Structure and backbone dynamics of a microcrystallinemetalloprotein by solid-state NMRMichael J. Knighta, Andrew J. Pella, Ivano Bertinib,1, Isabella C. Fellib, Leonardo Gonnellib, Roberta Pierattellib,Torsten Herrmanna, Lyndon Emsleya, and Guido Pintacudaa,1

aCentre de Résonance Magnétique Nucléaire à Très Hauts Champs, Unité Mixte de Recherche 5280 Centre National de la Recherche Scientifique / EcoleNormale Supérieure de Lyon, Université Claude Bernard Lyon 1, 5 rue de la Doua, 69100 Villeurbanne, France; and bDepartment of Chemistry“Ugo Schiff” and Centro di Risonanze Magnetiche, University of Florence, via Luigi Sacconi 6, 50019 Sesto Fiorentino, FI, Italy

Edited by* Harry B. Gray, California Institute of Technology, Pasadena, CA, and approved May 21, 2012 (received for review March 17, 2012)

We introduce a new approach to improve structural and dynamicaldetermination of large metalloproteins using solid-state nuclearmagnetic resonance (NMR)with 1H detection under ultrafast magicangle spinning (MAS). The approach is based on the rapid andsensitive acquisition of an extensive set of 15N and 13C nuclearrelaxation rates. The system on which we demonstrate thesemethods is the enzyme Cu, Zn superoxide dismutase (SOD), whichcoordinates a Cu ion available either in Cuþ (diamagnetic) or Cu2þ

(paramagnetic) form. Paramagnetic relaxation enhancements areobtained from the difference in rates measured in the two formsand are employed as structural constraints for the determinationof the protein structure. When added to 1H-1H distance restraints,they are shown to yield a twofold improvement of the precision ofthe structure. Site-specific order parameters and timescales of mo-tion are obtained by a Gaussian axial fluctuation (GAF) analysis ofthe relaxation rates of the diamagnetic molecule, and interpretedin relation to backbone structure and metal binding. Timescales formotion are found to be in the range of the overall correlation timein solution, where internal motions characterized here would notbe observable.

paramagnetism ∣ nuclear relaxation rates ∣ copper ∣ microcrystal

Structure determination of proteins plays a central role in un-derstanding key events in biology. Although the structure of

many proteins can be obtained from single-crystal X-ray diffrac-tion, or by solution-state nuclear magnetic resonance (NMR)spectroscopy, there is nevertheless a range of important sub-strates for which structures cannot be determined today. Theseinclude immobile systems lacking long-range order such as pro-tein aggregates, large complexes and membrane-bound systems.

Solid-state NMR has the unique potential to study, with atom-ic resolution, systems of this nature, and spectacular progress hasbeen made in this area over the last decade (1). There are today asmall handful of structures obtained by solid-state NMR, frommicrocrystalline samples to fibrils and membrane-associatedsystems (2). Additionally, solid-state NMR is uniquely sensitiveto site-specific protein dynamics over a broad range of timescales,and a number of demonstration studies have recently appearedfor model proteins (3).

The use of perdeuterated proteins has very recently opened theway to highly sensitive proton-detected solid-state experiments(4). Despite early proof-of-principle papers (5), this approach onlybecame popular with the realization that amide sites must be onlypartially reprotonated (typically 10–30% back exchange) (6–8) toyield well-resolved 1H spectra. This represented a significant com-promise in sensitivity to gain resolution, and effectively made thedetermination of internuclear distances impractical, with few ex-ceptions (9–11). We have recently shown how this problem can becompletely overcome by using 100% reprotonation of exchange-able sites, without loss of resolution, if perdeuteration is combinedwith ultrafast MAS (60 kHz) (12, 13). Well-resolved “fingerprint”spectra may then be acquired rapidly, opening up the way to thedetection of a range of structurally important parameters.

Paramagnetic effects have long been used and today play acentral role in solution-state NMR structural determinations inmetalloproteins (14–16), or in biomolecules covalently modifiedwith a spin-label (17–19). Importantly, paramagnetic effects canmanifest themselves as either shifts (contact or “pseudocontact”shifts) or enhanced relaxation (PREs), depending on the charac-ter of the metal center. These effects act up to very long distances,and carry a well-defined dependence on the nuclear positionwith respect to the paramagnetic center. In solids, measuringpseudocontact shifts is relatively straightforward, and when theyoccur, they can be used as invaluable structural constraints (20–22). However, measuring PREs in solids is a prime examplewhere traditional detection methods have difficulty to providesufficient sensitivity. While enhanced relaxation caused by a para-magnetic center can be exploited for fast recycling and condenseddata collection approaches relying on 13C detection (23–25), theaddition of paramagnetic dopants is not appropriate for the quan-titative measurement of relaxation times as a source of structuraland dynamic information. As a result, site-specific PREs in thesolid state have only been reported on one small model protein(GB1) by the use of cysteine-containing mutants with paramag-netic tags attached (26, 27).

Here we demonstrate a new approach to improving structuraland dynamical determination of large metalloproteins. Using 1H-detected multidimensional experiments at high magnetic fieldunder ultra-fast magic-angle spinning (MAS), combined withperdeuteration, we can efficiently measure a large number of 13Cand 15N nuclear relaxation rates with high sensitivity in a largemetalloprotein as well as 1H-1H distance restraints. This leads tothe complete description of its site-specific motions in the ns-usscale and to the evaluation of a large number of long-range para-magnetic restraints for the determination of the full structure ofthe molecule, including a well-defined active site.

We apply these methods to the enzyme superoxide dismutase(SOD) (28, 29), which coordinates a Cu and a Zn ion and hasthe physiological function of protecting cells from oxidative stressby catalyzing the dismutation of the superoxide anion. The Cu ionis essential to catalysis, and cycles between the (paramagnetic)Cu2þ state and (diamagnetic) Cuþ state during the reaction.

PREs are obtained from the difference in rates, and whenadded to 1H-1H distance restraints, are shown to improve the

Author contributions: I.B., I.C.F., R.P., T.H., L.E., and G.P. designed research; M.J.K., A.J.P.,I.C.F., L.G., R.P., and G.P. performed research; M.J.K., T.H., L.E., and G.P. analyzed data; andM.J.K., I.B., I.C.F., R.P., T.H., L.E., and G.P. wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Data deposition: The atomic coordinates and chemical shifts have been deposited inthe protein data bank www.pdb.org (PDB ID code 2LU5) and the BioMagResBankwww.bmrb.wisc.edu (accession no. 18509).1To whom correspondence may be addressed. E-mail: Ivanobertini@cerm.unifi.it orguido.pintacuda@ens-lyon.fr.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1204515109/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1204515109 PNAS ∣ July 10, 2012 ∣ vol. 109 ∣ no. 28 ∣ 11095–11100

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

CHEM

ISTR

Y

precision of the structure by a factor of two. Site-specific orderparameters and timescales of motion are obtained by a Gaussianaxial fluctuation (GAF) analysis of the relaxation rates of the dia-magnetic molecules.

Results and DiscussionNMR Experiments. Fig. 1 shows 1H-detected 15N-1H correlationspectra (15N-1H Cross-Polarization Heteronuclear Single Quan-tum Coherence or CP-HSQC, Fig. S1) of microcrystalline SODin its Cuþ,Zn2þ state (A-B, diamagnetic) and Cu2þ,Zn2þ state(C, paramagnetic) recorded using ultrafast MAS at 60 kHz at850 MHz.

Samples are U-½2H; 13C; 15N� and have been allowed to ex-change with H2O after expression, such that all exchangeablesites are protonated (100% back-protonation). As shown recentlyin the case of the analog Zn2þ-SOD, this protocol yields hereexcellent resolution and sensitivity with 1H coherence lifetimesof generally around 10 ms with line widths of around 100 Hz(13). These spectra can be acquired in 15 min using only 3.5 mg(<0.22 μmol) of protein in a 1.3 mm rotor.

In the case of Cuþ,Zn2þ-SOD (which is diamagnetic), 136backbone amide resonances out of 147 nonproline residues wereassigned using 1H-detected triple-resonance 3D experiments,as detailed in the SI Text. In the case of Cu2þ,Zn2þ-SOD, 116backbone amide resonances were assigned. In this latter case,we expect that nuclei particularly close to the Cu2þ ion are re-laxed very rapidly and thus evade detection. Details of resonanceassignment procedures are given in the SI Text.

103 resonances were resolved in the CP-HSQC, indicating thatthe CP-HSQC experiment is an efficient detection block for 15Nor 13C relaxation measurements. Thus, the 15N-1H CP-HSQCdipolar correlation module was combined with a 15N inver-sion-recovery block (30, 31), 15N spin-lock (32, 33), or additional13C-15N specific transfers (34) and 13C inversion-recovery (35)into a new set of extremely sensitive and resolved experiments.Pulse schemes are shown in Fig. S1.

Using this approach, we have measured 15N R1, 15N R1ρ aswell as 13CO R1 in the Cuþ,Zn2þ, Cu2þ,Zn2þ-SOD samples.We note that ultrafast MAS has an added advantage in relaxation

studies because the proton bath is well-decoupled, alleviatingpotential interfering effects of coherent contributions. This is par-ticularly relevant in the site-specific measurement of the 15N R1ρand 13CO R1, which are both particularly susceptible to residualeffects. It has been shown that these parameters can be accuratelymeasured under MAS by using MAS frequencies >40 kHz anda spin-lock nutation frequency >10 kHz (31, 33, 35).

Examples of the decay curves are shown in Fig. S2, andthe complete set of relaxation rates obtained for both Cuoxidation states is plotted in Fig. 2, and listed in Table S1. Inthe case of Cuþ,Zn2þ-SOD, 15N R1s were generally in the range0.01–0.05 s−1, with faster R1s generally in loop regions. In thecase of 13CO R1, most fell in the range 0.1–0.5 s−1. In the caseof 15N R1ρ, which is dependent upon slower frequencies ofmotion than R1, rates were 2–10 s−1 typically.

In the case of the paramagnetic Cu2þ,Zn2þ-SOD sample,substantial increases in 15N R1 and 13CO R1 were observed forcertain sites, whilst others remained consistent with the diamag-netic form. These differences are due to the PRE induced by theCu2þ ion by the Solomon mechanism (36), and will be strongestfor residues close to the metal center. The 15N R1ρ is affected bylarge diamagnetic contributions (notably influenced by fluctua-

10 9 8 7 61δ( H)/ppm

Cu2+

R79

I113

V14D76

R115

Q15

A123

V94

I17V119

130

120

110

K3A4

V5

A6

V7

L8

K9

G10

D11

G12

V14

Q15

G16

I17

I18

N19

F20

E21

Q22

V29

K30

V31W32

G33

S34

I35K36

G37

L38

T39

E40

G41

L42

H43

G44

F45

H46

V47

H48

E49

F50

G51

D52

N53

A55

G56

C57

T58A60

G61

H63

N65

L67

S68

K70

H71

G72

G73

K75

D76

E77

E78 R79

H80

G82

D83L84

G85

N86

V87

T88

A89

D90

K91

D92

G93

V94

A95

D96V97

S98

I99

E100

D101

S102

V103

S105

L106

G108

D109

H110

S111

I112

I113

G114

R115

T116

L117

V118

V119

H120

E121

K122

A123

D124

D125

L126

G127

K128

G129

N131

E132

S134

T137

G138

A140

G141

S142 R143

L144

A145

C146

G147

V148

I149

G150

I151

A152

Cu+

V14

Q15

I17A60

L67

D76 R79

D83L84

V94

I113

R115

V119

A123

K128B

100

15δ(

N)/

ppm

C

A

Fig. 1. Solid-state 15N-1H CP-HSQC spectra of SOD, recorded at 850 MHz 1HLarmor frequency and using 60 kHzMAS. (A) Cuþ,Zn2þ-SOD. (B) and (C) showclose-up regions of CP-HSQC spectra of Cuþ,Zn2þ-SOD and Cu2þ,Zn2þ-SOD,respectively.

0

0.2

0.4

0.6

0.8

15N

R1 (

s-1)

0

0.2

0.4

0.6

0.8

15N

R1 (

s-1)

Cu+

*** *β1 β2 β3 β4 β5 β6 β7 β8α1 α2

Cu2+

0

0.5

1

13C

O R

1 (s-1

)

0

0.5

1

13C

O R

1 (s-1

)

Cu+

Cu2+

B

F

A

0

5

10

15

15N

R1ρ

(s-1

)

0

5

10

15

15N

R1ρ

(s-1

)

Cu+

Cu2+D

C

E

0 20 40 60 80 100 120 140 153Residue number

Fig. 2. Relaxation rates for microcrystalline SOD. (A) 15N R1 for Cuþ, Zn-SOD,(B) 15N R1 for Cu2þ, Zn-SOD, (C) 15N R1ρ for Cuþ, Zn-SOD, (D) 15N R1ρ for Cu2þ,Zn-SOD, (E) 13CO R1 for Cuþ, Zn-SOD, (F) 13CO R1 for Cu2þ, Zn-SOD. Thesecondary structure is indicated above the figure, showing the positions ofthe Cu-coordinating histidine residues with asterisks. The areas in gray boxesare those for which the residue comes within 12 Å of the Cu2þ ion accordingto the single-crystal X-ray structure (PDB code: 1SOS).

11096 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1204515109 Knight et al.

tions in the μs-ms time scale). PREs are dominated by electronfluctuation in the ns range, and thus influence R1ρ less than R1.The 15N R1ρ in the Cu2þ and Cuþ oxidation states are thereforequite similar, as seen in Fig. 2.

Paramagnetic Relaxation Enhancements (PREs). In a paramagneticsystem, nuclear relaxation is made faster due to interactions be-tween the unpaired electron(s) of the paramagnetic center andthe nuclear spins (36). PREs can thus be measured in the casethat paramagnetic and diamagnetic forms of the sample are avail-able. PREs were obtained here by subtracting the diamagneticspin relaxation rates from those of the paramagnetic sample(Fig. S2 G and H), and translated into distances from the metalcenter using the well-known relationship (SI Text) and a Cu2þelectronic correlation time of 2.5 ns, based on literature values(36, 37). These 15N R1 and 13CO R1 PREs quantify distances asclose as 10 Å from the Cu ion and as far as 20 Å (Table S2). Con-tributions from intermolecular effects (38) were neglected. (thisapproximation was subsequently validated and shown to have noimpact on the analysis for this large protein).

In order to use these data as structural restraints, PREs wereincorporated into a structure determination protocol in combina-tion with 1H-1H distance restraints, chemical shift-derived dihe-dral angle restraints and ambiguous H-bond restraints. For eachdistance predicted from the PRE data we ascribed to the corre-sponding nucleus an upper distance limit to the Cu ion 3 Å great-er than the predicted distance, and a lower limit 3 Å less than thepredicted distance. In total, we made use of 85 13C PREs and 9015N PREs. In addition there were 25 1H-15N cross-peaks obser-vable in the diamagnetic but not paramagnetic form. In thesecases, it was assumed that this was due to broadening of the 1Hresonances beyond detection, and an upper distance limit to theCu ion of 10 Å was used, without lower limit (39). The remainingresidues corresponded to peaks for which accurate PRE determi-nation was not possible.

1H-1H distance restraints were measured through a 3DðHÞNHHRFDR experiment (13, 40), to which we applied the auto-matic assignment and structure calculation programs ATNOS/CANDID (41, 42), implemented in UNIO, to iteratively assignthe ðHÞNHHRFDR and obtain a fold. This process (13) uses cyclesof concerted peak assignment and structure calculation. In thefirst cycle (but not in subsequent cycles), the homologousCu2þ,Zn2þ-SOD crystal structure (PDB code 1SOS (29)) was usedto identify artifacts that might be mistaken for peaks in the spec-trum not consistent with any realistic assignment, and to ensurethat assignments consistent with the homologous structure werenot erroneously discarded. However, assignments of 1H-1H con-tact peaks were made independently of the homologous structurein all cycles. Details are found in the methods section.

A total of 257 unique and nontrivial proton-proton distancerestraints were obtained from the ðHÞNHHRFDR experiment per-formed on Cuþ,Zn2þ-SOD. Of these, 99 were between residuesboth in beta-strands, 63 between a beta-strand residue and a loopresidue, and 95 between residues both in a loop (Fig. S3 A and B).PRE-derived distance restraints were then added with no furtherassignment procedure and the structure re-calculated. In caseswhere the PRE-derived restraints were consistent violations(violations in more than half the members of the final ensemble),the distance bounds for these restraints were made wider andthe calculation repeated. The final bundles of 20 structures, withand without PRE-derived restraints, are shown in Fig. 3. Thestructure is composed of a well-defined beta-barrel, two shortalpha helices, and long loop regions containing the histidinescoordinating the Cuþ∕2þ and Zn2þ ions. No experimental con-straints other than PREs were used to determine the position ofthe Cuþ∕2þ ion in the molecule.

The qualitative improvement in the structure by the inclusionof PRE-derived restraints is immediately evident in this figure,

and is indicated quantitatively by the drop in backbone-heavyatom RMSD from 2.90 Å without PREs to 1.66 Å with PREs.In the beta-barrel, the RMSD dropped from 1.54 Åwithout PREsto 1.01 Å with PREs, whilse in the loops the RMSD dropped from3.42 Å to 1.91 Å (Fig. S3C). Indeed, the backbone structure ofthe loop regions in the vicinity of the Cu ion is well defined onlywhen PREs are used, despite quantitative PRE-derived restraintsbeing available only for regions at least 10 Å distal to the Cu ion.The Cu ion itself, with the inclusion of PRE-derived restraints,is positioned with an RMSD of 0.334 Åwhen the structure bundleis superimposed using all backbone heavy atoms. The use ofPRE-derived restraints has therefore enabled a much more de-tailed characterization of the active site of the enzyme than intheir absence. In particular, a lack of information around the Cuion has been overcome by multiple long-range restraints. We notethat, with only a single paramagnetic centre, in the absence of theshort-range RFDR-derived 1H-1H distance restraints, we did notobtain a well-defined structure.

In order to validate the implementation of PREs, Fig. 4 showsthe correlation between the PRE-derived distances from themicrocrystalline powder and the distances in the determinedstructure (a comparison with distances obtained from single-crys-tal X-ray diffraction on Cu2þ,Zn2þ-SOD is displayed in Fig. S3Dand E). The agreement is generally good, and the deviation be-tween the PRE-derived distances and NMR structure distances isnearly always less than 2.5 Å. This vindicates the upper and lowerdistance restraints of 3 Å above and below the predicted distancesrespectively in the structure refinement. Those for which an ad-justment to the upper and lower bounds was applied (on accountof being violations) are shown in gray in Fig. 4. These outliers aregenerally attributable to low signal intensity, or the peak beingpartly overlapped.

Making use of data acquired on the Cuþ and Cu2þ-boundstates assumes that the structural variations upon change in theoxidation state are minor, as established by several studies of glob-ular redox proteins in solution (43). In this case, comparison ofshifts (Fig. S4) shows that the two structures are very similar. PREsfor nuclei in sufficient proximity to the Cu ion to be sensitive to any

Fig. 3. The structure of SOD obtained from solid-state NMR, showing20-structure bundles overlaid by backbone heavy atoms. A and C structureobtained using automatically assigned distance restraints from a ðHÞNHHRFDR

spectrum and chemical-shift derived dihedral angle restraints, viewed fromtwo different angles. (B and D) The same restraints used as in A and C but in-cluding R1 PREs from 15N and 13CO. The Cu ion is shown as a red sphere.

Knight et al. PNAS ∣ July 10, 2012 ∣ vol. 109 ∣ no. 28 ∣ 11097

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

CHEM

ISTR

Y

major structural differences between the two states were not mea-sured, because they are rendered unobservable by large transversePREs, justifying the use of a point-dipole approximation for theparamagnetic effects, validating the method further.

In summary, we find that in concert RFDR and PRE-deriveddistance restraints (Fig. 4C) are able to determine the structurewith good precision (Fig. 3 B andD) and good accuracy (Fig. 4D),as shown by the comparison with the X-ray single crystal structure(PDB code: 1SOS).

Site-Specific Backbone Dynamics. Because the PRE method involvesmeasuring relaxation times in the diamagnetic form, a natural ex-tension of the protocol is to determine dynamics in concert withthe PRE-restrained structure.

Nuclear spin relaxation parameters determined by NMR spec-troscopy are powerful and widely used reporters of dynamicalinformation in studies of proteins, in which structural flexibilityis understood to be highly important (44). While solution-stateNMR methods to study dynamics are today very sophisticated,using up to 30 dynamical restraints per residue (45), extensivesite-specific relaxation studies of dynamics in solid proteins arerare (31, 33). Despite the technical obstacles, solid-state nuclearrelaxation times should be exquisitely sensitive to functionallyrelevant internal motions. These motions are generally preservedin crystals, while overall rotational diffusion, which dominatesrelaxation in the solution state, is absent. Because relaxation isdetermined by motion, not static disorder, the information pro-vided is inherently different from crystallographic B-factors.

In the case of the (diamagnetic) Cuþ,Zn2þ-SOD, we haveperformed a determination of the backbone dynamics from thesite-specific set of 15N R1 and R1ρ according to the recentlydemonstrated GAF approach (33). To calculate relaxation decays,we assume that 15N relaxation is dominated by its CSA and thedipolar coupling with the directly bonded proton (32, 33). To de-

scribe the dynamics, we have used the 1DGAF model (46, 47), forwhich we assume that the NH bond vector undergoes diffusivefluctuations in the peptide plane subject to a harmonic potential.In this model, dynamics are specified by the GAF diffusion time-scale and an order parameter related to the variance (equivalentlythe angular restriction) of the GAF motions. The relevant formu-lae can be found in the Methods section in the SI Text.

In Fig. 5, the order parameters and timescales determinedfor the 1D GAF model are plotted along the amino acid se-quence, and depicted on the dimeric structure of SOD. In thisrepresentation we explicitly assume that every copy of the mono-mers comprising each SOD dimer undergo identical motions.Our measurements inherently detect the average dynamics.Dynamics parameters are tablulated in Table S3, those obtainedwith different motional models are plotted in Fig. S5, and theform of the GAF correlation function is depicted in Fig. S6.

It is immediately seen that the solid-state NMR order para-meters are in qualitative agreement with general expectations ofprotein structure; high order parameters are seen in the beta-barrel, whereas lower order parameters are found in the loopregions. It is important to note that an order parameter is notaffected by static disorder, but is instead determined entirely bydynamics. The low order parameters observed in loops thereforeindicate that such regions are indeed flexible. The mean orderparameter obtained was 0.92 and varies between 0.97 and 0.65with a clear bias towards higher order parameters.

High order parameters were found for β-strands. Strands 4, 7,and 8, which constitute the core of the protein fold, have highervalues. The residues directly involved in metal-protein inter-actions (H46, H48, H120 coordinating Cuþ and H63, H71, H80

10 15 20 258

10

12

14

16

18

20

22

2415

N D

ista

nce

from

obs

erve

d P

RE

(Å

)

15N Distance from NMR bundle (Å)

10 15 20 258

10

12

14

16

18

20

22

24

13C

O D

ista

nce

from

obs

erve

d P

RE

(Å

)

13CO Distance from NMR bundle (Å)

A B

β1β2

β3

β4

β5

β6

β7

β8DC

Fig. 4. (A and B) Agreement between Cu-nucleus distance in the NMR struc-ture bundle and the PRE-derived distances for (A) 15N and (B) 13CO. (C) ThePRE-derived distances (blue lines) plotted on the X-ray crystal structure onwhich secondary structure elements are labeled. The Cu ion is shown inred. In all graphs, the dotted lines indicate the 2.5 Å deviation from perfectagreement. (D) The PRE-inclusive NMR structure (blue) overlaid with theX-ray single-crystal structure (PDB code 1SOS, magenta). In A and B the un-filled gray points are those identified as violations in the structure calcula-tion, and for which wider distance bounds were applied than the �3 Åbounds used generally.

0.6

0.8

1

S2

A

0

500

1,000

τ GA

F (

ns)B

C

0

20

40

60

τ eff (

ns)

0 20 40 60 80 100 120 140 153

*** *

Residue number

β1 β2 β3 β4 β5 β6 β7 β8α1 α2

D 1

0.6

0.8

0 ns

500 ns

E

Fig. 5. Internal dynamics of microcrystalline Cuþ, Zn-SOD as determinedby the 1D Gaussian axial fluctuation (1D GAF model). (A) Order parameters,(B) GAF timescales, (C) effective timescales of internal motion representingthe best mono-exponential approximation to the correlation function,equivalent to the Lipari-Szabo model-free effective correlation time,(D) order parameters projected onto the dimeric structure of SOD (PDB code:1SOS). Larger width and more red color denote a lower order parameter.(E) GAF timescales projected onto the dimeric structure of SOD. Larger widthand more turquoise color denote larger (slower) GAF timescales. In D and E,the Cu ion is shown as a gold sphere, and the Zn ion as a blue sphere.

11098 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1204515109 Knight et al.

and D83 coordinating Zn2þ) feature also generally very high(mostly >0.95) order parameters, indicating quite rigid bindingsites. A notable exception is H63, which is involved in Cu2þ andZn2þ coordination, but is not bound to Cuþ in the reduced state.Its relatively low order parameter (S2 ¼ 0.90, the lowest amongstthe metal-coordinating residues) may thus be the result of in-creased motility in the absence of side-chain metal binding.

If we consider the deviation between the NMR and the singlecrystal X-ray structures, in some regions where RFDR restraintsare sparse, the agreement can be poor, with the largest deviationsoccurring in the 22–24, 107–109, and 124–130 regions (Figs. 4Dand 5). The region 124–130 has generally high order parameters,indicating an absence of flexibility, despite the indication ofdisorder from the PRE analysis, and the variability amongst thevarious molecules in the asymmetric units (29), and between sev-eral X-ray crystal structures (48, 49) in this region. As for residues107–109, the order parameters are lower, suggesting that this re-gion is dynamic even in the crystal, also consistent with the highercrystallographic B-factors.

There is considerable variation in the timescale of motion,from 11 ns to 956 ns. The mean timescale of motion TGAF is222 ns. This corresponds to a “model-free” effective correlationtime TEFF of 25 ns, a timescale that is particularly difficult to mea-sure using solution-state NMR due to overall rotational diffusion,which occurs on a timescale of 25 ns for dimeric SOD (50). It ispossible, of course, that multiple timescales of motion occur, orthat collective motions are also present, methods for the evalua-tion of which exist (51). The evaluation of such models, however,is beyond the scope of the current study. In general, GAF time-scales are higher in beta-strands than in loops, implying greatercontributions of fast motions in loop regions relative to betastrands. In the beta-barrel, there is a clear separation of the time-scale of motion of the “outer” sheet (strands 1, 2, 3, and 6) andthose that form the “inner” sheet at the core of the molecule(strands 4, 5, 7, and 8). This observation may be indicative ofconcerted dynamics in each sheet, which has been suggested pre-viously (SOD solution) but not quantified due to the difficulty ofmeasuring this timescale by any other means. The high orderparameters show that such motions are of low amplitude, butare detectable here nonetheless, such is the sensitivity of thetechnique. However, these measurements are not unambiguouslydiagnostic of concerted motions, and extensive further experi-mentation would be required to validate such models.

The three Cuþ-coordinating residues have very consistentGAF timescales (H46: 491 ns, H48: 396 ns, H120: 435 ns), andall are well above (slower than) the mean timescale of 222 ns.H63, conversely, has a GAF timescale of 108 ns, which is one ofthe lowest (thus fastest) recorded. At the other extreme, D83,which is also involved in Zn coordination, shows the slowest mo-tion of all recorded, with a GAF timescale of 956 ns. As for theorder parameters, these timescales appear to correlate to thebinding pattern in the active site.

A model-free analysis of the backbone dynamics of Cuþ,Zn-SOD has also been published using solution NMR (50). Onthe whole, the solution-state order parameters are in good agree-ment with those obtained for our microcrystalline sample, suchthat the amplitude of dynamics may be similar between the twostates. A mean order parameter of 0.92 and standard deviationof 0.09 was reported in solution, which is in remarkably goodagreement with our solid-state analysis. However, the reportedtimescales of motion are different, but they are not comparablebecause they correspond to physically distinct models.

The range in which our GAF timescales fall is tens to hundredsof nanoseconds, similar to those obtained in a recent analysis ofmicrocrystalline GB1 (33). The R1ρ used in our solid-state studyare highly sensitive to this timescale, whereas these timescalesare essentially invisible to the methods applied in the solution-state study.

For example, consistently slower dynamics observed in theinner beta-strands here are not seen in the solution state. Thesolution-state analysis ascribes some fast motions on a low nano-second timescale, but such motions would likely be incorporatedinto our GAF timescales. Here we clearly show that solid-stateNMR relaxation can therefore provide dynamics information thatis unavailable from solution data.

ConclusionsIn conclusion, we have implemented a new approach that allowsdetermination of structure and dynamics of a large paramagneticmetalloprotein using solid-state NMR. We have shown that deu-teration, 100% amide reprotonation and ultrafast MAS allow therapid and sensitive acquisition of an extensive set of 15N and 13Crelaxation rates on SOD in its diamagnetic (Cuþ) and paramag-netic (Cu2þ) forms. From these data, PREs were obtained andemployed as structural constraints for the determination of theprotein structure. When added to 1H-1H distance restraints, theywere shown to yield a twofold improvement of the precision ofthe structure. Site-specific order parameters and timescales ofmotion were obtained by a Gaussian axial fluctuation (GAF) ana-lysis of the relaxation rates of the diamagnetic molecules, and in-terpreted in relation to backbone structure and metal binding.

The system we chose for this study is eminently suitable for thedevelopment of approaches for PRE determination and applica-tion in solid-state NMR on account of the ability to control theoxidation state of an intrinsically bound ion. Given that metallo-proteins are common in biochemistry, and include many large(or membrane-bound) proteins that have metal-containing cofac-tors that can be manipulated into diamagnetic and paramagneticforms, the ability to exploit quantitative PREs for structurerefinement is potentially very useful (52). The method developedhere is also directly compatible with the use of covalently at-tached paramagnetic tags (16, 19) in otherwise diamagnetic pro-teins. During the revision phase of this article, a paper hasappeared showing that PREs induced in the solid-state by attach-ing a paramagnetic chelator to multiple cysteine mutants can beused in addition to TALOS restraints to determine the fold of themodel protein GB1 (53). Moreover, the detail with which we havebeen able to characterize the internal dynamics is a clear indica-tion that solid-state NMR relaxation is a powerful techniquewhich we expect to make an impact in the study of systems notamenable to solution-state analysis. There exist many membrane-bound and fibrillar systems of medical or pharmaceutical interestin which such information may be very useful.

MethodsSample Preparation. The U-½2H; 13C; 15N� SOD samples were prepared essen-tially as previously described (13), but also metallated with Cu2þ. In order toobtain a sample of Cuþ,Zn2þ-SOD, the Cu2þ form was reduced by the addi-tion of 1.5 molar equivalents of iso-ascorbic acid under nitrogen atmosphereto avoid subsequent oxidation. The reduction was carried out prior to crystal-lization (also under nitrogen atmosphere) as in previous work.

NMR Spectroscopy. All relaxation measurements were performed using a19.9 T (1H Larmor frequency 850 MHz) wide-bore Bruker Avance III spectro-meter, and all experiments used 60 kHz MAS, with a sample temperatureof 13 °C. The pulse sequences used for measurement of relaxation ratesare shown in Fig. S1. The experimental PREs were obtained by subtractingthe diamagnetic R1 from the paramagnetic R1. The ðHÞNHHRFDR spectrum(τmix ¼ 3.3 ms) was recorded for Cuþ,Zn2þ-SOD only, and was automaticallyassigned according to the procedure described in (13). PRE-derived distancerestraints were then applied, and a final structure calculation performedusing the ðHÞNHHRFDR assignments as generated above.

For the fitting of dynamics models to relaxation data, for each site 15N R1

and 15N R1ρ were calculated according to well-known formulae for dipolarand CSA relaxation, reproduced in the SI Text (30, 33).

A full description of sample preparation, NMR spectroscopy, data analysisand structure calculation is provided as SI Text.

Knight et al. PNAS ∣ July 10, 2012 ∣ vol. 109 ∣ no. 28 ∣ 11099

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

CHEM

ISTR

Y

ACKNOWLEDGMENTS. Support from the Agence Nationale de la Recherche(ANR 08-BLAN-0035-01 and 10-BLAN-713-01), from Ente Cassa di Risparmiodi Firenze, from Egide (programme Galiée 22397RJ), from the Università Ita-

lo-francese (programma Galileo 11/12) and from Joint Research Activity andAccess to Research Infrastructures in the seventh Framework Programme ofthe European Community (EAST-NMR 228461, BioNMR 261863).

1. Bertini I, McGreevy KS, Parigi G, eds. (2012)NMRof Biomolecules: TowardsMechanisticSystems Biology (Wiley-VCH, Weinheim).

2. Renault M, Cukkemane A, Baldus M (2010) Solid-state NMR spectroscopy on complexbiomolecules. Angew Chem Int Ed Engl 49:8346–8357.

3. McDermott AE (2009) Structure and dynamics of membrane proteins by magic anglespinning solid-state NMR. Annu Rev Biophys 38:385–403.

4. Reif B (2012) Deuterated peptides and proteins: Structure and dynamics studies byMAS solid-state NMR. Methods Mol Biol 831:279–301.

5. Reif B, Jaroniec CP, Rienstra CM, Hohwy M, Griffin RG (2001) 1H-1H MAS correlationspectroscopy and distance measurements in a deuterated peptide. J Magn Reson151:320–327.

6. Paulson EK, et al. (2003) Sensitive high resolution inverse detection NMR spectroscopyof proteins in the solid state. J Am Chem Soc 125:15831–15836.

7. Chevelkov V, Rehbein K, Diehl A, Reif B (2006) Ultrahigh resolution in protonsolid-state NMR spectroscopy at high levels of deuteration. Angew Chem Int Ed Engl45:3878–3881.

8. Akbey U, et al. (2010) Optimum levels of exchangeable protons in perdeuteratedproteins for proton detection in MAS solid-state NMR spectroscopy. J Biomol NMR46:67–73.

9. Zhou DH, et al. (2007) Solid-state protein-structure determination with proton-detected triple-resonance 3D MAS NMR spectroscopy. Angew Chem Int Ed Engl46:8380–8383.

10. Linser R, Bardiaux B, Higman V, Fink U, Reif B (2011) Structure calculation from un-ambiguous long-range amide and methyl 1H-1H distance restraints for a microcrystal-line protein with MAS solid-state NMR spectroscopy. J Am Chem Soc 133:5905–5912.

11. Huber M, et al. (2011) A proton-detected 4D solid-state NMR experiment for proteinstructure determination. ChemPhysChem 12:915–918.

12. Lewandowski JR, et al. (2011) Enhanced resolution and coherence lifetimes in the so-lid-state NMR spectroscopy of perdeuterated proteins under ultrafast magic-anglespinning. J Phys Chem Lett 2:2205–2211.

13. Knight MJ, et al. (2011) Fast resonance assignment and fold determination of humansuperoxide dismutase by high-resolution proton-detected solid-state MAS NMRspectroscopy. Angew Chem Int Ed Engl 50:11697–11701.

14. Bertini I, Luchinat C, Parigi G, Pierattelli R (2005) NMR spectroscopy of paramagneticmetalloproteins. ChemBioChem 6:1536–1549.

15. Arnesano F, Banci L, Piccioli M (2005) NMR structures of paramagnetic metalloproteins.Q Rev Biophys 38:167–219.

16. Otting G (2010) Protein NMR using paramagnetic ions. Annu Rev Biophys 39:387–405.17. Clore G, Iwahara J (2009) Theory, practice, and applications of paramagnetic relaxa-

tion enhancement for the characterization of transient low-population states ofbiological macromolecules and their complexes. Chem Rev 109:4108–4139.

18. Su XC, Otting G (2010) Paramagnetic labelling of proteins and oligonucleotides forNMR. J Biomol NMR 46:101–112.

19. Keizers PM, Ubbink M (2011) Paramagnetic tagging for protein structure anddynamics analysis. Prog Nucl Magn Reson Spectrosc 58:88–96.

20. Balayssac S, Bertini I, Lelli M, Luchinat C, Maletta M (2007) Paramagnetic ions providestructural restraints in solid-state NMR of proteins. J Am Chem Soc 129:2218–2219.

21. Balayssac S, Bertini I, Bhaumik A, Lelli M, Luchinat C (2008) Paramagnetic shifts insolid-state NMR of proteins to elicit structural information. Proc Natl Acad Sci USA105:17284–17289.

22. Bertini I, et al. (2010) Ultrafast MAS solid-state NMR permits extensive 13C and 1Hdetection in paramagnetic metalloproteins. J Am Chem Soc 132:5558–5559.

23. Laage S, et al. (2009) Fast acquisition of multi-dimensional spectra in solid-state NMRenabled by ultra-fast MAS. J Magn Reson 196:133–141.

24. Wickramasinghe NP, et al. (2009) Nanomole-scale protein solid-state NMR by breakingintrinsic 1H T1 boundaries. Nat Methods 6:215–218.

25. Nadaud PS, Helmus JJ, Sengupta I, Jaroniec CP (2010) Rapid acquisition of multidimen-sional solid-state NMR spectra of proteins facilitated by covalently bound paramag-netic tags. J Am Chem Soc 132:9561–9563.

26. Nadaud PS, Helmus JJ, Hofer N, Jaroniec CP (2007) Long-range structural restraints inspin-labeled proteins probed by solid-state nuclear magnetic resonance spectroscopy.J Am Chem Soc 129:7502–7503.

27. Nadaud PS, Helmus JJ, Kall SL, Jaroniec CP (2009) Paramagnetic ions enable tuning ofnuclear relaxation rates and provide long-range structural restraints in solid-stateNMR of proteins. J Am Chem Soc 131:8108–8120.

28. McCord JM, Fridovich I (1969) Superoxide dismutase. An enzymic function for erythro-cuprein (hemocuprein). J Biol Chem 244:6049–6055.

29. Parge HE, Hallewell RA, Tainer JA (1992) Atomic structures of wild-type and thermo-stable mutant recombinant human Cu, Zn superoxide-dismutase. Proc Natl Acad SciUSA 89:6109–6113.

30. Giraud N, et al. (2005) Quantitative analysis of backbone dynamics in a crystalline pro-tein from nitrogen-15 spin-lattice relaxation. J Am Chem Soc 127:18190–18201.

31. Schanda P, Meier BH, Ernst M (2010) Quantitative analysis of protein backbonedynamics in microcrystalline ubiquitin by solid-state NMR spectroscopy. J Am ChemSoc 132:15957–15967.

32. Krushelnitsky A, Zinkevich T, Reichert D, Chevelkov V, Reif B (2010) Microsecond timescale mobility in a solid protein as studied by the 15N R1rho site-specific NMR relaxationrates. J Am Chem Soc 132:11850–11853.

33. Lewandowski JR, Sass HJ, Grzesiek S, Blackledge M, Emsley L (2011) Site-specific mea-surement of slow motions in proteins. J Am Chem Soc 133:16762–16765.

34. Baldus M, Petkova AT, Herzfeld J, Griffin RG (1998) Cross polarization in the tiltedframe: assignment and spectral simplification in heteronuclear spin systems.Mol Phys95:1197–1207.

35. Lewandowski JR, et al. (2010) Measurement of site-specific 13C spin-lattice relaxationin a crystalline protein. J Am Chem Soc 132:8252–8254.

36. Bertini I, Luchinat C, Parigi G (2001) Solution NMR of Paramagnetic Molecules: Appli-cations to Metallobiomolecules and Models (Elsevier, Amsterdam).

37. Bertini I, Luchinat C, Kowalewski J (1985) Nuclear-spin relaxation in paramagnetic(S > 1∕2) systems—A comparison of 2 new theoretical approaches. J Magn Reson62:235–241.

38. Nadaud PS, Sengupta I, Helmus JJ, Jaroniec CP (2011) Evaluation of the influence ofintermolecular electron-nucleus couplings and intrinsic metal binding sites on themeasurement of 15N longitudinal paramagnetic relaxation enhancements in proteinsby solid-state NMR. J Biomol NMR 51:293–302.

39. Bertini I, Felli IC, Luchinat C, Parigi G, Pierattelli R (2007) Towards a protocol for solu-tion structure determination of copper(II) proteins: The case of Cu(II)Zn(II) superoxidedismutase. Chembiochem 8:1422–1429.

40. Paulson EK, et al. (2003) High-sensitivity observation of dipolar exchange and NOEsbetween exchangeable protons in proteins by 3D solid-state NMR spectroscopy. JAm Chem Soc 125:14222–14223.

41. Herrmann T, Guntert P, Wuthrich K (2002) Protein NMR structure determination withautomated NOE assignment using the new software CANDID and the torsion angledynamics algorithm DYANA. J Mol Biol 319:209–227.

42. Herrmann T, Guntert P, Wuthrich K (2002) Protein NMR structure determination withautomated NOE-identification in the NOESY spectra using the new software ATNOS. JBiomol NMR 24:171–189.

43. Banci L, et al. (1994) The three-dimensional structure in solution of the paramagnetichigh-potential iron-sulfur protein I from Ectothiorhodospira halophila through nucle-ar magnetic resonance. Eur J Biochem 225:715–725.

44. Mittermaier A, Kay LE (2006) Review—new tools provide new insights in NMR studiesof protein dynamics. Science 312:224–228.

45. Salmon L, et al. (2011) Structure, dynamics, and kinetics of weak protein-protein com-plexes from NMR spin relaxationmeasurements of titrated solutions. Angew Chem IntEd Engl 50:3755–3759.

46. Bruschweiler R, Wright PE (1994) NMR order parameters of biomolecules—a new ana-lytical representation and application to the Gaussian axial fluctuation model. J AmChem Soc 116:8426–8427.

47. Bremi T, Bruschweiler R, Ernst RR (1997) A protocol for the interpretation of side-chaindynamics based on NMR relaxation: Application to phenylalanines in antamanide. JAm Chem Soc 119:4272–4284.

48. Strange RW, et al. (2003) The structure of holo and metal-deficient wild-type humanCu, Zn superoxide dismutase and its relevance to familial amyotrophic lateral sclerosis.J Mol Biol 328:877–891.

49. Strange RW, et al. (2006) Variablemetallation of human superoxide dismutase: Atomicresolution crystal structures of Cu-Zn, Zn-Zn and as-isolated wild-type enzymes. J MolBiol 356:1152–1162.

50. Banci L, et al. (2000) Backbone dynamics of human Cu, Zn superoxide dismutase and ofits monomeric F50E/G51E/E133Q mutant: The influence of dimerization on mobilityand function. Biochemistry 39:9108–9118.

51. Lewandowski JR, Sein J, Blackledge M, Emsley L (2010) Anisotropic collective motioncontributes to nuclear spin relaxation in crystalline proteins. J Am Chem Soc132:1246–1248.

52. Bertini I, Luchinat C, Parigi G (2011) Moving the frontiers in solution and solid-statebioNMR. Coord Chem Rev 255:649–663.

53. Sengupta I, Nadaud PS, Helmus JJ, Schwieters CD, Jaroniec CP (2012) Protein fold de-termined by paramagnetic magic-angle spinning solid-state NMR spectroscopy. NatChem 4:410–417.

11100 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1204515109 Knight et al.