Splice Variants of the Drosophila PS2 Integrins Differentially Interact with RGD-containing Fragments of the Extracellular Proteins Tiggrin, Ten-m, and D-Laminin α2*

Two new potential ligands of theDrosophila PS2 integrins have been characterized by functional interaction in cell culture. These potential ligands are a new Drosophila laminin α2 chain encoded by the wing blister locus and Ten-m, an extracellular protein known to be involved in embryonic pattern formation. As with previously identified PS2 ligands, both contain RGD sequences, and RGD-containing fragments of these two proteins (DLAM-RGD and TENM-RGD) can support PS2 integrin-mediated cell spreading. In all cases, this spreading is inhibited specifically by short RGD-containing peptides. As previously found for the PS2 ligand tiggrin (and the tiggrin fragment TIG-RGD), TENM-RGD induces maximal spreading of cells expressing integrin containing the αPS2C splice variant. This is in contrast to DLAM-RGD, which is the first Drosophila polypeptide shown to interact preferentially with cells expressing the αPS2 m8 splice variant. The βPS integrin subunit also varies in the presumed ligand binding region as a result of alternative splicing. For TIG-RGD and TENM-RGD, the β splice variant has little effect, but for DLAM-RGD, maximal cell spreading is supported only by the βPS4A form of the protein. Thus, the diversity in PS2 integrins due to splicing variations, in combination with diversity of matrix ligands, can greatly enhance the functional complexity of PS2-ligand interactions in the developing animal. The data also suggest that the splice variants may alter regions of the subunits that are directly involved in ligand interactions, and this is discussed with respect to models of integrin structure.

The integrins are a family of heterodimeric transmembrane glycoproteins, consisting of ␣ and ␤ subunits, that serve as receptors for extracellular matrix molecules and cell surface molecules of neighboring cells. Integrins have roles in diverse phenomena, such as cell adhesion, spreading, migration, and differentiation, as well as roles in the development and progression of numerous pathological states, such as cancer and cardiovascular disease (1)(2)(3)(4). As might be expected from these varied requirements, integrins not only provide mechanical linkages to the matrix and neighboring cells but also receive and transmit information from the cell exterior to the cell interior, and vice versa (5). The fruit fly, Drosophila melanogaster, provides a valuable genetic system in which to examine these integrin functions in the developing animal (6,7). As a complement to these genetic studies, we have utilized cultured cells, expressing various combinations of Drosophila PS integrin transgenes, to examine interactions of PS integrins and potential integrin ligands.
The PS1, PS2, and PS3 integrins of Drosophila consist of a common ␤ PS subunit paired with an ␣ PS1 , ␣ PS2 , or ␣ PS3 subunit, respectively. ␤ PS , ␣ PS1 , and ␣ PS2 were originally identified as position-specific (PS) 1 antigens in monoclonal antibody screens of imaginal discs (8,9). Subsequent biochemical and molecular analyses of these antigens indicated that they are members of the integrin family (10 -13). ␣ PS3 was identified only recently, and little is known of its ligand binding properties (14).
Both the ␤ PS and ␣ PS2 subunits may be alternatively spliced to generate proteins that vary in their extracellular domains. The ␤ PS subunit mRNA has been found in alternatively spliced forms to generate proteins referred to as ␤ PS4A and ␤ PS4B (15,16). These subunits differ in the utilization of different fourth exons, which encode 29 amino acids in the ligand binding "head" of the ␤ subunit. The ␣ PS2 subunit exists in splice forms called ␣ PS2C and ␣ PS2m8 (17), referring to the presence (C, canonical) or absence (m8, missing exon 8) of exon 8 . When present, the eighth exon encodes 25 amino acids, potentially located in a region that would be expected to influence ligand associations. Thus, there are at least four possible ␣/␤ heterodimer combinations for PS2 integrins: ␣ PS2C ␤ PS4A , ␣ PS2C ␤ PS4B , ␣ PS2m8 ␤ PS4A , and ␣ PS2m8 ␤ PS4B . These receptors may generate significant PS2 integrin functional diversity during development.
Ligands that support Drosophila PS2 integrin-mediated cell spreading include mammalian vitronectin and fibronectin (18,19) and the novel Drosophila extracellular matrix protein, tiggrin (20). A key feature of several vertebrate integrin ligands is the tripeptide sequence, Arg-Gly-Asp (RGD). This same tripeptide is apparently recognized by Drosophila PS2 integrins, as all previously identified PS2 ligands contain an RGD sequence, and PS2 integrin-mediated cell spreading is inhibited by soluble RGD peptides. Moreover, tiggrin polypeptides in which the RGD sequence has been changed to LGA no longer support cell spreading, and the RGD sequence is required for maximal rescue by transgenes of some tiggrin mutant phenotypes in situ (21). In contrast, PS1-expressing cells have been shown to spread on Drosophila heterotrimeric laminin, which does not contain an RGD motif (22), and this spreading is not inhibited by RGD peptides. 2 One approach for identifying additional PS2 ligands is to first search for candidate extracellular matrix molecules based on structure (e.g. an RGD sequence) or location (e.g. muscle attachment sites) and ask whether the purified proteins or protein fragments will support PS2-mediated cell spreading in culture. One such candidate is Ten-m (23), a protein with tenascin-type EGF repeats (Fig. 1). Ten-m contains a C-terminal RGD sequence, and earlier studies had suggested that it may function as a PS2 ligand. 3 Mutants for the ten-m gene display an early embryonic patterning defect of the "pair-rule" type (23,24). Another potential PS2 ligand is encoded by the wing blister locus, mutations in which can lead to wing blisters similar to those caused by loss-of-function integrin mutations (25). Recently, this gene was found to encode a new laminin ␣ chain, D-laminin ␣2, which, in contrast to the previously characterized Drosophila laminin ␣ chain (26 -28), contains an RGD motif (Fig. 1). 4 We have examined the ability of tiggrin and these newly characterized matrix components to support PS2-mediated cell spreading, utilizing S2 cell lines that express each of the different PS2 integrin ␣/␤ heterodimer combinations. Our results demonstrate that peptides from both Ten-m and D-laminin ␣2 can serve as integrin ligands in our in vitro assays. Moreover, we find that both ␣ and ␤ splice variants lead to ligand-dependent differences in integrin function.

MATERIALS AND METHODS
Cell Culture, Cell Transfection, and Cell Spreading Assays-Cell culture techniques and methods for transfection of cells have been previously described, as have Schneider's line 2 (S2) Drosophila cells that have been stably transfected with integrin transgenes under the regulation of the heat shock protein 70 promoter (18,19,29). The construct used to express the ␤ PS4B subunit, also under the regulation of the heat shock protein 70 promoter, is described in Ref. 16. Cell spreading assays were performed as described previously (19,20). Briefly, cells were treated with dispase/collagenase to remove existing matrix molecules and cell surface proteins. Cells were then heat shocked at 37°C for 30 min to induce expression of integrin transgenes and were then plated on TIG-RGD, DLAM-RGD, or TENM-RGD substrates (described below). 4 -6 h following the heat shock, the cells were fixed and quantified by scoring for spread cells using a Nikon phasecontrast microscope (Nikon Diaphot-TMD). Three fields of cells were counted for each well, and the numbers reported represent the averages (and standard errors) of three separate experiments.
TIG-RGD, TENM-RGD, and DLAM-RGD-TIG-RGD was a generous gift from Frances Fogerty and has been described (20). It is a polyhistidine-tagged bacterial fusion protein that contains 270 amino acids of tiggrin, residues 1891-2161, with the RGD sequence being residues 1989 -1991 (tiggrin has a total of 2186 amino acids).
TENM-RGD is a bacterial fusion protein that contains a polyhistidine tag fused to the final 212 amino acids of Ten-m. Ten-m has a total of 2515 amino acids (23), and the RGD sequence is 72 amino acids from the C terminus. This fusion protein was produced from the expression vector pTrcHisB (Xpress System TM , Invitrogen), into which was cloned an XhoI-HindIII fragment of the ten-m cDNA.
His-tagged DLAM-RGD was made by cloning a polymerase chain reaction product into pTrcHisA. Genomic DNA from Oregon-R flies was used as template, and sequencing of the wing blister gene showed that there are no introns in this interval. 4 The recombinant protein contains 340 amino acids of D-laminin ␣2. These are residues 492-832 (RGD is found at 689 -691) of a total of 3325 residues.
Protein induction was performed according to the manufacturer's protocols, and recombinant peptide was affinity purified using a nickel resin (Ni-NTA, Qiagen). Purified fusion proteins were dialyzed from a buffer containing 8 M urea stepwise into 50 mM Tris, 100 mM NaCl, pH 7.5. Protein concentrations were determined by SDS-polyacrylamide gel electrophoresis and comparison of fusion protein staining with that of protein molecular weight standards (Bio-Rad). Gels were stained with Coomassie Brilliant Blue. Total protein concentrations were determined by using a BCA protein assay (Pierce) with bovine serum albumin as a standard. FACS Analysis-Cells were prepared for flow cytometry following essentially the same procedure as for cell spreading; briefly, cells were protease-treated, heat-shocked, and allowed to recover for 3 h. Cells were then incubated with an anti-PS2 monoclonal antibody (CF.2C7), followed by staining with a fluorescein isothiocyanate-labeled antimouse secondary antibody (Jackson ImmunoResearch). Cells were then fixed in 3% formaldehyde. FACS analyses were performed at the Research Flow Cytometry Service Laboratory of the University of Arizona Cancer Center. Data were acquired with a FACScan device (Becton Dickinson), and data were analyzed using FACSvantage and Cell Quest software.

RESULTS
The alternative splicing to produce extracellular variants in ␣ PS2 and ␤ PS have been described previously (15)(16)(17). Recently, models for the structure of the ligand binding heads of both ␣ and ␤ integrin subunits have been proposed, and the positions of the variant residues with respect to these models are detailed under "Discussion." Cell Spreading on TIG-RGD Is Unaffected by ␤ Integrin Subunit Splice Variants-Fogerty et al. (20) showed that the novel Drosophila extracellular matrix protein tiggrin serves as a ligand in PS2 integrin-mediated cell spreading assays. A 270-amino acid C-terminal recombinant fragment containing the RGD sequence of tiggrin (here referred to as TIG-RGD) ( Fig. 1) also promoted cell spreading. The ␤ subunit of the integrin receptors used in those assays was ␤ PS4A . We extended the analyses of cell spreading on TIG-RGD to include integrin heterodimers composed of ␣ PS2 ␤ PS4B . As was seen previously (20), PS2C cells spread better on TIG-RGD than PS2 m8 cells (Fig. 2). Additionally, we found that PS2 integrin-expressing cell lines spread equally well on TIG-RGD regardless of the ␤ subunit splice variant of the integrin. All of the cell spreading on TIG-RGD was inhibited by soluble RGD peptides (Fig. 3).
A Ten-m Fragment Promotes PS2-mediated Cell Spreading-We generated a recombinant protein fragment of 212 amino acids of Ten-m, including the RGD sequence, and used this fragment (TENM-RGD) as a substrate for cell spreading assays. As shown in Fig. 4, TENM-RGD supported cell spreading for all tested PS2 integrin-expressing cell lines. Reminiscent of PS2-mediated cell spreading on TIG-RGD, PS2C cells spread better on TENM-RGD than did PS2m8 cells. Additionally, the alternative splice forms of the ␤ PS subunit made little or no difference in the levels of cell spreading on TENM-RGD. As with TIG-RGD, this cell spreading was inhibited by soluble RGD peptides (Fig. 3).
A Laminin Fragment Promotes PS2 m8-Mediated Cell Spreading-Analysis of the predicted coding sequence of Dlaminin ␣2 indicates that this protein is a member of the laminin ␣ chain family of extracellular matrix molecules, 4 and the putative protein domain structure may be grouped according to accepted laminin nomenclature (Fig. 1). Overall, the sequence of D-laminin ␣2 is similar to murine laminin ␣2 chains, and it contains 19 laminin EGF-like repeats, 5 laminin G domains, a laminin B motif, and a characteristic laminin N-terminal domain. D-laminin ␣2 also possesses a potential integrin-binding RGD sequence in the N-terminal quarter of the protein, in the IVb region between two blocks of EGF-like domains.
We generated a recombinant protein fragment that includes the D-laminin ␣2 RGD sequence (DLAM-RGD), and this 342-amino acid fragment was purified and plated on microtiter plates to be used as a ligand in integrin-mediated cell spreading assays. All PS2 integrin-expressing cells spread on D-laminin ␣2, in contrast to the parental S2 cell line (Fig. 5). However, ␣ PS2m8 ␤ PS4A cells spread 2-3 times better on D-laminin ␣2 than all other PS2 integrin-expressing cells, including ␣ PS2m8 ␤ PS4B cells. This was in contrast to PS2-mediated cell spreading on TIG-RGD and on TENM-RGD, where PS2C cells always spread better than PS2 m8 cells, and the ␤ subunit splice variant made little difference. Again, spreading on DLAM-RGD was inhibited by RGD peptides (Fig. 3).
Cell Spreading Is Not Correlated with Integrin Expression Level-Although it is formally possible that the differences in spreading observed between different cell lines are due to dif-ferences in integrin expression levels, this does not appear to be the case. Both FACS analysis (Fig. 6) and immunofluorescence (see, for example, Ref. 18) indicated that surface integrin expression on the cells was heterogeneous, but the large majority (typically 85% or more) of the induced cells expressed significant integrin for all of the lines. Most importantly, there was no obvious correlation between expression levels and spreading. For example, among the four transfected cell lines, the ␣ PS2C ␤ PS4A -expressing and ␣ PS2C ␤ PS4B -expressing cells exhibited the highest and lowest levels of surface integrin (displaying a difference of 2-fold or more in mean and median fluorescence values in two FACS experiments), but showed virtually identical, and very reproducible, levels of spreading on two peptide ligands. In general, it appears that once a relatively low level of surface integrin is present, further increases do not result in large changes in spreading behavior. Indeed, even uninduced cells (but not untransfected S2 cells), which contain very small amounts of integrin relative to heat shock-induced cells (see, for example, Fig. 1 of Ref. 18), will spread in culture if a suitable matrix is present.

DISCUSSION
Integrin Structure and ␣ PS2 Isoforms-Using structural homology arguments, Springer (30) has generated a model to describe the organization of the integrin ␣ subunit globular head. According to this model, seven repeat domains (termed FG-GAP repeats for the phenylalanyl-glycl and glycyl-alanylprolyl consensus sequences) are folded into a cyclic ␤-propeller, and each "blade" of the ␤-propeller is postulated to be composed of four strands of anti-parallel ␤ sheet. For ␣ PS2 , sequence alignments place the residues encoded by the alternatively spliced exon 8 in the loop connecting ␤ sheet strands two and three of the third propeller blade (Fig. 7). Recently, mutagenesis studies have demonstrated that residues in the corresponding loops of ␣ IIb , ␣ 4 , and ␣ 5 are critical for ligand binding (33)(34)(35), and one possibility is that the extra 25 amino acids extend this loop on the top of the ␤-propeller, providing a new surface for integrin-ligand interactions. Polypeptides that support good spreading of cells expressing PS2C (vitronectin, tiggrin, and TENM-RGD) also serve as ligands for PS2m8, albeit less well, and so the exon 8-encoded segment probably does not completely replace the normal site of ligand interaction on ␣ PS2m8 , but it may provide an additional surface that adds to the stability of binding (17). Alternatively, exon 8 could encode a new strand of ␤ sheet that directs the polypeptide chain down to the lower side of the ␤-propeller, leaving the ligand binding surface relatively intact but adding new residues to potential regulatory sites. For example, alternative splicing of the ␣ 7 subunit alters residues in the same part of the protein as does the ␣ PS2 alternative splicing, and the ␣ 7 isoforms have been shown recently to affect the activity state of the ␣ 7 ␤ 1 integrin (36). In this vein, it is noteworthy that PS2m8 and PS2C can display different cation requirements in cell spreading experiments (19).
In an attempt at gaining further insights into these possibilities, we have used a number of different algorithms to predict potential secondary structures in this region of ␣ PS2 . Unfortunately, no consistent pattern was seen in the predictions for the residues encoded by exon 8; of particular relevance to the above discussion, residues at the beginning of exon 8 show ␤ sheet potential in some predictive paradigms but not in others. Overall, there is no consistent pattern that would allow one to prefer one structural model over the other. In any case, the residues encoded by exon 8 are likely to be involved in specific protein interactions, since they are highly conserved in the distantly related dipteran Ceratitis capitata (17).
Integrin Structure and ␤ PS Isoforms-An overall similarity in hydropathy profiles suggested that the ligand binding domain of ␤ subunits would fold into a structure similar to the I domains of some ␣ subunits, with a cation-containing pocket that is expected to be directly involved in ligand association (37). Recently, models for ␤ subunit I domain-like structures have been proposed, and these models differ significantly in the predicted tertiary structure for the region encoded by ␤ PS exon 4. (A comparison of the sequences encoded by the alternatively spliced forms of ␤ PS exon 4 (15,16) is shown in Fig. 7.) In models that are driven primarily by secondary structure predictions from computer algorithms (Refs. 38 and 39; see also Ref. 40 for a non-I domain interpretation of secondary structure profiles), exon 4 encodes residues that form a loop on the top of the ␤ I domain and then run via a ␤ sheet to the lower part of the domain, including the beginning of a motif postulated to be important in integrin regulation (41). Another model makes adjustments to the secondary structure predictions in order to more closely copy the structure of ␣ subunit I domains (42). According to this model, the exon 4-encoded domain begins low in the structure, and then, via an ␣ helix and loop structure, moves across the top of the I domain, near the putative ligand binding region. Thus, in either model exon 4-encoded residues might be expected to interact directly with ligand, but they are likely to be different residues in the respective models. It is intriguing that the exon 4 residues include and connect domains that have been postulated to interact with ligand and mediate integrin regulation, based on mutagenesis and antibody binding studies (39,41). This region of ␤ PS should prove to be a fruitful location for more extensive site-directed mutagenesis studies.
Ten-m as a PS2 Integrin Ligand-Ten-m possesses some, but not all, of the features common to most vertebrate tenascins (reviewed in Ref. 43). For example, Ten-m is a secreted glycoprotein with eight tenascin-type EGF-like repeats and putative fibronectin-type III repeats (23). Ten-m lacks a tenascin Cterminal fibrinogen-like domain, and the Ten-m RGD sequence is found 72 amino acids from the C terminus. Recombinant protein fragments containing this RGD sequence promote RGDdependent, PS2 integrin-mediated cell spreading (Fig. 4), with PS2C cells spreading better than PS2m8 cells.
Levine et al. (24) reported a partial cDNA sequence from the ten-m gene (which they called odd Oz); this partial sequence stops short of the final 325 amino acids and thus does not include the RGD tripeptide near the C terminus, and it also includes 216 N-terminal residues not reported by Baumgartner et al. (23). Levine et al. (24) ascribed properties to the presumed polypeptide that are significantly different from those deduced by Baumgartner et al.; for example, they suggest that Odd Oz is a transmembrane phosphoprotein with tenascin homology in its putative extracellular domain, and they also propose that the polypeptide is cleaved into smaller mature proteins. These apparent discrepancies have yet to be resolved, and it is possible that the protein functions in different forms. In any case, Baumgartner et al. (23) found that a Ten-m polypeptide could be found in conditioned media from Drosophila cells, and so a secreted form is present in at least some instances.
Curiously, the ten-m gene is expressed in an embryonic pairrule pattern, and ten-m mutants display pair-rule patterning defects (23,24). Since the protein influences expression of downstream genes, it must communicate its presence to the cell nucleus. However, it does not appear that integrin signal transduction is important in early embryonic segmentation. PS integrins are not strongly expressed at this time, and, more FIG. 7. Structural predictions for the splice variants of PS2 integrin subunits. A, alignment of the four strands of ␤ sheet composing the third blade of the ␣ subunit ␤-propeller structure proposed by Springer (30). Underlined residues for human ␣ 4 , ␣ IIb, and ␣ M are predicted to have the potential to form ␤ sheet (30). For ␣ PS2m8 , the underlined residues scored above an arbitrarily chosen number for ␤ sheet prediction, using the PHD algorithm (31,32) in all tested sequence alignments. importantly, mutations in integrin subunit genes do not cause segmentation phenotypes (6,44).
Ten-m is later localized (among other places) at muscle attachment sites, where integrins are known to accumulate (11,23,45). This localization of Ten-m in vivo, as well as the demonstration of TENM-RGD interactions with PS2 integrins in vitro, suggests that Ten-m may function with PS2 integrins in muscle attachment. Interestingly, the heparan sulfate-containing protein D-syndecan also localizes to muscle attachments (46), and Ten-m contains a consensus heparin-binding sequence near the RGD, suggesting the potential of a Ten-msyndecan-integrin complex. Syndecan proteoglycans recently have been shown to be important in signal transduction in focal adhesions in vertebrate cells (47).
The available data, although very suggestive, do not demonstrate unequivocally that Ten-m serves as an integrin ligand at muscle attachment sites. One advantage to using Drosophila is that genetic approaches can often be employed to indicate functional interactions in situ. However, other potential PS2 ligands, such as tiggrin (20), also accumulate at muscle attachment sites, and genetic studies of tiggrin suggest considerable functional redundancy among the extracellular matrix components there (21). Because of this redundancy, a direct genetic demonstration of a role for Ten-m in muscle attachment may require simultaneous disruption of multiple genes encoding matrix proteins, and the early embryonic phenotype of ten-m mutants will further complicate such an analysis. One potential approach might be to demonstrate a dominant genetic effect of ten-m mutations in a background that has been sensitized for loss of function phenotypes by viable mutations in other genes that encode proteins important for muscle attachment or other integrin-dependent processes. Early attempts to do this for Ten-m have been unsuccessful. 2 D-Laminin ␣2 as a PS2 Integrin Ligand-The Drosophila wing blister locus encodes a new laminin ␣ chain. 4 Laminins have long been known to interact with integrins (reviewed in Ref. 48), and the previously characterized native heterotrimeric Drosophila laminin is a PS1 integrin ligand (22). Our experiments indicate that a fragment of D-laminin ␣2 can function as an RGD-dependent ligand for PS2 integrins.
Overall sequence comparisons indicate that D-laminin ␣2 resembles the mouse laminin ␣2 chain, as exemplified by the length of the protein and the signature laminin domain structure of the molecule. 4 For our purposes, however, a more notable comparison is with the recently cloned mouse laminin ␣5 chain (49). Although the overall domain structure of ␣5 more closely resembles the previously described Drosophila ␣ chain (28), potential integrin binding sites of D-laminin ␣2 and mouse ␣5 are very similar. In D-laminin ␣2, the RGD tripeptide sequence is in region IVb between the third and fourth laminin EGF-like repeats in the N-terminal (short arm) portion of the molecule. In murine laminin ␣5, there are two RGD sequences, located in domains IVa and IIIa. A 50-amino acid region of D-laminin ␣2 domain IVb and murine laminin ␣5 domain IVa show 39% identity and 54% similarity to each other when rooted in the RGD sequence (Fig. 1B), suggesting a functionally conserved portion of the proteins. To date, integrin association with this portion of vertebrate laminin ␣5 has not been reported.
Previous experiments using portions of the original Drosophila laminin chains as substrates for PS1-mediated cell spreading indicated the necessity for the native heterotrimeric molecule. 2 Our results with DLAM-RGD demonstrate that a portion of this laminin chain alone has an inherent PS2 integrin binding domain. Presumably, D-laminin ␣2 forms a trimer with laminin ␤ and ␥ chains in situ. The domain containing the RGD motif is in the exposed short arm of D-laminin ␣2 (based on homology to other laminins) and should not be obscured by association with other subunits.
As the name implies, mutations in the wing blister locus can lead to blistering of the wing, where the dorsal and ventral wing surfaces separate (25). This phenotype is frequently associated with mutations in integrins (50 -52), and D-laminin ␣2 and integrins co-localize in many fly tissues. 4 These observations suggest that the interaction between the RGD-containing domain of D-laminin ␣2 and PS2 integrins that we find in cell culture is important for morphogenesis in vivo. This proposal is supported strongly by genetic interactions between myospheroid (␤ PS ) and wing blister mutations in situ. 2 For example, wing blister mutations enhance the partial lethality of weak myospheroid mutations, and even wing blister heterozygosity can lead to blisters in myospheroid mutants that normally show insignificant frequencies of wing defects. Conversely, weak myospheroid alleles can greatly increase the blistering of wing blister flies. Thus, D-laminin ␣2 and PS2 integrins have the biochemical potential to recognize one another, are in many of the same places, and function in at least some of the same morphogenetic events. From this we infer that D-laminin ␣2 serves as a PS2 integrin ligand in vivo (see also below).
␣ PS2 Splice Variant Isoforms Affect Ligand Preference-Results presented here indicate that ␣ PS2 integrin subunit isoforms differ in their abilities to mediate cell spreading on fly polypeptides; suggestions that this might be true had earlier come from studies of PS2 interactions with vertebrate matrix proteins (19). As was seen previously (20), cells expressing PS2C integrins spread better on TIG-RGD than PS2m8 cells. The same is true for cell spreading on TENM-RGD. However, cells expressing ␣ PS2m8 ␤ PS4A integrins spread better than cells expressing any of the other PS2 subunit combinations on DLAM-RGD. This is the first Drosophila integrin ligand that appears to be preferred by a PS2m8 integrin over a PS2C integrin.
Levels of the ␣ PS2 alternatively spliced transcripts vary during development (17), which may imply different roles for the different ␣ PS2 isoforms. Genetic data also support the notion that ligand preferences will have functional significance in situ. Although transgenic expression of either form of the ␣ PS2 subunits in flies is sufficient for viability in ␣ PS2 (inflated) null backgrounds, the two isoforms are not equivalent (53). ␣ PS2C rescues overall viability better than ␣ PS2m8 , whereas expression of ␣ PS2m8 is more efficient at rescuing some specific mutant phenotypes, such as wing blisters. This latter result is particularly noteworthy, in light of the preference we find for a PS2m8 integrin in mediating cell spreading on polypeptides from the product of the wing blister (D-laminin ␣2) gene.
␤ PS Splice Variant Isoforms Affect Ligand Preference-S2 cells expressing ␣ PS2m8 ␤ PS4A integrins spread more efficiently on recombinant DLAM-RGD protein fragments than did cells expressing ␣ PS2m8 ␤ PS4B . One potential trivial explanation for the preference for ␤ PS4A is that the cells might be making more ␤ PS4A than ␤ PS4B . However, we did not see large differences in expression between ␣ PS2m8 ␤ PS4A and ␣ PS2m8 ␤ PS4B , and as discussed earlier, spreading does not generally appear to be sensitive to expression levels above a relatively low threshold. More importantly, there were no significant ␤ PS -related differences in cell spreading when the same cell lines were plated on TIG-RGD or TENM-RGD; this specificity indicates that the difference in spreading seen with DLAM-RGD reflects a genuine functional difference between the ␤ PS isoforms.
Although we can state unequivocally that the isoform of ␤ PS can affect function, it is difficult to apply any precise quantita-tive interpretations to these data. The S2 cell line makes a relatively small amount of endogenous ␤ PS , which appears to be mostly ␤ PS4A . Following the proteolysis and induction protocol, the ␤ PS produced from the multiple copies of the heat shock-induced transgenes would be expected to be present in large excess relative to that generated from endogenous genes, and this expectation is borne out by Western blot data (18). Our functional results further indicate that there is relatively little ␣ PS2m8 ␤ PS4A present on the surface of ␣ PS2m8 ␤ PS4B -transformed cells; otherwise, this line would be expected to spread much better on DLAM-RGD. It should also be noted that the data overall indicate that associations of the various ␣/␤ subunits are not grossly disturbed by isoform composition, since in vivo and in vitro, all combinations tested either rescue mutant phenotypes or demonstrate ability to spread on at least some ligands.
In flies carrying mutations in the ␤ PS subunit, rescue experiments with ␤ PS transgenes indicate that ␤ PS4A and ␤ PS4B are both capable of rescuing the postembryonic mutant phenotypes in the eye and wing (16). Rescue of embryonic lethality, on the other hand, is efficient only if both isoforms are expressed (16). This would lead one to expect that another ligand, as yet uncharacterized, may show preference for ␤ PS4B , in combination with one or more ␣ subunits.