Cutting Off Functional Loops from Homodimeric Enzyme Superoxide Dismutase 1 (SOD1) Leaves Monomeric β-Barrels*

Demetallation of the homodimeric enzyme Cu/Zn-superoxide dismutase (SOD1) is known to unleash pronounced dynamic motions in the long active-site loops that comprise almost a third of the folded structure. The resulting apo species, which shows increased propensity to aggregate, stands out as the prime disease precursor in amyotrophic lateral sclerosis (ALS). Even so, the detailed structural properties of the apoSOD1 framework have remained elusive and controversial. In this study, we examine the structural interplay between the central apoSOD1 barrel and the active-site loops by simply cutting them off; loops IV and VII were substituted with short Gly-Ala-Gly linkers. The results show that loop removal breaks the dimer interface and leads to soluble, monomeric β-barrels with high structural integrity. NMR-detected nuclear Overhauser effects are found between all of the constituent β-strands, confirming ordered interactions across the whole barrel. Moreover, the breathing motions of the SOD1 barrel are overall insensitive to loop removal and yield hydrogen/deuterium protection factors typical for cooperatively folded proteins (i.e. the active-site loops act as a “bolt-on” domain with little dynamic influence on its structural foundation). The sole exceptions are the relatively low protection factors in β-strand 5 and the turn around Gly-93, a hot spot for ALS-provoking mutations, which decrease even further upon loop removal. Taken together, these data suggest that the cytotoxic function of apoSOD1 does not emerge from its folded ground state but from a high energy intermediate or even from the denatured ensemble.

Cu/Zn-superoxide dismutase (SOD1) is a ubiquitous radical scavenger that is observed to misfold and aggregate into small intracellular inclusions in the motor neurons of patients with both familial and sporadic amyotrophic lateral sclerosis (ALS) 2 (1)(2)(3). As with other protein misfolding diseases (e.g. Alzheimer disease, Parkinson disease, and the prion diseases), the struc-tural events underlying the pathological aggregation (4,5) and also the role of this aggregation in the neurodegeneration are still obscure (6). Native SOD1 is a thermodynamically stable (7), long lived (8,9), and structurally robust (10) homodimer that coordinates one redox-active Cu 1ϩ/2ϩ ion and one structural Zn 2ϩ ion in each subunit (7,(11)(12)(13). The Cu 1ϩ/2ϩ ion is coordinated directly to the protein's immunoglobulin-like scaffold, whereas the neighboring Zn 2ϩ ion is coordinated more peripherally by the long loop IV (Fig. 1). Loop IV also stretches across the SOD1 surface to support the dimer interface, contains the conserved disulfide bond between Cys-57 and Cys-146, and forms together with loop VII the rigid encasement of the metallated active site (Fig. 1). Upon dissociation of the metal ions, however, the active-site loops loosen up and become more dynamic (14 -18), coupled to a radical loss of protein stability (7,9,18) and increased propensity to aggregate (19 -22). Consistently, several in vivo studies (23) and folding analysis of ALSassociated SOD1 mutations (24) implicate apoSOD1 as the precursor for pathological misfolding and aggregation. Moreover, there is congruent evidence that the core of the SOD1 aggregates involves sequence elements from the ␤-barrel, at least as they occur in aggregation assays in vitro (25) and during overexpression in bacteria and transgenic mice (26). The question is then from which state of the apoSOD molecule the misfolding commences. To shed light on this issue, we examine here the role of the long functional loops in modulating the stability and structure of the apoSOD1 barrel. Experimentally, we replace loops IV and VII with short Gly-Ala-Gly linkers and follow the structural consequences by NMR and folding analysis. The results show that the loop removal is energetically favorable in the sense that it increases the thermodynamic stability of the apoSOD1 monomer but has limited effect on the structure and dynamic motions of the barrel to which they are anchored. In essence, the apoSOD1 barrel retains ordered structure and cooperative folding transition in both the presence and absence of loops IV and VII. Together with the low thermodynamic stability of the immature apoSOD1 monomers, this intrinsic two-state nature of the SOD1 barrel points at the globally unfolded state as the starting material for gain of toxic function in ALS.

EXPERIMENTAL PROCEDURES
Gene Design-The monomeric reference protein, SOD1 pwt , was obtained from the human SOD1 wild-type sequence by introduction of the following mutations: C6A/C111A, which system from Bio-Rad and by electrospray ionization mass spectrometry performed at the Protein Analysis Center (Karolinska Institute, Solna, Sweden) after dialysis against milliQ H 2 O and centrifugation at 18,000 rpm in a Beckman Avanti J-25 centrifuge, JA-25.50 rotor. For the variant SOD1 ⌬IV,⌬VII S111A and S111A/S146C, mutagenesis was performed on the SOD1 ⌬IV,⌬VII gene using the QuikChange TM site-directed mutagenesis kit from Stratagene (Agilent Technologies, Santa Clara, CA) with primers from Eurofins MWG Operon (Ebersberg, Germany).
Equilibrium and Kinetic Measurements-Fast refolding and unfolding kinetics (log k obs Ͼ Ϫ2.5) were measured in 10 mM BisTris, pH 6.3, from Sigma-Aldrich at 25°C by Trp fluorescence using an Applied Photophysics PiStar-180 stopped-flow spectrometer (Leatherhead, UK). The excitation wavelength was 280 nm, and emission was collected with a 320-nm cut-off filter. The final protein concentration was 4 M. Equilibrium unfolding and slow kinetics (log k obs Ͻ Ϫ2.5) were measured on a Varian Cary Eclipse spectrophotometer (Santa Clara, CA) with excitation at 280 nm and emission collected at 360 nm. Ultrapure urea from MP Biomedicals Inc. (Solon, OH) was used in all denaturation experiments.
The SOD1 monomers were assumed to display two-state behavior, yielding the expression, where U and F represent the unfolded and folded monomers, respectively, and k u and k f are the unfolding and refolding rate constants, respectively (28,29). Protein stability, ⌬G U/F ϭ Ϫ2.3RT log K U/F , was further assumed to depend linearly on [urea], yielding the following, log K U/F ϭ log K H2O ϩ m U/F [urea] (Eq. 2) and log k u ϭ log k u H2O ϩ m u [urea] (Eq. 3) log k f ϭ log k f H2O ϩ m f [urea] (Eq. 4) where k f H 2O and k u H 2O represent the rate constants extrapolated to 0 M denaturant, and m u and m f are constants that reflect the sensitivity to the denaturant, commonly perceived as a measure of the change in solvent-accessible surface area in the activation process of unfolding or refolding, respectively (30).
T 1 , T 2 , and steady state heteronuclear NOE experiments were performed on a Bruker 600-MHz spectrometer equipped with a triple resonance probe. Spectra were transformed using nmrPipe and analyzed using the program Sparky. In the T 1 and T 2 experiments, the signal attenuation, from 10 different relaxation delays, was fitted to a single exponential decay, and the relaxation rates were determined. The fitting routine was performed using MATLAB (Mathworks, Natick, MA).
Diffusion experiments were performed using a pulse field gradient longitudinal encoding-decoding (PFG-LED) sequence (43) with a gradient prepulse on a Bruker 600-MHz spectrometer. The 1 H signal intensity was determined at 32 linearly spaced gradient strengths, and the attenuating intensity was fitted to the Stejskal-Tanner equation. The diffusion delay was 150 ms, and the gradient pulses were 5 ms. The gradient strength was calibrated using the water and ␣-cyclodextrin diffusion at 25°C.
H/D exchange was initiated by diluting the protein sample with D 2 O to 50% final D 2 O concentration. The exchange rate was detected as the attenuation of the peak intensity in 15 N-{ 1 H} HSQC experiments. The decay of the peak intensities was fitted to a single exponential decay with base line with a linear slope.

Loop Removal Diminishes the Dimer Interface and Leads to Soluble Monomers
To examine how the active-site loops influence the structural properties of the apoSOD1 molecule, we truncated them by protein engineering (Fig. 1). Loop IV was replaced with a short Gly-Ala-Gly linker between His-48 and Gly-82, removing Cys-57 in the native disulfide bond as well as the Zn 2ϩ ligands His-63, His-71, and His-80. Correspondingly, the electrostatic loop VII was replaced with a Gly-Ala-Gly linker between Ala-123 and Ala-140. These drastic alterations led to the excision of 49 of the 153 residues comprising the wild-type monomer. In addition, to avoid aggregation by disulfide cross-linking, we substituted Cys-6 with Ala, and Cys-111 and the leftover Cys-146 with Ser. The energy-minimized structure of the loop-depleted protein (SOD1 ⌬IV,⌬VII ) is shown in Fig. 1. Essentially, SOD1 ⌬IV,⌬VII comprises nothing but the naked, barrel scaffold of the wild-type monomer, including the poorly structured loop VI (15) that links ␤6 and ␤7 (Fig. 1). SOD1 ⌬IV,⌬VII was subsequently cloned and overexpressed with high yields in E. coli. The first notable effect of loop removal is that SOD1 ⌬IV,⌬VII migrates as a monomer in size exclusion chromatography (Fig.  2). Splitting of the dimer is fully consistent with structural predictions; excision of loop IV substantially reduces the dimer FIGURE 1. The structure of the native SOD1 dimer (Protein Data Bank entry 1HL5) and the loop regions removed by protein engineering. A, in the native SOD1 dimer, the long loops IV and VII adapt a compact and highly ordered structure around the active site, where loop IV also forms part of the dimer interface (green). The left-hand monomer is shown as accessible surface (1.4 Å probe radius), whereas the right-hand monomer is represented as a schematic diagram. Highlighted are the residues coordinating the active-site Cu 1ϩ/2ϩ and Zn 2ϩ ions and the Cys-57-Cys-146 disulfide linkage between loop IV and the central ␤-barrel. B, to examine the structural interplay between the active-site loops and the SOD1 barrel, we substituted loops IV and VII with short Gly-Ala-Gly linkers. The figure shows the energy-minimized structure of the loop-free construct apoSOD1 ⌬IV,⌬VII .
interface area (Fig. 1). Moreover, purification of SOD1 ⌬IV,⌬VII yields apoprotein without coordinated metals. As an additional control of the solution state of apoSOD1 ⌬IV,⌬VII , we measured the protein's translational diffusion coefficient (D t ), which reports directly on the hydrodynamic dimensions and overall fold of the structure (44,45). The analysis was done by PFG-NMR diffusion experiments. As references for monomeric and dimeric SOD1, we used the hydrodynamic radii R H ϭ 22.5 Å and R H ϭ 30.3 Å, determined previously for the apoSOD1 pwt monomer and dimer, respectively (11). The observed hydrodynamic radius of SOD1 ⌬IV,⌬VII is 19.7 Å (Fig. 2). This value is slightly lower than R H for the wild-type monomer (11) but is in good agreement with the predicted radius of 18.6 Å for a globularly folded protein of 110 residues (45). Accordingly, the solution state of apoSOD1 ⌬IV,⌬VII seems to be a folded monomer, fully consistent with the dimensions of the naked SOD1 barrel in Fig. 1. The conclusion is further supported by the spectral difference between apoSOD1 ⌬IV,⌬VII and the apoSOD1 pwt monomer as measured by CD, which shows a coil-like component consistent with the x-ray structure of the active-site loops (see supplemental material S1).

Chevron Data Show That ApoSOD1 ⌬IV,⌬VII Is a Two-state Folder with Enhanced Protein Stability
Stopped-flow analysis (28,29) shows that the refolding (k f ) and unfolding (k u ) rate constants of apoSOD1 ⌬IV,⌬VII yield a v-shaped chevron plot (Fig. 3). Moreover, the values of k f and k u match the equilibrium constant (K U/F ) obtained from equilibrium denaturation data according to Equation 1 (Table 1 and Fig. 3). The chevron plot of apoSOD1 ⌬IV,⌬VII lacks further the Cys-57-Cys-146 of apoSOD pwt is intact. As a consequence of these kinetic changes, SOD1 ⌬IV,⌬VII becomes more stable than SOD1 pwt and requires higher urea concentration to unfold. The chevron data are fitted to Equation 5, and the rate constants are in units of s Ϫ1 . Thermodynamic parameters obtained from kinetic measurements are in agreement with those obtained from equilibrium unfolding experiments by fitting of Equation 6 (  curved unfolding limb characteristic for the dimeric protein (24,29) and is qualitatively the same as that for the apoSOD1 pwt monomer (46) (Fig. 3). Such folding behavior constitutes the hallmark for globular proteins that obtain their structures in cooperative two-state transitions (28,29), where U represents the unfolded state, ‡ is the folding transition state, F is the folded protein, and k down is the downhill rate constant. As an independent test of two-state folding, we have recorded the NMR HSQC spectrum at the midpoint of the urea equilibrium transition. Consistent with Scheme 1, these data show only a clean mixture of random coil and folded state cross-peaks and no trace of populated intermediates (data not shown). Removal of loops IV and VII thus seems to obliterate dimerization without affecting the two-state folding behavior of the SOD1 monomers. Even so, the chevron plots of apoSOD1 ⌬IV,⌬VII and the apoSOD1 pwt monomer display two interesting differences. First, loop removal induces a nearly 10-fold increase of the folding rate constant; log k f H 2O increases from Ϫ1.12 Ϯ 0.1 to Ϫ0.28 Ϯ 0.03 (Table 1). The corresponding effect on the unfolding rate constant, however, is much smaller, with extrapolated log k u H 2O values of Ϫ3.91 Ϯ 0.05 and Ϫ4.03 Ϯ 0.04 for apoSOD1 ⌬IV,⌬VII and apoSOD1 pwt , respectively (Table  1). Coupled to these changes of k f and k u is a gain in protein stability of ⌬⌬G U/F ϭ Ϫ1.31 Ϯ 0.17 kcal/mol (Equations 2-4 and Table 1). The stability gain is mainly due to the replacement A111S in SOD1 ⌬IV,⌬VII , which accounts for Ϫ1.29 kcal/mol (supplemental Fig. S2 and Table S1). The source of this stability gain seems to be better space filling at the protein surface around position 111; Ser mimics more closely the native Cys than the smaller Ala. The contribution from the mutation C146A in SOD1 ⌬IV,⌬VII , on the other hand, seems relatively small, judged by the double mutation SOD1 ⌬IV,⌬VII C146A/ S111A, which displays a stability overall similar to that of the single mutant SOD1 ⌬IV,⌬VII S111A (supplemental Fig. S2 and Table S1). The effects of mutation in positions 111 and 146 of SOD1 ⌬IV,⌬VII are in good agreement with previous results for the wild type-like apo monomer (29). On this basis, we conclude that loops IV and VII contribute relatively little to the stability of the apo monomer, despite being an integral part of the native SOD1 structure. More detailed interpretation of the energetics of loop removal is currently precluded by the unknown impact of the Gly-Ala-Gly linkers; in addition to reducing the sequence separation between the anchoring strands, these could destabilize the SOD1 ⌬IV,⌬VII by steric strain. The second effect of loop removal is a pronounced decrease of the slope of the unfolding limb (m u ); the urea dependence of log k u goes down (Table 1). Such a selective decrease of the m u value indicates that the folded state (F) has reduced in size (i.e. loop removal leads to smaller exposure of surface area in the global unfolding process) (47,48). Consistently, an identical change of the m u value is observed for a SOD1 variant in which loops IV and VII are locally unfolded due to mis-ligation of Zn 2ϩ (31). Loops that remain unfolded throughout the folding reaction do not contribute to burial of solvent-accessible surface area. Taken together, these observations suggest that the solution structure of the loop-truncated apoSOD1 ⌬IV,⌬VII species is indeed a naked barrel, as depicted in Fig. 1.

Loop Removal Has Minor Effects on the Chemical Shifts of the SOD1 Barrel
The 15N-{1H}-HSQC spectrum of SOD1 ⌬IV,⌬VII shows well dispersed cross-peaks at positions typical for a folded protein (Fig. 4). A notable feature of this spectrum is that many of the cross-peak overlaps caused by loops IV and VII in the apoSOD1 pwt monomer are eliminated, facilitating analysis of the barrel scaffold. Assignment of the apoSOD1 ⌬IV,⌬VII backbone was done by a standard set of three-dimensional NMR spectra and allowed determination of the NH cross-peaks of 107 of the 108 non-proline residues (supplemental Table S2).
As a first, tentative, test of structural alterations upon loop removal, we compared the chemical shifts of apoSOD1 ⌬IV,⌬VII with those determined previously for the apoSOD1 pwt monomer (15). The weighted difference of the 1 H N and the 15 N H chemical shifts is shown in Fig. 4. On the whole, the differences are very small. As expected, significant changes in the chemical shifts occur in the regions close to the insertion of the Gly-Ala-Gly linkers between His-48 and Gly-82 and between Ala-123 and Ala-140 (Fig. 4). The primary reason for this change is unlikely to be alterations of the three-dimensional structure but rather the altered chemical identity of the residue neighbors. Likewise, large chemical shift differences are found around the mutation C146S in apoSOD1 ⌬IV,⌬VII and for Ile-112 and Ile-113 just after the mutation C111S. These differences arise because the chemical shifts of Ile-112 and Ile-113 in the apoSOD1 pwt monomer are unusual and deviate significantly from the expected values of a folded protein. Why this local irregularity appears is not yet clear; the only difference is that in apoSOD1 pwt , Cys-111 is substituted for Ala (29).

Secondary Chemical Shifts Reveal Native-like ␤-Structure
To identify the parts of apoSOD1 ⌬IV,⌬VII that are constrained by secondary structure formation, we employed secondary chemical shift analysis (49 -51) Figs. 1 and 5). Notably, the secondary chemical shift deviations from random coil values of ␤4, ␤5, and ␤8 are less pronounced than for the other strands. Together with a somewhat lower intensity of the cross-peaks, this indicates less complete structure formation in these parts of the protein (data not shown) (i.e. dynamic motions reduce the cross-peak intensity by exchange broadening). A very similar distinction between ␤4 and ␤5 and the rest of the barrel has been observed earlier for the apoSOD1 pwt monomer by Banci et al. (15). Thus, in terms of ␤-structure content, the solution structure of apoSOD1 ⌬IV,⌬VII is indistinguishable from the scaffold of the apoSOD1 pwt monomer.

Evidence for Ordered Barrel Structure
Extensive Network of Long Range NOEs-In earlier studies, it has been suggested that the ␤-sheet facing the active-site loops gains increased conformational freedom and becomes partly disordered upon demetallation of the SOD1 dimer (14). To see whether such structural loosening is triggered also by loop removal, we mapped out the sequential and long range NOEs within the apoSOD1 ⌬IV,⌬VII structure. The presence of long range NOEs indicates that the contacts mediating the dipolar couplings are close, relatively long lived, homogeneous, and stable (52). Such robust long range contacts constitute the hallmark for ordered tertiary structure in globular proteins (53). As expected, sequential NOEs are found throughout the backbone of the apoSOD1 ⌬IV,⌬VII structure. In addition, we observe an extensive network of long range NOEs joining together all strands of the ␤-barrel into a native-like tertiary topology (Fig.  6). The only gap in this pattern is the absence of NOEs between the ends of ␤5 and ␤6 that splay apart in the crystallographic structure of the wild-type protein. This slit in the barrel hydrogen bonding closes up toward the short linkage between ␤5 and ␤6, which contains the "torsioned" residue Gly-93, a hot spot for ALS-associated mutations. Another notable region is the interface between ␤4 and ␤5, which comprises relatively few NOEs. These NOEs are mainly between the adjacent pairs Gly-41 ␤4 -Ala-89 ␤5 and His-43 ␤4 -Val-87 ␤5 at the top of the barrel, whereas the lower interface between ␤4 and ␤5 lacks NOEs. The pattern is, again, consistent with the slight divergence of ␤4 and ␤5 at these positions seen in the x-ray structure. Moreover, it can be noted that the Zn 2ϩ ligand Asp-83 ␤5 imposes a twist to the C-terminal end of ␤5, which could further compromise the interactions with ␤4. Strand 5 can thus be seen as moving away from the edge of the active-site sheet to partly fill the slit to ␤6. It is possible that this frustrated arrangement of ␤5 compromises its structural rigidity in the absence of coordinated metals. Overall, these NOE data match the subset of NOEs observed for the ␤-barrel of the apoSOD1 pwt monomer (15), showing that the truncation of loops IV and VII has no major effect on the SOD1 scaffold. On this basis, we conclude that the solution structure of apoSOD1 ⌬IV,⌬VII is a rigid ␤ barrel, in good accord with its two-state folding behavior (Fig. 3). 15 N NMR Relaxation Data-Evidence for an ordered SOD1 barrel is further provided by data from 15 N NMR relaxation measurements (54, 55) (supplemental material S5). The heteronuclear 15 N-{ 1 H} NOE values are overall high (0.81 Ϯ 0.1) and constant along the apoSOD1 ⌬IV,⌬VII backbone, with only slight dips in the loop regions connecting the ␤ strands. Such high order of the backbone is characteristic for well packed, folded structures (56). For comparison, the more dynamic loops IV and VII of apoSOD1 pwt show average NOE values of 0.44 and 0.27, respectively (supplemental material S5). As a final test of the conformational state of apoSOD1 ⌬IV,⌬VII , we used the ratio of the relaxation rates R 2 /R 1 to estimate the rotational correlation time ( c ) (57), which constitutes a sensitive measure of the protein's hydrodynamic dimensions (supplemental material S5). By omitting the relaxation rates from the flexible loop regions, we calculated c of the apoSOD1 pwt monomer to 9 ns, corresponding to a globular protein with a hydrodynamic radius R H ϭ 21.5 Å. This estimate is in good agreement with the previously published value of 22.5 Å (11). Correspondingly, the c and R H values of apoSOD1 ⌬IV,⌬VII were determined to be 6.7 ns and 19.5 Å, respectively, which are in excellent agreement with the results from the pulse field gradient-NMR diffusion experiments in Fig. 2. Taken  mental material S4). The numbers are expressed as deviation from random coil values. Overall, the secondary structure propensity of apoSOD1 ⌬IV,⌬VII is very similar to those of apoSOD1 pwt (data not shown) and indicates retained native-like ␤-structure. Inserted is a schematic outline of the ␤-structure in the native holoSOD1 dimer (Protein Data Bank entry 1HL5) (cf. Fig. 1).
together, these data provide strong evidence that the solution structure of apoSOD1 ⌬IV,⌬VII is a rigid ␤-barrel, in good accord with its two-state folding behavior (Fig. 3).

Detection of Rare Structural Fluctuations by Hydrogen Exchange Experiments
A common method for studying rare fluctuations in protein structures is native state H/D exchange of the backbone amides (58,59). The method is based on the observation that amide protons are protected from H/D exchange when they are fixed in stable intramolecular H-bonds. As long as an individual backbone H-bond is protected in this manner, it is considered to be in its closed state (C).  This is known as the EX1 limit, and exchange through this mechanism occurs typically from the unfolded state following global unfolding (58). Alternatively, if closing is rapid and only a fraction of the opening transitions lead to exchange (k close Ͼ k ex int ), Equation 7 can be rewritten as follows, where the stability of the open state estimated from ⌬G O/C is ϪRTlnK O/C . This is known as the EX2 regime, and exchange through this mechanism is often found for local fluctuations on the native side of the folding barrier (58). Following standard procedures, we measured in this study k ex obs by HSQC NMR (supplemetal material S6) and calculated the values of k ex int from model peptide data using the protocols by Englander and co- FIGURE 6. Schematic outline of the apoSOD1 ⌬IV,⌬VII structure showing the long range NOE connectivity as measured by 15 N-filtered NOESY experiments. The contacts with high intensity NOEs are indicated by dotted lines, and the included NOEs are those that connect residues i and j if ͉i Ϫ j͉ Ͼ 3. Shown as a reference is the crystallographic secondary structure of the native holoSOD1 dimer (Protein Data Bank entry 1HL5). NOEs are found to connect all of the crystallographic turns and ␤ strands (except for the edge strands ␤5 and ␤6), suggesting that the apoSOD1 ⌬IV,⌬VII monomer has a folded and relatively rigid structure. The lack of NOEs between the flanks of ␤5 and ␤6 corresponds to the gap between these stands observed in 1HL5 (Fig. 1). The figure design was adapted from Ref. 10. workers (60). The results for apoSOD1 pwt and apoSOD1 ⌬IV,⌬VII are listed in Table 2.

Data Define Three Levels of H/D Exchange Rates
The three levels are (i) slowly exchanging amide protons with k ex obs values similar to or somewhat slower than the rate constant for global unfolding k u (Ϫ4.9 Ͻ log k ex obs Ͻ Ϫ3.9); (ii) amide protons that exchange on intermediate time scales (Ϫ3.9 Ͻ log k ex obs Ͻ Ϫ2.8); and (iii) amide protons that exchange rapidly in the dead time of the experiment (log k ex obs Ͼ Ϫ2.8). The slowly exchanging amides, we conclude, represent stable H-bonds that exchange through global unfolding close to the EX1 limit where k ex obs Ϸ k u . Consistent with this, the k ex obs values for these positions are overall lower for apoSOD1 ⌬IV,⌬VII than for apoSOD1 pwt ( Table 2). The reason for this difference is 2-fold. First, apoSOD1 ⌬IV,⌬VII unfolds somewhat slower than apoSOD1 pwt with log k u H 2O values of Ϫ4 and Ϫ3.9, respectively. Second, the higher refolding rate of apoSOD1 ⌬IV,⌬VII , log k f H 2O ϭ Ϫ0.28 versus log k f H 2O ϭ Ϫ1.12 for apoSOD1 pwt , ensues at several positions a kinetic competition between k close and k ex int ( Table 2). This decreased probability of exchange during the globally unfolded state decreases k ex obs and biases the system toward the EX2 regime (Equation 7 and supplemental material S6). Correspondingly, the amides with intermediate k ex obs values (Ϫ3.9 Ͻ log k ex obs Ͻ Ϫ2.8) indicate backbone positions that exchange by local fluctuations in the clean EX2 regime. Finally, the amide protons that exchange rapidly (log k ex obs Ͼ Ϫ2.8) are either exposed to the solvent or involved in relatively weak hydrogen bonds, which offer little protection. As observed for other proteins (61), the exchange behavior of SOD1 under close to physiological conditions is thus a mix between EX1 and EX2. As an independent test of EX1 and EX2 behavior, we determined the exchange rates at three different pH values, pH 5.4, 6.3, and 7.2 (supplemental Table S3). Amide protons exchanging in the EX2 regime are expected to display exchange rate constants that follow the pH dependence of k ex int (Equation 9) (i.e. they should increase by a factor of 10 per increased pH unit (60). The results show that, consistently, the amide protons with fast and intermediate exchange at pH 6.3 exchange rapidly in the experimental dead time at pH 7.2, whereas they are considerably slowed down at pH 5.4 (supplemental Table S3). At pH 5.4, we can also resolve some of the amide protons that exchange too rapidly for detection at pH 6.3 (e.g. 41, 43, and 74 -76) (supplemental Table S3). The amide protons exchanging with rate constants close to global unfolding, on the other hand, show much weaker dependence on pH, corroborating the idea that these positions are in the EX1 regime (supplemental Table S3). Phenylalanine 45 is also detected at low pH, whereas the cross-peak is not detectable at higher pH. The measured exchange rate at pH 5 is similar to that of His-46, suggesting that Phe-45 forms a hydrogen bond with Ala-55 and exchanges by the same dynamic events as His-46. Toward a final test of the slowly exchanging positions, we determined the pH dependence of log k u H 2O from chevron data. The results shows that log k u H 2O has a maximum around pH 6.3 and decreases slightly both upon raising and lowering the pH (supplemental Fig. S6 and Table S4). This pH dependence matches that of the average of the slowly exchanging amides (i.e. the pH dependence of log k u H 2O follows that of log k ex obs ) as is expected for EX1 exchange rate-limited by global unfolding (supplemental Fig. S7). Accordingly, the pH data corroborate our conclusion that the slowly exchanging amides reflect global unfolding and also that the protected core is flanked by residues that exchange in the EX2 regime (Table 2 and supplemental Table S3).
The structural locations of the amide protons with slow, intermediate, and rapid exchange are shown in Figs. 7 and 8. From these data, it is apparent that the exchange patterns of apoSOD1 ⌬IV,⌬VII and apoSOD1 pwt are very similar; most of the slowly exchanging amides are confined to the H-bond network defining the central ␤ barrel, whereas the rapidly exchanging amides are mainly in the loops connecting the ␤ strands. However, there are two regions of the SOD1 barrel that stand out as structurally deviant. Below is a detailed description of these regions and how they respond to loop removal and metallation. The effects of metallation are deduced by comparison with previously published H/D exchange data from the holoSOD1 dimer (10).

Weak Protection of ␤2 at the Center of the Major Sheet
For both apoSOD1 ⌬IV,⌬VII and apoSOD1 pwt , the amide protons of Gly-16 ␤2 and Ile-17 ␤2 exchange rapidly in the dead time of the experiment (log k ex obs ϾϪ2.8), despite forming H-bonds with the neighboring strands in the crystal structure (Table 2 and Figs. 7 and 8). Next to Ile-17 ␤2 is the rapidly exchanging Val-7 ␤1 facing a local gap in the H-bonding to Val-148 ␤8 , which yields a contiguous region of weakly protected H-bonds at the center of the main sheet. Interestingly, a similar exchange pattern is seen in the holoSOD1 dimer (10), suggesting that the weak protection of the N-terminal end of ␤2 is a built-in feature of the SOD1 barrel rather than an effect of metal loss or dimer splitting. The explanation for the weak protection could be the notably long H-bond distances in this region coupled to increased backbone flexibility (10). In the x-ray structure of holoSOD1 pwt (Protein Data Bank entry 2XJK), the distance between the Ile-17 ␤2 amide and the Ser-34 ␤3 carbonyl is 2.46 Å compared with an average of 1.9 Ϯ 0.2 Å for the rest of the sheet. The apparent flexibility of ␤2 disappears at the highly protected residue Ile-18 ␤3 , which anchors the strand into the core at the packing layer of the folding nucleus (46) ( Table 2 and Figs. 7 and 8). Such closely packed layers of hydrophobic side chains are not unique for the core of SOD1 but appear to be a general feature of immunoglobulin-like folds (62). Before this position, ␤2 is poorly anchored to the SOD1 core, mainly because the conserved position Gly-16 ␤2 lacks side chain. The role of this seemingly conserved "imperfection" in the SOD1 barrel, if any, is not yet clear. Although data from the holoSOD1 dimer indicate that the protection of ␤2 undergoes a small change upon mutation G93A (10), we do not detect any corresponding effects upon loop removal in the apo monomer (Table  2 and Figs. 7 and 8). We note, however, that ␤2 is one of the few segments in the SOD1 sequence with a propensity to fibrillate in vitro (25,26). In combination with its low protection factors, this local sequence property underlines ␤2 as a putative aggregation hot spot. Also, this rapidly exchanging region of the sheet borders in the wild-type protein the buried Cys-6 ␤1 ,  .89 a Amino acid; the subscript ␤i denotes the secondary structure element in which the amino acid residue is located. b Calculated from the fit of I ϭ I 0 exp(Ϫt⅐k ex ) to the time dependence of the attenuation of NH cross-peak intensity. c Error is the S.D. in the fitted k ex from 10 4 fits with randomized normal distributed noise estimated from the spectra. d Intrinsic exchange rates were calculated from model-peptide data using the protocols by Englander and co-workers (60). e log PF ϭ log(k int /k ex ). f Numbering of the SOD1 ⌬IV,⌬VII sequence. g Not completely exchanged during the experimental dead time but too few data points for reliable fitting. h The pH dependence of the exchange rate suggests that these amide protons exchange in the EX2 regime. i Substitution of the loop IV residues 49 -81 to Gly-Ala-Gly. j Substitution of the loop IV residues 124 -139 to Gly-Ala-Gly. which is implicated in aggregation by disulfide cross-linking (14).

Weak Protection of ␤4, ␤5, and ␤7 at the Interface to the Active Site
The second region of the SOD1 barrel with poorly protected H-bonds is the segment comprising ␤4, ␤5, and the beginning of ␤7. From a structural perspective, these strands are interesting because they anchor the functional loop IV (␤4 and ␤5) and harbor several of the metal-binding ligands (i.e. the Cu 1ϩ/2ϩ ligands His-46 ␤4 , His-48 ␤4 , and His-120 ␤7 and the Zn 2ϩ ligand Asp-83 ␤5 ) (Fig. 7). Although ␤4 stretches between Leu-42 ␤4 and His-48 ␤4 in the crystal structure, its only position with slowly exchanging amide is His-46 ␤4 (log k ex obs ϭ Ϫ4.42), which forms an H-bond with the carbonyl of V118 ␤7 (Table 2 and Figs. 6 and 7). This is also the only position with backbone NOEs between ␤4 and ␤7 in apoSOD1 ⌬IV,⌬VII (Fig. 6). The amides of ␤4 facing the carbonyls in ␤5 all exchange rapidly (log k ex obs Ͼ Ϫ2.8). Even so, the amides of Val-87 ␤5 and Ala-89 ␤5 , which bind back to the carbonyls of His-43 ␤4 and Gly-41, display relatively slow exchange rates (Fig. 7). In apoSOD1 pwt , the log k ex obs values of Val-87 ␤5 and Ala-89 ␤5 are Ϫ4.13 and Ϫ3.53, respectively (Table 2). This large variation of the k ex obs values in the sheet, taken together with the intermediate exchange rate of Ala-89 ␤5 , suggests that ␤4 and ␤5 exchange protons by local fluctuations in, or close to, the EX2 regime (Equation 9). The extent of structural opening required for such local exchange is believed to be small (63), ranging from backbone breathing sufficient to allow the penetration of individual water molecules (64,65) to local unfolding of a few neighboring H-bonds (66,67). In apoSOD1 ⌬IV,⌬VII , the exchange rates of Val-87 ␤5 and Ala-89 ␤5 increase to log k ex obs ϭ Ϫ3.36 and log k ex obs Ͼ Ϫ2.8, respectively ( Table 2). The increase supports not only the interpretation that ␤4 and ␤5 exchange by local fluctuations but shows further that these fluctuations are facilitated by loop removal. Following loop removal, we observe also increased exchange rates for Asp-90, Gly-93, and Val-94 (loop V) and for Thr-116 ␤7 that H-bonds to His-48 ␤4 , next to the Gly-Ala-Gly insert ( Table 2 and Figs. 7 and 8B). Structurally, the increased flexibility of the ␤4-␤5 region could have several causes. It could arise from loss of the loop-packing interface, loss of backbone H-bonding between Lys-122 (loop VII) and the twisted Gly-44 ␤4 , or conformational strain introduced by the short Gly-Ala-Gly linker between His-48 ␤4 and Gly-82. Comparison with the H/D exchange data from the holoSOD1 dimer suggests that the high exchange rates of ␤4, ␤5, and ␤7 in apoSOD1 pwt and apoSOD1 ⌬IV,⌬VII are induced by demetallation (Fig. 7). In the holoSOD1 dimer, the majority of positions in this barrel segment as well as around the Zn 2ϩ -binding site display slow exchange rates (10). We conclude from these results that the loss of metal ions from the SOD1 structure not only sets free rapid dynamic motions of the functional loops IV and VII (14 -18) but also enhances the slower breathing motions of the underlying ␤ sheet and loop V. However, these breathing motions of the apoSOD1 barrel are relatively rare and are not in conflict with the minimal two-state folding behavior depicted in Scheme 1. For example, the intermediate exchange rate of Ala-89 ␤5 in apoSOD1 pwt corresponds to an open state that is populated only 1 ⁄ 2200 of the time (K O/C ϭ 4.4⅐10 Ϫ4 ; Equation 9), yet the interaction between Ala-89 ␤5 and Gly-41 ␤4 yields strong backbone NOEs in apoSOD1 ⌬IV,⌬VII , where the H/D exchange rate is even faster (Fig. 6).

Stable Stacking of Loops 3 and 5 Constitutes a Hot Spot for ALS-associated Mutations
The amide of Leu-38 (loop III) forms a highly protected H-bond to the carbonyl of Gly-93 (loop V). This indicates high structural rigidity in the stacking of loops III and V, which is also a hot spot for ALS-associated SOD1 mutations (2). An analogous, but not identical, strong protection pattern of this region is reported for the holoSOD1 dimer by Jonsson and coworkers (10), who observe further that the local ALS mutation G93A, which strains loop V sterically, has long range effects on the metal binding sites (10) at the other ends of ␤4 and ␤5. In this study, we find reciprocally that substitution of the metal-supporting loops IV and VII leads to increased H/D exchange rates in loop V (Fig. 7). An explanation for this apparent allosteric effect could be that the stacked loops III and V play a role in supporting the structure of the bridging and relatively dynamic strands ␤4 and ␤5. Whether the observed phenomenon is just a trivial effect of loop removal, stems from intrinsic structural frustration in the active-site region (31,68), or reflects a functional feature of the SOD1 molecule is not yet clear.

Loops IV and VII Have Modest Influence on the Barrel Stability in the Absence of Coordinated Zn 2؉
The immunoglobulin-like barrel represents one of the most common folds in living organisms (69), indicating viable and useful structural properties. In the case of SOD1, it provides a scaffold for biological function by anchoring the active-site loops IV and VII (Fig. 1). Besides encapsulating the redox-active Cu 1ϩ/2ϩ ion, these loops seem to contain all of the features necessary for a concerted activation and maintenance of the protein's enzymatic function (i.e. interaction with the Cu 1ϩ/2ϩinserting chaperone CCS) (70), control of the dimer interface (71), the Cys-57-Cys-146 disulfide linkage, and the site for the thermodynamically stabilizing Zn 2ϩ ion. It is possible that this complex functional optimization of the SOD1 loop region has occurred, at least partly, in conflict with the structural properties of the barrel scaffold (31,72,73). The existence of such a conflict could have direct bearing on the susceptibility of SOD1 to misfold and the involvement of SOD1 in ALS (31). Consistent with the idea that the functional features exert a burden on the SOD1 scaffold, apoSOD1 ⌬IV,⌬VII is slightly more stable than apoSOD1 pwt (Table 1). Upon reductive cleavage of the apoSOD1 pwt Cys-57-Cys-146 disulfide bond (29), the gain of loop removal increases to Ϫ3.6 Ϯ 0.2 kcal/mol (30). Bearing in mind that the removed loops represent 28% of the original sequence and also form an extensive interface with the barrel structure ( Fig. 1), such a minor effect on protein stability is remarkable; the well packed x-ray structure of holoSOD1 would rather suggest a detrimental loss of global stability. One explanation for the destabilizing effect of loops IV and VII is that they, for functional reasons, contain too few hydrophobic side chains for proper close packing in the absence of coordinated Zn 2ϩ . As a result, the loop structures will become dynamic and penalize the folded apo structure entropically. This idea is supported by the observation of augmented loop dynamics in both the monomeric (15,17) and dimeric (14) apoSOD1 structures. However, the situation changes abruptly upon coordination of the native Zn 2ϩ ion. Binding of Zn 2ϩ drives loops IV and VII into a compact tertiary topology and boosts the monomer stability by Ͼ3 kcal/mol (74). Accordingly, loops IV and VII not only orchestrate biological function but constitute also a powerful modulator of global stability that enhances the integrity of the native holoSOD1 structure.

Loop Removal Has Limited Effect on the Apo-barrel Dynamics
In good accord with stability data, long range NOEs (Fig. 6), 15 N NMR relaxation (supplemental material S5), and H/D exchange rates (Fig. 7) show that the removal of loops IV and VII also has a very limited effect on the dynamic properties of the SOD1 barrel; the SOD1 barrel stands out as an independent structural domain that retains its cooperative folding behavior also in the absence of the functional loops. Correspondingly, the folding behavior of the titin domain TI I27, another protein with an immunoglobulin-like structure (69), has turned out relatively insensitive to loop insertions, pointing to the possibility of a general decoupling between folding and function of this common structural motif (75). The modular and seemingly straightforward architecture of the SOD1 molecule sheds new light on the identity of the precursor state for toxic misfolding and aggregation. A common view is that the misfolding commences from a high energy intermediate in the form of a locally unfolded barrel, as has been implicated in aggregation of ␤2-microglobulin (4,76). The most apparent region of the apoSOD1 barrel to undergo such fluctuations would then be the weakened strands ␤4 and ␤5, which also anchor the dynamic loop IV (Fig. 1). This region, containing the edge protection of the sheet, has also been implicated as a misfolding hot spot in previous NMR (15,17) and folding (46) studies. Given the cooperative folding behavior of the SOD1 barrel, however, a more reductionist conclusion would be that the precursor for misfolding is the denatured ensemble (24, 77) (cf. U in Scheme 1). At physiological temperatures, where the disulfide-reduced apo monomers are already in the unfolding transition region because of their low thermodynamic stability (29,77), the occupancy of globally denatured species is likely to exceed that of any high energy intermediate on the native side of the folding barrier. Under the same conditions, some of the severely destabilizing ALS-associated SOD1 mutations (e.g. G41D ␤4 , H43R ␤4 , and G93A) will even populate the denatured ensemble close to 100% in the absence of coordinated metal ions (24). The aggregation precursor of SOD1 could then be structurally similar to those of the A␤ peptide associated with Alzheimer disease and the disordered protein ␣-synuclein associated with Parkinson disease (i.e. an overall flexible chain that samples in a dynamic way aggregation-prone microstates) (6). Such a scenario would also explain how so many destabilizing, but structurally disparate, SOD1 mutations can still result in a common ALS phenotype; they all shift the folding equilibrium toward the denatured ensemble (24).