A Comprehensive Model of the Spectrin Divalent Tetramer Binding Region Deduced Using Homology Modeling and Chemical Cross-linking of a Mini-spectrin*

Spectrin dimer-tetramer interconversion is a critical contributor to red cell membrane stability, but some properties of spectrin tetramer formation cannot be studied effectively using monomeric recombinant domains. To address these limitations, a fused αβ mini-spectrin was produced that forms wild-type divalent tetramer complexes. Using this mini-spectrin, a medium-resolution structure of a seven-repeat bivalent tetramer was produced using homology modeling coupled with chemical cross-linking. Inter- and intramolecular cross-links provided critical distance constraints for evaluating and optimizing the best conformational model and appropriate docking interfaces. The two strands twist around each other to form a super-coiled, rope-like structure with the AB helix face of one strand associating with the opposing AC helix face. Interestingly, two tetramer site hereditary anemia mutations that exhibit wild-type binding in univalent head-to-head assays are located in the interstrand region. This suggests that perturbations of the interstrand region can destabilize spectrin tetramers and the membrane skeleton. The α subunit N-terminal cross-links to multiple sites on both strands, demonstrating that this non-homologous tail remains flexible and forms heterogeneous structures in the tetramer complex. Although no cross-links were observed involving the β subunit non-homologous C-terminal tail, several cross-links were observed only when this domain was present, suggesting it induces subtle conformational changes to the tetramer site region. This medium-resolution model provides a basis for further studies of the bivalent spectrin tetramer site, including analysis of functional consequences of interstrand interactions and mutations located at substantial molecular distances from the tetramer site.

by the membrane skeleton, a two-dimensional protein network on the cytoplasmic face of the membrane. The membrane skeleton major component is spectrin (1), a flexible, rod-like molecule present in normal erythrocytes at ϳ240,000 copies per cell (2). Spectrin is comprised of two homologous subunits, a 281-kDa ␣ subunit and a 246-kDa ␤ subunit, which associate side to side in an anti-parallel orientation to form a long, flexible, heterodimer. Most of the mass of the spectrin dimer is composed of tandem, homologous, ϳ106-amino acid-long segments that fold into three helix bundles commonly referred to as "spectrintype motifs" or "repeats" (3). Spectrin heterodimer assembly is initiated by a rapid, high affinity interaction of complementary spectrin-type repeats near the actin binding end of the molecule (4,5), which is primarily enthalpically driven. A structural model of the erythrocyte spectrin heterodimer initiation site (␣20 -21/␤1-2), determined by homology modeling and chemical cross-linking, supports the hypothesis that initial docking of the correct ␣ and ␤ repeats from among many similar repeats in both subunits is driven primarily by long-range electrostatic interactions (6,7).
Spectrin tetramers subsequently are formed by head-to-head association of two heterodimers (Fig. 1A), and tetramers predominate on intact erythrocyte membranes (8,9), although hexamers and higher oligomers form readily in vitro (10). The hexagonal structures observed in spread red cell membranes result from association of the distal ends of spectrin tetramers with protein complexes consisting of short actin oligomers in complex with numerous other proteins (11)(12)(13)(14). Formation of tetramers involves a moderate affinity association of heterodimers that is dynamic and a critical feature of red cell membrane deformability and elasticity. That is, when red cells encounter sheer stress in the microvasculature, local breaking and the reformation of spectrin tetramers is thought to play a critical role in providing elasticity without membrane breakage observed under normal conditions (15).
Hereditary elliptocytosis (HE) 2 and hereditary pyropoikilocytosis (HPP) are related, common, inherited blood disorders where red blood cell membranes are destabilized and cells exhibit abnormal shapes and increased cellular lysis. Importantly, the severity of the disease appears to be directly related to the extent of decreased membrane mechanical stability. The mechanistic basis for this decreased membrane stability is weakened lateral interactions in the membrane skeleton due to either defective spectrin tetramerization or defective interactions at the spectrin-actin junctional complex. The most common genetic defects are mutations of ␣or ␤-spectrin that destabilize tetramer formation (16,17). Many of these mutations are located within the tetramer binding domain, although, surprisingly, some very common mutations are located at large molecular distances from any known functional site.
Recently, we systematically studied structural and functional properties of the 14 known HE/HPP mutations located in the ␣-spectrin tetramer binding site using a univalent tetramer binding assay (18). The results showed that all ␣0 HE/HPP mutations exert their destabilizing effects through molecular recognition rather than structural destabilization of individual components of the tetramer site. Interestingly, however, the R34W and K48R mutations showed wild-type binding affinity using the univalent tetramer binding. We subsequently solved the crystal structure of the univalent, head-to-head tetramer complex, and most ␣0 HE/HPP mutations mapped to the tetramer binding site interface (19). However, consistent with the observed wild-type binding affinity of R34W and K48R mutants, these residues exhibited only minimal contacts within the tetramer interface. This suggests these mutations may perturb tetramer formation through an alternative mechanism such as disruption of the interstrand association in full-length tetramers, as full-length tetramers involve bivalent, head-tohead binding associations with concurrent lateral association of the two strands of the molecule. The recombinant peptide univalent tetramer assay also is inadequate for studying several common ␣-subunit HE/HPP mutations that are located large distances from the tetramer site because they exhibit wild-type tetramer binding in univalent tetramer binding assays (data not shown).
We recently engineered, expressed, and characterized a mini-spectrin recombinant protein as a template for studying the closed-open dimer equilibrium, properties of weak lateral associations in the tetramer binding region, and effects of distal mutations on bivalent tetramer complex formation (20). In this construct, the ␣ and ␤ chain components of the tetramer binding site were sequestered in close proximity using a flexible glycine linker (Fig. 1A, lower panel). Characterization of this fused mini-spectrin showed that it was well folded and mimicked structural and functional properties of intact full-length dimers and tetramers, including lateral associations of the ␣ and ␤ subunits in the dimer and in the bivalent tetramer. It also exhibited formation of closed dimers and wild-type tetramer binding affinity (20).
In the current study, we predicted the structures of the seven spectrin repeats in the mini-spectrin tetramer binding site region, predicted interstrand docking of laterally associated domains, and confirmed and refined these predictions using chemical cross-linking data. The results show that the two strands twist around each other to form a super-coiled, ropelike structure. The R34W and K48R HE mutations that exhib-ited wild-type binding in a univalent tetramer assay are located in the interstrand region, suggesting these mutations may destabilize membranes by disrupting the interstrand interaction. However, there are less extensive interstrand contacts compared with the spectrin heterodimer initiation site and the central ␣-actinin region, which form higher affinity lateral associations than those in the tetramer region. Cross-linking of the ␣ subunit N-terminal to multiple sites on both strands demonstrates this non-homologous tail remains flexible. Interestingly, although no cross-links were observed involving the ␤ subunit non-homologous C-terminal tail, several cross-links were observed only when this domain was present, suggesting it induced a conformational change in adjacent regions of the tetramer.
Cross-linking Reactions-For cross-linking reactions with EDC/sulfo-NHS, 5 l of a freshly prepared aqueous cross-linker solution was added to 300 l of either ␣0-5Ϸ␤16-17 or ␣0-5Ϸ␤16-17C (0.38 -0.41 mg/ml), with final concentrations of EDC and sulfo-NHS of 10 and 5 mM, respectively. The reaction mixtures were incubated at 4°C, and 100-l aliquots were taken after 1, 2, and 4 h. The reactions were quenched by the addition of 5 l of aqueous dithiothreitol (DTT) solution to each aliquot (final DTT concentration ϭ 20 mM). Before SDS-PAGE, solutions were concentrated by ultra filtration using a Microcon YM-10 filter (Millipore).
SDS-PAGE and Trypsin Digestion-Reaction mixtures were analyzed by one-dimensional SDS-PAGE on 3-8% Tris acetate mini-gels (Invitrogen), and gels were stained with Colloidal Blue. The bands of interest were excised and digested in gel as described previously using modified trypsin (Promega) (21).
Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS)-Locations of cross-linked residues were determined using LC-MS/MS on a hybrid linear quadrupole ion trap FT-ICR (LTQ-FT Ultra TM ) mass spectrometer (Thermo Fisher Scientific), equipped with a NanoLC TM pump and autosampler (Eksigent Technologies). Tryptic peptides were separated by RP-HPLC on a PicoFrit (New Objective) 75 m id x 15 cm nanocapillary column packed with 5 m MAGIC C18 resin (Michrom BioResources). Solvent A was 0.1% formic acid in Milli-Q water (Millipore), and solvent B was 0.1% formic acid in acetonitrile. Peptides were eluted at 200 nl/min using an acetonitrile gradient consisting of 3-28% B over 42 min, 28 -50% B over 26 min, 50 -80% B over 5 min, and 80% B for 5 min before returning to 3% B in 1 min. The LTQ-FT Ultra TM mass spectrometer was set to perform a full MS scan (m/z 400 -2000) in the FT-ICR cell. The resolution at 400 m/z was set to 1 ϫ 10 5 for the MS scan, and the six most intense ions exceeding a minimum threshold of 800 were selected for MS/MS in the linear ion trap using an isolation width of 2.5 Da. Monoisotopic precursor selection was disabled, and singly charged ions were excluded from MS/MS analysis. Ions subjected to MS/MS were excluded from repeated analysis for 45 s.
Identification of Cross-linked Peptides-Cross-linked and control mini-spectrin tetramer samples were analyzed in parallel using LC-MS/MS. For data analysis, the LC-MS patterns for the control and cross-linked sample were compared using Rosetta Elucidator software (Version 3.2) (Rosetta Biosoftware) to identify all ions unique to the cross-linked sample. MH ϩ values for all remaining features that had a charge Ͼ2 then were compared with the MH ϩ for all theoretical cross-linked peptides, which were calculated using GPMAW Version 7.5. Ions specific to cross-linked samples and within 2 ppm of a theoretical cross-linked peptide were selected for further analysis. The Fuzzy Ions program in the SEQUEST Browser software (Thermo Fisher Scientific) and de novo sequencing were used to verify cross-linked peptides and to identify specific amino acid residue positions involved in the cross-links.
Homology Modeling-The final sequence alignment was submitted to MODELLER 9v7 for generating separate homology models of ␣4-5, ␤16-17/␣0, and ␣1-3. Homology modeling and refinement were performed simultaneously by including known intrastrand cross-links. Molecular graphics were illustrated using PyMOL Version 0.99 (24), which also was used to calculate distances between cross-linked C-atoms of the carboxylic acid of glutamate, aspartate, or the protein C terminus and the N-atoms of amines of lysine or the protein N terminus. The docking models were predicted by ClusPro Version 1.0 (25). Interstrand cross-links then were used to test alternative docking models and subsequently to refine the model that was most consistent with the observed cross-link. The dimer interaction was analyzed using PROTORP (26).

RESULTS
Chemical Cross-linking Reaction-Two forms of wild-typefused mini-spectrin consisting of ␣0-5 connected to either ␤16-17 or ␤16-17C ( Fig. 1) were purified, and the mini-spectrin "tetramer" fractions were isolated using analytical HPLC gel filtration immediately before cross-linking. Although the ␣0-5Ϸ␤16-17 and ␣0-5Ϸ␤16-17C fusion proteins are single polypeptide chains and the head-to-head complexes of two chains are technically dimers, we will refer to the single chain fusion protein as a mini-spectrin "dimer" and the head-to-head complex as a mini-spectrin tetramer because they represent structures equivalent to truncated spectrin dimers and tetramers, respectively (Fig. 1A). The tetramer fractions from HPLC gel filtration of ␣0-5Ϸ␤16-17 and ␣0-5Ϸ␤16-17C were crosslinked in parallel reactions using "zero-length" cross-linking with EDC/sulfo-NHS, which functions by converting carboxyls (Asp, Glu, and the C terminus) into amine-reactive isourea intermediates that create an amide bond with elimination of an H 2 O molecule to lysine residues or other available primary amines. The reaction mixtures were incubated at 4°C to minimize thermal motion and dissociation of tetramer complexes. Parallel control experiments confirmed that there were negligible amounts of dimer in the tetramer fraction under the conditions used for cross-linking. After quenching the cross-linking reaction, the reaction mixtures were separated by onedimensional SDS-PAGE, and the gels were stained with Colloidal Blue. A range of cross-linking reaction conditions and reaction times was evaluated. An optimal reaction condition was determined to be a 1-h reaction using 10 mM EDC and 5 mM sulfo-NHS, because approximately half the dimers had at least one interchain cross-link, as indicated by a shift to the covalent tetramer position on the gel (ϳ180 kDa), with minimal formation of larger covalent aggregates (Fig. 1B). After SDS-PAGE separation of the cross-linked mixtures, the cross-linked tetramer band was excised from the reduced gel and digested in-gel with trypsin for analysis by LC-MS/MS.

Characterization of Inter-and Intramolecular Cross-linked Peptides Using Ion Trap FT-ICR MS-LC-MS/MS
data were obtained as previously described (6). Raw MS spectra of crosslinked mini-spectrin tetramers (␣0-5Ϸ␤16-17 and ␣0-5Ϸ␤16-17C) and uncross-linked tetramer controls were imported into the Rosetta Elucidator software, and the chromatograms were aligned and compared. Intensity filters were used to remove background noise, and a precursor ion charge filter of zՆ3 was used to simplify the dataset samples because prior studies showed that all cross-linked peptides using this protocol could be detected by ions with zՆ3 (6). MH ϩ values for all features unique to the cross-linked sample and with a charge Ն3 then were compared with a list of MH ϩ values for all theoretical cross-linked peptides using a 2-ppm tolerance. The list of all theoretical cross-linked peptides was produced by the GPMAW program by in silico prediction of all theoretical tryptic peptides with up to two missed cleavages and prediction of all possible single cross-links between amino and carboxyl groups on different tryptic peptides. These analyses reduced the number of MS/MS spectra to be considered from an initial Ͼ2000 MS/MS spectra to Ͻ100 MS/MS spectra per run. The selected putative cross-linked peptide spectra were interpreted and verified using Fuzzy Ions and manual de novo sequencing. We identified 14 cross-linked peptides in the ␣0-5Ϸ␤16-17 tetramer. The same 14 cross-links plus two additional cross-links were identified in two separate ␣0-5Ϸ␤16-17C reactions ( Table 1). The sequences of the cross-linked peptides and highest quality MS/MS spectra for each cross-link are shown in supplemental Figs. 1-16, and the peak assignments are shown in supplemental Tables 1-16.
Sequence Alignment and Models of Laterally Associating ␣ and ␤ Repeats in the Tetramerization Site Region (␣4-57 ␤16-17/␣0)-These modeling studies were started before the crystal structure of the univalent spectrin tetramer was completed (19). Hence, we initially modeled all domains in the minispectrin tetramer. To increase the accuracy of homology models, we used multiple templates that were repeats 16 -17 of chicken brain ␣-spectrin (PDB code 1CUN), repeats 14 -16 of ␤2-spectrin (PDB code 3EDV), repeats 15-16 of human brain ␣-spectrin (PDB code 3FB2), and repeats 8 -9 of erythrocyte ␤-spectrin (PDB code 1S35). Sequence alignments were produced with ClustalW 2.0.12 (Fig. 2), and modeling and optimization were performed with MODELLER 9v7. The quality of the homology models was tested by PROCHECK (27). Ramachandran maps of the ␤16-17/␣0 model revealed that it contained 98% of non-Gly-non-Pro residues in most favored, 2.0% in additional allowed, and 0.0% in generously allowed and disallowed regions. The ␣4-5 model contained 96.5% of non-Glynon-Pro residues in most favored, 3.0% in additional allowed, 0.0% in generously allowed, and 0.5% in disallowed regions.
To model alternative ␣4-57 ␤16-17/␣0 docking interfaces, the ␣4-5 and ␤16-17/␣0 homology models were docked using the ClusPro server (25). Thirty docking models were predicted using DOT (28) and ZDOCK (29,30). Only one docking model was consistent with previous experimental data and our current cross-linked data, as illustrated in Fig. 3A, which compares one of the docking models from ClusPro that is inconsistent with an observed cross-link with the single docking model that fits the cross-linking data. Both illustrated docking models are in agreement with previous experimental data; that is, paired repeats that align side by side in an anti-parallel orientation, but FIGURE 1. The mini-spectrin tetramer. A, shown is the relationship between full-length spectrin tetramers and the mini-spectrin tetramer. The ␣ subunit is shown in gold, and the ␤ subunit is shown in cyan. ABD, actin binding domain; EF, EF hand regions; triangle ϭ the SH-3 motif, which is located in the BC loop of ␣9 and for historical reasons is called ␣ motif 10. Spectrin-type, three-helix bundle motifs are shown as rounded rectangles numbered from N-to C-terminal, and adjacent non-homologous tails are shown as squiggles. The black boxes near the actin binding domains indicate the heterodimer initiation site. The spectrin domains included in the mini-spectrin are outlined in red. The mini-spectrins (lower panel) were created by linking ␣0-5 and ␤16-17 using a Factor Xa site flanked on both sides by six Gly; this linker is indicated by a gray loop. Two versions of the protein were constructed; one has the ␤ subunit C-terminal non-homologous tail and the other does not. B, cross-linking of the mini-spectrin tetramer with EDC/sulfo-NHS is shown. SDS-PAGE of mini-spectrin tetramer cross-linked with EDC/sulfo-NHS at 4°C for the indicated times (C, control; D, dimer; T, tetramer) is shown. Left panel, mini-spectrin without the ␤ subunit C-terminal nonhomologous tail is shown. Right panel, mini-spectrin with ␤ subunit C-terminal non-homologous tail is shown. FIGURE 2. Sequence alignments of human erythrocyte ␤16-17/␣0, ␣4-5, and ␣1-3. Alignments to sequences from known crystal structures were produced with ClustalW 2.0.12, and sequence-structure alignments were performed using MODELLER alignment.align2d. in the model on the left the distance between the cross-linked Glu-523 and Lys-1905 (cross-link 14) is 35.3 Å. In contrast, in the correct docking model shown in the right panel, the crosslinked residues are 9.6 Å apart, indicating good agreement of this model with the cross-linking data. Specifically, previous analyses of cross-links in proteins with known crystal structures have shown that residues up to ϳ10 Å apart in the crystal structure could be cross-linked by the zero-length EDC crosslinker due to the length and flexibility of the side chains involved (31). These data illustrate the high discriminatory power of chemical cross-links for distinguishing between alternative conformational and docking models.
Sequence Alignment and Models of Laterally Associating ␣7 ␣ Repeats in the Tetramer Region (␣1-27 ␣2-3)-Conformational and docking models of the ␣1-27 ␣2-3 complex also were predicted using MODELLER and ClusPro server, as described above. Ramachandran maps of the ␣1-3 model revealed that it contained 96.3% of non-Gly-non-Pro residues in most favored, 3.0% in additionally allowed, 0.0% in generously allowed, and 0.7% in disallowed regions, indicating a good model. To model alternative ␣1-27 ␣3-2 docking interfaces, the appropriate sections of the ␣1-3 homology model were docked using the ClusPro server. Thirty docking models were predicted using DOT and ZDOCK. Similar to the other laterally associated repeats described above, only one docking model is consistent with the interstrand cross-link between Lys-79 and Glu-172 (cross-link #11) as shown in Fig. 3B.
A Comprehensive Model of the Divalent Mini-spectrin Tetramer Complex-As described above, we initially modeled individual two-and three-repeat segments of the tetramer site because the longest available spectrin crystal structure templates are three repeats in length. Subsequently, we independently laterally docked two sections of the molecule and then attempted to combine these laterally associated pieces into a larger model. However, this resulted in either a large gap in one chain or extensive overlaps in the other strand.
While the above modeling studies were in progress, we obtained a crystallographic structure of the ␣0 -1/␤16-17 univalent tetramer site complex (19). A comparison of the C␣ coordinates of the crystal structure and our homology-based model of ␣0/␤16-17 (supplemental Fig. 17) showed a root mean square deviation for superimposition of 0.87 Å, indicating very good agreement between the experimental and predicted models. Most important, the crystal structure overlapped with our model of ␣1-3. Hence, we superimposed the common region of the crystal structure (␤16-17/␣0 -1) and the ␣1-3 homology model to produce a contiguous five-repeat strand. The ␣4-5 and ␣2-3 models then were associated laterally with the fiverepeat strand using ClusPro server, and distance constraints were used in MODELLER to connect the C terminus of ␣2-3 with the N terminus of ␣4-5. This resulted in ␣2-5 laterally associated with ␤16-17-␣0-3. Because the two sides of the minitetramer exhibit mirror symmetry about the central laterally paired ␣2-␣2 repeats (see Fig. 1A), we duplicated this model and superimposed the copy onto common components, removed the redundant repeats, and connected appropriate amino acids using MODELLER. Additional cross-links then were used to further refine this model, which was designated model A and evaluated using PROCHECK. The superimposition of the preliminary comprehensive model A and the final refined model A (Fig. 3C) show that at this stage the introduction of additional cross-link constraints had a relatively minor effect on the overall structure. The final model A structure shows that as previously hypothesized, the two strands twist around each other with a periodicity of approximately eight repeats, and the overall molecular shape exhibits substantial curvature (Fig. 4, A and B).
Several other strategies for combining the structural components into a comprehensive model were explored, but only one alternative approach yielded a feasible structure that was consistent with all observed cross-links and other known features of the spectrin molecule. This alternative model, designated model B, was constructed as described above for model A, except that the gap between ␣2-3 and ␣4-5 was closed by combining the two sequence files for ␣2-3 and ␣4-5 into a single sequence alignment file in MODELLER rather than using the distance constraint between the polypeptide ends that needed to be connected. The two models are quite similar (Fig. 4C), and the most notable difference is that in model B, a twist has been introduced between ␤16 and ␤17 relative to the crystal structure of the univalent tetramer complex (Fig. 4D).
Cross-links Involving the ␣ Subunit Non-homologous N-terminal Tail-In addition to the intrastrand and interstrand cross-links described above, a third group of cross-links involved the ␣-amino group of the N-terminal amino acid of ␣0. Analogous to the previously observed propensity of this flexible tail to cross-link to multiple sites in the mini-spectrin dimer (20), the N-terminal residue cross-linked to multiple sites in both strands of the tetramer (Fig. 5). These observed cross-links are consistent with NMR structures of the isolated ␣0 -1domain, which showed that the first ϳ20 residues of the ␣ subunit form highly flexible, extended structures that occur in many orientations (32). One such orientation is reproduced in Fig. 5A, with residues 1 and 21 highlighted to illustrate the distance between the N-terminal residue and the N-terminal residue of the ␣0 C-helix. The spatial distance between these two residues in this conformation is ϳ50 Å, whereas the contour length after the amino acid backbone is ϳ85 Å. If this region of the molecule remains flexible in the tetramer complex, we would expect that the N-terminal amino group could cross-link to residues up to 50 -85 Å away from the end of the ␣0 C helix. The actual locations of residues cross-linked to the ␣-subunit N-terminal are shown in Fig. 5B, and the approximate distances between these cross-links and the N-terminus of the ␣0 C helix are shown in Table 2 for models A and B. For both models, the distances are less than 50 Å, indicating that this tail remains flexible and adopts multiple conformations in the tetramer analogous to the ␣0 -1 domain alone.

DISCUSSION
Chemical cross-linking using a zero-length cross-linker coupled with homology modeling has enabled development of a medium-resolution, comprehensive model of a two-stranded mini-spectrin tetramer. This laterally associated complex, totaling 14 spectrin repeats, is by far the largest spectrin structure determined to date. Importantly, it provides multiple, novel insights into the relationship between spectrin structure and function, including the nature of the interstrand interface over the entire length of the tetramer site region defined by a fused mini-spectrin and possible mechanism of membrane destabilization by tetramer site mutations with apparent normal affinity. Other interesting features obtained from the current study include further structural insights into the nonhomologous spectrin "tails" that are located in close proximity to the tetramer binding site.
The N-terminal ␣-spectrin flexible region cross-links to multiple sites on both chains in the tetramer, indicating that this region of the molecule, which has been shown previously to be flexible in NMR studies (33), maintains this flexibility in the spectrin tetramer. Our prior analysis of mini-spectrin dimers showed that this region cross-linked to multiple sites, but some sites were distinct and depended upon whether the dimer was in an opened or closed conformation. These two groups of crosslinks in dimers provided useful insights into multiple conformations of the molecule that were difficult to identify and study by alternative methods. In the case of the current tetramer analysis, it is important to note that tetramers can form if a single head-tohead association occurs and, therefore, the second site could either be in a head-to-head complex or in a dissociated conformation equivalent to the structure of an open dimer. All of the cross-links observed in this study are consistent with the tetramers existing in a state where both head-to-head associations have formed, thereby providing the first direct experimental evidence that both head-to-head association sites are predominantly in the associated conformation in spectrin tetramers.
The structural and functional roles of the ␤-subunit nonhomologous C-terminal tail also have been ambiguous. This region contains six sites that are sequentially phosphorylated in  a These cross-links involve the ␣-subunit ␣-amino terminus, which is connected to the first helix that starts with Ala-21 by a long flexible linker. Distances shown are between Ala-21 and the residue that is cross-linked to the ␣-amino terminus. Cross-link numbers correspond to those in Table 1. red cells (34), and changes in phosphorylation levels in intact red cells affect cell membrane stability (35). This region is expected to have a relatively disordered structure due to the presence of a large number of prolines, and such a disordered structure is supported by its high protease susceptibility. Multiple prior studies have shown this region is not required for tetramer binding, and it does not appear to affect tetramer affinity. Although no direct cross-links involving this region were identified in the current study, as shown in Table 2, we did identify two additional cross-links that were specific to the mini-spectrin with the non-homologous C-terminal ␤-spectrin domain. These two cross-linked peptides show that ␤16A interacts with the ␣5 helix B in a manner that is at least slightly different from that of the interaction when the ␤ C-terminal domain is missing. This indicates that the ␤ subunit non-homologous tail directly interacts with homologous spectrin domains and may play a subtle role in further stabilization of the two-strand spectrin tetramer. In contrast to the dimer initiation site, where laterally paired repeats are aligned in direct register with each other (6), the tetramer region laterally associated domains are slightly staggered. In addition, the two subunits twist around each other, with a periodicity of a complete twist requiring approximately eight repeats, and the overall molecular shape shows substantial curvature rather than a linear extended structure (Fig. 4).
The nature of the tetramer region interface is substantially different from the high affinity lateral association of ␣-actinin dimers, where the terminal repeats in the four-repeat bundle involve B-C helix interactions on both chains (Fig. 6). However, the two central ␣-actinin repeats do resemble both the weak lateral associations in the spectrin tetramer region as well as the high affinity spectrin dimer initiation site lateral associations in that one subunit contributes an A-B helix and the complementary subunit always contributes an A-C helix. Hence, with the exception of the terminal ␣-actinin repeats, all spectrin-type lateral associations are AB/AC. Although in the central ␣-actinin repeats and the spectrin dimer initiation site region, the interacting surfaces within a chain flip-flop between adjacent repeats. That is, the next repeat within a chain switches from an A-B interaction surface to an A-C interaction surface. In contrast, as shown in Fig. 6A, all the weak lateral interactions in the tetramer binding site region involve an A-C helix in one strand interacting with an A-B helix in the opposite strand, but the interacting surface does not alternate in sequential repeats. Instead, there is a single transition within each of the two strands from an AC interface to an AB interface. Interestingly, in model A, this transition occurs between the ␣2 and ␣3 repeats in one strand and between the ␣1 and ␣2 repeats in the opposing strand. As a consequence, the two laterally associated ␣2 motifs do not involve identical surfaces, as shown in Fig. 6C, and are, therefore, not equivalent. Specifically, the AB interface of one ␣2 repeat interacts with the AC interface of the opposing ␣2 repeat. In contrast, the interstrand interface of model B differs slightly from model A because the ␣2-␣2 interface in this model is AC/AC. The functional consequences of these relatively subtle differences in the nature of the interstrand interfaces remain to be determined. However, they might play a role in ensuring that the correct isoforms assemble into laterally associated dimers in cells where multiple spectrin and ␣-actinin isoforms are expressed.
An important consequence of the spectrin tetramer region interstrand interaction that is not significantly affected by the relatively minor differences between models A and B is that the R34W and K48R HE mutations are located in the interstrand interface (Fig. 5C). Fig. 5D shows the electrostatic surfaces of intermolecular interaction of ␤16-17/␣0/␣4-5 complex. As described above, these two ␣0 mutations associated with HE/HPP are intriguing because they exhibited wild-type binding in univalent tetramer binding assays (18). Their locations in the interstrand interface suggest that they might destabilize spectrin tetramer binding through perturbation of the interstrand interaction, although such a mechanism of tetramer disruption is surprising because the lateral association between strands in this region is known to be relatively low affinity. The concept that interstrand interactions outside the dimer initiation site are weak and, therefore, relatively unimportant is based primarily upon the general observations that 1) the two strands in electron micrograms of spectrin tetramers are often "blown apart" except at the end near the actin binding domain, and 2) only peptides containing the dimer nucleation site can laterally assemble with the opposing subunit in peptide binding assays. However, these earlier assessments of interstrand associations in spectrin dimers and tetramers appear to need revision. In addition to the apparent functionally significant perturbation of the interstrand interface by R34W and K48R described above, we recently showed that simply sequestering the N-terminal region of the ␣ subunit next to the C-terminal region of the ␤ subunit via a flexible linker was sufficient for FIGURE 6. Comparisons of interstrand interfaces in spectrin and ␣-actinin. The gray-shaded bars represent the helices involved in interstrand interfaces. For the continuous helix between the C-helix and the A-helix of the next repeat, the short helical linker is delineated and is not shaded to highlight this boundary. A, docking interfaces of model A determined in this study for the bivalent mini-spectrin tetramer are shown. The interaction surfaces of model B are the same as the illustrated model except the ␣2/␣2 docking interface is AC/AC (data not shown). B, docking interfaces for human skeletal muscle ␣-actinin are indicated by the crystallographic structure of an anti-parallel ␣-actinin central domain (36). C, shown are docking interfaces for the human red cell spectrin dimer initiation site previously determined using homology modeling and chemical cross-link analysis (6). efficient lateral association of the subunits via lateral "weak" association of less than two full repeats (20). If a small pool of dimers had not been laterally associated, long linear polymers should have formed rapidly. However, no polymerization was detected with mini-spectrin dimers under any conditions tested. Taken together, these observations suggest that spectrin repeats outside the dimer initiation site can engage in functionally important lateral associations with affinities strong enough to form and preserve the laterally associated state if they are simply sequestered in close proximity. The preferential dissociation of these associations in electron microgram images may primarily be due to selective structural perturbations during specimen preparation that make them appear weaker than they are in the physiological state.
Of course, a critical factor to consider when using models based on structural predictions is the reliability of these predictions, as molecular models not supported by experimental data are often inaccurate. The important difference in studies of this type is that multiple chemical cross-links were used to distinguish between quite erroneous versus feasible models as illustrated in Fig. 3. Furthermore, once a feasible model is obtained, applying more space constraints from additional cross-links can be used to further refine the molecule, although once a feasible model supported by a number of cross-links is obtained, use of additional cross-links for further refinement often will only subtly further improve the structure as illustrated in Fig. 3C. Of course, this method is unlikely to yield high resolution structures where all subtle details are correct. The differences in models A and B highlighted in Fig. 4C illustrate the level of uncertainty that can be expected using currently available homology templates and experimental data. As noted above, all available cross-link information is compatible with both models (Table 2), and whereas some details of the existing model must be considered approximate, these two very similar models share all of the major structural features described above with the major difference being a kink between the ␤16 and ␤17 repeats. Because model A is more consistent with the recently solved crystal structure (19) of the univalent tetramer site, this model is preferred and has been emphasized here. Nonetheless, a kink or a twist in a larger construct relative to a smaller crystal structure also is quite feasible, and hence, the model B variant cannot be discounted based on available data.
A factor that can limit the accuracy of some details of spectrin homology models is the lack of availability of appropriate crystal structures that are larger than three spectrin repeats for use as modeling templates. Hence, one way to improve upon the current structure in the future may be if larger crystallographic structures are solved, although so far crystallizing and solving larger spectrin structures has been quite challenging, presumably due at least in part to the increasingly flexible nature of larger spectrin fragments. Another strategy for further improving the current model is to obtain additional chemical cross-links for further structural refinement. However, there are several technical limitations. First, zero-length crosslinks provide the most precise distance constraints, but the number of cross-links that can be obtained in any given protein complex are limited, and most cross-links are far below stoichiometric. More extensive reaction conditions can increase yields of very low-level cross-links, but extensive cross-linking is likely to lead to distortion of the native conformation, thereby providing incorrect distance constraints. Therefore, our future efforts will focus on more effective data mining of LC-MS/MS data to identify additional cross-links present at very low stoichiometric levels as well on combining data from multiple types of cross-linkers with minimal length cross-link distances.
A related challenge encountered when analyzing zero-length cross-links is the relative lack of effective computational tools for reliably identifying the small number of cross-links in a very large dataset. For example, each of the analyses performed in this experiment produced ϳ10,000 MS/MS spectra. The labelfree LC-MS pattern comparisons used here, together with selection of higher charge states (z Ն 3), allowed us to identify rapidly ϳ100 spectra per run that contained the potential chemical cross-links. However, a subsequent review of these spectra to verify specific putative cross-links is primarily dependent upon tedious manual de novo sequence analysis, which because of the nature of cross-linked molecules is substantially more challenging than analysis of linear peptides. Hence, future development of improved software tools for identifying and interpreting MS/MS spectra from protein cross-linking experiments will further facilitate development of high-confidence medium-resolution molecular models that are supported by experimentally determined distance constraints.