation of theMicrociona aggregation factor revealed that the variability detected on Southern blots at least partially reflects the individual variability of aggregation factor protein components. Polyclonal antibodies raised against MAFp3 strongly cross-reacted with a 68-kDa protein localized in sponge cell membranes. Immunohistochemical use of the anti-MAFp3 antibodies strongly stained a cell layer along the line of contact in allogeneic grafts. We show that the transcription level of the MAFp3/MAFp4 mRNA in sponge allo- and isografts is clearly increased in comparison with non-grafted tissue. These data are discussed with respect to a possible evolutionary relationship between cell adhesion and histocompatibility systems.

Specific cell adhesion in the marine sponge Microciona prolifera is mediated by an extracellular aggregation factor complex, whose main protein component, termed MAFp3, is highly polymorphic. We have now identified MAFp4, an ϳ400-kDa protein, from the aggregation factor that is translated from the same mRNA as MAFp3. The existence of multiple potential sites for Nglycosylation and calcium binding suggests a direct involvement of MAFp4 in the species-specific aggregation of sponge cells. The deduced partial polypeptide consists of a 16-fold reiterated motif that shows significant similarity to a repeat in an endoglucanase from the symbiontic bacterium Azorhizobium caulinodans and to the intracellular loop of mammalian Na ؉ -Ca 2؉ exchangers. Restriction fragment length polymorphism analysis indicated that the genomic variability of MAFp4 is high and comparable to that of MAFp3. Their combined polymorphism correlates with allogeneic responses studied in a population of 23 sponge individuals. Peptide mass fingerprinting of tryptic digests of the polymorphic MAFp3 bands observed on polyacrylamide gels after chemical deglycosylation of the Microciona aggregation factor revealed that the variability detected on Southern blots at least partially reflects the individual variability of aggregation factor protein components. Polyclonal antibodies raised against MAFp3 strongly crossreacted with a 68-kDa protein localized in sponge cell membranes. Immunohistochemical use of the anti-MAFp3 antibodies strongly stained a cell layer along the line of contact in allogeneic grafts. We show that the transcription level of the MAFp3/MAFp4 mRNA in sponge allo-and isografts is clearly increased in comparison with non-grafted tissue. These data are discussed with respect to a possible evolutionary relationship between cell adhesion and histocompatibility systems.
Species-specific cell recognition in sponges is mediated by proteoglycan-like extracellular complexes termed aggregation factors (AFs), 1 unidentified so far in other phyla, although Northern blots suggested the existence, in several human tissues, of proteins related to those found in the AF of the marine sponge Microciona prolifera (1). Cell adhesion in this species is mediated by calcium-dependent carbohydrate interactions of homologous Microciona AF (MAF) molecules via N-linked glycans. A proposed model suggested that species specificity could be at least partially provided by the different spacing of glycosaminoglycan chains on the protein core (2), thus postulating the existence of a different AF core protein in each species. Unexpectedly, an extraordinarily high intraspecific variability was found for the MAF core protein (1), suggesting that AFs might also be involved in individual specificity.
The immune and cell adhesion systems are clearly related, as deduced from phylogenetic relationships between molecules of both groups (3) and as expected from the fact that the cells of the immune system depend on regulated interactions with other cells to be activated (4). An evolutionary connection between cell adhesion and histocompatibility, however, has yet to be demonstrated. Presumably, it was around the time when metazoans appeared that the primitive machinery regulating cell interactions had to evolve and diversify into the whole array of molecules that allowed the far more complex structures leading to multicellularity first and then to tissues and organs. A second possibility is that the cell adhesion system of higher animals appeared de novo, and therefore, vertebrate and invertebrate histocompatibility would be the result of convergent evolution rather than of a common ancestry. The finding, in an ever increasing number of invertebrates including sponges, of Ig-like domains (5,6) and of extracellular matrix components common to all pluricellular animals (7) suggests that the latter view is less likely.
A surprising characteristic of sponges, considering their phylogenetic position as the most primitive extant metazoans, is that they possess a highly evolved histocompatibility system. When tissues from different individuals of a given sponge species are brought into contact, they either fuse or reject through cellular events similar to those observed in vertebrate grafts, which include (i) an inflammation-like massive migration of certain cell types toward the region of contact (8), (ii) phagocytosis and/or cytotoxic reactions (9,10), and (iii) the layering of a collagen barrier separating the apposed tissues (8). Any grafting between two genetically different sponge individuals is almost invariably incompatible in the many species investigated (1,9,11,12), exhibiting a variety of transitive qualitatively and quantitatively different responses, which can only be explained by the existence of an extensive genetic polymorphism at the locus or loci controlling graft acceptance and rejection. In the protochordate Botryllus schlosseri, contact between two genetically distinct individuals leads to the resorption of one of them by the other through a phenomenon that is controlled by highly polymorphic loci (13). The frequency of resorption in Botryllus is comparable to that of rejection in sponges, suggesting that both might be analogous manifestations of a sophisticated histocompatibility system in inverte-brates that would have been an ideal building block for the vertebrate major histocompatibility complex. And yet, despite the relevant cytological and genetic studies outlined above, there is an absolute lack of sequence and evolutionary data concerning invertebrate molecules involved in allorecognition.
Individual variability of sequences related to the main protein of MAF (MAFp3) matched the elevated sponge alloincompatibility (1), although from Southern blot results, it could not be established whether such variability resided in the AF proteins themselves or if it also represented sequence-related molecules not belonging to the AF. MAFp3 and another protein suspected to belong to the AF seemed to be translated together as a longer polypeptide from the same mRNA (1,14). Here, we characterize this new component and start studying the significance of the variability in MAFp3. We also present the first attempts to study what role, if any, AFs might play in sponge histocompatibility.

EXPERIMENTAL PROCEDURES
Sponges, Cell Membrane Preparation, and Western Blots-Live specimens of the red beard sponge (M. prolifera) were collected in the Woods Hole area by staff of the Aquatic Resources Division of the Marine Biological Laboratory. To isolate cell membranes, dissociated sponge cells obtained during AF preparation (see below) were washed twice with cold phosphate-buffered saline and incubated on ice for 10 min in hypotonic salt solution (5 mM KCl, 1 mM EDTA, and 10 mM HEPES, pH 7.5). Cells were then opened with 10 strokes in a Dounce homogenizer (type B pestle). The suspension was cleared for 5 min at 400 ϫ g, and the resulting supernatant was spun at 100,000 ϫ g for 20 min. The pelleted membranes were resuspended in storage buffer (1 mM MgCl 2 , 1 mM CaCl 2 , 10% glycerol, and 20 mM HEPES, pH 7.5) and stored at Ϫ80°C. Peripheral membrane proteins were extracted by incubation of dissociated sponge cells in 70 mM NaCl on ice for 1 h (15). After removing the cells by a short centrifugation, the resulting supernatant was immediately analyzed by Western blotting, which was performed as described (1). The rabbit polyclonal antibody raised against squid actin was provided by Prof. Dr. Ute G. Stewart (Institut fü r Zoologie, Technische Hochschule, Darmstadt, Germany).
cDNA Sequencing and Polyclonal Antibodies-To extend sequences in the 5Ј-direction, rapid amplification of cDNA ends (RACE) was performed using two different templates. Preexisting cDNA libraries yielded short 5Ј-RACE products with primer T3 and a gene-specific primer corresponding to an antisense oligonucleotide close to the 5Ј-end of the longest sequences available (1). Based on Northern blot results, total RNA samples were then chosen for their content of the largest forms of the MAFp3 mRNA. Poly(A) ϩ RNA was selected with standard methods (16) and used to construct oligo(dT) cDNA libraries with the Marathon TM cDNA amplification kit (CLONTECH). 5Ј-RACE products were obtained with antisense gene-specific primers and the adaptor primer supplied by the manufacturer. Polymerase chain reaction amplifications were done with the Expand TM Long Template PCR system (Boehringer Mannheim). The products were analyzed on 1% agarose gels and excised, and the DNA was extracted (QIAEX II gel extraction kit, QIAGEN Inc.), subcloned in pCR TM II (TA cloning kit, Invitrogen), and sequenced with an automated laser fluorescent sequencer (Amersham Pharmacia Biotech) using fluorescein-labeled primers. Between the gene-specific primer 3Ј-end and the 5Ј-end of the region to be extended, a stretch of at least 25 bases was left for comparison, only accepting those new sequences that showed a 100% match. Chicken polyclonal IgYs against the recombinant 170-amino acid N-terminal region of MAFp3 were produced and purified according to standard protocols (17,18). The fusion protein was generated with the glutathione S-transferase gene fusion system (Amersham Pharmacia Biotech). For the purification of the specific antibodies against MAFp3, the fusion protein was cross-linked with dimethyl pimelimidate to glutathione-Sepharose 4B as described (18). The preparation eluted from the antigen affinity column was monitored for purity by SDS-polyacrylamide gel electrophoresis and for its antigen-binding activity, and finally, it was quantitated. According to all these controls, we had an extremely pure specific antibody, which was able to detect its antigen on Western blots at a concentration of Ͻ1 ng of purified antibody/ml.
Southern and Northern Blots and in Situ Hybridization-Sponge genomic DNA was isolated as described (14), DraI-digested, electrophoresed, transferred to a membrane, and hybridized with probes labeled with alkali-labile digoxigenin as specified before (1). Probes I and III (both ϳ1 kb; defined in the legend to Fig. 1) were polymerase chain reaction-amplified from subcloned regions lacking DraI sites. Probe I was derived from an intronless cDNA sequence, and probe III from a genomic DNA sequence containing at least one intron. Before reprobing, the membranes were treated twice with 1.5 M NaCl and 0.3 M NaOH for 10 min each and reequilibrated with 1.5 M NaCl, 1 M Tris-HCl, pH 7.4, and 2ϫ SSC for 10 min each. To establish grafts, 1-cm-long sponge papillae were pushed together on a 0.5-mm-thick stainless steel insect pin until the contact between the tissues was as intimate as possible. During these operations, the sponge tissue was always kept under seawater. The pins were stuck into the underside of a Styrofoam rack left to float on a tank with running seawater at 20°C for the times indicated. Total RNA from grafted tissues was prepared using standard protocols (19). Briefly, the grafted papillae with a mean volume of 125 mm 3 were removed from the pins and gently squeezed with flat-tip forceps for 30 s in 1 ml of Ca 2ϩ -and Mg 2ϩ -free artificial seawater at 0°C in order to release the cells, which were immediately pelleted for 1 min at 200 ϫ g. 0.9 ml of the cleared supernatant was then removed, and the cell pellet was quickly but gently resuspended in the remaining 100 l before adding 6 volumes of 4 M guanidinium thiocyanate, 25 mM sodium citrate, pH 7, 0.5% sarcosyl, and 0.1 M 2-mercaptoethanol. The viscous solution was gently mixed until it became homogeneous, and the pH was then lowered with 0.1 volume of 2 M sodium acetate, pH 4. The proteins were extracted with 1 volume of phenol equilibrated with 0.2 M sodium acetate, pH 4, chloroform, and isoamyl alcohol (200:40:1). After 1 h on ice, the samples were centrifuged at 20,000 ϫ g for 30 min, and the RNA contained in the water phase was precipitated with 0.2 volume of isopropyl alcohol, washed twice with 80% ethanol, and resuspended in diethyl pyrocarbonate-treated H 2 O. For Northern blotting, 5 g of total RNA were loaded per lane on denaturing 0.8% agarose gels containing formaldehyde. RNA probes were labeled with the digoxigenin RNA labeling kit (Boehringer Mannheim) as described (14). In situ hybridization of 20-m-thick sections was performed as described (20) without prior desilicification of the tissue.
Immunohistochemistry-Grafted tissues were fixed with 3.7% formaldehyde in artificial seawater at the indicated times after being pinned together. Without removing the needle, grafts were immersed in the fixing solution and incubated overnight at 4°C. To dissolve the spicules, grafted tissues were desilicified according to Humason (21). Grafts were washed twice with artificial seawater and then with progressively increasing amounts of ethanol in artificial seawater (from 30 to 70%), with a final wash in 70% ethanol in water. Spicules were then removed by treatment with 1.5 N hydrofluoric acid in 70% ethanol for 4 -7 h at room temperature. Grafts were stored in 70% ethanol at 4°C until processing. Tissues were processed routinely with Richard-Allan dehydrants, cleared in xylene, and embedded in Fisher PARAPLAST ® X-TRA paraffin. The needles were removed once the paraffin hardened. 6-m sections were deparaffinized, rehydrated, and washed twice with Tris/NaCl buffer (30 mM Tris-HCl and 0.9% NaCl, pH 7.5). After a wash in 0.1 mM citrate buffer, pH 6.0, the sections were incubated in 0.3% H 2 O 2 for 30 min and washed again with Tris/NaCl buffer before blocking with 1.5% normal goat serum (Vector Laboratories, Inc., Burlingame, CA) in Tris/NaCl buffer for 30 min. The chicken polyclonal antibody Bertha 1 against MAFp3 or the mouse monoclonal antibody Block 2 against a carbohydrate epitope in MAF was applied overnight at 4°C in the blocking solution at a final concentration of ϳ1 g/ml. Subsequently, the sections were washed twice, incubated with second antibodies (biotinylated goat anti-chicken and goat anti-mouse, respectively; Vector Laboratories, Inc.) at a final concentration of ϳ10 g/ml for 30 min, rinsed with Tris/NaCl buffer, and incubated for 30 min with avidin/biotinylated peroxidase according to the instructions in the VEC-TASTAIN ® ABC kit (Vector Laboratories, Inc.). After washing with Tris/NaCl buffer, the sections were incubated with the peroxidase substrate diaminobenzidine (DAB single solution, Zymed Laboratories, Inc., South San Francisco, CA) for 1-5 min at room temperature before stopping the reaction in H 2 O. The sections were finally mounted with Kaiser's glycerol gelatin (Merck, Darmstadt, Germany).
Peptide Mass Fingerprinting and Amino Acid Sequencing-The AF isolated as described (22) was deglycosylated with trifluoromethanesulfonic acid according to standard procedures (23). After reduction and alkylation with iodoacetamide, the proteins were analyzed by SDSpolyacrylamide gel electrophoresis. Selected Coomassie Blue-or Alcian blue-stained bands were excised, and the protein therein was trypsinized. Tryptic peptides were analyzed by LC-MS as described (24) using a Rheos 5000 separation system equipped with a Vydac C8 column (1 ϫ 2500 mm). Peptides were eluted in a linear gradient of 5-50% buffer B (0.045% trifluoroacetic acid and 80% acetonitrile in H 2 O) in buffer A (0.05% trifluoroacetic acid and 2% acetonitrile in H 2 O) at a flow rate of 50 l/min. An aliquot of 10% of the effluent was directed on-line to an API 300 triple quadrupole mass spectrometer (PE Sciex, Concord, Canada), and 90% was collected for further analysis. The ion spray voltage of the mass spectrometer was set to 5000 V, and a mass range of 300 -2400 Da was scanned with a step size of 0.5 Da and a dwell time of 0.75 ms/mass. Sequence analysis was carried out on a Model 477A protein sequencer (Applied Biosystems, Inc., Foster City, CA) according to the recommendations of the manufacturer.

RESULTS
A New Protein Component of MAF-Of the several cDNA-derived forms described for the main protein of MAF, we decided to elongate, in the 5Ј-direction, MAFp3C and MAFp3D as representatives of two distinct groups identified by sequence relatedness (1). The sequences shown in Fig. 1a could not be extended further due to mismatches between the existing and newly obtained sequences. We interpret this result as a consequence of the variability of these proteins, which reduces the probability of amplifying the same form used to generate the antisense gene-specific primers.
Chemical deglycosylation of MAF yielded, as the main product, MAFp3, whose first identified cDNA predicted a polypeptide of 35 kDa (14), beginning with Pro-1883 in MAFp3C. Since this residue is immediately preceded by Asp, and the Asp-Pro bond is especially sensitive to acidic conditions (25) such as those of the deglycosylation reaction, the possibility exists that MAFp3 could be the result of chemical cleavage of a bigger protein rather than an independent entity in vivo. However, 36 other Asp-Pro bonds are present in the sequences from Fig. 1a, although we never detected any peptides arising from cleavage at those positions. This supports the view that our deglycosylation conditions do not cause extensive damage to the protein backbone, and therefore, MAFp3 is likely to be translated as part of a single peptide, which is then processed into at least two proteins. Later in this work, we confirm that the cDNA sequences extended in the 5Ј-direction correspond to a longer mRNA coding for a new protein component of MAF that we will term MAFp4.
MAF-mediated species-specific cell adhesion depends on polyvalent interactions between N-linked glycans (2,26). The high number of NX(S/T) consensus sequences found on MAFp4 confirms the expectation of extensive N-glycosylation, a hypothesis backed by the sequencing of several tryptic peptides that yielded a blank instead of a potentially glycosylated Asn (see Figs. 3b and 5c) as result of the inefficient cleavage by trifluoromethanesulfonic acid of protein-bound GlcNAc (23), which accounts for ϳ20% of the total carbohydrate content of MAF (22). MAF self-interaction is Ca 2ϩ -dependent, unlike its binding to the cell receptor (27). The abundant potential Ca 2ϩbinding sites of the type DXD (28) that are present in MAFp4 indicate that self-interaction might not involve only the acidic glycans, but also the protein core. MAFp4 contains a reiterated motif of a mean length of 117 residues (Fig. 1b). Data base searches revealed two significant matches (Fig. 1c). As already described by us (14), the MAFp4 repeat shares an ϳ30% identity with a stretch of the intracellular loop of the Na ϩ -Ca 2ϩ exchanger protein from mammals. In addition to this, we have now identified an 8-fold reiterated motif in a cDNA-deduced endoglucanase gene from the symbiontic bacterium Azorhizobium caulinodans ORS571 sharing a similar degree of identity with the MAFp4 repeats. The bacterial repeats are ϳ115 amino acids long and were also described to show significant similarity to the Na ϩ -Ca 2ϩ exchangers (29).
MAFp4 Is Highly Polymorphic-Allogeneic rejection in Microciona is clearly identified after ϳ6 h with the naked eye due to a yellowish line originating from the massive migration to the region of contact of gray cells, which have been suggested to be the functional immunocytes of sponges (8). We studied allo-geneic reactions in a population of 23 M. prolifera individuals with the result that, besides the control autografts, only individuals 14 and 15 did not reject each other (1). To test the genomic variability of MAFp4, the DraI-digested genomic DNA of this population was subjected to restriction fragment length polymorphism analysis using probes I and III, corresponding to MAFp3 and MAFp4, respectively. The result shown in Fig. 2 reveals an elevated variability in both regions. Individuals 14 and 15 are the only individuals sharing identical fingerprints on Southern blots using MAF-related probes. In our studies, we considered a sponge individual all the substance contained by a continuous pinacoderm, the epithelial cell layer. This assumption, which until now was supported only by studies of the sponge structure and function (30), is confirmed by our observation that DNA isolated from cells of different areas within a continuous pinacoderm yielded an identical restriction fragment length polymorphism pattern (data not shown).
Anti-MAFp3 Antibodies-We raised polyclonal antibodies against the amino-terminal half of MAFp3 and tested them on trifluoromethanesulfonic acid-deglycosylated MAF. As shown in Fig. 3a, anti-MAFp3 antibody recognized all major bands between 60 and 30 kDa that were stained with Coomassie Blue. This region of the gel had been already described to contain MAFp3 (14). Immunostaining was also observed in a wide region between 400 and 200 kDa and especially at the origin of the gel, where aggregated material was retained. These two areas were stained also with Alcian blue, indicating that they still contained carbohydrate. Based on standard protein and carbohydrate determination, however, the glycan content of the deglycosylated protein was Ͻ2%. Trypsin digestion of the weak Alcian blue-stained band of ϳ400 kDa, whose amino terminus was blocked, yielded complex LC-MS chromatograms (data not shown). 10 out of 11 of the most prominent peptides could be assigned to MAFp4C (Figs. 1a and 3b). The few mismatches are probably a reflection of the variability of the protein.
When anti-MAFp3 antibody was used to decorate Western blots of M. prolifera cell membranes, in ϳ20% of the individuals tested, a strong cross-reactivity was observed with a 68-kDa protein (Fig. 3c). Together with its scarcity, the apparent sensitivity of the 68-kDa protein to standard storage conditions prevented further analysis. Coomassie Blue or Amido Black staining indicated that the protein exists in very little amounts, and therefore, it is unlikely that the strong signal observed on Western blots results from unspecific binding to a totally unrelated protein. Whereas the 68-kDa protein was not detected on our Western blots after it had been freeze-thawed a few times, deglycosylated MAFp3 was not affected by this procedure, thus reinforcing the view that both proteins are different entities. A 43-kDa protein was also immunodetected by anti-MAFp3 antibody on Western blots of sponge cell membranes and in a fraction obtained upon incubation of the sponge cells in 70 mM NaCl. This protein could be easily isolated from SDS-polyacrylamide gel electrophoresis fractionation of the hypotonic salt-released proteins and in-gel digested with trypsin, and the resulting peptides were analyzed by LC-MS. Sequencing of the most prominent peptides confirmed that the 43-kDa protein was actin. As expected, sequence comparisons established its closest known relatives among echinoderm actins (Fig. 3d). Polyclonal rabbit antibodies raised against squid actin strongly cross-reacted with its sponge homolog (Fig. 3c). Although its finding in a hypotonic fraction that was supposed to contain external peripheral membrane proteins suggests that actin might be present extracellularly in sponges, we believe it is more likely that some cells burst during the exposure to 70 mM NaCl, releasing cytosolic proteins. However, the existence of extracellular actin has been reported in several , and therefore, it would not be surprising that a small fraction of actin molecules play some extracellular function in sponges. According to Coomassie Blue and Amido Black staining, actin is the most abundant protein in both cell membranes and 70 mM NaCl fractions, a circumstance that favors its apparent cross-reactivity with the purified anti-MAFp3 antibodies.
To investigate the existence of any structural features revealed by antibodies raised against MAF epitopes, 6-m sections of allogeneic and isogeneic M. prolifera grafts were incubated either with anti-MAFp3 antibody or with the monoclonal antibody Block 2, raised against a carbohydrate epitope of MAF (2). Anti-MAFp3 antibody clearly revealed a thin layer of cells all along the line of contact of allogeneic grafts after several hours of grafting (Fig. 4a). This reaction was not observed in isogeneic grafts (Fig. 4b). Allogeneic grafts at times under 10 h did not show such a pattern (Fig. 4c), indicating that the staining does not come from a mere apposition of epithelial cell layers that might be expressing the epitope before contact. On the contrary, the pinacoderm was not particularly stained by anti-MAFp3 antibody (Fig. 4, b-d). 24 h after grafting, the border line of cells was still easily identified (Fig. 4d), although the time interval during which the signal persisted was found to depend on the individual combination of the grafted tissues.
The staining observed with anti-MAFp3 antibody was not due to the accumulation of MAF, as indicated by routine controls done with monoclonal antibodies raised against carbohydrate epitopes of MAF (Fig. 4e). Additional controls were performed those reported in Figs. 3 and 5, identified after trypsin digestion. Polyclonal anti-MAFp3 IgYs were raised against the sequence marked in boldface. Sequence C was chosen to design probes I (between amino acids 1883 and 2205) and III (between amino acids 1 and 207) and the RNA sense and antisense probes for in situ hybridization (between amino acids 2053 and 2205). The antisense RNA corresponding to probe I was also generated for its use in Northern blotting. b, alignment of the 16 tandem repeats in MAFp4C. The scheme at the top shows the relative positions of the 16 consecutive repeats in the partial cDNA-deduced MAFp4 sequence. Bars indicate lengths in amino acid residues. In yellow are marked the positions where at least 8 of 16 residues are identical. c, alignment of repeat 6 from A. caulinodans endoglucanase (Egl 6; GenBank TM accession number Z48958), repeat 2 from MAFp4C (MAFp4C 2), and a region in the rat Na ϩ -Ca 2ϩ exchanger (NaCa rat; GenBank TM accession number P70549). In yellow are marked the positions where at least 2 of 3 residues are identical. ). An identical gel was stained first with Coomassie Blue and later with Alcian blue to reveal protein-and glycan-containing bands, respectively. The ϳ400-kDa Alcian blue-stained band was subjected to peptide mass fingerprinting. b, the sequences of the most prominent tryptic peptides were determined either by MS analysis and comparison with the cDNA-deduced sequences (in italics) or by direct sequencing (all others). The residues found in the sequence of MAFp4C (Fig. 1a) are indicated in boldface. Dashes indicate blanks in the sequence probably due to the presence of a modified amino acid. c, Western blots of M. prolifera cell membrane proteins (Membranes) and of peripheral cell membrane proteins (70 mM NaCl extract) fractionated on a 12.5% SDS-polyacrylamide gel and stained with Amido Black or decorated with anti-MAFp3 IgYs (anti-MAFp3) or with rabbit polyclonal antibodies raised against squid actin (anti-actin). d, alignment of the complete actin sequence from the sea urchin Strongylocentratus purpuratus (GenBank TM accession number P53472) with the main tryptic peptides obtained from M. prolifera actin excised from SDS gels. Nonconserved amino acids are boxed in the sponge sequence.
with sections prepared as those for in situ hybridization, which were not treated with hydrofluoric acid. The results were identical to those obtained with desilicified sections (data not shown), thus indicating that the treatment with diluted hydrofluoric acid, which is a deglycosylating agent, did not significantly remove the carbohydrate epitope recognized by Block 2 from fixed tissue.
Different MAFp3 Forms in AFs of Different Individual Origin-To investigate the possibility that MAFs of different origin might have different core proteins, we proceeded to chemically deglycosylate MAF preparations obtained from different individuals. A series of protein bands between 38 and 52 kDa were detected (Fig. 5, a and b). Samples of different individual origin treated in parallel reproducibly revealed specific combinations of bands, indicating that the observed polydispersity is a reflection of the in vivo situation rather than an artifact of deglycosylation. Anti-MAFp3 antibody decorated all bands with an intensity proportional to that of Coomassie Blue staining, suggesting that all the forms shared a significant degree of homology. To confirm this, we separately excised single bands from SDS gels, digested them with trypsin, and performed LC-MS analysis. The chromatograms shown in Fig. 5c confirm the existence of multiple related proteins. cDNA-deduced MAFp3 forms had been identified whose molecular masses differ between 1 and 5.4 kDa, mainly due to different carboxylterminal ends (1), a region for which we could not obtain peptide sequences from our trypsin digests. It is likely that major variability in this region will be responsible for the differences in electrophoretic mobility observed. Amino-terminal analysis of the 39-, 46-, 49-, 51-, and 52-kDa bands yielded, for all of them, the sequence of MAFp3 (PLFTVPI).
Accumulation of MAFp3/MAFp4 mRNA-Northern blots of sponge total RNA showed variable amounts of the MAFp3/ MAFp4 mRNA (Fig. 6a). The use of probe I revealed a main transcript of ϳ12 kb, in accordance with a translated ϳ400-kDa protein, which might correspond to full-length MAFp4 as suggested by the Alcian blue-stained gel in Fig. 3a. Often, the MAFp3-specific probe recognized abundant RNAs below 12 kb (Fig. 6a, top), suggesting the existence, under still undetermined conditions, of a great variability of MAFp3-related sequences. Occasionally, bands of ϳ1 kb were also observed, which are thought to represent small transcripts containing only the MAFp3 sequence (14). Accordingly, the use of probes corresponding to MAFp4 revealed the ϳ12-kb band, but never any signal around 1 kb (data not shown). To study the effect of tissue grafting on the transcription of MAFp3/MAFp4 mRNA, we chose those individuals whose levels of expression in nongrafted controls were constant and low during the duration of the experiment. A typical result is presented in Fig. 6b. Grafting of isogeneic or allogeneic tissue coincides with an accumulation of the MAFp3/MAFp4 mRNA. As shown by the ␤-actin control, this does not seem to be the consequence of a generalized enhanced transcription. Occasionally, the control samples removed at the end of the experiment from parts of the animal ϳ10 cm apart showed a dramatic accumulation of the MAFp3/ MAFp4 mRNA throughout the body, whereas tissue separated from the animal at the beginning of the experiment and kept under the same conditions as the grafts exhibited relatively constant low levels (Fig. 6c). Such accumulation was generally observed in the healthy areas of sponge individuals with patches of decaying tissue. In situ hybridization (Fig. 6, d and  e) showed that the MAFp3/MAFp4 mRNA was homogeneously distributed in the grafted tissues.

DISCUSSION
The correlation between the variability of MAF proteins as deduced from restriction fragment length polymorphism analysis and the allogeneic behavior within a sponge population supports a possible involvement of the cell adhesion system in sponge histocompatibility reactions (1). The accumulation in grafts of the mRNA coding for the protein core of the main adhesion molecule of the sponge adds to this hypothesis. The phenomena occurring in M. prolifera grafts can be first identified under the microscope ϳ1 h after grafting, when the pina-coderm begins to be disrupted in both rejecting and accepting grafts (data not shown). It is also at this moment when MAFp3/ MAFp4 mRNA starts to accumulate. The function of MAF in tissue grafts is at present obscure, although the increased transcription of its mRNA under adverse conditions such as necrosis in nearby tissue suggests some role in defense processes. The observations that a cell function such as transcription is so homogeneous in areas of the animal separated by Ͼ10 cm and that it can change dramatically in a time scale of hours indicate that sponges, despite lacking circulatory and nervous systems, must have an efficient way to quickly exchange information between cells separated by large distances.
In different individuals, the polymorphic MAFp3 forms were isolated in variable amounts relative to each other (see 51-and 49-kDa bands in individuals 1-3; Fig. 5, a and b), an unexpected result if all the cells of the species were always producing the same type of MAF. This result could indicate that the production of a given MAFp3 type depends on particular physiological states of the sponge, and therefore, certain MAFp3 forms would be preferentially present at different times of the life cycle. An alternative explanation suggests that different types of cells within the same individual produce different types of MAF, and therefore, different dosages of cell types in each preparation would explain the fluctuations in the relative amounts of MAFp3 forms. Atomic force microscope studies have shown that Block 2 binds, with very different affinities, individual AF molecules that were purified from a single sponge (32), and this is in agreement with the assumption that different types of AF molecules exist that exhibit distinct specific activities. Peptides containing sequences exclusive of MAFp3D or MAFp4D have never been found in the many AF preparations analyzed. The presence in sponge cell membranes of a 68-kDa protein related to the core protein of a highly glycan-substituted component of the extracellular AF would substantiate the hypothesis of the existence of proteins phylogenetically related to those in the AF, but that might have a function other than promoting species-specific cell adhesion. The appearance in allogeneic grafts of a cell layer turned on to express an epitope recognized by polyclonal antibodies raised against MAFp3 could represent the tissue distribution of the elusive 68-kDa protein. The absence of colocalization with monoclonal antibodies against MAF glycans suggests that the protein being detected does not belong to MAF and might be expressed only under particular physiological conditions such as allogeneic contact.
Mixtures of cells dissociated from different individuals of the freshwater sponge Ephydatia fluviatilis segregated into clumps containing cells of only one individual several days after a mixed aggregate was formed (33). Such individual specificity could be also observed between cell fractions containing only one cell type (34). Different sponge cell types also show selective reaggregation and different adhesive properties (35). Therefore, dissociated sponge cells exhibit species-, individual-, and cell-type specificities in their reaggregation properties. In rejection studies done by grafting in situ a population of the marine sponge Hymeniacidon sp., "interaction modulation factors" were purified from each individual, and measurements of their adhesion efficiency were made on cells of all the sponges among which grafting had been done (36). While the adhesiveness of those cells treated with a factor preparation made from the same sponge or from another strain whose grafts were accepted was always appreciable, factor from those strains that were not graft-compatible with the strain type of the cells led to a very considerable decrease in cell adhesion. The procedure described to prepare the interaction modulation factors was equivalent to the method used to extract MAF, suggesting that both molecules are analogous. If, then, AFs were directly involved in allorecognition, would they have any structural features resembling histocompatibility molecules of higher animals?
The reiterated domain found in MAFp4 has several features that vaguely recall Ig-like domains. It remains a consensus that, to be included in the Ig superfamily, a molecule has to fulfill certain criteria (37). Although the first condition sine qua non is the presence of a domain-sized sequence with significant similarity to Ig-related domains, there should also be the probability that the sequence shares key structural characteristics of the Ig fold. Although an adequate statistical test like the Dayhoff scoring matrix (38) is widely available to evaluate sequence homologies, structural proof for an Ig-related domain can only be convincingly established by tertiary structure determination. Ig chains are thought to have evolved from a primordial gene coding for ϳ100 amino acids, with a characteristic intrachain disulfide bond. Preceding this, there may have been a half-domain structure, as suggested by the existence of genes that have introns in the sequence between the conserved cysteines. Genes with the intradomain introns might therefore be more directly derived from ancient ancestor genes. The disulfide bond of the Ig fold stabilizes two ␤-sheets that consist of anti-parallel ␤-strands containing 5-10 amino acids. Last, but not least, Ig-related molecules diverge very rapidly, as indicated by the percentage of identity between species. For instance, the variable-like domain of CD8 chain I is only 42% identical between rodents and humans.
The positions of introns in the carboxyl-proximal region of MAFp4 were determined in our previous work (1). The introns are placed in the boundaries between domains and, interestingly, in the middle of the domains (Fig. 7). The reiterated motifs in MAFp4 are not classical Ig-like domains, but the short ␤-strand-forming regions in the MAFp4 repeats (Fig. 7) could evolve toward the ␤-barrel of the most classical Ig folds with just a few mutations, easily achievable during the period of time between the appearance of sponges 600 million years ago and the earliest occurrence of a typical Ig domain, identified by sequence analysis in the insect proteins fasciclin II (39) and amalgam (40). An evolutionary connection, however, is difficult to prove with only the basis of protein sequence comparison due to the enormous variability of the molecules belonging to the Ig superfamily. Considerable variation in sequence length of typical Ig domains occurs between the conserved disulfide bond, ranging from 40 to 75 amino acids. Often, ␤-sheets contain ␤-breaking residues like a proline in ␤-strand D of several C1 set sequences (37). Some Ig folds like those in the T cell adhesion molecules CD2 and LFA-3 even lack the disulfide bond. Finally, despite that most of the superfamily consists of membrane-bound members, non-cell surface molecules like ␣ 1B -glycoprotein and the basement membrane link protein are also included. Based on simple alignment programs, polymorphic Ig-like featuring genes have been described in the marine sponge Geodia cydonium (41), suggesting that the evolutionary origin of the Ig fold might have to be pushed back to the lowest contemporary invertebrate phylum, if not beyond. Alternatively, the Ig system may have arisen independently, and the similarity of the Ig domain may be the result of convergent evolution imposed by its functional properties.
The similarities in the domain structure and sequence between the sponge adhesion protein MAFp4 and a bacterial endoglucanase reinforce the hypothesis that the reiterated do- mains might be involved in the binding of other proteins and in the attachment of the endoglucanase to the bacterial surface (29). Such modular nature is a common feature of many cell adhesion proteins (42). MAFp3, on the other hand, does not have significant similarities either to the A. caulinodans endoglucanase or to any other protein or translated DNA sequences in data bases. In our previous work, however, we reported the existence of human nucleic acids hybridizing with MAFp3specific probes on Southern and Northern blots (1). As a rule, DNA regions can be identified by annealing if the nucleotide sequences are Ͻ30% different (38), thus suggesting that still unknown proteins that are highly homologous to MAFp3 exist in higher vertebrates.
Although the role of AFs in species specificity has been clearly established, we propose here that also individual specificity might be based on AF-related molecules or even on the AF itself. This scenario suggests that modern histocompatibility systems might have evolved from primitive cell adhesion molecules, a hypothesis supported by the occurrence of putative ancestors of typical Ig domains in MAFp4, a protein involved in sponge cell adhesion. If, on the other hand, the histocompatibility system of sponges developed after their split from the vertebrate evolutionary line, it would represent a fascinating alternative to allorecognition strategies.