The Alloantigenic Sites of α3α4α5(IV) Collagen

Anti-glomerular basement membrane (GBM) antibody nephritis is caused by an autoimmune or alloimmune reaction to the NC1 domains of α3α4α5(IV) collagen. Some patients with X-linked Alport syndrome (XLAS) develop post-transplant nephritis mediated by pathogenic anti-GBM alloantibodies to collagen IV chains present in the renal allograft but absent from the tissues of the patient. In this work, the epitopes targeted by alloantibodies from these patients were identified and characterized. All XLAS alloantibodies recognized conformational epitopes in the NC1 domain of α5(IV) collagen, which were mapped using chimeric α1/α5 NC1 domains expressed in mammalian cells. Allograft-eluted alloantibodies mainly targeted two conformational alloepitopes mapping to α5NC1 residues 1-45 and 114-168. These regions also encompassed the major epitopes of circulating XLAS alloantibodies, which in some patients additionally targeted α5NC1 residues 169-229. Both kidney-eluted and circulating alloantibodies to α5NC1 distinctively targeted epitopes accessible in the α3α4α5NC1 hexamers of human GBM, unlike anti-GBM autoantibodies, which targeted sequestered α3NC1 epitopes. The results identify two immunodominant α5NC1 epitopes as major alloantigenic sites of α3α4α5(IV) collagen specifically implicated in the pathogenesis of post-transplant nephritis in XLAS patients. The contrast between the accessibility of these alloepitopes and the crypticity of autoepitopes indicates that distinct molecular forms of antigen may initiate the immunopathogenic processes in the two forms of anti-GBM disease.

Glomerulonephritis is an important cause of renal injury leading to end-stage kidney failure. Some of the most aggressive forms of rapid progressive glomerulonephritis are medi-ated by IgG anti-glomerular basement membrane (anti-GBM) 2 antibodies, which circulate in patient plasma and bind along the glomerular capillary loops in a linear pattern. Although anti-GBM antibody nephritis can be triggered by either autoimmune or alloimmune mechanisms, both autoantibodies and alloantibodies target the ␣3␣4␣5(IV) collagen network of the GBM, producing similar pathology findings and clinical presentations (1).
Studies of the epitopes targeted by anti-GBM antibodies and their relationship to the tertiary and quaternary structures of the antigen can provide insights into the pathogenic mechanisms at the molecular level. For instance, most patients with autoimmune anti-GBM nephritis and its variant, Goodpasture disease, have circulating autoantibodies targeting two conformational epitopes, E A and E B , within the non-collagenous (NC1) domain of ␣3(IV) collagen (2). These autoepitopes are structurally sequestered (cryptic) in the GBM by quaternary interactions among NC1 domains (3,4), which form hexamer complexes in basement membranes. In the ␣3␣4␣5 NC1 hexamer found in the GBM, the accessibility of the ␣3NC1 autoepitopes is limited by their proximity to the ␣4NC1 and ␣5NC1 subunits (5). Abnormal exposure of these cryptic epitopes by putative pathogenic factors has been postulated to circumvent immune self-tolerance to ␣3(IV) collagen, inducing autoantibodies and autoimmune glomerulonephritis.
Anti-GBM alloantibodies also target the NC1 domains of ␣3␣4␣5(IV) collagen but are elicited by a distinct mechanism, an alloimmune reaction to "foreign" collagen IV chains in the renal allograft of transplanted Alport patients. In these patients, mutations in the COL4A3 or COL4A4 genes (in autosomal recessive Alport syndrome) or in the COL4A5 gene (in X-linked Alport syndrome, XLAS) prevent the normal assembly of ␣3␣4␣5(IV) collagen in the GBM (6), causing progressive glomerulonephritis and eventually end-stage renal failure. Among patients receiving a kidney transplant, about 3-5% develop Alport post-transplant nephritis mediated by anti-GBM alloantibodies, a serious complication resulting in allograft loss in the majority (ϳ88%) of cases (7). The alloantibodies target various subunits of the ␣3␣4␣5 NC1 hexamer, depending on the underlying genetic defect. Autosomal recessive Alport syndrome patients produce alloantibodies to ␣3NC1 and ␣4NC1 domains (8 -11). XLAS patients tend to elicit alloantibodies to ␣5NC1, although other NC1 subunits may also be targeted (8,9,12). Thus far, the alloantigenic sites recognized by XLAS alloantibodies have not been identified, although the majority of cases of post-transplant nephritis occur in patients with XLAS, the most common form of Alport syndrome (ϳ85% cases).
The aim of this study was to identify and characterize the major alloantigenic sites of ␣3␣4␣5(IV) collagen in XLAS patients with post-transplant anti-GBM nephritis. Using newly engineered chimeric ␣1/␣5 NC1 domains, two immunodominant conformational alloepitopes within the NC1 domain of ␣5(IV) collagen were identified as major targets of kidneybound and circulating XLAS alloantibodies. The accessibility of the XLAS alloepitopes within the ␣3␣4␣5 NC1 hexamers of human GBM, contrasting with the crypticity of autoepitopes, implicates distinct molecular forms of the antigen as initiators of the pathogenic process in alloimmune versus autoimmune anti-GBM disease.

EXPERIMENTAL PROCEDURES
Patients-All alloantibodies characterized in this work were from transplanted male XLAS patients with clinically proven Alport post-transplant nephritis (APTN). Patient APTN-1 had hearing loss and developed end-stage kidney disease at age 20; acid-eluted alloantibodies were available from the fourth allograft. From patient APTN-2 (same as "AS-I" in Ref. 13), the second allograft was stored frozen after nephrectomy. Pieces of the allograft were thawed, homogenized, and washed with cold phosphate-buffered saline, pH 7.2, and GBM-bound alloantibodies were eluted with 0.1 M glycine at pH 2.8 and 2.2. The acid eluates were pooled and neutralized with Tris buffer, pH 8.0. By SDS-PAGE analysis, the wash fraction contained plasma proteins (albumin, IgG) trapped in the blood vessels of the renal allograft, including alloantibodies not bound to the allograft GBM. Alloantibodies in the wash fraction had distinct properties from those eluted at low pH, likely representative of the circulating alloantibodies of this patient. From patient APTN-3 (same as "patient c" in Ref. 9), protein A-bound antibodies from therapeutic immunoabsorption were used. Serum positive for anti-GBM alloantibodies by indirect immunofluorescence was available from patient APTN-4, who developed proteinuria and hematuria at 35 days after transplantation; the initial biopsy showed crescentic glomerulonephritis, and linear IgG along the allograft GBM was found at 4 and 7 months post-transplant (14). Another serum was also available from patient APTN-5, described in Ref. 15, and "patient h," described in Ref. 9. Sera or plasma exchange fluid from patients with autoimmune anti-GBM (Goodpasture) disease were previously described (2) or purchased from Wieslab AB (Lund, Sweden).
DNA Constructs-The cDNAs for human ␣1NC1 and ␣5NC1 were amplified by PCR using Pfu polymerase (Stratagene) for its low error rate. The ␣1NC1 cDNA was subcloned with the 5Ј end into the NheI and the 3Ј end into the SacII site of a pRC-X expression vector containing the BM40 signal peptide and FLAG peptide (2). The ␣5NC1 cDNA was ligated into the NheI and ApaI sites of the same vector. Because ␣1NC1 cDNA contained unique restriction sites for ApaI and NarI and ␣5NC1 cDNA for KpnI (see Fig. 1), cleavage sites for these restriction enzymes and for XbaI were introduced at the appropriate positions into the ␣1NC1 or ␣5NC1 cDNA by polymerase chain reaction using the primers listed in Table 1. Segments of ␣5NC1 cDNA between consecutive restriction sites were ligated into the respective sites of the ␣1NC1 cDNA. Restriction enzymes and ligases were purchased from New England Biolabs. All cDNAs in pBluescript SK-vector were sequenced to verify the sequence of each construct and then subcloned into pRC-X expression vector.
Cell Transfection and Expression of Recombinant Proteins-Human embryonic kidney (HEK)-293 cells were cultured in 90-mm cell culture plates in Dulbecco's modified Eagle's medium/F12 medium supplemented with 5% fetal bovine serum. For each construct, 4 g of plasmid DNA was transfected by calcium phosphate co-precipitation into 70% confluent HEK 293 cells. After 2 days, transfected cells were seeded onto a new plate and selected with 250 g/ml G418. Resistant cells were screened for expression of recombinant proteins by Western blot using anti-FLAG mAb M2. After G418-resistant cells reached confluence, culture supernatants were collected every 2 days. Recombinant proteins were affinity-purified on immobilized anti-FLAG M2 mAb columns, concentrated by ultrafiltration, and stored frozen in aliquots. Electrophoresis and Immunoblotting-Proteins were separated by SDS-polyacrylamide gel electrophoresis in 4 -20% gradient gels under non-reducing conditions followed by staining with Coomassie Brilliant Blue or transfer to Immobilon P membranes for Western blotting with antibodies to FLAG or ␣1-␣6 NC1 domains.
Enzyme-linked Immunosorbent Assays (ELISA)-Indirect ELISA was performed in 96-well plastic plates (Nunc-Immuno) coated overnight with purified antigens (100 ng/well) in 50 mM sodium carbonate buffer, pH 9.5, and blocked with 1% bovine serum albumin. In some experiments, the antigen was reduced prior to coating by treatment with 50 mM Tris(2-carboxyethyl) phosphine hydrochloride for 10 min at 60°C. Binding of human IgG or mouse mAbs to the immobilized antigens was detected with alkaline phosphatase-conjugated antibodies to human or mouse IgG. Hydrolysis of p-nitrophenyl phosphate was monitored at 405 nm using a SpectraMax 190 plate reader (Molecular Devices). For inhibition ELISA, alloantibodies were preincubated overnight at room temperature with various concentrations of NC1 antigens before measuring binding to immobilized ␣5NC1 or chimeras. Capture ELISA was adapted from Ref. 4, using different pairs of capture and detecting antibodies. NC1 hexamers from human kidney cortex (1 g), native or dissociated by brief exposure to low pH, were incubated in wells precoated with 300 ng of mAb 8D1 (recognizing ␣3NC1 epitopes accessible in NC1 hexamers). The binding of human anti-GBM allo-or autoantibodies to captured antigen was detected with alkaline phosphatase-conjugated anti-human IgG.
ELISA measurements were performed in duplicate determinations and repeated three times. The results are shown as means and standard deviations. For each alloantibody, the significance of differences in binding to various antigens was analyzed by one-way analysis of variance followed by Dunnett's test for multiple comparisons (versus control ␣1NC1). For overall comparisons among all alloantibodies, the binding data were analyzed by repeated measures analysis of variance. Significance was inferred when p was Ͻ 0.05. GraphPad Prism version 4.03 was used for all statistical analyses.

Anti-GBM Alloantibodies from XLAS Patients with Posttransplant Nephritis Primarily Target Conformational Epitopes within the NC1 Domain of ␣5(IV)
Collagen-Before selecting a strategy for mapping the XLAS alloepitopes, the specificity of alloantibodies for recombinant human NC1 monomers expressed in HEK-293 cells was determined by indirect ELISA (Fig. 1). Allograft-eluted XLAS alloantibodies from patient APTN1 (Fig. 1a) reacted strongly and specifically with ␣5NC1, whereas those from APTN-2 ( Fig. 1b) also exhibited weak binding to ␣3NC1 and ␣4NC1 (ϳ6% relative to ␣5NC1). Circulating XLAS alloantibodies from all four patients (Fig. 1, c-f) bound to ␣5NC1, and two patients (APTN-4 and -5) also had alloanti-bodies to ␣3NC1. Reduction of disulfide bonds in the reactive NC1 monomers strongly decreased the binding of XLAS alloantibodies from all patients by 50 -90% (Fig. 1, solid bars), verifying the conformational requirements of the alloepitopes. For comparison, the binding of Mab5 to ␣5NC1 was not affected by reduction (less than 3% decrease in reactivity; data not shown), indicating that Mab5 recognizes linear epitopes. An overall comparison among all patients revealed that the binding of XLAS alloantibodies to ␣5NC1 was significantly higher than that to any other NC1 domain; other differences were not statistically significant. To map the conformational ␣5NC1 alloepitope(s), a strategy based on chimeric ␣1/␣5 NC1 domains was chosen, analogous to those used for mapping conformational ␣3NC1 epitopes (2,17). The ␣1NC1 was selected as inert scaffolding because it is highly homologous to ␣5NC1 (ϳ83% sequence identity) but did not bind XLAS alloantibodies.
Because serum IgG from patient APTN-3 reacted with three ␣1/␣5 chimeras, the existence of alloantibodies targeting three distinct epitopes was verified by reciprocal inhibition. The binding of APTN-3 serum IgG to each of the immobilized chimeras 5111, 1151, and 1115 was fully inhibited only by the same chimera in soluble form, which was as effective as a mixture of all chimeras or the whole ␣5NC1 (Fig. 5, e-g). The binding of APTN-3 to immobilized ␣5NC1 (Fig. 5h) was more strongly inhibited by chimera 1151 (46%) than by 5111 (22%) or 1115 (21%). The combined inhibition by all chimeras was 65%, as compared with 90% by ␣5NC1, indicating that ϳ28% of APTN-3 alloantibodies target composite ␣5NC1 epitopes.

Binding of XLAS Alloantibodies to Human NC1 Hexamers and the Accessibility of Alloepitopes in ␣3␣4␣5NC1
Hexamers-Differences in epitope specificity between kidney-eluted and circulating alloantibodies, as in patient APTN-2, suggest that subsets of alloantibodies with higher affinity for the native form of the alloantigen are enriched in the allograft GBM. Consistent with this hypothesis, human kidney NC1 hexamers strongly inhibited the binding to immobilized ␣5NC1 of kidney-eluted alloantibodies (Fig. 6). The binding to ␣5NC1 of circulating alloantibodies was also inhibited but at higher concentrations of hexamer, indicat- Conserved amino acid residues are indicated by asterisks. Among residues specific for ␣5NC1, those indicated in bold are predicted to be accessible in the NC1 hexamer based on the analysis of the x-ray structure of homologous ␣1␣2NC1 hexamers. Arrowheads indicate the boundaries of regions swapped in ␣1/␣5NC1 chimeras. The location of ␣3NC1 epitopes E A and E B is indicated by solid and interrupted lines, respectively. B, schematic representation of cloning strategy for constructing human ␣1/␣5NC1 chimeras. The ApaI, KpnI, NarI, and XbaI cleavage sites were introduced into the ␣1NC1 and ␣5NC1, and then quarter-domains of ␣5NC1 were substituted into the corresponding position within NC1 to produce four chimeric constructs, designated 5111, 1511, 1151, and 1115. The chimeric constructs consisted of FLAG peptide followed by two or four Gly-X-Y repeats (for ␣1NC1 or ␣5NC1-derived amino termini, respectively) and the full NC1 sequence cloned into the pRC-X expression vector. UTR, untranslated region; aa, amino acids. APRIL 6, 2007 • VOLUME 282 • NUMBER 14 ing lower affinity of interaction. Because NC1 hexamers could inhibit the binding to ␣5NC1 of kidney-bound and circulating XLAS alloantibodies, the ␣5NC1 alloepitopes must be accessible in the quaternary structure of the NC1 hexamers.

DISCUSSION
This study identifies for the first time the major alloantigenic sites of ␣5(IV) collagen, targeted by both kidneybound and circulating anti-GBM alloantibodies from XLAS patients with post-transplant nephritis. All XLAS alloantibodies characterized in this work reacted predominantly (and some exclusively) with conformational epitopes within the ␣5NC1 domain. Within the ␣5NC1 domain, three alloreactive regions were identified based on the binding of XLAS alloantibodies to ␣1/␣5 NC1 chimeras. All three XLAS alloepitopes were conformation-dependent, as shown by the loss of alloantibody binding to ␣1/␣5 chimeras after the reduction of disulfide bonds. Furthermore, the XLAS alloepitopes were found to be accessible in the native ␣3␣4␣5NC1 hexamers, and thus, in the ␣3␣4␣5(IV) collagen network occurring in the normal human GBM.
Within the two nephritogenic regions of ␣5NC1, the candidate residues that form the XLAS alloepitopes were identified (Fig. 9). This was achieved by comparing the primary structure of ␣5NC1 with that of the non-immunoreactive ␣1NC1 to identify the ␣5NC1-specific residues followed by an analysis of the accessibility of each residue within the NC1 hexamer by comparison with the x-ray structure of the homologous ␣1␣2 NC1 hexamers (18). The candidate alloepitope residues are: Asp-3, Thr-15, Ala-17, Gln-22, Leu-25, Gln-26, Glu-20, Val-116, His-129, Gln-132, and Asp-135. Notably, most of these residues clustered in two narrow regions of the ␣5NC1 domain, residues 15-29 and 125-139, designated E A and E B (Fig. 9), based on their homology to the respective epitope regions of ␣3NC1 (2,3).
A distinctive feature of the XLAS alloepitopes is their accessibility in the ␣3␣4␣5 NC1 hexamer. This was demonstrated by three lines of evidence: (a) the ability of human kidney NC1 hexamers to inhibit alloantibody binding to immobilized ␣5NC1; (b) the complete binding of ␣3-␣6NC1 monomers and dimers to XLAS alloantibodies; and (c) strong reactivity of XLAS alloantibodies with native rather than dissociated ␣3␣4␣5NC1 hexamers in capture ELISA. Except for kidneyeluted alloantibodies from one patient, APTN-2 in this study, the accessibility of XLAS alloepitopes in NC1 hexamers has not previously been investigated.
A comparison of the properties of the ␣5NC1 alloepitopes with those of ␣3NC1 autoepitopes reveals commonalities and distinctions. The immunodominant alloepitopes and autoepitopes both reside at homologous E A and E B regions. However, the alloepitopes are accessible on the surface of the two known subsets of ␣3␣4␣5 hexamers, non-cross-linked and cross-linked, whereas the autoepitopes are sequestered within the cross-linked hexamers (4). The contrast between the accessibility of alloepitopes and the crypticity of autoepitopes implicates distinct molecular forms of antigen as initiators of the pathogenic process in anti-GBM disease.
In XLAS patients who develop post-transplant nephritis, the likely antigen is the native form of ␣3␣4␣5(IV) collagen present in the renal allograft. Because this network is not assembled in the basement membranes of the affected XLAS patients, B cells specific for epitopes within the ␣3␣4␣5NC1 hexamers (whether accessible or sequestered in the quaternary structure) cannot undergo deletion, anergy, or receptor editing. In contrast, anti-GBM autoantibodies do not target exposed epitopes in the ␣3␣4␣5NC1 hexamer, suggesting that central or peripheral tolerance to this epitope is normally established in the B cell compartment. Autoimmune anti-GBM (or Goodpasture) disease may be triggered if immune self-tolerance to ␣3(IV) collagen is circumvented by putative pathogenic modifications that expose cryptic autoepitopes in the NC1 hexamers, for instance by reactive oxygen species (19).
The repertoire of alloreactive T helper cells in XLAS patients may also shape the specificity of the alloantibody response. Because the defect in the COL4A5 gene that causes the loss of ␣5(IV) chains also prevents the stable incorporation of ␣3, ␣4, and ␣6(IV) chains in the basement membranes of XLAS patients, it is unlikely that B cells reactive with any of these chains undergo negative selection. However, XLAS alloantibodies consistently target ␣5NC1 epitopes and only occasionally react with ␣3NC1, whereas ␣4NC1 or ␣6NC1 do not appear to be significant targets. The overall pattern of alloantibody specificity is consistent with the primary defect in the COL4A5 gene, leading to a failure to establish T cell tolerance to the missing ␣5(IV) collagen chain. However, operational tolerance to ␣3, ␣4, and ␣6(IV) chains may still be established in the T cell compartment because at least some, if not all, of these chains continue to be produced intracellularly (20). Alloantibodies to the ␣3NC1 domain found in some XLAS patients may FIGURE 9. The location of the XLAS alloepitopes in a molecular model of the ␣5NC1 domain. A homology model of ␣5NC1 monomer was built based on the crystal structure of ␣1NC1. The ␣5NC1 regions corresponding to various chimeras are shown in yellow (5111), blue, (1511), green (1151), and red (1115). The ␣5NC1-specific residues are indicated by spheres (colored for residues accessible in the NC1 hexamers and gray for buried residues). The alloreactive regions are indicated by Y symbols representing XLAS alloantibodies. Within the immunodominant regions, most ␣5NC1-specific residues potentially important for alloantibody binding are contained in two short sequences designated E A and E B (outlined) based on their homology to the epitope regions of ␣3NC1 (2, 3). be elicited by epitope spreading: for instance, if B cells recognizing ␣3NC1 epitopes process ␣3-␣5NC1 dimers and present ␣5NC1 peptides to activated Th cells specific for ␣5NC1 epitopes. This mechanism is consistent with the observation that XLAS alloantibodies are initially specific for ␣5NC1 in the early stages of post-transplant nephritis, but then reactivity to ␣3NC1 arises as the disease progresses (21).
Detailed knowledge of the B cell and T cell alloepitopes implicated in the pathogenesis of post-transplant nephritis may facilitate the development of preemptive therapeutic strategies aiming to restore immune tolerance to normal ␣3␣4␣5(IV) collagen in X-linked Alport patients prior to a renal transplant. Protocols for inducing immune tolerance are widely used to prevent the production of inhibitory alloantibodies against transfused factor VIII in patients with hemophilia A (22). In principle, similar strategies may be developed to induce tolerance in Alport patients before a kidney transplant. Such therapeutic approaches would be particularly beneficial for Alport patients with a history of post-transplant anti-GBM nephritis, who have a very high risk of disease recurrence in subsequent allografts. The availability of dog and mouse models of Alport syndrome provides the necessary platform for experimental studies seeking to understand how immune tolerance to ␣3␣4␣5(IV) collagen can be modulated for therapeutic purposes.