A Structural Model of the Erythrocyte Spectrin Heterodimer Initiation Site Determined Using Homology Modeling and Chemical Cross-linking*

Spectrin assembles into an anti-parallel heterodimeric flexible rod-like molecule through a multistep process initiated by a high affinity interaction between discrete complementary homologous motifs or “repeats” near the actin binding domain. Attempts to determine crystallographic structures of this critical dimer initiation complex have so far been unsuccessful. Therefore, in this study we determined the subunit-subunit docking interface and a plausible medium resolution structure of the heterodimer initiation site using homology modeling coupled with structural refinement based on experimentally determined distance constraints. Intramolecular and intermolecular cross-links formed by the “zero length” cross-linking reagent, 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide were identified after trypsin digestion of cross-linked heterodimer complex using liquid chromatography-tandem mass spectrometry analysis. High confidence assignment of cross-linked peptides was facilitated by determination of cross-linked peptide masses with an uncertainty of a few parts per million using a high sensitivity linear ion trap mass spectrometer equipped with a Fouriertransform ion cyclotron resonance detector. Six interchain cross-links distinguished between alternative docking models, and these distance constraints, as well as three intrachain cross-links, were used to further refine an initial homology-based structure. The final model is consistent with all available physical data, including protease protection experiments, isothermal titration calorimetry analyses, and location of a common polymorphism that destabilizes dimerization. This model supports the hypothesis that initial docking of the correct α and β repeats from among many very similar repeats in both subunits is driven primarily by long range electrostatic interactions.

Spectrin is an actin cross-linking protein that lines the intracellular side of the plasma membrane of most cell types where it is a major component of a submembraneous scaffold. This membrane skeleton plays important roles in maintaining mem-brane integrity and organization, particularly in red cells. Red cell spectrin is composed of a 281-kDa ␣ subunit and a 246-kDa ␤ subunit that associate side-to-side in an anti-parallel orientation, and two heterodimers associate head-to-head to form a tetramer, which cross-links actin oligomers. The spectrin-actin membrane skeleton is linked to the membrane through several indirect linkages to transmembrane proteins as well as direct interactions with lipid head groups (1)(2)(3)(4)(5)(6).
Most hereditary hemolytic anemias, which are quite common in the human population, destabilize the red cell membrane by disrupting either the spectrin-actin scaffold or its linkages to the plasma membrane. For example, most hereditary elliptocytosis and hereditary pyropoikilocytosis patients have mutations in one of the spectrin genes, and many of these mutations disrupt spectrin self-assembly into actin cross-linking tetramers (7)(8)(9). Spectrin mutations are usually heterozygous, and clinical severity of these mutations is frequently modulated by a common polymorphism, ␣ LELY , which affects the proportion of ␣ chains from each allele that are incorporated into the red cell membrane by affecting heterodimer assembly (10). Hence, both heterodimer and tetramer assembly are critical steps in producing a functional membrane skeleton and disruption of one or both of these processes is frequently the cause of hereditary anemias.
The mechanism of spectrin heterodimer assembly is of particular interest because both subunits are primarily comprised of many tandem homologous "spectrin type" motifs or "repeats" ϳ106 residues in length. There are 20 repeats in the ␣ subunit and 17 repeats in the ␤ subunit (see Fig. 1). All of these repeats are expected to have similar three-helix bundle conformations (11), and all repeats laterally associate with a homologous repeat in the complementary subunit. Despite the presence of many possible similar structures that could laterally dock with each other (Fig. 1) correct alignment and pairing of complementary repeats readily occurs. However, the structural features driving this assembly are only partially elucidated.
In earlier studies, we showed spectrin heterodimer assembly required a small region near the actin binding end of the molecule (␤1-2 and ␣20 -21) that initiated rapid, high affinity assembly (12,13). Any spectrin fragments containing these sites could rapidly associate in the correct orientation with the complementary subunit, whereas fragments lacking these regions could not assemble. After dimer initiation, adjacent repeats rapidly laterally paired with an inter-chain partner, although these associations were very weak and were not detected in solution unless the dimer nucleation site was covalently linked to the weaker repeats (12). This series of studies resulted in a "zipper" model of spectrin heterodimer assembly, which consisted of three discrete steps: 1) rapid, high affinity binding of the dimer initiation site; 2) rapid, low affinity lateral association of complementary repeats along the length of the monomer; and 3) latching the zipper by forming a closed dimer through a slow, moderate affinity, temperature-dependent association (14). We subsequently showed dimer initiation was weakened in high ionic strength buffers. In the same study homology modeling of the ␣20 and ␤2 repeats suggested their AB faces had complementary electrostatic surfaces that could drive initial docking of correct repeats through electrostatic interactions (15). Unfortunately, our repeated attempts to obtain a crystallographic structure of this complex have so far been unsuccessful, which is not surprising because crystallization of even single chain spectrin repeats has proven to be very challenging, presumably due to high molecular flexibility.
An alternative approach to structure determination is to use chemical cross-linking combined with identification of specific cross-linked sites using tandem mass spectrometry (MS/MS) 2 (16,17). Although chemical cross-linking is a very old technique, it has historically been very challenging to identify crosslinked peptides, because they are usually substoichiometric and difficult to separate, detect, and identify (17). This situation has improved substantially over the past several years as high sensitivity, high speed mass spectrometers capable of very precise mass determinations have become available. Although only a few cross-links are likely to occur under mild cross-linking conditions, these distance constraints can be of great value in distinguishing between alternative docking orientations as well as to test and refine structures based on homology modeling (16).
In this study, the docking interfaces of the red cell spectrin heterodimer initiation site region were experimentally determined using a "zero length" cross-linker coupled with analysis of cross-linked tryptic peptides using liquid chromatography-MS/MS analysis on an LTQ FT-ICR mass spectrometer. Distance constraints from both intrachain and interchain crosslinks were also used to further refine an initial homology-based structure. The final model correlates well with available biochemical and thermodynamic data, and it supports the hypothesis that initial alignment and pairing of correct ␣ and ␤ repeats from among many very similar repeats in both subunits is driven primarily by long range electrostatic interactions.

Construction of Expression Plasmids
The following recombinant peptides were used in this study: ␣20 -21 (repeats 20 and 21 of the human red cell spectrin ␣ subunit, residues 2033-2259) and ␤1-2 (repeats 1 and 2 of the ␤ spectrin subunit; residues 293-528). The repeats described are structural domains that use the phasing of Yan et al. (18). The design and construction of the ␣20 -21 and ␤1-2 pGEX-2T expression plasmids have been previously described (13,15).

Expression and Purification of Recombinant Erythrocyte Spectrin Peptides
The glutathione S-transferase-spectrin fusion proteins were expressed and purified as described (13,15), except that ␤1-2 was expressed at 30°C, ␣20 -21 was expressed at 18°C, and both of the proteins were purified from the soluble fraction after cell lysis. After initial purification on a glutathione-Sepharose column, peptides were cleaved from the glutathione S-transferase moiety using bovine thrombin (Sigma) at 37°C for 3 h. NaCl was added to decrease the formation of secondary cleavage products at a final concentration of 0.5 M for ␤1-2. The cleaved spectrin recombinants were purified by rechromatography on a glutathione-Sepharose column. The ␣20 -21 was then separated from residual glutathione S-transferase by anion-exchange chromatography on a 1-ml HiTrap-Q column (GE Healthcare). The ␣20 -21 was bound to the column in Buffer A (65 mM NaCl, 5 mM sodium phosphate, 2.5 mM EDTA, 75 M phenylmethylsulfonyl fluoride, pH 7.3), washed with Buffer A, and eluted with Buffer B (1 M NaCl, 5 mM sodium phosphate, 2.5 mM EDTA, 75 M phenylmethylsulfonyl fluoride, pH 7.3). The final purification step for both spectrin recombinants was HPLC gel filtration on two preparative TSKgel columns (G3000SW plus G2000SW) in series in phosphatebuffered saline (10 mM sodium phosphate, 130 mM NaCl, 1 mM EDTA, 0.15 mM phenylmethylsulfonyl fluoride, 1 mM ␤-mercaptoethanol, pH 7.4).
Freshly prepared ␣20 -21 and ␤1-2 were mixed at a 1:1 ratio, and the mixture was incubated on ice for 30 min. The mixture was loaded onto a HPLC gel filtration column to separate the ␣20 -21/␤1-2 complex from monomers and aggregates.

Cross-linking Reactions
For cross-linking reactions with EDC/sulfo-NHS, 10 l of a freshly prepared aqueous cross-linker solution, containing 0.02, 0.1, 0.2, or 1 M EDC in addition to 0.5 M sulfo-NHS, was added to 990 l of ␣20 -21/␤1-2 complex (2 M). Thus, molar excesses of 100, 500, 1000, and 5000 of EDC over the complex concentration were evaluated. For a control, 10 l of water was added, instead of the cross-linker, to the complex solution. The reaction mixtures were incubated at room temperature, and FIGURE 1. Model of the human erythrocyte spectrin anti-parallel heterodimer. The structural domains of the anti-parallel ␣ and ␤ subunits are schematically represented as follows: numbered rectangles, homologous "spectrin type" repeats; loop, labeled as repeat 10 is actually a src SH-3 homology domain; diamonds, EF hands; large rectangle, actin binding domain. The "␣0" represents a partial repeat at the N terminus of ␣ spectrin involved in forming head-to-head tetramers as well as the closed hairpin form of the dimer (shown here). The "tail" at the C-terminal end of ␤ spectrin is the nonhomologous phosphorylated region.
200-l aliquots were taken after 5, 15, 30, 60, and 120 min. The reactions were quenched by adding 20 l of 110 mM aqueous hydroxylamine solution to each aliquot (final concentration, 10 mM). Before SDS-PAGE, the solutions were dialyzed against 25 mM NH 4 HCO 3 overnight and lyophilized.

SDS-PAGE and Trypsin Digestion
Following separation of the reaction mixtures by one-dimensional SDS-PAGE, the bands of interest were excised and digested in gel as described previously using modified trypsin (Promega, Madison, WI) (19).

Ion Trap FT-ICR MS
Liquid chromatography-MS/MS experiments were performed on a hybrid ion trap FT-ICR mass spectrometer (LTQ-FT, Thermo Electron, San Jose, CA), equipped with a NanoLC pump and autosampler (Eksigent Technologies, Livermore, CA). Specifically, tryptic peptides were separated by reversedphase HPLC on a 75-m inner diameter ϫ 15-cm PicoFrit (New Objective, Woburn, MA) nanocapillary column packed with 5 m of MAGIC C18 AQ resin (Michrom BioResources, Auburn, CA). Solvent A was 0.58% acetic acid in Milli-Q water, and solvent B was 0.58% acetic acid in acetonitrile. Peptides were eluted at 200 nl/min using an acetonitrile gradient consisting of 1-50% B over 15 min, 50 -80% B over 5 min, 80% B for 10 min before returning to 1% B in 1 min. The LTQ-FT mass spectrometer was set to perform a full MS scan (m/z 375-2000) in the FT-ICR with resolution at 400 m/z set to 100,000. Simultaneously, the six most intense ions exceeding a minimum threshold of 1000 were selected for MS/MS in the linear trap. The normalized collision energy was set to 30%, and ions subjected to MS/MS were excluded from repeated analysis for 30s. Singly charged ions were not subjected to MS/MS.

Identification of Cross-linked Peptides
Cross-linked peptides were identified using a combination of GPMAW (General Protein Mass Analysis for Windows) software, version 7.10 (Lighthouse Data, Odense, Denmark), SEQUEST Browser software (ThermoFisher Scientific), and the ASAP (Automatic Spectrum Assignment Program) software packages (16). MS/MS spectra of cross-linked peptide candidates were further analyzed by MS2Assign (16) and ProteinXXX, which is the crosslinking feature of GPMAW v. 7.10.
Homology Modeling-The final sequence alignment was submitted to MODELLER 8v2 for generating a homology model of ␣20 -21 and ␤1-2. Molecular graphics were illustrated using PyMOL (23). The dimer interaction was analyzed using the Protein-Protein Interaction Server (24).

Determination of Distances between Atoms in Homology
Models of the ␣20 -21 and ␤1-2 Complex-Swiss-PdbViewer (25) was applied to calculate distances between 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.

RESULTS
Chemical Cross-linking Reaction-Monomers of the recombinant proteins were verified by MALDI MS, and potential irreversible aggregation of purified samples was measured shortly before the cross-linking reaction using analytical HPLC gel filtration (data not shown). These results together with prior characterization of these recombinants using CD, sedimentation equilibrium, and differential scanning calorimetry provided assurance these recombinant proteins were properly folded prior to cross-linking experiments.
The ␣20 -21/␤ 1-2 heterodimer was purified using HPLC gel filtration ( Fig. 2A), and cross-linked using "zero length" cross-linking with EDC/sulfo-NHS, which cross-links acid groups (Asp, Glu, and the C terminus) to amines (⑀-amino group of Lys and the N terminus) by creating an amide bond with elimination of an H 2 O molecule. Both intra-and interchain cross-links are expected, because these two proteins contain high concentrations of dispersed glutamic acid, aspartic acid, and lysine residues. After the cross-linking reaction, the reaction mixtures were separated by one-dimensional SDS-PAGE, and the gels were stained with Coomassie Brilliant Blue. A range of cross-linking reaction conditions were evaluated as described under "Experimental Procedures," and the optimal reaction condition was determined to be a 60-min reaction using a 1:5000 molar excess of EDC, because this produced a substantial amount of a cross-linked dimer complex without significant evidence of aggregation (Fig. 2B). The cross-linked ␣20 -21/␤1-2 complex is expected to migrate on SDS gels at ϳ55,000 Da depending on the extent of chemical cross-linking. Aggregation of proteins caused by excessive cross-linking was not observed for the cross-linked product under the 1:5000 molar ratio condition as evidenced by the absence of gel bands above the cross-linked dimer or at the top of the gel. Following SDS-PAGE separation of the cross-linked reaction mixtures, the cross-linked complex was excised from the gel, subjected to in-gel digestion with trypsin, and analyzed on a hybrid ion trap-FT-ICR mass spectrometer.
Characterization of Inter-and Intramolecular Cross-linked Peptides Using Ion Trap FT-ICR MS-Non-cross-linked tryptic peptides from ␣20 -21 and ␤1-2 were identified using the SEQUEST browser, and all MS/MS spectra corresponding to high confidence assignments from this analysis were removed from the dataset. The experimentally obtained monoisotopic masses for all remaining MS/MS spectra were compared with calculated masses of predicted tryptic peptides and theoretical cross-linked peptides using GPMAW 7.10 and ASAP (16). An initial mass error cut-off of 5 ppm was employed for this comparison.
MS/MS spectrum assignments of putative cross-linked peptides were performed with the aid of MS2Assign and the Pro-teinXXX program as well as manual de novo sequencing. Analysis of all spectra representing potential cross-linked peptides yielded high quality, high confidence matches for nine different masses arising from six specific intermolecular cross-links as summarized in Table 1. Fig. 3 shows the correlation of two representative MS/MS spectra with the assigned sequences, and assignments of observed ions are summarized in Tables 2 and 3. The nomenclature of cross-linked peptide fragment ions follows that proposed by Schilling et al. (26). The larger peptide is indicated by a subscript "A" and the smaller peptide by a subscript "B." In the instrument used here, MS/MS spectra primarily result from random fragmentations at peptide bonds, and by convention, ions from the N-terminal region of a peptide are b-ions while the complementary C-terminal fragments are y-ions. In most cases, b-and y-ions can arise from either peptide, and some ions will include an intact cross-linked peptide. For example, ion 28 in Fig. 3A and Table 2 is designated as (A int b3 B ) ϩ2 , which indicates it is a doubly charged b3 ion from peptide B that has an intact peptide A cross-linked to it via the lysine side chain at its N-terminal (see sequence diagram inset in Fig. 3A).
Sequence Alignment and Model of Laterally Associating ␣ and ␤ Repeats in the Dimer Initiation Site-An accurate sequence alignment is of utmost importance in building a useful three-dimensional homology model. The crystal struc-  Table 2 for the MS/MS assignments; B, cross-linked peptides MEENLSE 39 PVHCVSLNEIR and K 140 AAMR with precursor m/z 679.0865 and a charge state of ϩ4. See Table 3 for the MS/MS assignments. Most ion assignments were from MS2Assign and ProteinXXX. The superscript numbers in the peptide sequences represent the cross-linked amino acid residue positions. The locations of the peptide bonds fragmented to produce the numbered ions are indicated on the sequences of the cross-linked peptides. ture of human skeletal muscle ␣-actinin 2 (PDB ID: 1HCI) with 28% sequence identity to ␣20 -21 and 42% sequence identity to ␤1-2, was the best template for building homology models. Sequence alignments were produced with ClustalW-1.83, and MODELLER alignment.align2d, which is preferred for aligning a sequence with structure(s) in comparative modeling, was used to determine sequence-structure alignment (Fig. 4A). An additional consideration when modeling a non-covalent complex is how the two structures dock with each other. For our initial homology model of the ␣20 -21/␤1-2 complex we aligned ␣20 -21 to the ␣-actinin repeats 3 and 4 (R3-4) and ␤1-2 with ␣-actinin repeats 1 and 2 (R1-2) using the ␣-actinin docking interface from its dimer crystal structure. This alignment and docking mapped each spectrin heterodimer repeat to its most homologous repeat in the ␣-actinin antiparallel homodimer (Docking A in Fig. 4B). Based upon homology alone, this docking interaction seemed to be the most reasonable. However, this docking interface, which contains CA/AB and BC/BC interaction surfaces (Fig. 4B, Docking A), did not agree with our previous protease protection analyses of the heterodimer initiation site complex, which indicated AB faces were protected in the heterodimer interface region (15).
To model alternative heterodimer docking interfaces, the ␣20 -21 and ␤1-2 monomers were docked using the ClusPro server (27). When using the ClusPro program, the user has the option of selecting DOT (28) or ZDOCK (29,30) to perform rigid body docking, both of which are based on fast Fourier transform correlation techniques. Applied to a benchmark set of 2000 conformations, the algorithm predicts at least one experimentally relevant complex structure within the top 30 predictions, and, in ϳ30% of the cases, the best prediction is ranked first. Ten docking models were obtained using DOT and ZDOCK, respectively (data not shown). The returned models contained three kinds of inter-chain interactions: parallel interaction, "X"-like interaction, and anti-parallel side-by-side interaction. Because many previous experiments had shown the ␣ and ␤ subunits associate anti-parallel and side-by-side along the length of both chains to form a long flexible rod-like heterodimer, the first two types of interactions were discarded as unrealistic predictions. The third group of models had the correct anti-parallel orientation of subunits, but the predicted docking faces were inconsistent with the protease protection data (15). Therefore, none of the models predicted by ClusPro fit available experimental data.
We therefore reconsidered all published crystallographic structures of spectrin-type repeats. Of those available structures, only ␣-actinin R1-4 (PDB ID: 1HCI) and ␣-actinin R2-3 (PDB ID: 1QUU) form a physiologically relevant homodimer. Chicken brain spectrin ␣15-17 (PDB ID: 1U4Q) crystallized as an anti-parallel homodimer, however the interchain repeats are not aligned in register, as is expected for the ␣20 -21/␤1-2 dimer initiation site (15). Hence, we returned to using ␣-actinin as a docking template. The crystal structure of the ␣-actinin dimer contains three possible anti-parallel dimer docking orientations for two tandem repeats as illustrated in Fig. 4B. Development of the model using Docking A was described above. Models using Dockings B and C were obtained by first aligning ␣20 -21 to R3-4 and ␤1-2 with R1-2 to build homology models of the spectrin monomers as described above. Alternative alignments were regarded as less favorable due to both the substantially higher homologies of the above pairings and differing lengths of repeats and locations of gaps/insertion. Alternative models of the spectrin dimer were then obtained by superimposing the monomer models onto the crystal structure of the human skeletal muscle ␣-actinin as illustrated in Fig. 4B (Docking models B and C). Modeling and optimization were performed with MODELLER. Fig. 5 shows locations of observed cross-linked amino acid residues in models using the three alternative docking orientations. In the models using Docking A and B, cross-linked residues were often on opposite faces of the putative dimer (Fig. 5,  A and B), and these distance constraints could not be used to refine the structures without extensive distortion of the prototypical three helix bundle observed in all crystal structures of spectrin type repeats determined to date. In contrast, the Docking C model is consistent with these cross-linked data (Fig. 5C), and the distance between the two residues in each identified cross-link (see Table 1) could be minimized during model refinement without excessively distorting the basic three helix bundle structures of these repeats. The quality of the Docking C-based structure was subsequently tested by PROCHECK (31). Ramachandran maps of the model revealed that the ␣20 -  In the modeled complex (Docking C model), the interacting surface on ␣20 -21 consists of the AB face of ␣20 and the AC face of ␣21, which interact with the AC face of ␤2 and the AB face of ␤1, respectively. The interface surface along the long axis of the molecule buries 12.6% (1929 Å 2 ) of the accessible  1HCI). B, alternative docking orientations for the two-repeat spectrin heterodimer initiation site using the ␣-actinin homodimer crystal structure as a template. The crystal structure of the central domain of ␣-actinin consists of 4 spectrin-like repeats (R1, R2, R3, and R4) in an anti-parallel orientation. The three possible alternative pairings of the two-repeat spectrin dimer initiation site on the ␣-actinin dimer template and the helices that participate in the resulting dimer interfaces (docking surface) are shown.
surface area of ␣20 -21 and 12.8% (1841 Å 2 ) of the accessible surface area of ␤1-2 (determined with a probe radius of 1.4 Å). The interface involves 46% polar and 54% nonpolar atoms for ␣20 -21, and 43% polar and 57% nonpolar atoms for ␤1-2. The hydrophobic effect is not a dominant driving force in the interaction of ␣20 -21 with ␤1-2 because no significant hydrophobic patches were located in complementary positions on the protein interaction surfaces as indicated by LIGPLOT (32) analysis of the refined Docking C model. In contrast, the model reveals 10 intermolecular hydrogen bonds (H-bond) (3.9 Å cutoff) and 10 putative salt bridges ( Table 4). The negatively charged residues are primarily contributed by ␣20 -21, and the positively charged residues are primarily contributed by ␤1-2 for these electrostatic interactions (Fig. 6A and Table 4). These complementary electrostatic interactions are very likely to drive recognition and pairing of the correct ␣ and ␤ repeats during heterodimer assembly as we previously proposed (15).
Another interesting comparison is to evaluate the location of the six residue deletion in the ␣21 repeat associated with the ␣ LELY polymorphism in the alternative docking models. This polymorphism is very common in the human population, and it leads to low incorporation into the red cell membrane of the protein from the allele carrying this mutation. The polymorphism actually involves three linked mutations, which include: a point mutation at codon 1857 that changes a Leu to a Val, a point mutation in intron 45, and a point mutation in intron 46. Presumably as a result of one or both intron mutations, there is a 50% skipping of incorporation of exon 46 (residues 2177-2182) in the expressed protein (10). We previously showed the Leu 3 Val substitution did not affect incorporation of ␣ subunits into dimers and hence into red cell membranes, but the deletion of the 6 amino acids encoded by exon 46 prevented these subunits from being incorporated into the membrane, by reducing dimer binding affinity (33). These six amino acids are located in the ␣21 A helix, which is in the dimer contact site in our Docking C model (Fig. 6B). Similarly, the ␣21 A helix is in the dimer interface in the B model (see Figs. 4B and 6B). In contrast, in model A, these residues are not in or near the dimer interface region (Fig. 6B). Hence, both Models B and C, but not model A, are consistent with the observed abolishment of heterodimer assembly by the exon 46 deletion. Although the location of the ␣ LELY polymorphism is consistent with models B and C, Docking model B is not consistent with the cross-linking data, so the Docking C model involves the best docking orientation of the spectrin heterodimer initiation site.

DISCUSSION
The structural features that determine correct docking of the specialized spectrin repeats in the dimer initiation site were previously ambiguous. Three feasible models of the interchain contact surfaces were suggested by the crystal structure of the ␣-actinin homodimer (34). To further characterize this critical spectrin self-assembly site and to distinguish the most feasible inter-subunit docking surfaces, we exploited the improved  capacity of high mass accuracy hybrid ion trap mass spectrometers to identify interchain cross-links in complex protease digests.
The use of chemical cross-linking to either identify binding partners or to obtain low to moderate resolution insights into protein conformation is a conceptually simple, classic strategy (16), but the purification and identification of cross-linked peptides using conventional protein chemical methods requires large amounts of sample and is usually very challenging. Crosslinking reactions are frequently very incomplete and attempts to drive them to completion increases the likelihood that native conformation will be disrupted and artifactual cross-links might be identified. In addition, the sub-stoichiometric crosslinked peptides are often especially difficult to purify and char-acterize. The application of mass spectrometry to this problem has greatly improved the feasibility of obtaining useful information on small amounts of cross-linked proteins, and recently the combination of high resolution FT-ICR MS with cross-linking has been demonstrated to be a fairly robust strategy for rapidly defining interfaces between proteins (35).
So that distance constraints would be as specific as possible, we used the "zero length" cross-linker, EDC, to distinguish among three feasible docking orientations of the spectrin dimer initiation site (Fig.  4B). As summarized above, the identified cross-links are incompatible with the docking orientations of the A and B models (Fig. 5). In contrast, the Docking C model is highly consistent with both the cross-links identified in the current study (Fig.  5C) and prior protease protection experiments (15). That is, cleavage sites unaffected by dimerization compared with monomer in exposed sites in our Docking C model (Fig. 6C, top panel), whereas sites protected in the heterodimer map to buried regions in the dimer interface (Fig. 6C, middle panel). Interestingly, several cleavage sites were shown to increase in yield in dimer compared with monomers; specifically the ␤T2 and ␤T3 sites become more sensitive to trypsin digestion in the dimer complex (15), presumably due to conformational changes induced by association of ␣20 -21 with ␤1-2. These sites do map to relatively exposed regions in our heterodimer model, and therefore, substantial proteolytic cleavage of these sites in the dimer is consistent with model C (Fig. 6C, bottom panel).
As noted above, ␣-actinin, a member of the spectrin superfamily, crystallized as an anti-parallel homodimer of two monomers, each consisting of four spectrin-like repeats (34). The four red cell spectrin repeats that are most homologous to ␣-actinin are the repeats that constitute the spectrin dimer initiation site. Based on sequence homology, the ␣-actinin dimerization site resembles two mirror image spectrin initiation sites. Hence it was reasonable to speculate that this analogy might extend to structural and binding properties of the dimerization sites of these two proteins; a hypothesis further supported by the fact that the 2 ϩ 2 repeat interaction of the spectrin dimer initiation site has a K d in the low nanomolar range, whereas the FIGURE 6. Evaluation of the Docking C model. A, electrostatic properties of the docking surface. The electrostatic potentials of the spectrin heterodimer initiation site in Docking Model C are displayed on the solventaccessible surface using the GRASP program (42). Positive regions are depicted in blue and negative regions are shown in red. B, locations of the six amino acids encoded by exon 46 (residues 2177-2182), which are deleted in 50% of the ␣-chains from alleles with the ␣ LELY polymorphism are shown in magenta for the three alternative docking models. C, location of protease-sensitive sites as previously determined in Begg et al. (15) are mapped onto the Docking C model by highlighting the residue on the N-terminal side of the cleaved bound using a magenta colored side chain. Labels indicate fragments of ␣20 -21(yellow) or ␤1-2 (cyan) digested with trypsin (T) or proteinase K (P), and peptide yields in the heterodimer/monomer are from Begg et al. (15). 4 ϩ 4 repeat ␣-actinin interaction has a K d in the low picomolar range (36 -38). However, a prior study showed recombinant complementary ␣-actinin peptides could not dimerize with high affinity unless all four spectrin-type repeats were present (38). These relative binding affinities are roughly reflected in the amount of solvent accessible surface area buried by each of these complexes. That is, the interface between the two 4-repeat ␣-actinin monomers buries a total of ϳ6244 Å 2 of solventaccessible surface area per dimer, and therefore a 2 ϩ 2 repeat ␣-actinin interaction, which is very weak, buries ϳ3122 Å 2 , while the low nanomolar 2 ϩ 2 repeat spectrin interaction buries ϳ3770 Å 2 (1929 Å 2 for ␣20 -21 and 1841 Å 2 for ␤1-2). The only other dimeric structure of spectrin-type repeats is the crystallographic structure of chicken brain ␣-spectrin repeats 15-17, which forms an anti-parallel homodimer (39). The interface between the two monomers buries a total of ϳ3080 Å 2 of solvent-accessible surface area for this 3 ϩ 3 repeat interaction or ϳ513 Å 2 per individual repeat compared with 943 Å 2 per repeat in the high affinity spectrin dimer initiation site. It is currently unclear whether the CB␣15-17 homodimer has a much smaller buried surface area, because it is not a physiological heterodimer or because these and other spectrin repeats outside the dimer initiation site laterally associate with weak affinity.
In a separate study we recently used isothermal titration calorimetry to characterize binding interactions of ␣and ␤-spectrin dimer initiation site peptides that contained the full dimer initiation site plus differing numbers of unpaired and laterally paired repeats (␣20 -217␤1-2, ␣20 -217␤1-3, ␣20 -217␤1-4, ␣18 -217␤1-2, ␣18 -217␤1-3, and ␣18 -217␤1-4) (40). Using global analysis of these data, we determined the intrinsic enthalpy change and intrinsic entropy change involved in lateral pairing of the dimer initiation site as well as lateral pairing of up to two additional repeats next to the dimer initiation site. Heterodimerization of all combinations tested are both enthalpically and entropically driven. However, interaction of ␣20 -21 and ␤1-2 is dominated by a strongly favorable enthalpy change, at 30°C, with a smaller contribution coming from a favorable entropy term. A large negative enthalpy change often implicates hydrogen bonding as a substantial contributor to the strength of the interaction. Because a favorable enthalpy change is the major component of the Gibbs free energy for the association of ␣20 -21 and ␤1-2, this suggests the inter-chain interaction force primarily comes from hydrophilic interactions. Our refined Docking C model reveals 10 H-bonds, 10 salt bridges, and a minor hydrophobic effect. It is often assumed that oppositely charged atoms in close proximity can form a salt bridge if they are less than or equal to 4.0 Å apart (41). In our model, we used a 6.0-Å cut-off for estimating intermolecular salt bridges due to likely lower accuracy of the model compared with crystallographic structures. In agreement with calorimetric results, the current model of the spectrin heterodimer initiation site indicates H-bonds and salt bridges are key forces in stabilizing spectrin dimer initiation.
In summary, nine intermolecular ion pairs in the ␣20 -21/ ␤1-2 complex were identified using chemical cross-linking and characterization of resulting tryptic peptides by FT-ICR MS. These intermolecular distance constraints provide unambigu-ous experimental evidence for the docking interface within the spectrin dimer initiation site. Our model of the complex, which used these cross-link distance constraints to refine the most plausible homology model, provides a reliable medium resolution conformational model for this critical heterodimer complex. Structural features of this model are consistent with prior protease protection analyses, thermodynamic properties of heterodimer assembly for wild type subunits, and inability of ␣ subunits containing the ␣ LELY 6-residue deletion to heterodimerize. This structure is consistent with a mechanistic model for spectrin heterodimer assembly where long range complementary electrostatic interactions on the ␣20 -21/␤1-2 repeats ensure alignment of the correct specific dimer initiation site repeats despite the fact both subunits have many tandem homologous repeats capable of improper low affinity interaction. After the long range electrostatic interactions within the dimer initiation site align these complementary repeats in proper register, this interaction is stabilized by extensive hydrophilic interactions in an interface that buries a substantial amount of surface area.