Identification of Noncollagenous Sites Encoding Specific Interactions and Quaternary Assembly of α3α4α5(IV) Collagen

Defective assembly of α3α4α5(IV) collagen in the glomerular basement membrane causes Alport syndrome, a hereditary glomerulonephritis progressing to end-stage kidney failure. Assembly of collagen IV chains into heterotrimeric molecules and networks is driven by their noncollagenous (NC1) domains, but the sites encoding the specificity of these interactions are not known. To identify the sites directing quaternary assembly of α3α4α5(IV) collagen, correctly folded NC1 chimeras were produced, and their interactions with other NC1 monomers were evaluated. All α1/α5 chimeras containing α5NC1 residues 188-227 replicated the ability of α5NC1 to bind to α3NC1 and co-assemble into NC1 hexamers. Conversely, substitution of α5NC1 residues 188-227 by α1NC1 abolished these quaternary interactions. The amino-terminal 58 residues of α3NC1 encoded binding to α5NC1, but this interaction was not sufficient for hexamer co-assembly. Because α5NC1 residues 188-227 are necessary and sufficient for assembly into α3α4α5NC1 hexamers, whereas the immunodominant alloantigenic sites of α5NC1 do not encode specific quaternary interactions, the findings provide a basis for the rational design of less immunogenic α5(IV) collagen constructs for the gene therapy of X-linked Alport patients.

Type IV collagen is a major constituent of basement membranes, evolutionarily conserved across animal phyla. In mammals, three pairs of genes (COL4A1-COL4A6) encode six homologous chains (␣1-␣6) that form heterotrimeric molecules and networks. Specificity of assembly is encoded by noncollagenous (NC1) 2 domains at the carboxyl terminus of each chain (1). Interactions among NC1 domains direct the assem-bly of trimeric molecules and mediate dimerization of molecules via their NC1 ends, forming NC1 hexamers. Analysis of NC1 hexamers from tissues has identified three collagen IV networks with different chain composition and tissue distribution. The ubiquitous network composed of ␣1␣2␣1(IV) trimers is essential for development (2). A network containing ␣5␣6␣5(IV) trimers (3) occurs in skin, smooth muscle, and Bowman capsule basement membranes. A network composed of ␣3␣4␣5(IV) trimers (4) forms the scaffolding of the glomerular basement membrane (GBM) and maintains the integrity of the blood ultrafiltration barrier.
Mutations preventing the normal assembly of ␣3␣4␣5(IV) collagen in the GBM cause Alport syndrome (5), the most prevalent inherited glomerular disease leading to renal failure. A majority (ϳ85%) of patients have X-linked Alport syndrome (XLAS) because of mutations in the COL4A5 gene. The rarer autosomal recessive Alport syndrome is due to mutations in both alleles of the COL4A3 or COL4A4 gene. These mutations cause the persistence of the embryonic (␣1) 2 ␣2(IV) collagen in the Alport GBM, deterioration of GBM ultrastructure, and eventual failure of the glomerular filtration barrier. Patients progress slowly but inexorably to end-stage renal disease, requiring dialysis or kidney replacement.
No specific treatments for Alport syndrome currently exist. Among prospective remedies, gene therapy holds the promise of correcting the underlying molecular defect (6). The feasibility of delivering ␣5(IV) collagen to glomerular cells was demonstrated in pigs (7). Moreover, intramuscular injection of an adenoviral ␣5(IV) construct restored expression of the ␣5(IV) chain and corrected the assembly of (␣5) 2 ␣6(IV) collagen in the smooth muscle of XLAS dogs (8). However, a potential complication of gene therapy is the immunogenicity of exogenous collagen IV chains. Alloimmune reactions occur in Alport patients after a kidney transplant, causing Alport posttransplant nephritis and loss of the allograft in ϳ3-5% of cases (9). Alport posttransplant nephritis is mediated by alloantibodies targeting the NC1 domains of ␣3␣4␣5(IV) collagen in the allograft GBM (10 -13). In XLAS patients with Alport posttransplant nephritis, serum and allograft-bound alloantibodies are directed against two major alloepitopes mapping to ␣5NC1 residues 1-45 and 114 -168 (14). These sites are also likely to elicit an alloimmune reaction after gene therapy, as found in XLAS dogs injected with an adenoviral ␣5(IV) collagen construct. 3 Editing out the antigenic sites of ␣5(IV) collagen delivered by gene therapy is a possible strategy for diminishing its immunogenicity in Alport recipients, but the impact of such modifications on the ability of ␣5(IV) chain to co-assemble with ␣3 and ␣4(IV) chains is not known. We conjectured that the sites of ␣5NC1 important for its quaternary assembly are distinct from the alloantigenic sites. To test this hypothesis, chimeric ␣1NC1/␣5NC1 domains were used to identify which ␣5NC1 regions encode specific interactions with ␣3NC1 and ␣4NC1 monomers. The carboxyl-terminal residues 188 -227 of the ␣5NC1 domain were found to be necessary and sufficient for specific binding to ␣3NC1 monomers and subsequent assembly into ␣3␣4␣5NC1 hexamers. The finding that the alloantigenic regions of ␣5NC1 are not essential for its quaternary interactions provides a basis for the rational design of improved ␣5(IV) collagen constructs for the gene therapy of X-linked Alport syndrome.

EXPERIMENTAL PROCEDURES
Materials-Recombinant ␣1/␣5NC1 chimeras were produced by swapping fragments of ␣5NC1 and ␣1NC1 cDNA using the ApaI, KpnI, and NarI restriction sites occurring naturally in the cDNA sequence or introduced by PCR. This strategy, previously used for making chimeras 5111, 1511, 1151, and 1115 (14), was employed to produce new ␣1/␣5NC1 chimeras (see Fig. 1A) with longer ␣5NC1 inserts. To make 1555, ␣5NC1 cDNA was amplified with primers 5Ј-ata ttc tag act tgg gga cgg ctg gca gct-3Ј and 5Ј-cag cga gct cta gca ttt agg-3Ј, and ␣1NC1 cDNA with primers 5Ј-agg ccc aag ctt ctg cct gcc gcc t-3Ј and 5Ј-caa ttc tag acc atg ggc ccg ttc att gcc t-3Ј. For 1155, ␣5NC1 cDNA was amplified with primers 5Ј-ata agg cgc cag ctg tgg tga tcg cag ttc a-3Ј and 5Ј-gta tcc gcg gtg tcc tct tca tgc aca ctt g-3Ј. For 5511, ␣5NC1 cDNA was amplified with primers 5Ј-aga ccc aag ctt ctg cct gcc gcc t-3Ј and 5Ј-agt tgg cgc ctc aca tac tgc aca tcg act a-3Ј. For 5551, the ␣5NC1 cDNA fragment digested with KpnI was subcloned into the KpnI site of the 1151 construct. All cDNAs in the pBluescript SK vector were sequenced using ABI 310 to verify the sequence of the constructs and then subcloned into the pRC-X expression vector. After transfection of HEK-293 cells (70% confluent) by calcium phosphate co-precipitation with 4 g of plasmid DNA, resistant clones were selected with G418 (250 g/ml) and screened for production of recombinant protein. NC1 chimeras were purified from the cell culture supernatant by affinity chromatography on immobilized anti-FLAG mAb M2. Chimera 5511 was not expressed by the transfected cells; other recombinant chimeras migrated at the expected size of 25-30 kDa by SDS-PAGE and were judged to be pure by protein staining and immunoblotting with anti-FLAG mAb (see Fig. 1, B and C). Recombinant human NC1 monomers (1) and ␣1/␣3NC1 chimeras 331, 311, 131, 113, and D3-D7 (15)(16)(17) were produced as described.
ELISA-Purified NC1 monomers or chimeras in 50 mM sodium carbonate buffer, pH 9.5, were coated overnight onto plastic microtiter plates (Nunc-Immuno TM modules). For some assays, the NC1 domains were misfolded prior to coating by the reduction of disulfide bonds with 50 mM tris(2-carboxyethyl)phosphine hydrochloride for 10 min at 60°C. The binding of mAbs to immobilized NC1 domains was detected with alkaline phosphatase-conjugated secondary antibodies to rat or mouse IgG, followed by chromogenic substrate (p-nitrophenol phosphate). For solid-phase binding assays, immobilized NC1 domains were incubated for 1 h at room temperature with ␣3NC1, ␣4NC1, and ␣5NC1 monomers tagged by mAbs 8D1, RH41, and b14, respectively. The binding to the solid phase of the tagging mAb via the soluble NC1 domain was then detected  with the appropriate secondary antibody. Each measurement was performed in duplicate, and all experiments were repeated at least three times with similar results.
Hexamer Assembly-In vitro co-assembly of recombinant NC1 monomers and chimeras was performed as described (1,4). Briefly, equal amounts of NC1 domains were mixed for 3 min at pH 5.0; the pH was adjusted to neutral with 1.5 M Tris, pH 8.0; and 1 M sodium chloride was added to dilute the NC1 domains to ϳ1 M each. After an overnight incubation at room temperature, the reaction products were separated according to their size by HPLC gel filtration on a Bio-Sil TSK250 column.
The composition of NC1 hexamers assembled in vitro was determined by immunoprecipitation and immunoblot analysis. The NC1 complexes eluted at hexamer size were incubated with mAb 8D1 for 1 h; immune complexes were isolated on protein G-Sepharose beads and separated by SDS-PAGE in 4 -20% gradient gels under nonreducing conditions; and then the NC1 domains transferred to Immobilon-P membranes were identified by staining with mAbs H31, H43, and H52.
Analysis of the Contact Interfaces in NC1 Hexamers-The surface area of residues buried by interactions between subunits of the NC1 hexamer of ␣1␣2(IV) collagen (Protein Data Bank code 1T61) (21) was calculated using PISA (Protein Interfaces, Surfaces and Assemblies) at the European Bioinformatics Institute (www.ebi.ac.uk/msd-srv/prot_int/pistart.html) (22). The three-dimensional model of the NC1 hexamer was rendered with Rasmol version 2.0.

Expression and Correct Folding of Recombinant ␣1/␣5NC1
Chimeras-To identify which regions of ␣5NC1 are important for specific assembly into quaternary structures, a strategy based on chimeric ␣1/␣5NC1 domains was used. This approach relies on the inability of ␣1NC1 to co-assemble with ␣3 and ␣4 NC1 domains despite its high homology to ␣5NC1. Chimeric ␣1/␣5NC1 domains were generated by swapping varying lengths of ␣1NC1 sequences at the amino-or carboxylterminal ends of ␣5NC1 (Fig. 1). The chimeras contained an amino-terminal FLAG sequence for affinity purification and were expressed in HEK-293 cells for correct folding.
Analysis of Binary Interactions among ␣3NC1, ␣4NC1, and ␣5NC1 Monomers by Sandwich ELISA-Chain-specific mAbs to NC1 domains are valuable tools for analyzing binary NC1-NC1 interactions. By tagging a soluble NC1 domain with a specific mAb, its binding to surface-bound NC1 domains could be assayed in a sandwich ELISA format (4). To minimize the interference of the tagging antibody on the NC1-NC1 interactions, all mAbs used for solid-phase binding assays targeted epitopes accessible in the NC1 hexamers. Soluble ␣3NC1 tagged by mAb 8D1 bound to immobilized ␣4NC1 and ␣5NC1 but not to ␣1NC1 or to misfolded ␣4NC1 or ␣5NC1 (Fig. 3A). Soluble ␣4NC1 tagged with mAb RH41 bound ␣3NC1 but not ␣5NC1 or ␣1NC1 (Fig. 3B). Soluble ␣5NC1 tagged with mAb b14 bound to ␣3NC1 but not to ␣4NC1 or ␣1NC1 monomers (Fig.  3C). These results demonstrate interactions between correctly folded ␣3NC1 and ␣4NC1 and between ␣3NC1 and ␣5NC1, whereas ␣4NC1 and ␣5NC1 do not interact significantly in the solid-phase binding assay. Moreover, ␣3NC1, ␣4NC1, and ␣5NC1 did not bind to ␣1NC1, thus validating the choice of ␣1NC1 as an inert scaffolding for ␣1/␣3NC1 and ␣1/␣5NC1  A, a solid-phase binding assay was used to measure the binding of soluble ␣3NC1 (5 g/ml) tagged by mAb 8D1 (black bars) or mAb 8D1 alone (white bars) to immobilized ␣1/␣5NC1 chimeras and control NC1 monomers. B, the same solid-phase binding assay was performed using ␣1/␣3NC1 chimera 331 tagged by mAb 8D1. DECEMBER 12, 2008 • VOLUME 283 • NUMBER 50 JOURNAL OF BIOLOGICAL CHEMISTRY 35073 chimeras. Therefore, these chimeras were used to map the NC1 sites important for the ␣3NC1-␣5NC1 interaction, as follows.

DISCUSSION
The collagen superfamily currently includes 46 genetically distinct chains classified into 28 types (23). Homotypic assembly of collagen chains into trimeric molecules is directed by interactions among terminal noncollagenous domains characteristic for each type (24). Collagen types comprising several chains can theoretically form numerous homo-and heterotrimeric molecules, yet only a few occur naturally. The small number of extant trimeric combinations suggests that the permissible interactions may be specified by yet-to-be identified recognition sequences.
Type IV collagen provides a paradigm for deciphering how the specificity of collagen assembly is encoded at the molecular level. First, of 66 possible trimeric combinations of six collagen IV chains, only three are known to exist. Association of molecules into networks also exhibits a high degree of specificity. Second, the assembly of collagen IV chains is driven by specific interactions among independently folded NC1 domains, which can be purified from tissues or expressed as recombinant proteins for in vitro assembly studies, producing native-like structures. Third, the crystal structure of the NC1 hexamers of ␣1␣2(IV) collagen has been solved (21,25,26), providing a framework for interpreting experimental results. In silico analyses of hexamer interfaces (27,28) predicted two sites within NC1 domains likely to encode specific interactions: variable region VR3 and a ␤-hairpin (Fig. 9). Fourth, defective assembly of collagen IV is the underlying cause of several inherited diseases in human patients and animal models. Given the importance of ␣3␣4␣5(IV) collagen in the glomerular homeostasis and its implication in the pathogenesis of Alport syndrome, we sought to identify which NC1 regions mediate specific assembly of ␣3␣4␣5NC1 hexamers.
In this work, we show that ␣5NC1 residues 188 -227 encompass a site encoding specific binding to ␣3NC1, whereas the 58 amino-terminal residues of ␣3NC1 encompass a site encoding specific binding to ␣5NC1. This is the first report that experimentally identifies the location of recognition sequences directing specific interactions among collagen chains. The results uniquely determine the relative orientation of ␣3, ␣4, and ␣5 chains within the ␣3␣4␣5(IV) collagen heterotrimer, verifying the predicted arrangement of subunits in the ␣3␣4␣5NC1 hexamers as deduced from the inhibition of ␣3NC1 interactions by blocking antibodies (4,29). By reference to the homologous (␣1) 2 ␣2 NC1 hexamers, ␣3NC1 is equivalent to ␣1NC1 subunits A and D, ␣5NC1 corresponds to ␣1NC1 subunits B and E, and ␣4NC1 corresponds to ␣2NC1 subunits C and F.
Although about half of the residues in the NC1 hexamer are involved in quaternary interactions, relatively few residues appear to encode the specificity of assembly. Chimera 1115, the smallest emulating the ability of ␣5NC1 to bind ␣3NC1 and co-assemble into hexamers, differs from the nonreactive ␣1NC1 at only 13 positions (Fig. 9, lower box). Most of these 13 ␣5NC1-specific residues map to the variable region VR3, one of two regions predicted in silico to encode specific NC1 interactions (27). Judging from the large surface area buried upon complexation, ␣5NC1 residues Val 196 , Ser 207 , Glu 208 , and Arg 227 are most likely to impart specific binding to ␣3NC1. The close proximity of these ␣5NC1 residues to the ␣3NC1 subunit is apparent in the NC1 hexamer model depicted in Fig. 10.
Surprisingly, no ␣5NC1-specific residues before position 188 were required for assembly into ␣3␣4␣5NC1 hexamers, even though the amino-terminal half of ␣5NC1 interfaces with the carboxyl-terminal half of ␣4NC1. Thus, substitution of specific ␣5NC1 residues by ␣1NC1 residues at the interface with ␣4NC1 can be tolerated without compromising hexamer assembly. Because ␣4NC1 and ␣5NC1 monomers have no intrinsic affinity for one another, their interaction contributes least toward the specificity of hexamer assembly and is probably the last step in the assembly of ␣3␣4␣5(IV) collagen trimers.
The 58 amino-terminal residues of ␣3NC1 were found to encode specific binding to ␣5NC1. Of these, 23 are ␣3NC1specific (Fig. 9, upper box). Eight of 10 amino-terminal residues are ␣3NC1-specific, and some of these may be important for ␣5NC1 binding, judging from the low activity of chimeras D3 and D4 relative to D5. In addition, the weak binding of ␣5NC1 to the D6 chimera suggests that other ␣3NC1-specific residues among Phe 31 , Gln 40 , Leu 50 , Gln 55 , and Arg 56 also impart specificity for ␣5NC1. In particular, Gln 55 and Arg 56 have the largest surface area buried upon complexation (Fig. 9) and are proximal to several ␣5NC1specific residues (Fig. 10). Although predicted in silico to encode specific NC1 interactions (27), the ␤-hairpin region of ␣3NC1 was not necessary for ␣5NC1 binding. Nevertheless, the variable region VR3 of ␣3NC1 was likely important for specific quaternary assembly because chimera 331 neither bound ␣4NC1 nor formed hexamers. FIGURE 10. Three-dimensional model of the NC1 hexamer depicting the location of residues involved in the interaction between ␣3NC1 and ␣5NC1 subunits. The ␣3, ␣4, and ␣5 NC1 subunits are shaded in light red, blue, and green, respectively. The ␣3NC1-specific residues within chimera D5 are shown in purple, and the ␣5NC1-specific residues within chimera 1115 are shown in dark green. Other residues at the ␣3NC1-␣5NC1 interface are colored in red for the ␣3NC1 subunit and in green for the ␣5NC1 subunit. For optimal views, left and right panels visualize the ␣3NC1-␣5NC1 interaction after the hexamer was rotated by about 40°in the direction marked by the arrows. Lower panels illustrate the footprint (dotted line) on the neighboring NC1 subunit after removing the ␣3NC1 monomer (left) or the ␣5NC1 monomer (right). Arrows point to buried interfacing residues.
Identification of NC1 regions specifying the quaternary interactions among collagen IV chains may have potential applications in the gene therapy of Alport syndrome. In animal models of genetic diseases, introduction of the gene encoding the normal protein often causes the undesirable activation of alloreactive B and/or T cells and production of antibodies against the transgenic protein (30 -33). It is therefore not surprising that XLAS dogs injected with an adenoviral ␣5(IV) collagen construct produced alloantibodies against the ␣5NC1 alloepitopes targeted by sera from XLAS patients with posttransplant nephritis. 3 Editing out such alloantigenic sites without affecting the intended function is one strategy to minimize the immunogenicity of proteins delivered by gene therapy. For instance, mutagenesis of a B cell epitope in factor VIII significantly reduced the production of inhibitory antibodies in hemophilia A mice without affecting the coagulant activity (34). Here, we show that engineering ␣5NC1 by editing out its alloantigenic sites preserves its ability to bind ␣3NC1 and coassemble into hexamers with ␣3NC1 and ␣4NC1 monomers, a key step in the quaternary assembly of ␣3␣4␣5(IV) collagen. This finding affords a strategy for the rational design of an improved ␣5(IV) collagen for gene therapy of XLAS patients. Thus, the immunodominant alloantigenic sites of ␣5NC1 not important for tissue assembly of ␣3␣4␣5(IV) networks could be replaced by homologous ␣1NC1 residues, less immunogenic because of the ubiquitous expression of endogenous ␣1(IV) collagen. The antigenicity of ␣1/␣5NC1 chimeras is presently being tested in Col4a5-null mice, a murine model of XLAS (35).