Cbln1 Is Essential for Interaction-Dependent Secretion of Cbln3^▿

Antibodies and reagents.

Mouse monoclonal antibody to β-actin and Flag tag were from Sigma (St. Louis, MO). Mouse monoclonal anti-V5 tag was from Invitrogen (Carlsbad, CA). An antiserum to a synthetic peptide corresponding to amino acids 95 to 112 of mouse Cbln1 was generated and described previously (1) and is referred to as anti-Cbln1. Chloroquine was from Sigma. Lactacystin, MG132, and Z-Ile-Glu(OtBu)-Ala-Leucinal (ZAL) were from Calbiochem (La Jolla, CA).

Plasmids and constructs.

Standard molecular cloning and sequencing techniques were used to isolate and validate cDNA clones carrying full-length cbln1, cbln2, cbln3, and cbln4 from brain total RNA. These cDNAs were subcloned into the BamHI and XbaI sites of the pcDNA3.1 and pcDNA3.1V5-His vectors (Invitrogen). The cDNA of cbln1 was also inserted into the p3xFLAG-CMV-9 vector (Sigma). All truncation and mutation constructs were generated by standard PCR-based methods. All plasmids were purified with a Midi-Prep kit (QIAGEN, Valencia, CA) before use. Primer synthesis, DNA sequencing, and bioinformatics support were provided by the Hartwell Center for Bioinformatics and Biotechnology at St. Jude Children's Research Hospital.

Generation of cbln3^−/− mice.

A cbln3 targeting vector was constructed by replacing all coding sequences, except for the first two codons of cbln3 plus the intervening sequences, with a fragment containing the LacZ coding sequence from pMC1847 and a neo cassette derived from the PGK.neo.TK plasmid. The LacZ was placed in frame with the start codon of cbln3. The left arm is a 2.28-kb XbaI-KpnI (KpnI was introduced by PCR following the second codon) fragment, and the right arm is a 5.98-kb ApaI fragment (see Fig. 2A). A thymidine kinase expression cassette was placed downstream to the right arm for negative selection. AB2.2 prime embryonic stem cells were transfected with targeting vector and screened for positive clones harboring disrupted cbln3 alleles using Southern blot analysis. The external probe for Southern blotting was a 0.57-kb XhoI-XbaI genomic DNA fragment upstream to the left arm, and the internal probe was a 1.45-kb XbaI-BamHI fragment cloned in pBlueScript. Positive clones were microinjected into blastocysts. Two chimeras underwent germ line transmission and generated founder strains.

The generation of cbln1^−/− mice has been described previously (9). All mice were maintained at St. Jude Children's Research Hospital and had ad libitum access to food and water. All procedures involving animals were performed in accordance with the National Institutes of Health guidelines and were approved by the Animal Care and Use Committee of St. Jude Children's Research Hospital.

Production of polyclonal anti-Cbln3 antiserum.

An antiserum was raised to an epitope in Cbln3 located in the equivalent site to the cerebellin peptide motif in Cbln1. The synthetic peptide APPGRVAFAAVRSHHH, corresponding to amino acids 59 to 74 of mouse Cbln3, was coupled to keyhole limpet hemocyanin. Rabbits were immunized with 0.5 mg of keyhole limpet hemocyanin-peptide conjugate in complete Freund's adjuvant and given three booster immunizations with 0.25 mg of conjugate in incomplete Freund's adjuvant (Rockland, Gilbertsville, PA) at 1-week intervals. Antisera were further purified with an affinity column comprising Sepharose coupled to the peptide antigen. The specificity of the anti-Cbln3 antiserum was established by the immunoblotting of cerebellar extracts from cbln3^−/− mice (Fig. 2D) and lysates of COS-7 cells transfected with recombinant vectors expressing Cbln1, Cbln2, Cbln3, or Cbln4 (see Fig. 5B). Further, the immunoreactive bands in lysates of cbln3-transfected cells were competed when the anti-Cbln3 antiserum was preadsorbed with excess immunogen peptide but not an irrelevant peptide (data not shown).

Immunoblotting.

Culture medium, cell lysates, and cerebellar extracts were run on sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis gels and electrotransferred to polyvinylidene difluoride membranes (Bio-Rad, Hercules, CA). Subsequently, membranes were processed according to standard techniques and bound antibody was detected by the ECL chemiluminescence system (Amersham, Piscataway, NJ). For quantification, the intensity of each band was measured using FluorChem 8900 densitometric software (Alpha Innotech, San Leandro, CA) and normalized to β-actin level. Student's t test was used for statistical comparison.

Northern blotting.

The method and probes for Northern blotting were described previously (27). Briefly, total RNA from mouse cerebellum was extracted using RNAzol B (Tel-Test, Friendswood, TX) and hybridized to ³²P-labeled probes from either cbln3 or cbln1. Blots were subsequently stripped and rehybridized to a ³²P-labeled probe to β-actin.

Histological methods.

One-month-old mice were perfused transcardially with buffered 4% paraformaldehyde, and the cerebella were removed and postfixed as previously described (9). Frozen sections (16 μm) were stained using cresyl violet for standard histological analysis. For immunohistochemistry, a rabbit polyclonal antiserum to PEP-19 (the product of the pcp4 gene) was used to immunostain cerebellar Purkinje cells as described previously (45).

Rota-rod test.

Wild-type, cbln1^−/− and cbln3^−/− mice (gender and age matched; n = 6 to 8) were tested on an accelerating Rota-rod (San Diego Instruments, San Diego, CA). The Rota-rod was programmed to accelerate from 0 to 40 rpm in 4 min. Each mouse was tested over 4 consecutive days, with two 4-min trials each day. The latency of the mice to fall from the rod was scored as an index of their motor coordination.

Deglycosylation.

Cerebellar extracts were subjected to deglycosylation using endoglycosidase H (endo-H) (Roche, Indianapolis, IN) and N-glycosidase F (Calbiochem, La Jolla, CA). One hundred micrograms of extract was heated to 95°C for 5 min in reaction buffer (20 mM Tris-HCl, 50 mM sodium chloride, 0.2% SDS, 1% β-mercaptoethanol, pH 6.5), and subsequently, endoglycosidase H (10 mU) or N-glycosidase F (5 mU) was added and samples were incubated at 37°C for 4 h. The reaction was terminated by the addition of an equal volume of 2× sample buffer and then subjected to SDS-polyacrylamide gel electrophoresis and immunoblotting.

Cell cultures and transfection.

Primary dissociated cerebellar neuron cultures were prepared as previously described (1). After 3 weeks of culture, the medium and cell lysate were subjected to immunoblotting. African green monkey kidney COS-7 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Invitrogen) in 5% CO₂ at 37°C. Cells were transfected using the FuGENE 6 transfection reagent (Roche, Indianapolis, IN) according to the manufacturer's instructions. Forty-eight hours after transfection, cells were scraped into SDS sample-loading buffer (50 mM Tris-HCl, pH 6.8, 2% SDS, 10% glycerol, 0.01% bromophenol blue) and subjected to immunoblotting.

Coimmunoprecipitation.

Cerebellar extract or culture medium was precleared by adding 50 μl protein G-agarose beads (Roche) per 1 ml extract for 1 h at 4°C. Ten micrograms of immunoglobulin was added to supernatant samples, and the incubation continued overnight at 4°C. Subsequently, 50 μl protein G-agarose beads was added and the incubation continued for 1 h at 4°C. Protein G-agarose beads were then washed four times with a buffer containing 150 mM NaCl, 50 mM Tris-HCl, pH 6.8, 5 mM EDTA, 0.5% NP-40, 0.1% deoxycholate, and EDTA-free protease inhibitor cocktail (Roche). Beads were boiled for 5 min in 50 μl of 2× SDS sample-loading buffer, and supernatants were analyzed by immunoblotting as described above.

Yeast two-hybrid analysis.

The methods for yeast two-hybrid analysis were described previously (27). Briefly, cDNAs of interest were introduced into the yeast-Escherichia coli shuttle vector pSD10a and/or the Y.LexA vector. Saccharomyces cerevisiae strain S260 (lacking URA3 and TRP1), which contains a genomic lexA-operator-lacZ reporter gene integrated into the URA3 locus (16), was cotransformed with the LexA and VP16 fusion constructs. Transformants were selected on plates lacking Trp and Ura and transferred onto HybondN filters (Amersham, Piscataway, NJ). Filters were transferred to galactose medium to induce the expression of the fusion proteins, and LacZ-positive colonies were identified and scored in a β-galactosidase assay using 5-bromo-4-chloro-3-indolyl-β-d-galactopyranosidase (X-Gal) from Promega (Madison, WI) as a substrate. Results of protein-protein interactions are shown as “++,” indicating that yeast colonies turned dark blue within 1 h; “+,”indicating that yeast colonies turned blue between 1 and 2 h; “± (trace activity),” indicating that yeast colonies are just above background at 2 h; and “− (no activity),” indicating that yeast colonies are at background at 2 h (27).

Structure modeling.

The structural model of the C1q globular domains of Cbln1 and Cbln3 was generated with the Swiss-PdbViewer (Swiss Institute of Bioinformatics, Basel, Switzerland). We used the crystal structure of Acrp30 (32) as a template.

Cbln3 is a secreted glycoprotein that participates in heteromeric complexes with Cbln1.

As a prior study established that Cbln1 binds Cbln3 in yeast two-hybrid assays (27), experiments were conducted to establish whether Cbln3 is secreted and assembles into heteromeric complexes with Cbln1.

Cbln1 is secreted from cerebellar granule cells in primary dissociated cultures (1). Using the identical culture, the secretion of Cbln3 was assessed by immunoblotting using a Cbln3-specific antiserum. A major immunoreactive band of ∼28 kDa was detected in medium and cell lysate, along with several minor bands of lower molecular masses (Fig. 1A). The major immunoreactive band (28 kDa) has a molecular mass greater than that predicted for Cbln3 (∼20 kDa), suggesting posttranslational modification, possibly N-linked glycosylation. A cerebellar lysate was subjected to deglycosylation using N-glycosidase F to remove all asparagine-linked carbohydrates (1, 38). Following treatment, the major 28-kDa immunoreactive band was lost and a new major band of Cbln3-like immunoreactivity appeared at around 21 kDa (Fig. 1B), indicating that Cbln3 is an asparagine-linked glycoprotein.

To establish whether Cbln3 forms complexes with Cbln1, coimmunoprecipitation was performed on cerebellar lysates. Whereas Cbln3 exists as a single prominent species in cerebellum (Fig. 1B and C), multiple Cbln1-like immunoreactive bands, many of submonomeric size, are detected (Fig. 1C). As shown in Fig. 1C, complexes immunoprecipitated with a rabbit anti-Cbln3 antiserum (anti-Cbln3), but not an irrelevant rabbit antiserum (rabbit immunoglobulin G), contained a single Cbln1-like immunoreactive band at the correct molecular mass (∼35 kDa) (1). Although the anti-Cbln1 antiserum detects multiple Cbln1-like immunoreactive bands in cerebellar extracts, only the full-length, secreted form of Cbln1 coprecipitates with Cbln3, suggesting that the heteromeric complexes are composed of the secreted forms of the proteins. Thus, Cbln1 and Cbln3 are present in the same molecular complex in vivo.

Various technical and theoretical limitations preclude us from answering several important questions regarding the composition and relative abundance of Cbln-containing complexes in brain. Our lack of knowledge of the precise subunit stoichiometry of Cbln-containing complexes in vivo and technical limitations intrinsic to coimmunoprecipitation immunoblotting from whole tissues make it impossible to unambiguously determine the relative proportions of Cbln1 homomeric and Cbln1-Cbln3 heteromeric complexes in the cerebellum. Similarly, technical constraints involving the immunodepletion of whole-brain extracts make it impossible to assess unequivocally whether Cbln3 is found only in complex with Cbln1. Nevertheless, we established that full-length Cbln1 can participate in homomeric complexes as well as oligomeric complexes with mature Cbln3.

Phenotype of cbln3-deficient mice.

To study the function of Cbln3, we used homologous recombination to eliminate all three exons of cbln3 in the mouse genome (Fig. 2A). Successful targeting of cbln3 was confirmed by Southern blotting using external (probe A) and internal (probe B) probes (Fig. 2B). Northern blotting and Western blotting were performed on cerebella of cbln3^+/+, cbln3^+/−, and cbln3^−/− mice. Homozygous cbln3^−/− mice exhibited a complete loss of cbln3 mRNA (Fig. 2C) and protein (Fig. 2D), whereas heterozygous mice had intermediate levels of mRNA and protein (Fig. 2C and D).

The cbln3^−/− mice are viable and fertile and have normal life spans and no obvious gross anatomical anomalies. As cbln3 is almost exclusively expressed in cerebellum (27), a more detailed study of this brain region was undertaken. The cerebella of cbln3^−/− mice had normal foliation and laminated structures (Fig. 2E). All principal neuronal types were present in the cerebella of cbln3^−/− mice (Fig. 2E), and Purkinje cells appeared normal when immunostained for PCP4/PEP-19 (Fig. 2E, lower panel), a Purkinje cell marker (45). In contrast to that with ataxic cbln1^−/− mice (1), no behavioral deficit was observed in cbln3^−/− mice based on general activity, gait analysis (data not shown), and performance on the accelerating Rota-rod (Fig. 2F).

We next tested whether there was a genetic interaction between cbln1 and cbln3 at the level of the cerebellar phenotype. cbln1^+/− cbln3^+/− and cbln1^+/− cbln3^−/− mice had normal phenotypes (data not shown). Thus, a partial or complete loss of Cbln3 does not produce a cbln1-null-like phenotype in a cbln1 heterozygous animal. Furthermore, the neuroanatomical and behavioral phenotypes of cbln1^−/− cbln3^−/− mice were indistinguishable from those of cbln1^−/− cbln3^+/+ mice (data not shown). Thus, although Cbln1 and Cbln3 are coexpressed and secreted from cerebellar granule cells in the same complex, null alleles of the two genes have distinct phenotypes, with no obvious genetic interaction between the two.

Inverse relationship between the levels of Cbln1 and Cbln3.

One explanation for the discordance in phenotypes between cbln1 and cbln3 knockout mice is functional compensation at the transcriptional or posttranslational level. Therefore, Northern and Western blotting was performed on cbln1^−/− and cbln3^−/− cerebella. No compensatory changes in mRNA levels for cbln1 or cbln3 were observed in either knockout strain (Fig. 3A). In striking contrast, there were profound changes in the levels of the two proteins (Fig. 3B). In cbln1-null mice, the level of Cbln3 was barely detectable (Fig. 3B). In contrast, Cbln1 levels were greatly elevated in cbln3-null mice (Fig. 3B). Quantitative immunoblotting revealed a gene dosage-dependent effect. Cbln3 levels were reduced by ∼46% in cbln1^+/− mice and ∼93% in cbln1^−/− mice (Fig. 3C and D). In contrast, cbln3^+/− mice had 170% the level of Cbln1 and cbln3^−/− mice had a 640% increase (Fig. 3E and F). Thus, cbln1^−/− mice essentially represent a double knockout of Cbln1 and Cbln3. This potentially explains why cbln1 cbln3 double-knockout mice are no more affected than cbln1-null mice.

Additional experiments elucidated the mechanisms underlying the interdependence between Cbln1 and Cbln3 protein levels.

Cbln1 is required for ER export and secretion of Cbln3.

The levels of many secreted glycoproteins are regulated by the quality control mechanism for endoplasmic reticulum (ER) exit (6, 39). Therefore, the influence of Cbln1 on ER retention and the secretion of Cbln3 was assessed.

A hallmark of proteins located or retained in the ER is the presence of asparagine-linked, high-mannose-containing carbohydrate chains that can be specifically cleaved by endo-H (14). Therefore, the glycosylation state of Cbln1 and Cbln3 was determined following endo-H digestion of cerebellar extracts (Fig. 4A). Cbln1 was insensitive to endo-H digestion in both wild-type and cbln3^−/− mice (Fig. 4A, upper panel). Whereas Cbln3 was also endo-H insensitive in wild-type mice (Fig. 4A, left side of lower panel) the residual Cbln3 in cbln1^−/− mice was completely endo-H sensitive (note the generation of the lower-molecular-weight band indicated in Fig. 4A). To confirm that the molecular weight shift was not due to another enzymatic activity in the endo-H, lysates were treated with N-glycosidase F (PNGaseF) that removes all asparagine-linked carbohydrates (38). Both Cbln1 and Cbln3 were sensitive to PNGaseF (Fig. 4A). The presence of immature Cbln3 in the cbln1-null cerebellum suggests that Cbln1 is essential for Cbln3 to exit the ER.

To assess whether Cbln1 influences the secretion of Cbln3, primary cultures of wild-type, cbln1^−/−, and cbln3^−/− cerebellar neurons were analyzed by Western blotting. As with Cbln1 (1), cultures from wild-type mice secrete large amounts of Cbln3 (Fig. 4B). When corrected for loading, approximately 95% of total Cbln3 is in the medium. Mirroring the situation in vivo, there is little Cbln3 present in cell lysates from cultures generated from cbln1-null mice (Fig. 4B). Moreover, only a trace of Cbln3 is present in the medium of these cultures (Fig. 4B). Based upon the ratio of Cbln3 in medium to lysate from wild-type (20:1) versus cbln1-null (3:1) cells, this result is consistent with impaired secretion of Cbln3 in cbln1-null granule cells. However, this result is ambiguous given the low levels of Cbln3.

To establish whether Cbln1 modifies the secretion of Cbln3, recombinant Cbln1 and Cbln3 were expressed singly or in combination in COS-7 cells and medium and cell lysate were analyzed by immunoblotting. Cbln1 and Cbln3 were detected in the lysates of transfected cells (Fig. 4C). Unlike primary granule cells, neither protein appeared to influence the level of the other. In marked contrast, whereas Cbln1 was robustly secreted into the medium, Cbln3 was secreted only when coexpressed with Cbln1 (Fig. 4C). Thus, despite equivalent levels of Cbln3 within the intracellular compartment, Cbln3 is secreted only when Cbln1 is present, potentially because Cbln1 binds to Cbln3.

Flag-tagged Cbln1 (Flag-Cbln1) and V5-tagged Cbln3 (Cbln3-V5) were expressed in COS-7 cells and heteromeric complexes characterized by sequential immunoprecipitation and immunoblotting with antibodies to the respective epitope tags (Fig. 4D). Cbln3-V5 was detected in the medium only when coexpressed with Flag-Cbln1 (Fig. 4D, bottom panel). Furthermore, the Cbln3-V5 present in the medium from cotransfected cells was coimmunoprecipitated with Flag-Cbln1 (Fig. 4D, first and third panels). Noticeably, whereas submonomeric fragments of Cbln1 were detected in transfected cells, only the mature form of the protein was present in the secreted complexes (Fig. 4C). Thus, full-length secreted Cbln1 and Cbln3 form heteromeric complexes both in vivo and in heterologous systems.

Proteins retained in the ER by the quality control mechanism are subsequently translocated to the cytosol for degradation, predominantly in the proteasome (39). To determine whether Cbln3 is subject to similar processing, COS-7 cells expressing Cbln1 or Cbln3 were treated with a proteasome inhibitor, lactacystin, MG132, or ZAL, or the lysosome inhibitor chloroquine. Whereas the levels of Cbln1 increased modestly in the presence of proteasome inhibitors, the levels of Cbln3 increased dramatically (Fig. 4E). Despite the marked increase in Cbln3, it was still not secreted (Fig. 4E). In the case of Cbln3, the proteasome inhibitors increased the intensity of multiple bands that were also present in untreated cells and, in addition, generated new bands of submonomeric masses (Fig. 4E). Whereas proteasome inhibitors also increased the intensity of existing bands in Cbln1-transfected cells, they did not generate any prominent new bands. Chloroquine caused a small increase in Cbln3 levels (Fig. 4E). In contrast, although not as potent as the proteasome inhibitors, chloroquine caused a relatively large increase in Cbln1-like immunoreactive bands (Fig. 4E). These data suggest that Cbln3 is degraded predominantly via the proteasome, whereas Cbln1 is eliminated in both the proteasome and the lysosome.

Specificity of Cbln1 family member interactions.

All Cbln1 family members share a C1q globular domain in their C termini, raising the possibility of additional homomeric and heteromeric complexes. Therefore, protein-protein interactions among the four family members were assessed using the yeast two-hybrid assay. All family members interacted with all other family members (Fig. 5A). In addition, all family members, with the striking exception of Cbln3 (Fig. 5A), interacted with themselves. Thus, Cbln3 is unique in being unable to undergo homomeric association.

The data also raise the question of whether Cbln2 or Cbln4 can rescue the secretion of Cbln3. Therefore, we examined the secretion of Cbln3 in COS-7 cells cotransfected with cbln2 or cbln4. Cbln4, a secreted protein, rescued the secretion of Cbln3 (Fig. 5B). In contrast, Cbln2, a nonsecreted transmembrane protein (41), did not (Fig. 5B). Thus, heteromeric binding of Cbln3 to a family member is essential for secretion but is insufficient if the interacting partner is not itself secreted.

The C1q domain of Cbln3 is essential for heteromeric interactions.

The C1q globular domain in Cbln1 is essential for mediating homomeric protein-protein interactions (1). However, the structural bases of heteromeric interactions among Cbln1 family members and the lack of homomeric interactions of Cbln3 are undetermined. Therefore, structure-binding analyses were performed to assess whether the C1q domains in Cbln1 and Cbln3 were required for heteromeric association and whether sequences outside of the C1q domain in Cbln3 inhibited its homomeric binding.

N-terminal truncations of Cbln1 and Cbln3 were generated with either LexA or VP16 fused to their N termini (Fig. 6A) in yeast expression plasmids. The choice of truncations is based upon a structure-binding analysis of homomeric interactions in Cbln1 (1). Cbln1Δ21N and Cbln3Δ24N are the secreted mature forms of Cbln1 and Cbln3, respectively. Cbln1Δ54N is the product generated by cleavage at a putative tumor necrosis factor (TNF) alpha-cleaving enzyme-like site in Cbln1 responsible for liberating the peptide, cerebellin (1, 33, 34, 40), and Cbln3Δ57N is the equivalent fragment in Cbln3. Cbln1Δ61N and Cbln3Δ63N constitute just the C1q domains of the two family members. Cbln1Δ72N is the predicted product of cleavage at the C terminus of the cerebellin peptide sequence in Cbln1 (1), and Cbln3Δ74N is the equivalent fragment of Cbln3; these products lack the first beta strand (designated A) of the C1q motif and abolish homomeric interactions in Cbln1 (1).

For the assays summarized in Fig. 6A and B, we assessed the binding of mutant (Mt) forms of Cbln1 or Cbln3 to wild-type Cbln1 or Cbln3 in both LexA and VP16 configurations. In addition, homomeric binding of mutant forms of Cbln1 and Cbln3 was also assessed. Full-length Cbln1, Cbln1Δ21N, Cbln1Δ54N, and Cbln1Δ61N bound to Cbln3, whereas Cbln1Δ72N, which deletes the A strand of the C1q globular domain, did not (Fig. 6A). In assays of Cbln3 mutants, full-length Cbln3, Cbln3Δ24N, and Cbln3Δ57N bound to Cbln1 and Cbln3Δ63N showed weaker or marginal binding, whereas Cbln3Δ74N did not bind to Cbln1 (Fig. 6B). Thus, the C1q globular domains in both Cbln1 and Cbln3 are essential for heteromeric interactions. In addition, eliminating all sequences N terminal to the C1q domain of Cbln3 (mt-Cbln3Δ63N with mt-Cbln3Δ63N) does not confer the ability to undergo self-interaction (Fig. 6B). Therefore, the structural basis for the inability of Cbln3 to form a homomeric complex must lie within the C1q domain.

Arg119 in Cbln3 determines homomeric binding and secretion.

To identify amino acids in Cbln3 that influence homomeric assembly, two assumptions were made. First, amino acids that preclude protein-protein association should lie at or near subunit interfaces and, second, they should differ between Cbln3 and other family members. To identify interfaces, structural models of the trimeric C1q domains of Cbln1 and Cbln3 were generated using Swiss-PdbViewer software (Fig. 7A). The C1q domain contains 10 beta strands (designated A to H) oriented in a so-called 10-fold beta sandwich configuration, the structural hallmark of the C1q/TNF superfamily of proteins (32). The C strand lies at the interface of the three subunits (Fig. 7A) and is the most highly conserved region in terms of amino acid sequence in not only the Cbln1 family but also the entire C1q/TNF superfamily (1, 11, 13). Surveying a range of C1q/TNF superfamily members revealed that Cbln3 was unique in possessing a charged arginine (R119) residue within the C strand (1, 11, 13) (Fig. 7A). Within the predicted model of Cbln3 homotrimers, the side chains of arginine 119 converge at the center of the structure (Fig. 7A and B). Charged residues can provide intersubunit electrostatic repulsion, thereby regulating the assembly and stability of some proteins (4, 5). Noticeably, the equivalent amino acid in Cbln1 (asparagine 115) does not constitute a steric clash in a predicted Cbln1 homotrimer, and it can also potentially accommodate the arginine residue in a Cbln1+Cbln3 heterotrimer (Fig. 7B).

Arginine- and lysine-containing motifs have been implicated in ER retention and retrieval, especially in the immune and nervous systems (7, 20, 29, 30, 37). We therefore used mutational analysis to establish whether arginine 119 prevents Cbln3 from assembling into homotrimers and ultimately undergoing secretion. Cbln1, Cbln3, and the point mutant, Cbln3(R119A) were expressed in LexA and VP16 fusion vectors and subjected to yeast two-hybrid analysis (Fig. 7C). Cbln3 did not undergo self-association but did bind to Cbln1. However, Cbln3(R119A) not only bound to Cbln1 but also interacted with itself and Cbln3 (Fig. 7C), indicating that Arg119 is a determinant for assembly. To establish whether this was also true for secretion, Cbln3 and Cbln3(R119A) were expressed in COS-7 cells in the presence or absence of Cbln1 and medium and cell lysates were analyzed by Western blotting. Whereas Cbln3 was not secreted into the medium in the absence of Cbln1, Cbln3(R119A) was robustly secreted (Fig. 7D, left panel). Neither Cbln3 nor Cbln3(R119A) markedly influenced the secretion of Cbln1 in cotransfected cells (Fig. 7D, right panel). Thus, a single arginine residue in Cbln3 appears responsible for its inability to undergo homomeric association and secretion.