Combinatorial homophilic interaction between γ-protocadherin multimers greatly expands the molecular diversity of cell adhesion

Abstract

The specificity of interactions between neurons is believed to be mediated by diverse cell adhesion molecules, including members of the cadherin superfamily. Whereas mechanisms of classical cadherin adhesion have been studied extensively, much less is known about the related protocadherins (Pcdhs), which together make up the majority of the superfamily. Here we use quantitative cell aggregation assays and biochemical analyses to characterize cis and trans interactions among the 22-member γ-Pcdh family, which have been shown to be critical for the control of synaptogenesis and neuronal survival. We show that γ-Pcdh isoforms engage in trans interactions that are strictly homophilic. In contrast to classical cadherins, γ-Pcdh interactions are only partially Ca²⁺-dependent, and their specificity is mediated through the second and third extracellular cadherin (EC) domains (EC2 and EC3), rather than through EC1. The γ-Pcdhs also interact both covalently and noncovalently in the cis-orientation to form multimers both in vitro and in vivo. In contrast to γ-Pcdh trans interactions, cis interactions are highly promiscuous, with no isoform specificity. We present data supporting a model in which γ-Pcdh cis-tetramers represent the unit of their adhesive trans interactions. Unrestricted tetramerization in cis, coupled with strictly homophilic interactions in trans, predicts that the 22 γ-Pcdhs could form 234,256 distinct adhesive interfaces. Given the demonstrated role of the γ-Pcdhs in synaptogenesis, our data have important implications for the molecular control of neuronal specificity.

Members of the cadherin superfamily mediate Ca²⁺-dependent cell adhesion and cell signaling crucial to the development and function of multicellular organisms (1). Whereas adhesive properties of the “classical” cadherins have been extensively characterized (2), much less is known about the protocadherins (Pcdhs), the largest group in the superfamily. Pcdhs can be divided into two major groups: “clustered” Pcdhs, whose ~60 genes are found in three tandem arrays (Pcdh-α, -β, and -γ) on a single chromosome in mammals (Fig. S1A), and “nonclustered” Pcdhs (3). The former in particular have attracted attention due to their diversity, their synaptic localization, and the demonstration that individual neurons express distinct repertoires of Pcdh-α and -γ genes (4–8). We previously demonstrated critical roles for the 22-member γ-Pcdh family in central nervous system development through analysis of mice in which the entire gene cluster has been deleted (5). These null mutants exhibit severe neurologic defects and die shortly after birth, exhibiting severe apoptosis of spinal cord interneurons and concomitant loss of synapses (5, 9, 10). An increase in neuronal apoptosis is also observed in the postnatal retina (11) and hypothalamus (12) when the γ-Pcdhs are disrupted using a conditional allele. Crossing null mutants with Bax^−/− mice rescues neurodegeneration but not synapse density, indicating a primary role for the γ-Pcdhs in synaptogenesis (9), in part through mediation of perisynaptic astrocyte–neuron contacts (13).

In contrast to this extensive genetic analysis highlighting essential functions for the γ-Pcdhs, less is known about their biochemical mode of action. Structural similarity to classical cadherins suggests adhesive roles for these proteins, and some in vivo functional data are consistent with this. However, roles in cell signaling distinct from, or as a consequence of, adhesion are also likely. Interaction of the γ-Pcdh C-terminal constant domain with FAK and PYK2 was demonstrated to inhibit the activity of these tyrosine kinases (14), and the γ-Pcdhs also can inhibit canonical Wnt signaling in vitro (15). Cell adhesion mediated by trans interactions between clustered Pcdhs has been reported in several studies (7, 16–21). With one exception (19), the sparse extant data are consistent with a cadherin-like homophilic mode of Pcdh interaction. However, none of these studies performed a detailed examination of γ-Pcdh adhesive interactions, or addressed the possibility of heterophilic interactions between γ-Pcdhs. Furthermore, there are no data on the mechanisms of γ-Pcdh interaction either in cis or in trans.

Here we show that γ-Pcdhs can mediate cell aggregation through homophilic trans interactions between cis-multimers. The specificity of trans interactions requires that both the extracellular cadherin (EC)2 and EC3 domains match and, in contrast to the classical cadherins, does not depend on the identity of the EC1 domain. Given that cis interactions are promiscuous and trans interactions are strictly homophilic, the diversity of adhesive interfaces mediated by the 22 γ-Pcdhs could be on the order of 10⁵, which, given their combinatorial expression and demonstrated role in synaptogenesis, has important implications for the molecular control of neuronal specificity.

Results

γ-Pcdhs Mediate trans interactions with Properties Distinct from Those of Classical Cadherins.

Previous studies of interactions among clustered Pcdhs used a wide variety of methods, including aggregation of transfected L929 cells (16, 17), K562 cells (20), or HEK293 cells (7); recruitment of Pcdhs to cell–cell contacts in HEK293 cells (18, 21); and aggregation of beads coated with ectodomains (19). We tested several of these methods and found that only cell aggregation assays using the human leukemia line K562 were suitable for our experiments. Considerable endogenous expression of γ-Pcdhs by L929 and HEK293 cells (Fig. S1B) complicates the interpretation of studies using these cell lines. Attempts to use purified γ-Pcdh ectodomains in combination with an ELISA-based assay failed to detect any trans interactions.

K562 cells are an appropriate venue for adhesion studies, because they lack endogenous expression of the γ-Pcdhs (Fig. S1B), as well as of any classical cadherins (22). Transfection of K562 cells with constructs encoding GFP-tagged N-cadherin (N-cad), full-length γ-Pcdh-A3 (A3FL-GFP), or γ-Pcdh-A3 lacking the cytoplasmic domain (A3Δcyto-GFP) leads to cell aggregation, indicating the homophilic interaction of γ-Pcdh-A3 in trans (Fig. 1A). Under higher magnification, localization of both A3FL and A3Δcyto to sites of cell–cell contact, similar to that of N-cad, was seen (Fig. 1A). Deletion of the cytoplasmic tail of γ-Pcdh-A3 did not disrupt the formation of cell aggregates; in fact, aggregates appeared to be larger than those formed by A3FL-expressing cells (Fig. 1A). A similar heightened aggregation was observed for Δcyto versions of γ-Pcdh-B2 and -C3 as well. This is likely due to the lower cell surface expression of FL γ-Pcdhs, some of which are retained in intracellular compartments (21), compared with Δcyto mutants (Fig. S1 C and D). Unless noted otherwise, we used Δcyto constructs because of their more efficient surface expression, although in no case did results differ qualitatively between Δcyto and FL constructs.

Fig. 1.

K562 cell aggregation mediated by γ-Pcdhs. (A) Aggregation of K562 cells after transfection with γ-Pcdh-A3 (Upper, GFP fluorescence; Lower, DIC). Higher-magnification images (Lower) show that in K562 cell aggregates, γ-Pcdhs localize at cell–cell junctions, as does N-cad. (B) Schematic representation of the quantitative assay used in these studies (see Materials and Methods for details). (C) Effect of EDTA on interaction between K562 cells nucleofected with N-cad, γ-Pcdh-A3Δcyto, or GFP only. Data are presented as normalized to β-gal levels in a sample of unmixed reporter cells (Upper), or as relative to untreated N-cad levels (Lower). Bars show the mean ± SD of four experiments performed in duplicate.

To investigate the adhesive properties of the γ-Pcdhs, we developed a sensitive cell aggregation assay that allowed us to determine the effect of various experimental manipulations in a rigorous, quantitative manner that was not dependent on the counting of cell aggregates in micrographs (Fig. 1B). One group of K562 cells (“bait” cells) was transfected with a given γ-Pcdh construct bearing an extracellular, N-terminal HA-tag, whereas a second group (“reporter” cells) was cotransfected with two plasmids, one encoding β-galactosidase (β-gal) and another encoding either the same or a distinct γ-Pcdh isoform as the bait cells. The two groups of cells were mixed and allowed to aggregate, after which bait cells and any adherent reporter cells were isolated using magnetic beads bound to anti-HA antibodies. We then quantified β-gal activity (normalized for expression level) which is proportional to the size of cell aggregates and thus to the strength of trans interactions. Importantly, we confirmed that the presence of small N-terminal tags, such as HA or Myc, did not interfere with the adhesive properties of the γ-Pcdhs (Fig. S2). Because classical cadherin adhesion is disrupted when only a few extra amino acid residues are attached to EC1 (23–25), this result hinted that molecular mechanisms of γ-Pcdh adhesion would differ from those of the classical cadherins.

Calcium dependency is a hallmark of adhesion mediated by classical cadherins (26–28). Using our quantitative assay, we examined the Ca²⁺ dependence of the trans interaction between γ-Pcdhs (using the A3 isoform as an exemplar) and compared it with that of N-cad. When cell mixes were preincubated with 5–10 mM EDTA for 30 min before isolation, adhesion mediated by N-cad was almost completely abolished, as expected (Fig. 1C). In contrast, aggregation mediated by γ-Pcdh-A3 was reduced by only 30–40% (Fig. 1C). The commonly accepted model of classical cadherin adhesion, supported by a wide range of functional and structural data, involves a trans interaction between the EC1 domains of cadherins on adjacent cells (29, 30). To investigate the role of the γ-Pcdh EC1 domain, we constructed a ΔEC1 mutant of γ-Pcdh-A3 and tested its ability to form cell aggregates both with itself and with WT A3 (Fig. S2). As it does for classical cadherins, deletion of the EC1 domain of γ-Pcdh-A3 completely abolishes its cell aggregation ability (Fig. S2 A and B). This could not be accounted for by a lack of cell surface expression; whereas this was lower for the ΔEC1 mutant than for WT A3, staining of nonpermeabilized cells for the extracellular HA or Myc tag revealed abundant surface protein. Whereas the WT protein was tightly localized to cell–cell contacts, the ΔEC1 mutant was evenly distributed across the cell surface (Fig. S2C), emphasizing the role of the EC1 domain in γ-Pcdh trans interaction. A comparison of amino acid sequences in the EC1 domains of clustered Pcdhs shows two strongly conserved sites: a tyrosine in position 30 (numbered as in mouse γ-Pcdh-A3), which is a part of the hydrophobic pocket (19) and a CX₅C motif (C93/C99). We created a Y30A point mutation in γ-Pcdh-A3, and found that this led to the complete loss of cell aggregation activity (Fig. S2D) and reduced (but still detectable) surface expression. Mutation of cysteines in the CX₅C motif essentially abolished surface expression (see below), making examination of the role of this conserved motif in trans interactions infeasible using our assays.

γ-Pcdh Interaction in trans Is Strictly Homophilic.

The mouse γ-Pcdh gene locus encodes 22 different isoforms with unique extracellular domains (Fig. S1A). One of the most important questions regarding the adhesive properties of γ-Pcdhs is the specificity of their trans interactions. Whereas classical cadherins are primarily homophilic, the degree of heterophilic interaction among them is a controversial issue (2). Whereas some reports indicate exclusively homophilic binding for at least some classical cadherins (31, 32), others suggest broad-range cross-reactivity between them, particularly for the type II cadherins (33–35). To investigate the specificity of interaction between different γ-Pcdh isoforms, we used both our quantitative cell aggregation assay as well as standard qualitative microscopic analysis of aggregates formed by mixing two fluorescently labeled cell groups. Both experimental approaches demonstrated a highly specific, strictly homophilic interaction between the seven different γ-Pcdh isoforms tested (A3, A10, A12, B1, B2, B6, and C3) (Fig. 2). First, K562 cells were transfected with constructs encoding an individual γ-Pcdh isoform tagged with either GFP or RFP at the C terminus, mixed and allowed to aggregate, and examined visually. Mixed aggregates composed of both GFP- and RFP-positive cells were observed only when both groups expressed the identical isoform (e.g., A3 in Fig. 2A). In every case where cells expressed different isoforms, GFP-positive and RFP-positive cells segregated into exclusively homogeneous aggregates (Fig. 2A). The quantitative cell aggregation assay confirmed this highly specific, strictly homophilic interaction between γ-Pcdh isoforms (Fig. 2B). Surprisingly, even highly similar isoforms within the same subfamily such as A3 and A10, which share ~70% identity in their ectodomains, showed no heterophilic interaction in either assay.

Fig. 2.

Trans interactions between individual γ-Pcdh isoforms are strictly homophilic. (A) Aggregates of K562 cells transfected with γ-Pcdh-A3-Δcyto fused to GFP (green) or the indicated γ-Pcdh-Δcyto constructs fused to RFP (red). (B) Quantitative analysis of the interaction specificity between K562 cells expressing different γ-Pcdh-isoforms (x-axis) with A3 (yellow), B2 (blue), or C3 (green). Bars show mean ± SD of four experiments performed in duplicate.

Specificity of Interaction Between γ-Pcdh Isoforms Is Mediated Through the EC2 and EC3 Domains.

What is responsible for this remarkable specificity of trans interaction between γ-Pcdh family members? To address this question, we constructed a series of EC-domain swapping chimeric constructs using γ-Pcdh-A3 and -C3 (Fig. 3A). The EC1 domain of classical cadherins is known to be the primary determinant of trans interaction specificity (31, 35, 36), and swapping the EC1 domain of a given cadherin for that of another is sufficient to respecify its adhesive interactions (35, 36). Surprisingly, a chimeric construct encoding γ-Pcdh-C3 with its EC1 domain replaced by that of A3 exhibited no change in specificity, interacting with WT C3 but not A3 (Fig. 3). Thus, whereas the EC1 domain is required for any γ-Pcdh trans interaction to occur (Fig. S2), it is not responsible for specificity, at least in the case of these two isoforms. We next examined the specificity of a chimeric γ-Pcdh-C3 construct containing EC1 and EC2 from γ-Pcdh-A3 (Fig. 3). Not only did swapping EC1 and EC2 fail to switch specificity to A3, this chimeric construct no longer interacted with its parent C3 (Fig. 3C). However, swapping EC1–3 did result in a specificity switch; a C3 chimera in which EC1–3 was swapped for that of A3 now interacted with WT A3 but not C3 (Fig. 3 B and C). Taken together, these findings indicate that γ-Pcdh interaction specificity is controlled by either EC3 or by EC2 and EC3 together. To examine these two possibilities, we constructed an additional chimeric C3 construct in which EC1 and EC3 (but not EC2) was swapped for those of A3 (Fig. 3A). This chimera failed to interact with either WT A3 or C3, indicating that the specificity of trans interaction is controlled by EC2 and EC3 together. Remarkably, all chimeric constructs were able to interact homophilically (Fig. S3).

Fig. 3.

EC2 and EC3 mediate the specificity of trans interaction between γ-Pcdhs. (A) Schematic representation of EC-domain swapping mutants used in the experiments, along with a summary of results from cell aggregation assays. (B) Formation of cell aggregates in K562 cells transfected with γ-Pcdh-A3Δcyto -GFP or γ-Pcdh-C3Δcyto -GFP containing WT ectodomains (green) or different EC-domain swapping mutants (red). (C) Interaction between the WT γ-Pcdh-A3 and γ-Pcdh-C3 with EC-domain swapping mutants as measured by quantitative assays. Bars show mean ± SD of four experiments performed in duplicate.

γ-Pcdhs Can Form Stable cis-Multimers via Covalent and Noncovalent Mechansims.

Classical cadherins are known to form stable homodimers and heterodimers in cis, which is important for the strength of their trans interaction (37). Some studies support the formation of homomultimers and heteromultimers between γ-Pcdhs and other members of the clustered Pcdhs (38–41). Here we sought to verify the ability of γ-Pcdhs to form cis-multimers and, more importantly, to investigate the specificity of such cis interactions, which has yet to be addressed. Lysates from H1299, HEK293, or COS7 cells cotransfected with HA-tagged A3 and Myc- or GFP-tagged A3, B2, or C3 were immunoprecipitated with HA antibody and detected by Western blot analysis with Myc or GFP antibodies. All three Myc- or GFP-tagged γ-Pcdh proteins, but not control proteins (ALCAM, N-cad, or E-cad), coprecipitated with HA-A3 (Fig. 4A and Figs. S4D and S5). When cells were separately transfected with either HA-A3 or Myc-A3, -B2, or -C3 and then mixed and cultured together before immunoprecipitation (IP) with HA-antibody, Myc-tagged proteins did not coimmunoprecipitate with HA-A3 (Fig. S4D). This suggests that the proteins in the first set of co-IP experiments were interacting in cis. Because constructs used in this experiment lacked cytoplasmic domains, and because secreted ectodomains also could interact (Fig. S5B), we can conclude that the extracellular domains are responsible for this cis interaction. Remarkably, in contrast to their highly specific homophilic interaction in trans, cis interactions exhibited no specificity between different members of γ-Pcdh family members, with all Myc-tagged isoforms tested showing robust co-IP with HA-A3 (Fig. 4A).

Fig. 4.

γ-Pcdhs form cis-multimers promiscuously. (A) Lysates from HEK293 cells cotransfected with γ-Pcdh-A3Δcyto bearing a N-terminal HA-tag (HA-A3) and different Myc-tagged γ-Pcdh isoforms were immunoprecipitated using anti-HA antibody and probed with anti-Myc or anti-HA antibodies. (B) Lysates from TR cells or mouse brain were resolved on a nonreducing SDS/PAGE gel, and analyzed by Western blot with antibody raised against the constant domain of γ-Pcdhs. (C) Size-exclusion chromatography of lysates from HEK293 cells transfected with HA-tagged γ-Pcdh-A3Δcyto construct. The approximate size range of multimers is most consistent with tetramers (~550 kDa).

Paraxial-protocadherin (PAPC/Pcdh8/Arcadlin) has been reported to form covalent multimers in cis via disulfide bonds (42). The extracellular domains of γ-Pcdhs, like PAPC, contain several cysteine residues as potential sites for disulfide bond formation. To test whether γ-Pcdhs can multimerize via disulfide bond formation, we designed constructs encoding His/Myc-tagged ectodomains of γ-Pcdh-A3, -B2 and -C3 (A3-, B2-, and C3-EC-His/Myc), transfected HEK293 cells, and purified the secreted proteins from conditioned media. We used the extracellular domains of PAPC, N-cad, and E-cad as controls. In reducing SDS/PAGE gels, the γ-Pcdh EC-His/Myc constructs migrate with an apparent molecular weight of ~100 kDa, corresponding to the size of monomers (Fig. S4A). Under nonreducing conditions, however, a portion migrates at ~300–400 kDa, suggesting that, similar to PAPC, γ-Pcdhs form multimers via disulfide bonds (Fig. S4A). Neither E-cad nor N-cad exists as higher–molecular weight aggregates under nonreduced conditions. The presence of endogenous disulfide-linked γ-Pcdh multimers also was seen in lysates of mouse brain and of neocortical progenitor-like TR cells (Fig. 4B) (43). Indeed, in the brain, the vast majority of γ-Pcdh protein exists as disulfide-linked multimers. Size-exclusion chromatography using lysates from transfected cells confirmed that γ-Pcdhs exist under native conditions as multimers, which, based on apparent size, appear to be tetramers (Fig. 4C and Fig. S4F).

We further investigated the role of particular cysteine residues for the formation of disulfide bonds in γ-Pcdh-A3. Like most of the γ-Pcdhs, A3 contains four cysteines in its extracellular domain: two (C93 and C99) that are highly conserved in all Pcdhs located in EC1, and two others (C224 and C515) located in EC2 and EC5, respectively (Fig. S4B). Mutation of C93, C99, or both C224/C515 greatly reduced the formation of tetramer-like bands in nonreducing, denaturing SDS/PAGE gels (Fig. S4C), confirming that these cysteines can participate in intermolecular disulfide bonding. A similar result was obtained for PAPC, where the mutation of cysteines in the EC1 domain also has been reported to influence surface expression (42). We found that the surface expression of γ-Pcdh mutants lacking C93 and/or C99 was strongly reduced, whereas mutation of C224 and C515 had no effect (Fig. S6 A and B). C93 and C99 mutants accumulated in the endoplasmic reticulum, based on cell staining and on the sensitivity of these mutant proteins to EndoH glycosidase (Fig. S6 B and C). Do γ-Pcdh cis interactions occur exclusively through the formation of disulfide bonds? To address this question, we performed co-IP experiments using an HA-A3ΔCyto mutant lacking all four cysteines in its ectodomain [Cys(−)]. Lysates from cells cotransfected with this mutant and with Myc-A3-, B2-, or C3-Δcyto were immunoprecipitated with anti-HA and blotted with anti-Myc (Fig. S6D). All Myc-tagged constructs could still be coprecipitated with the HA-A3-Δcyto-Cys(−) mutant, demonstrating that cis interaction does not require formation of disulfide bonds between individual γ-Pcdhs. Because mutation of C224 and C515 (unlike C93/99) does not influence the surface expression of γ-Pcdh-A3 (Fig. S6A), we tested the C224/515S mutant's cell aggregation activity, and found no difference from WT A3 (Fig. S6E).

Combinatorial Multimerization of γ-Pcdhs Can Create Diverse Adhesive Interfaces.

Our experiments demonstrate that γ-Pcdh isoforms interact with remarkable specificity in trans, while at the same time showing no specificity for cis interactions. Size-exclusion chromatography indicates that under native conditions, γ-Pcdhs form what appear to be tetramers (Fig. 4C and Fig. S4F). The promiscuity of cis interaction among γ-Pcdh family members in our experiments suggests that natively formed tetramers can comprise different isoforms in vivo. Thus, the question arises of whether multimer composition, determined solely by individual isoform expression levels, is the primary “unit” that determines trans interaction specificity?

To address this question, we performed three sets of experiments. First, we measured cell aggregation between K562 bait cells transfected with equal amounts of plasmids encoding two different γ-Pcdh isoforms (2 μg each of A3 and C3; 4 μg total) and reporter cells transfected with varying amounts of the same constructs (A3 and C3 at 4:0, 3:1, 2:2, 1:3, and 0:4 μg) (Fig. 5A). Given the promiscuity of cis interactions, the composition of multimers should be roughly proportional to the ratio of plasmid concentration; for example, cells transfected with equal amounts of A3 and C3 should have a preponderance of equally mixed multimers, whereas cells transfected with A3 and C3 at a 3:1 ratio should have multimers with a composition weighted more heavily toward A3. We predicted that if multimers represent the unit of adhesive specificity, then the strongest interactions should be measured when reporter cells expressed A3 and C3 in the same ratio as the bait cells (Fig. 5A). In contrast, if aggregation occurs primarily between independent monomers, then we might expect to find much less variation in aggregation with differing reporter cell ratios, because the overall concentration of binding partners would remain constant (assuming both steady-state conditions and similar binding strengths for different γ-Pcdh isoforms; Fig. 5A). Our experimental results clearly support the former model, because the highest aggregation levels were obtained when the reporter cell transfection ratio, and the resulting γ-Pcdh protein level ratio, matched those of the bait cells (Fig. 5B and Fig. S7).

Fig. 5.

Evidence that cis-multimers represent the units of γ-Pcdh interaction specificity. (A) Model for experiment shown in B and hypothetical outcomes. See the text. (B) Results of experiment modeled in A. Cell aggregation is highest when bait and reporter cells express γ-Pcdh-A3 and -C3 in the same ratio. (C) Results of introducing a third γ-Pcdh isoform (i.e., B2 or A10) into reporter cells, but not into bait cells. (D) Results of an experiment in which bait cells expressed γ-Pcdh-A3, -B2, -C3, and -A10. Reporter cells also expressed four γ-Pcdh isoforms, none, one, two, three, or four of which matched those of the bait cells. (E) Proposed model of γ-Pcdh interactions. See main text.

To confirm that multimer composition is the primary determinant of γ-Pcdh trans interaction specificity, we performed a second set of experiments in which we examined aggregation between bait cells transfected as above and reporter cells now transfected with the same amount of γ-Pcdh-A3 and -C3 plus either GFP, γ-Pcdh-B2, or -A10 at a ratio of 2:2:1. We predicted that the coexpression of B2 or A10 (but not GFP) would increase the proportion of reporter cell multimers that did not match those of the bait cells, and thus decrease aggregation. Indeed, this is precisely what we observed (Fig. 5C). In a final set of experiments, we asked how varying the number of matching γ-Pcdh isoforms expressed by two sets of cells affects their coaggregation. We transfected bait cells with equal amounts of plasmids encoding four γ-Pcdh isoforms: A3, B2, C3, and A10. We also transfected reporter cells with four γ-Pcdh isoforms, of which none, one, two, three, or four matched those of the bait cells. The two cell groups were mixed, and interaction was measured by a quantitative aggregation assay. Aggregation did not differ from control levels until at least three out of four γ-Pcdh isoforms matched, and the strongest aggregation was obtained only when all four isoforms matched (Fig. 5D).

Discussion

Here we present eight key findings: (i) Cell aggregation can occur through trans interactions between antiparallel γ-Pcdhs; (ii) γ-Pcdh trans interactions are much less Ca²⁺-dependent than those of classical cadherins; (iii) trans interactions are strictly homophilic between γ-Pcdh isoforms; (iv) trans interactions require the presence of EC1, but interaction specificity is mediated not through this domain as in classical cadherins, but rather through EC2 and EC3; (v) γ-Pcdhs also interact in the parallel cis orientation to form multimers; (vi) cis interactions can be mediated both covalently, through disulfide bonds, and noncovalently; (vii) in contrast to trans interactions, cis interactions are highly promiscuous, with no isoform specificity observed; and (viii) multimers (most likely tetramers) represent the “unit” of γ-Pcdh homophilic adhesion, and thus the strength and specificity of interaction between two γ-Pcdh–expressing cells are determined by the repertoire and relative levels of expressed isoforms.

Despite their overall similarity, the adhesive properties of γ-Pcdhs and classical cadherins differ in at least two major respects. The first of these is the differential requirement for Ca²⁺. Two previous reports suggested that γ-Pcdh adhesion is Ca²⁺-dependent (7, 17), but neither presented any quantitation of this. Our quantitative aggregation assay, which uses β-gal enzymatic activity as a readout, may be more sensitive than visual inspection and thus detect reduced, but still significant, interaction. Frank et al. (7) transfected HEK293 cells with constructs encoding γ-Pcdh-A3 or -C3 and found that this increased the size of cell aggregates. But because these cells express high levels of endogenous N-cad, the adhesion of which is stronger than that of γ-Pcdhs and completely Ca²⁺-dependent, it is not surprising that EDTA disrupted aggregation in that assay. The requirement for Ca²⁺ in classical cadherin interaction has been explained by its role in stabilizing the rigid rod-like structure of the ectodomain, allowing EC1 domains to interact (27). Ca²⁺-binding sites are present in clustered Pcdhs (19), but given the important role of EC2 and EC3 compared with EC1 in γ-Pcdh specificity, it is possible that these proteins bind in a different orientation that does not require such rigidification. A second major difference concerns the role of the EC1 domain. In classical cadherins, EC1 is critical for both cis and trans interactions and is the domain that controls the specificity of adhesion (31, 35, 36). We found that the γ-Pcdh EC1 domain is also necessary for trans interactions, but is dispensable for cis interactions and does not determine the specificity of homophilic adhesion. This implies that the γ-Pcdh EC1 domain might not directly participate in recognition, which is in agreement with data showing that EC1 from α-Pcdh-A4 has no adhesive properties (19). The requirement for both EC2 and EC3 in γ-Pcdh specificity is reminiscent of interactions between isoforms of Drosophila Dscam, in which homophilic binding specificity is determined by three alternatively spliced Ig domains, all of which must match for interaction to occur (44). Although our experiments focused on two isoforms, it is important to note that all chimeric γ-Pcdhs, including two that had hybrid EC2/EC3 pairs, were able to mediate homophilic cell aggregation comparably with that of WT proteins. Thus, γ-Pcdh specificity mediated by EC2 and EC3 is likely a generalizable phenomenon. Consistent with this, alignments of ectodomains of human α-Pcdhs (45) and mouse γ-Pcdhs (Fig. S8) show that EC2 and EC3 are most diverse among family members.

In contrast to the remarkable homophilic specificity of their trans interactions, the γ-Pcdhs that we tested showed complete promiscuity of interaction in the cis orientation. We found that cis multimerization can occur covalently, through disulfide bond formation between cysteines located in the extracellular domain, as well as through noncovalent mechansims. Notably, the human ortholog of mouse γ-Pcdh-A3 is missing the equivalent of C224 in EC2 and C515 in EC5, consistent with these residues being dispensible for protein functionality. We found that the conserved CX₅C motif in EC1 is critical for surface expression of γ-Pcdh-A3 (and presumably other γ-Pcdhs); however, this precluded us from performing functional tests with mutants lacking the motif. We did not precisely determine the domain(s) mediating noncovalent cis interactions, but did find that a γ-Pcdh lacking EC1–3 was still able to interact, albeit more weakly. It is likely that multiple sites can contribute to cis interactions.

We found that γ-Pcdh isoforms exhibit highly specific homophilic interaction in trans, but no specificity for multimerization in cis. This has important implications for any potential role for γ-Pcdhs in the control of synaptic specificity, which remains to be demonstrated but would be consistent with some (although not all; see ref. 11) extant functional data in vivo (9, 10, 12, 13, 46). Several reports indicate that single neurons of the same cell type can express different subsets of clustered Pcdh isoforms (4, 8, 46–48). Yagi and colleagues (8, 47) performed RT-PCR on RNA from single cerebellar Purkinje cells and estimated that each neuron expresses approximately five or six γ-Pcdhs (as well as four or five α-Pcdhs). Our experiments support a model in which the composition of γ-Pcdh cis-tetramers determines the specificity of their trans interactions, which are strictly homophilic (Fig. 5E). Because cis interactions are promiscuous, the diversity of these tetramers in a given neuron are controlled only by the repertoire and expression levels of the various γ-Pcdh isoforms. Because a total of 234,256 different tetramers possibly could be formed from the 22 γ-Pcdh isoforms (i.e., 22⁴), this would greatly increase the diversity of the adhesive interfaces that could be specified molecularly. Given that γ-Pcdhs can interact in cis with α-Pcdhs (39) and with the much less widely studied β-Pcdhs (41), the adhesive diversity mediated by the clustered Pcdhs could be exponentially greater. Indeed, the possible combinations would far outnumber even those of Drosophila Dscam, making clustered Pcdhs ideal candidates for providing a unique recognition system for neuronal specificity in the mammalian central nervous system. The data presented here represent an important proof of principle using a variety of biochemical and in vitro, nonneuronal assay systems. The challenge for future studies will be to develop genetic tools through which the γ-Pcdh repertoire can be selectively manipulated during the development of discrete neuronal populations with well-characterized patterns of connectivity.

Materials and Methods

See SI Materials and Methods for more detailed information.

Cell Aggregation Assays.

Human leukemia cell line K562 (ATCC CCL243) was nucleofected with appropriate constructs encoding various γ-Pcdhs fused to GFP, RFP, HA, or Myc using an Amaxa Nucleofector II device (Lonza). Cells (10⁶) and 2–5 μg of plasmid DNA were resuspended in 100 μL of nucleofection solution V and nucleofected using program T-016. Up to 90% transfection efficiency could be achieved. The day after nucleofection, cells were collected and mixed in various combinations in six-well plates in 2 mL of total volume per well (~500,000 cells/well). Cell aggregates were examined by microscopy 24–36 h later, or were subjected to the quantitative binding assay. Cells expressing a given HA-tagged γ-Pcdh (“bait” cells) were mixed with cells transfected with the same or a different γ-Pcdh plus β-gal (“reporter” cells). The next day, 750 μL of cell suspension was transferred to a 2-mL tube, 0.3–0.5 μg of biotin-conjugated anti-HA antibody (Roche, clone 3F10, rat) and 10 μL of streptavidin-magnetic beads (Pierce) were added, and cells were incubated on an overhead shaker at room temperature (at ~60 rpm) for 30 min. Magnetic beads were isolated and washed once with 1.5 mL of warm DMEM. Subsequently, bound cells were lysed in 100 μL of passive lysis buffer (Promega). The cell lysates were used for measurement of β-gal activity; 25 μL of the lysate was mixed with 285 μL of ONPG-substrate buffer [100 mM phosphate buffer (pH 7.4), 2 mM MgCl₂, 0.5 mg/mL of ONPG, 50 mM β-mercaptoethanol] and incubated at 37 °C for 30–60 min. Reactions were stopped by the addition of 200 μL of 1 M NaCO₃, and the absorbance was measured at 420 nm. In addition, cell pellets from 250 μL of each mixed cell suspension were lysed in 150 μL of passive lysis buffer, and 25 μL was used for the determination of β-gal activity. This crude lysate measurement was used to normalize the results of the bait–reporter HA-pull-down β-gal activity, to control for variations in β-gal expression within each reporter cell set; that is, the pull-down β-gal activity measurement was divided by the matched crude lysate activity measurement. When “relative” measurements are graphed, the first bar was set to 1 relative unit, to allow comparison of results from multiple experiments.