A Role for Collagen IV Cross-links in Conferring Immune Privilege to the Goodpasture Autoantigen

The detailed structural basis for the cryptic nature (crypticity) of a B cell epitope harbored by an autoantigen is unknown. Because the immune system may be ignorant of the existence of such “cryptic” epitopes, their exposure could be an important feature in autoimmunity. Here we investigated the structural basis for the crypticity of the epitopes of the Goodpasture autoantigen, the α3α4α5 noncollagenous-1 (NC1) hexamer, a globular domain that connects two triple-helical molecules of the α3α4α5 collagen IV network. The NC1 hexamer occurs in two isoforms as follows: the M-isoform composed of monomer subunits in which the epitopes are accessible to autoantibodies, and the D-isoform composed of both monomer and dimer subunits in which the epitopes are cryptic. The D-isoform was characterized with respect to quaternary structure, as revealed by mass spectrometry of dimer subunits, homology modeling, and molecular dynamics simulation. The results revealed that the D-isoform contains two kinds of cross-links as follows: S-hydroxylysyl-methionine and S-lysyl-methionine cross-links, which stabilize the α3α5-heterodimers and α4α4-homodimers, respectively. Construction and analysis of a three-dimensional model of the D-isoform of the α3α4α5 NC1 hexamer revealed that crypticity is a consequence of the following: (a) sequestration of key residues between neighboring subunits that are stabilized by domain-swapping interactions, and (b) by cross-linking of subunits at the trimer-trimer interface, which stabilizes the structural integrity of the NC1 hexamer and protects against binding of autoantibodies. The sequestrated epitopes and cross-linked subunits represent a novel structural mechanism for conferring immune privilege at the level of quaternary structure. Perturbation of the quaternary structure may be a key factor in the etiology of Goodpasture disease.

Autoimmune disease results from a failure of the adaptive immune system to differentiate between self-and nonself-antigens. During the early steps of development, the immune system learns to recognize self by inducing "tolerance" against selfantigens. Several mechanisms, including clonal deletion, clonal anergy, and receptor editing, have been described for inducing self-tolerance of potential autoreactive lymphocytes (1). However, certain T cell self-epitopes, termed cryptic epitopes, are unable to induce tolerance during lymphocyte development. Although the mechanisms underpinning this remain elusive, under certain pathophysiological conditions such as inflammation, cryptic T cell self-epitopes may become apparent to the immune system and trigger a response that may lead to autoimmunity (2)(3)(4).
There is emerging evidence for the existence of cryptic B cell epitopes that may also play an important role in autoimmunity. In Goodpasture disease (GP), 3 the GP autoantibodies target the glomerular and alveolar basement membranes, specialized forms of extracellular matrix, causing rapidly progressive glomerulonephritis and pulmonary hemorrhage, respectively (5,6). The GP epitopes are cryptic within the native autoantigen, as first described by Wieslander et al. (7). The autoantigen is now known as the ␣3␣4␣5 noncollagenous-1 (NC1) hexamer, which connects two triple-helical protomers within the ␣3␣4␣5 network of collagen IV of certain basement membranes (6). The GP epitopes, designated E A and E B , are cryptic within the hexamer and reside in the ␣3 NC1 subunit (8).
The crypticity of GP epitopes is a feature of the quaternary structure of the ␣3␣4␣5 NC1 hexamer, which occurs in M-and D-isoforms. The M-isoform is composed of only NC1 monomer subunits. The epitopes are accessible to the GP autoantibodies that induce NC1 hexamer dissociation into monomer subunits (9). In contrast, the D-isoform is composed of both NC1 monomer and dimer subunits, but the epitopes are inaccessible (cryptic), indicating a role of dimers in the mechanism of crypticity (9). The epitopes can be exposed, however, by experimental dissociative conditions, such as low pH or protein denaturants that perturb the NC1 hexamer quaternary structure (7,9). Of particular note, immunization with ␣3 NC1 monomer and dimer subunits, having exposed B cell epitopes, induces antibody-mediated glomerulonephritis in rats, rabbits, and mice (10 -13). In contrast, their cryptic equivalents, ␣3-containing NC1 hexamers, are nonpathogenic (10 -12).
Cryptic B cell epitopes have also been implicated in Graves disease and anti-phospholipid disease. In Graves disease, the most common form of hyperthyroidism, autoantibodies target the thyrotropin receptor, which induces hyperactivity of the thyroid and results in goiter and thyrotoxicosis (14). The epitope is partially sequestered between the extracellular A subunit and membrane-bound B subunit of the thyrotropin receptor, rendering it inaccessible for binding of autoantibodies. The epitope, however, can be exposed as a result of intramolecular proteolytic cleavage of the receptor (15). Immunization with the A subunit, which bears the exposed epitope, induces disease. However, immunization with intact thyrotropin receptor, in which the epitope is cryptic, does not induce an immunogenic response. Another example is anti-phospholipid disease where the autoantigen is plasma ␤2-glycoprotein I in which the cryptic epitope is sequestered by a carbohydrate moiety (16,17). Interactions with cell membrane phospholipids induce a conformational change in ␤2-glycoprotein I structure that exposes the cryptic epitope by relocating the carbohydrate moiety away from the autoantigen, allowing binding of pathogenic autoantibodies.
Collectively, the crypticity and pathogenicity of B cell epitopes of the three different autoantigens described above suggest that the exposure of cryptic epitopes upon perturbation of tertiary or quaternary structure is a critical feature that triggers an autoimmune response. An understanding of the structural basis of the crypticity of various autoantigens can provide new insights into the mechanisms of autoimmunity.
In this study, the GP autoantigen, ␣3␣4␣5 NC1 hexamer, was investigated to determine the molecular basis of the crypticity of epitopes of the D-isoform. This included the elucidation of features of the quaternary structure of the ␣3␣4␣5 NC1 hexamer by mass spectrometry analysis of NC1 dimer subunits, homology modeling, and molecular dynamics simulation. The findings reveal a novel structural mechanism for immune privilege involving sequestration of epitopes within the autoantigen quaternary structure and stabilization of the structure by crosslinking of subunits.

MATERIALS AND METHODS
Bovine testes were purchased from Pel-Freeze biologicals (Rodgers, AR). Bacterial collagenase (CLSPA) was purchased from Worthington.
Immunoprecipitation and Western Blotting-Analysis of TBM hexamers by immunoprecipitation with Mab1 and Mab3 (Wieslab AB, Lund, Sweden) was performed as described previously (18). Antibody-antigen complexes were precipitated with protein G-agarose beads, solubilized in Laemmli sample buffer, and analyzed by immunoblotting. All samples were resolved into NC1 monomers and dimers by SDS-PAGE in replicate 4 -22% gradient gels under nonreducing conditions. After transfer to Immobilon P and blocking with casein, ␣1-␣5 NC1 domains were identified by immunoblotting with specific monoclonal antibodies, as described previously (18). Affinitypurified GP IgG (10 g) was incubated with native TBM NC1 hexamers (5-10 g) in 200 l of Tris-buffered saline, pH 7.4, overnight at room temperature. NC1 subunits, prepared by incubating hexamers with 6 M GdnHCl for 15 min at 60°C, were diluted at least 20-fold prior to incubation with GP IgG for 1 h. The precipitated NC1 domains were detected by Western blotting using monoclonal antibodies as described above.
Enzyme-linked Immunosorbent Assay (ELISA)-The assay was performed as described before (9), but with modifications. Briefly, 96-well Maxisorp microtiter plastic plates were coated overnight with an anti-␣3␣4␣5 NC1 hexamer monoclonal antibody (200 l of a 1 g/ml solution in carbonate buffer, pH 9.6) and blocked with 2 mg/ml bovine serum albumin. Samples containing TBM NC1 hexamers were diluted with incubation buffer (TBS, pH 7.4, with 1 mg/ml bovine serum albumin and 0.05% Tween 20) and allowed to bind to immobilized antibody for 1 h at 37°C. To dissociate TBM NC1 hexamers, some antigen samples were treated with 6 M GdnHCl for 15 min at 56°C before incubation with the immobilized antibody. The ␣3␣4␣5 NC1 hexamers captured by the immobilized antibody were incubated with three sera of GP patients. Binding of GP autoantibodies was detected by using an alkaline phosphatase-conjugated anti-human antibody (Sigma) and p-nitrophenyl phosphate as chromogenic substrate. Absorbance at 405 nm was measured with a Spectramax 190 plate reader (Molecular Devices, Sunnyvale, CA).
Mass Spectrometry Analysis of ␣3␣5-heterodimers and ␣4␣4-homodimers-Mass spectrometry analyses were performed on a ThermoFinnigan LTQ linear ion trap mass spectrometer equipped with a Surveyor LC pump and a microelectrospray source controlled by Xcalibur 1.4 software as described before (21). LC-ESI/MS 3 analysis was performed using data-dependent scanning in which one full MS spectrum (full mass range of 400 -2000 atomic mass units) was followed by one MS/MS spectrum. The three most intense ions in each MS/MS spectrum were subjected to an additional fragmentation (MS 3 ) analysis.
Software for Sequence and Post-translational Modification Analysis-Peptides derived from collagen IV NC1 domains were identified by searching the bovine subset of the Uniref100 data base using the TurboSEQUEST version 27 (revision 12) algorithm (Thermo Electron, San Jose, CA) on a high speed, multiprocessor Linux cluster in the Advanced Computing Center for Research and Education at Vanderbilt University. The search files were adapted for the identification of peptides that included the amino acid modifications of interest i.e. Ϫ48 atomic mass units on Met 93 , ϩ46 atomic mass units on Lys 211 , or ϩ62 atomic mass units on Lys 211 (46 ϩ 16 atomic mass units in the case of Hyl 211 ) as well as carboxyamidomethylation of cysteine (ϩ57 atomic mass units) and oxidation of methionine (ϩ16 atomic mass units). Theoretical protein, peptide, and peptide fragment ion masses were generated using General Protein Mass Analysis for Windows version 6.01.1 (Lighthouse Data, Odense, Denmark).
Homology Modeling of the D-␣3␣4␣5 NC1 Hexamer-A model of the ␣3␣4␣5 NC1 hexamer was constructed by a combination of homology modeling and molecular dynamics simulation. The crystal structure of bovine ␣1␣2␣1 NC1 hexamer (Protein Data Bank accession code 1M3D) was used as the starting template (22). The sequences of human ␣3, ␣4, and ␣5 NC1 domains were aligned with bovine ␣1, ␣2, and ␣1 NC1 sequences, respectively, using ClustalW (23). Of the available ␣1␣2␣1 NC1 trimers in the asymmetric unit, the one having the overall lowest B-factor average was chosen. Chains A (␣1), B (␣1), and C (␣2) were mutated to human ␣3, ␣5, and ␣4 sequences, respectively. In those cases where there were insertions or deletions in the primary sequence, the sequence variations were reiteratively modeled and stereochemically refined, using the molecular graphics software O, version 8 (24). The initial human ␣3␣4␣5 NC1 trimer model was duplicated to form a "head-to-head" NC1 hexamer using the symmetry operator obtained by the superposition of the two NC1 trimers of the ␣1␣2␣1 NC1 hexamer crystal structure. The resulting NC1 hexamer model was subjected to molecular dynamics simulation using VMD (25) and NAMD 2.5 (26).
Molecular Dynamics Simulation of the ␣3␣4␣5 NC1 Hexamer-The homology model was placed in a sphere of water using the SOLVATE module of VMD (25). The molecu-lar dynamics simulations were carried out with the program NAMD (26) utilizing the CHARMM27 force field (27). The boundary conditions started with the first boundary potential of 10 kcal/mol⅐Å 2 at a distance of 75 Å, with the potential exponent set to 2. Force field parameters were chosen in such a way that both van der Waals and electrostatic interactions between 1 and 4 atoms were taken into account. A switching function was employed to smooth van der Waals and electrostatic interactions to zero between 10 and 15 Å. The initial temperature of the system was set at 310 K, and the simulation was run at a step size of 2 fs/step (10 steps per cycle). The bonds were treated as rigid, including those involving hydrogen atoms, and Langevin dynamics were used with a dampening coefficient of 5/ps, set at 310 K (hydrogen atoms excluded). The resulting simulation provided a minimized model after 400 iterations. General analysis of the final model was performed using Chimera (28).
Model Analysis of the Epitopes on ␣3␣4␣5 NC1 Hexamer-Model analysis and graphics utilized PyMOL 1.0 and MMTSB toolkit (29). Surface area analysis was performed with GETAREA 1.1 (30) and NACCESS 2.1.1 (31). The accessible surface area of the ␣3 NC1 monomer was analyzed within and outside of the NC1 hexamer. We were primarily interested in the accessible surface area of the epitopes on the ␣3 NC1 domain for GP autoantibodies. This semi-quantitative analysis identified residues that are sequestered by neighboring NC1 monomers.

RESULTS
Strategy for Obtaining the Quaternary Structure of the ␣3␣4␣5 NC1 Hexamer-The quaternary structure of the ␣3␣4␣5 NC1 hexamer was further defined as the foundation to delineate the molecular basis of the crypticity of epitopes of the D-isoform. In our previous study, we proposed a quaternary structure of the ␣3␣4␣5 NC1 hexamer based on a composition of (␣3) 2 (␣4) 2 (␣5) 2 , which allows for a total of 18 possible combinations of different NC1 subunit arrangements (32). The proposed quaternary structure was based on two-dimensional electrophoresis and immunoblotting as well as immunoprecipitation studies using monoclonal antibodies and GP autoantibodies. These studies clearly demonstrated a counterclockwise arrangement of the ␣3, ␣4, and ␣5 within each NC1 trimer. However, the spatial orientation and connections between NC1 subunits across the trimer-trimer interface were only speculated because the existence of a ␣3␣5-heterodimer had not been unambiguously established.
The identification of an ␣3␣5-heterodimer, with respect to its constituent NC1 monomers and the chemical nature of a cross-link that connects the NC1 monomers, can provide novel information that further defines the quaternary structure of the ␣3␣4␣5 NC1 hexamer. An example is the ␣1␣1-homodimer, first observed by SDS-PAGE (33,34), that is derived from the ␣1␣2␣1 NC1 hexamer. This NC1 dimer is stabilized by a crosslink between Met 93 and Lys 211 of two ␣1 NC1 monomers juxtaposed at the trimer-trimer interface of the NC1 hexamer. The cross-link was first proposed by Than et al. (35), based on electron density connectivity in the crystal structure of the ␣1␣2␣1 NC1 hexamer. Subsequently, we (21) used mass spectrometry analysis as an independent chemical approach to identify and characterize the putative cross-link and showed it to be a novel covalent bond between Met 93 and hydroxylysine 211 (Hyl 211 ) rather than lysine. The cross-link is characterized by a sulfonium ion linkage and was termed S-hydroxylysyl-methionine. Upon NC1 hexamer dissociation, the ␣1␣1-homodimer retains information about the spatial orientation and connectivity between NC1 monomers at the trimer-trimer interface of the parent NC1 hexamer.
In this study, we hypothesized that the S-hydroxylysyl-methionine cross-links connect the putative ␣3␣5-heterodimers and the ␣4␣4-homodimers in the D-isoform of the ␣3␣4␣5 NC1 hexamer. Because large quantities were required for this study, the ␣3␣4␣5 NC1 hexamer was isolated from TBM, which contains an abundance of the ␣3 chain of collagen IV (19) in comparison with that of GBM. The NC1 dimer subunits were characterized to acquire new information about the quaternary structure, spatial orientation, and connectivity between monomers within the parent ␣3␣4␣5 NC1 hexamer.
Isolation and Characterization of ␣3␣4␣5 NC1 Hexamers from Testis Basement Membranes-TBM NC1 hexamers required chemical and immunochemical characterization to establish that they display cryptic epitopes as did those of GBM (9). TBM NC1 hexamers, consisting of a potential mixture of ␣3␣4␣5 and ␣1␣2␣1 NC1 hexamers, were isolated as described previously on a gel filtration column (see Fig. 2A in Ref. 19). The crypticity of epitopes in the TBM NC1 hexamers for autoantibodies from three GP patients was evaluated by an ELISA. In each case, the epitopes were cryptic in the native NC1 hexamers, but upon dissociation by GdnHCl the epitopes became exposed and reacted with GP autoantibodies (Fig. 1).
Collectively, the immunoprecipitation studies demonstrate that the GP epitopes are cryptic within the ␣3␣4␣5 NC1 hexamer of testis basement membrane, as illustrated in Fig. 2D. The results are consistent with the ELISA results presented in Fig. 1, where reactivity was observed only after NC1 hexamer dissociation. Therefore, the ␣3␣4␣5 NC1 hexamers in TBM exist in the D-isoform, harboring cryptic epitopes to GP autoantibodies. In contrast, both M-and D-isoforms occur in GBM NC1 hexamers, where the latter is predominant. Moreover, the D-hexamers from TBMs were isolated in milligram quantities that were required for crosslink analysis.
Strategies for Cross-link Analyses of NC1 Dimers-We utilized a strategy of proteolytic digestion coupled with mass spectrometric analysis, as we described previously to characterize the cross-link of the ␣1␣1-homodimers (21). We searched for peptides encompassing the conserved Met 93 and Lys 211 residues, which are presumed to harbor a cross-link (Fig. 4). The quaternary structure of ␣3␣4␣5 D-hexamers is presumed to be composed of up to three NC1 dimers, which result from the covalent interaction of NC1 monomers at the trimer-trimer interface; thus, two ␣3␣5-heterodimers and one ␣4␣4-homodimer are expected upon dissociation of the ␣3␣4␣5 NC1 hexamer (Fig. 5A) (32). The in silico analysis of the sequences of ␣3, ␣4, and ␣5 NC1 domains indicates that trypsin would generate three peptides containing Met 93 and three peptides containing Lys 211 , all of which have different lengths and amino acid compositions because of local sequence differences between three NC1 domains ( Fig. 4 and Fig. 5A). A random combination between Met 93 with Lys 211 containing peptides from any of the three NC1 domains would generate a total of nine different peptide complexes. However, quaternary structure constraints (21,32,35) would only allow the formation of cross-links between ␣3 and ␣5 NC1 domains to form ␣3␣5-heterodimers and between two ␣4 NC1 domains to a form ␣4␣4-homodimer. As illustrated in Fig. 5A, trypsin digestion of ␣4␣4-homodimer should generate the T-6498 complex. Likewise, trypsin digestion of ␣3␣5-heterodimers would generate the T-4481 and T-3959 complexes, each connected by the S-hydroxylysyl-methionine (Met-Hyl) cross-link (Fig. 5A).
Each of these peptide complexes (T-6498, T-4481, and T-3959) should display a fragmentation pattern by mass spectrometry that is characteristic of the Met-Hyl cross-link, as described previously for the ␣1␣1-homodimers (21). Fragmentation of the Met-Hyl cross-link by collision-induced dissociation (CID) breaks the bond between C-␥ and S-␦ of Met 93 resulting in a mass loss of 48 atomic mass units from Met 93 (thiomethyl group) and a concomitant mass gain 4 We used the term "␣3␣5-containing NC1 dimers" here to indicate that the composition of these NC1 dimers is uncertain prior to the findings of the present study. The NC1 dimers could represent ␣3␣5-heterodimers as well as ␣3␣3or ␣5␣5-homodimers co-migrating in SDS-PAGE.
. . of 46 atomic mass units by Hyl 211 (Fig. 5B). This characteristic fragmentation profile was used to identify tryptic peptide complexes containing the Met-Hyl cross-link by search-ing for cross-link-derived peptides containing either of these two modified amino acids using the SEQUEST algorithm (36) as indicated under "Materials and Methods." Mass Spectrometry Analysis of ␣3␣5-Containing Dimers-The ␣3␣5-containing dimers were digested with trypsin and analyzed by ESI-MS 3 . Fig. 6 illustrates representative spectra for each of the peptide complexes and their fragmentation (MS/MS) profiles. Fig.  6A (left panel) shows a full scan ESIϩ mass spectrum containing the multiple charged ions (ϩ3, ϩ4, ϩ5, and ϩ6) that represent the T-4481 complex, as was predicted in   derived from the ␣3 NC1 domain plus 46 atomic mass units, indicating the characteristic fragmentation of the Met-Hyl cross-link (Fig. 6A, right panel).
To confirm that these ions correspond to the peptide sequences predicted in Fig. 5A, including the chemical modifications of Met 93 and Hyl 211 , 823.8 m/z and 1007.87 m/z ions were selected for further fragmentation (MS 3 ) (supplemental Fig. 9A). The b-and y-series of the MS 3 fragmentation profile for the 828.8 m/z ion are consistent with the sequence for the ␣5 Met 93 -containing peptide lacking the 48 atomic mass units at Met 93 (T*-2466 peptide). Similarly, the MS 3 fragmentation profile of 1007.87 m/z ion is consistent with the sequence for the ␣3 Hyl 211 -containing peptide in which Hyl 211 has gained 46 atomic mass units (T*-2013 peptide). Fig. 6B (left panel) shows a full scan ESIϩ mass spectrum containing the multiple charged ions (ϩ3, ϩ4, ϩ5, and ϩ6) that represent a T-3959 complex, as was predicted in Fig. 5A for a second cross-link in the ␣3␣5-heterodimer. The average mass obtained by deconvolution of the ESIϩ mass spectrum is equivalent to that calculated for the T-3959 complex. Fragmentation of the quadruply charged ion (990.76 m/z, Fig. 6B, right panel) Fig. 9B).
Thus, mass spectrometric analysis demonstrates the presence of both T-4481 and T-3959 complexes, as predicted in Fig.  5A, in which peptides derived from ␣3 and ␣5 NC1 monomers are linked by the Met-Hyl cross-link. The presence of the Met-Hyl cross-link in the T-4481 and T-3959 complexes establishes a covalent interaction between ␣3 and ␣5 NC1 monomers, which unambiguously establishes the existence of an ␣3␣5-heterodimer, and verifies the identity of a ␣3-containing-heterodimer first observed by Kleppel et al. (37). Moreover, the location of the cross-link reveals that the spatial orientation and connectivity of the ␣3 and ␣5 NC1 monomers is homologous to that established previously for the ␣1␣1 NC1 homodimer. In addition, among the two possible cross-linking sites per dimer (35), the ␣3␣5-heterodimer provides the first chemical evidence that Met-Hyl cross-links occur at both sites (Fig. 5A), a structural feature that was not possible to discern in the previous analysis of ␣1␣1-homodimers (21).
Mass Spectrometry Analysis of ␣4␣4-Homodimers-The ␣4␣4-homodimers were digested with trypsin and analyzed by mass spectrometry as described above for the ␣3␣5-heterodimers. However, none of the peptide sequences containing Met 93 Ϫ48 atomic mass units or Hyl 211 ϩ 46 atomic mass units were found by SEQUEST, possibly because of the large size of the cross-linked peptides. Therefore, the predicted T-6498 complex (Fig. 5A) could not be detected in the ESI-MS 3 analysis.
For further MS analyses, the tryptic complexes were further digested with Asp-N endoproteinase in an attempt to yield the smaller predicted TA-3063 complex (Fig. 5A). The digested products were examined by an ESI-MS 3 experiment in which the peptide sequences containing Met 93 Ϫ48 atomic mass units and Hyl 211 ϩ46 atomic mass units were searched with SEQUEST. Although a SEQUEST search did not find the predicted TA-3063 complex, it did reveal an unexpected TA-3665 complex composed of two ␣4-derived peptides as follows: one peptide encompassing Lys 211 and the other Met 93 . The mass difference between the predicted TA-3063 complex and the observed TA-3665 complex reflects the replacement of Hyl 211 with Lys, together with additional residues at the amino or carboxyl terminus, indicating that cleavage occurred at sites other than those predicted (see below). Fig. 6C, left panel, shows the full scan mass spectrum containing the multiple charged ions (ϩ3, ϩ4, and ϩ5)  Hydroxylysine was expected at position 211 of ␣4 NC1 domain because it occurs in the ␣1␣1-homodimer (21) and in the ␣3␣5-heterodimer (see above). The absence of hydroxylysine, however, is consistent with the amino acid composition of the NC1 domains reported many years ago (7), which demonstrated that all the NC1 domains contain hydroxylysine except for M3, which would later be named the ␣4 NC1 domain. Collectively, these results demonstrate that the TA-3665 complex is composed of two peptides that are connected by a covalent cross-link between the side chains of Met 93 and Lys 211 , which unambiguously establishes the existence of an ␣4␣4-homodimer. Furthermore, the results establish that the NC1 monomers of the ␣4␣4-homodimer are covalently connected by a novel cross-link involving Met 93 and unmodified Lys 211 , designated herein as S-lysyl-methionine (Met-Lys). The amino acid sequence of the ␣4 NC1 domain may provide an explanation for the absence of hydroxylation of Lys 211 as discussed below.

DISCUSSION
In this study, the Goodpasture autoantigen, the ␣3␣4␣5 NC1 hexamer, was characterized with respect to its quaternary structure, as revealed by mass spectrometry analysis of crosslinks in NC1 dimer subunits, homology modeling, and molecular dynamics simulation. 5 This information provided the framework for interpreting the structural basis for the crypticity of B cell epitopes of the GP autoantigen. The findings reveal 5 The ␣3␣4␣5 NC1 hexamer model obtained via homology modeling and molecular dynamics simulation is available at the Vanderbilt Center for Matrix Biology website.
new molecular features of the quaternary structure of the GP autoantigen that may provide clues about the etiology of GP disease.
Cross-links of the ␣3␣4␣5 NC1 D-isoform-Using mass spectrometry, we characterized NC1 dimer subunits in search of a putative cross-link analogous to that previously found for the ␣1␣1-homodimer (21). We searched for tryptic peptide complexes that correspond to regions of the ␣3, ␣4, and ␣5 NC1 domains encompassing the conserved Met 93 and Lys 211 residues, which were presumed to harbor a cross-link (Fig. 4). The analyses revealed the presence of Met-Hyl cross-links in the ␣3␣5-heterodimers, analogous to that found in the ␣1␣1-homodimers, as described previously (21). Surprisingly, the findings revealed a Met-Lys cross-link in the ␣4␣4-homodimers, indicating that Lys 211 of the ␣4 NC1 monomer does not undergo post-translational modification to hydroxylysine. Typically, hydroxylation of lysine is catalyzed by lysyl hydroxylase, which recognizes the X-Lys-Gly motif in the triple-helical domain of various collagens (38). It also occurs in nonhelical regions of collagen I at the end of the triple-helical domain that contains an X-Lys-(Ala/Ser) sequence (38). At these nonhelical regions, the lysine and hydroxylysine residues undergo enzymatic oxidation, which induces formation of aldehyde-derived cross-links that stabilizes the collagen fibril. This sequence is identical to the X-Lys 211 -Ala sequence present in the NC1 domains of ␣1, ␣2, ␣3, ␣5, and ␣6 chains of collagen IV (Fig. 4), and which undergo hydroxylation and form the Met-Hyl crosslinks. However, this sequence differs in the NC1 domain of the ␣4 chain, where it is X-Lys 211 -Glu (Fig. 4), a distinction that may explain the lack of hydroxylation of the ␣4␣4-homodimer.
The existence of the Met-Lys rather than Met-Hyl cross-link in the ␣4␣4-homodimer indicates that the hydroxylation of Lys 211 is not required for cross-link formation. However, this hydroxyl group in the ␣3␣5-heterodimers and ␣1␣1-ho-modimers may contribute to the stability of the cross-link, as in the case of collagen I where the hydroxylysine-derived crosslinks are more stable than those derived from lysine residues (39). It may also serve as an attachment site for carbohydrate units, galactose, or the disaccharide glucosyl-galactose and play a role in regulating the self-assembly of protomers into collagen IV networks (21). Clearly, hydroxylation of collagen IV molecules is crucial for normal basement membrane function as was recently shown from studies of lysyl hydroxylase null mice (40,41).
Three-dimensional Model of the D-isoform of the a3a4a5 NC1 Hexamer-The identification of the ␣3␣5-heterodimer and the ␣4␣4-homodimer together with the chemical nature and location of cross-links establishes the spatial arrangement and connectivity of the NC1 monomers at the trimertrimer interface of the ␣3␣4␣5 NC1 hexamers, as illustrated in Fig. 7. The direct contacts between ␣3 and ␣5 NC1 monomers via the Met-Hyl cross-link, as well as two ␣4 NC1 monomers via the novel Met-Lys cross-link, are illustrated. Notably, the proposed chemical structure for the Met-Lys cross-link is also shown (Fig. 7). The findings verify the quaternary structure and the chain composition of the protomers of the ␣3␣4␣5 network of collagen IV, as previously proposed (32), and establish the end-to-end (NC1-to-NC1) connections between protomers of this network.
trimers interact through an extensive planar trimer-trimer interface to form an NC1 hexamer. Each monomer of the ␣3␣4␣5 NC1 hexamer contains a large domain swapping ␤-hairpin element that associates with a complementary binding site on an adjacent NC1 monomer, a feature that is critical in the assembly of NC1 trimers (22,42). Furthermore, the NC1 hexamer is stabilized by up to four Met-Hyl and two Met-Lys cross-links. These cross-links can be envisioned to function as "molecular fasteners" that reinforce the quaternary structure of NC1 hexamers, which in turn strengthen the connection between two adjoining triple-helical molecules of the ␣3␣4␣5 collagen IV networks.

Location and Structural Features of E A and E B Epitopes-
The three-dimensional model (see above) provides the framework for describing the location and tertiary and quaternary structures of the GP epitopes, as presented in Fig. 8. In previous studies, the residues that encompass the E A and E B epitopes on the ␣3NC1 domain were determined (8,32,(43)(44)(45)(46). The Molecular Dynamics-minimized homology model revealed that a number of these residues are located on the surface but that residues Val 27 , Leu 29 , Tyr 30 , and Gln 57 are sequestered under the neighboring ␣5 NC1 monomer at the bottom of a surface cleft, formed by the lateral interaction between ␣3 and ␣5 NC1 monomers (Fig. 8, left panel). The sequestration is sta-bilized by domain swapping interactions between neighboring NC1 monomers. Of the 1164 Å 2 accessible surface area of the E A epitope, 130 Å 2 (11%) is buried for these four residues. The supplemental Table 1 provides a summary of the change in accessible surface area for each of these epitope residues.
The E B epitope, encompassing residues Thr 127 through Lys 141 , as defined in previous studies (8,32,43,44), is located on a ridge of the ␣3 NC1 monomer adjacent to the ␣4 NC1 monomer. MD-minimized homology modeling revealed that a number of these residues are located on the surface but that residues Pro 133 , His 134 , and Trp 136 are sequestered by the neighboring ␣4 NC1 monomer (Fig. 8, right panel). Of the 1,030 Å 2 accessible surface area of the E B epitope, 87 Å 2 (9%) is buried for these three residues (supplemental Table 1).
Structural Basis for the Crypticity of E A and E B Epitopes of the D-isoform-The structural basis for the crypticity of epitopes of the D-isoform was determined by a comparison of the structural and immunochemical features of the M-and D-isoforms, together with an analysis of the quaternary structure of the E A and E B epitopes within the ␣3␣4␣5 NC1 hexamer (see below). As we described in a previous study (9), the epitopes of the M-isoform are accessible to GP autoantibodies, which disrupt interactions between neighboring subunits, whereas the epitopes of the D-isoform are inaccessible (cryptic) and the iso- In each case, the NC1 hexamer was rotated about 120°in the direction marked by the arrows for optimal view. Lower insets represent the same view except that the ␣5 (left) or ␣4 (right) NC1 monomers were removed leaving a footprint (dotted line) on ␣3 NC1 monomer to visualize the residues that were sequestered by the neighboring ␣5 or ␣4 NC1 domain. Sequestered residues are labeled in white captions. Molecular surfaces images were produced using University of California, San Francisco, Chimera (28).
form is impenetrable by GP autoantibody. The structural feature that distinguished the M-and D-isoforms was the presence of NC1 dimer subunits in the latter, a feature that correlated with the crypticity of epitopes (9). As reported herein, the ␣3␣5-heterodimer and the ␣4␣4-homodimer are held together by the Met-Hyl and Met-Lys cross-links, respectively. Thus, the cross-links play a key role in the crypticity of epitopes, as envisaged below.
In the absence of cross-links, as in the M-isoform, certain epitope residues are accessible on the surface for interaction with GP autoantibodies. This interaction, in turn, leads to the displacement of neighboring subunits, exposing the sequestered residues (Fig. 8) and allowing binding of GP autoantibodies to all residues that constitute the epitope. However, the cross-links present in the D-isoform render it impenetrable to GP autoantibodies such that sequestered residues do not become exposed for binding of GP autoantibodies. Thus, the cross-links provide a belt-like structural reinforcement that increases the structural integrity of the ␣3␣4␣5 NC1 hexamer, which protects against displacement of neighboring NC1 subunits (Fig. 8). Furthermore, that the GP autoantibodies do not bind to the D-isoform indicates that there are key residues within the epitope that are inaccessible because they are sequestered by neighboring subunits. These residues are Val 27 , Leu 29 , Tyr 30 , and Gln 57 for the E A epitope and Pro 133 , His 134 , and Trp 136 for the E B epitope, as identified in the three-dimensional model (Fig. 8). It follows that these same residues are also sequestered in the M-isoform, but in the absence of cross-links, they are accessible to GP autoantibody.
A Novel Molecular Mechanism for Conferring Immune Privilege to GP Epitopes, a Role for Collagen IV Cross-links-Immune privilege originally implied a location anatomically protected from attacks by the immune system, such as eye, brain, and testis (47). More recently, other cellular and molecular mechanisms have been described that confer immune privilege. These include cytokines that inhibit innate or adaptive immunity, systemic regulatory T cells, and local cytotoxic mechanisms producing apoptosis of effector T cells induced by local FasL-Fas interactions (48,49). Here we describe a novel mechanism for conferring immune privilege at the level of quaternary structure to an extracellular matrix protein. This structural mechanism involves two key features as follows: (a) the sequestration of key residues of an epitope within quaternary structure, and (b) the further stabilization of the integrity of the quaternary structure by cross-links. These structural features may prevent the immune system from recognizing the otherwise pathogenic epitopes that reside in the ␣3␣4␣5 collagen IV network of certain basement membranes.
The proposed mechanism also suggests that perturbation of the quaternary structure of the ␣3␣4␣5 NC1 hexamer exposing the pathogenic epitopes could be a key factor in the etiology of Goodpasture disease. Perturbations could include a disruption of domain-swapping interactions, which leads to the exposure of sequestered residues, eliciting the production of pathogenic antibodies, and/or a disruption of cross-links, which render the NC1 hexamer more susceptible to disruption, exposing immunogenic residues. Several environmental factors, including cigarette smoking, exposure to organic solvents, reactive oxygen species, viruses and bacteria, have been associated with the onset of Goodpasture disease (50 -52). Such factors could play a role in the loss of structural immune privilege of the GP autoantigen.