Chimeric Avidin – NMR Structure and Dynamics of a 56 kDa Homotetrameric Thermostable Protein

Chimeric avidin (ChiAVD) is a product of rational protein engineering remarkably resistant to heat and harsh conditions. In quest of the fundamentals behind factors affecting stability we have elucidated the solution NMR spectroscopic structure of the biotin–bound form of ChiAVD and characterized the protein dynamics through 15N relaxation and hydrogen/deuterium (H/D) exchange of this and the biotin–free form. To surmount the challenges arising from the very large size of the protein for NMR spectroscopy, we took advantage of its high thermostability. Conventional triple resonance experiments for fully protonated proteins combined with methyl–detection optimized experiments acquired at 58°C were adequate for the structure determination of this 56 kDa protein. The model–free parameters derived from the 15N relaxation data reveal a remarkably rigid protein at 58°C in both the biotin–bound and the free forms. The H/D exchange experiments indicate a notable increase in hydrogen protection upon biotin binding.


Introduction
Chicken egg-white avidin and its bacterial analogue streptavidin from Streptomyces avidinii bind their natural ligand biotin with an extremely high affinity (dissociation constant K d ,10 215 M). In addition, they are remarkably stable against heat and harsh conditions such as proteolysis, denaturants and extremes of pH. These exceptional properties are widely employed in (strept)avidin biotechnological applications which typically rely on bridging a biotinylated target molecule binder to (strept)avidin [1,2] often in solution conditions very unnatural to proteins. Chemical and genetical engineering of avidin and streptavidin have further extended the diversity of the techniques [3].
Despite low sequence similarity, proteins of the avidin family have a remarkably similar molecular structure composed of four identical subunits (of 128 residues in avidin, Figure 1). The monomeric unit consists of an antiparallel eight-stranded b barrel each of which accommodates one biotin molecule at one end of the barrel. The four avidin subunits are arranged in a dimer of dimers [4]. This quaternary structure results in three distinct interfaces: the 1-4 interface is characterized by hydrophobic and polar interactions of such extent that the dimer can be considered as a single structural unit whereas the 1-3 interface is the weakest, composed only of three residues in avidin. The 1-2 interface is important for the tetramer stability and biotin binding affinity. There, a crucial tryptophan residue interacts with biotin bound in the adjacent subunit [4,5]. Excitingly, the structural similarity encompasses also the biotin-free forms of the proteins. The binding site is preformed in the free form, and no significant tertiary or quaternary structure rearrangements are needed in order to achieve the tight protein-ligand interaction. The stability of the free form is, however, markedly lower than that of the bound form. This is reflected in lower unfolding and oligomer dissociation temperatures [6].
ChiAVD(I117Y), hereafter referred to as ChiAVD, is a product of rational protein engineering [7,8]. It is a hyperthermostable hybrid of avidin and avidin-related protein 4 (AVR4) [9][10][11] obtained by replacing a 23-residue segment in avidin with the corresponding segment found in AVR4 and additionally introducing an Ile to Tyr, p-p,1-3 interface-stabilizing point mutation. ChiAVD is remarkably resistant to heat with a transition midpoint temperature, T m , for thermal unfolding of 111.1uC in the free and ,130uC in the bound form. In the presence of SDS, it dissociates into monomers only at ,95uC and 110uC in the free and bound form, respectively. ChiAVD is the most thermostable avidin studied to date. It is also resistant to harsh conditions such as extremes of pH and various organic solvents, even at high temperature. The biotin-binding properties of ChiAVD are comparable to those of AVR4 [7]. Since ChiAVD has successfully been applied in novel approaches in biotechnology [12,13], the understanding of the molecular properties of the protein is of our special interest. The structure of ChiAVD in the biotin-free form has recently been solved by X-ray crystallography [8]. In this study we present the solution NMR structure of ChiAVD in the biotin-bound form determined at 58uC. Remarkably, the structure of this 56 kDa protein was solved from a uniformly 13 C/ 15 N-labelled sample using triple resonance experiments designed for fully protonated samples together with a set of experiments optimized for detection of methyl-containing residues [14]. The work stands out as a cost-effective approach for the structure determination of large proteins via combination of an optimized measurement temperature with experiments for efficient assignment of residues serving long-range NOEs.
NMR is unique in providing access to residue-specific protein dynamics on a wide time scale [15]. Here we explore the backbone motions of the free and bound ChiAVD through 15 N relaxation and hydrogen/deuterium (H/D) exchange experiments. The order parameters, S 2 , derived from the relaxation data, reporting on the backbone nano-to picosecond motion, reveal a remarkably stable protein at 58uC in both the free and biotin-bound forms. The H/D exchange experiments indicate a notable increase in hydrogen protection upon biotin binding.

NMR spectroscopy, experiments and data analysis
Expression and purification of the 13 C/ 15 N-labelled ChiAVD as well as resonance assignment have been described previously [16]. Spectra for structure determination were acquired with a Varian INOVA 800 MHz spectrometer equipped with a cryogenic probehead, at 58uC. The spectra were processed with Vnmr 6.1C (Varian Inc.) and analysed with Sparky 3.110 (T. D. Goddard and D. G. Kneller, University of California, San Francisco).
Distance restraints were obtained from NOE peaks picked from 13 C, 1 H NOESY-HSQC, and 15 N, 1 H NOESY-HSQC spectra acquired from a sample dissolved in 92/8% H 2 O/D 2 O and 13 C, 1 H HSQC-NOESY acquired from a sample in 100% D 2 O, the latter being especially useful for NOEs arising from methyl groups. By inspection of available avidin crystal structures in the RCSB protein data bank (PDB), intersubunit (1-2 and 1-4) NOE peaks were identified and manually assigned. With Cyana [17] version 2.1, two hundred 1-4 dimer structures were calculated with automatic assignment carried out for the intrasubunit peaks, and manually assigned NOE peak lists for 1-4 intersubunit restraints. In addition to distance restraints, H-bond (from H/D exchange experiments), h/y restraints from TALOS [18], x 1 restraints deduced from J(C-C9) and J(C-N)-coupling spectra and 1 H-15 N residual dipolar couplings (RDCs) from spectra acquired from a sample in dilute solution of bicelles at 40uC (see next subheading) were used. Twenty structures with the lowest target function were selected. An initial tetrameric structure was built by duplicating a dimer structure and positioning the two dimers at an approximately correct orientation. From each starting tetramer a set of 10 structures was calculated with XPLOR-NIH [19] version 2.29 using all the available restraints. Of the resulting 200 structures, twenty lowest-energy structures were minimized with Amber 8 [20] and selected to represent the ChiAVD structure in solution. The coordinates of the final ensemble have been deposited to the RCSB Protein Data Bank (http://www.rcsb.org/pdb/) with the accession code 2mf6. Structure figures were created with UCSF Chimera [21]. 15 N longitudinal relaxation time (T 1 ) and transverse relaxation time (T 2 ) data were acquired with the following time points: 10, 60, 110, 330, 660, 900, 1100, 1500, 1800, 2600, 3500 ms (T 1 ) and 10, 30, 50, 70, 90, 110 ms (T 2 ). Duplicate spectra were acquired for estimation of uncertainties. Recycle delays were set to 3.1 s. R 1 and R 2 values were obtained by non-linear least-squares fitting of peak heights to a one-parameter exponential function using Curvefit (A. G. Palmer III, Columbia University). Uncertainties in the fitted parameters were obtained with Jackknife simulations (A. G. Palmer III, Columbia University). The { 1 H}-15 N heteronuclear nuclear Overhauser enhancement (hetNOE) values were determined as the peak intensity ratio observed in NOE spectra acquired with and without 1 H saturation. A recycle delay of 5.1 s was used in the hetNOE experiments. Proton saturation was applied for 5 s. An estimate of the error was obtained from the rms noise in the two spectra. Relaxation data were acquired at 58uC for the biotin-free form and at 40 and 58uC for the bound form.
Local correlation times were calculated with the program r2r1_tm (A. G. Palmer III, Columbia University) from trimmed R 2 /R 1 data [22]. The global isotropic correlation time was calculated as the mean of these residue-specific values. The relative moments of the inertia tensor determined from the bound ChiAVD structure are 1.00:0.79:0.73. The molecular diffusion tensor was determined from a subset of residues in secondary structure regions and with no large-amplitude motions [22]. For the 800 MHz, 58uC, biotin-bound protein data this set included 51 out of 96 residues having relaxation data. The best fit to the experimental R 1 and R 2 values is obtained with an axially symmetric (oblate) tensor. The components of the tensor are D H 0.13610 27 and D I 0.12610 27 s 21 , h-1.8u, Q 28.4u, resulting in a small degree of anisotropy (D I /D H ) of 0.90. The principal axes of the diffusion tensor are almost collinear with those of the inertia tensor, the maximum angle between the axes being ,2.5u. Model-free analysis of the relaxation data at the two magnetic fields was performed with the program Tensor [23], with a version of the program allowing the number of residues to be up to 1000, The structure, here represented with biotin-bound avidin (PDB identifier 2AVI), is a homotetramer composed of units of ,128 residues. Subunits are numbered according to [4]. Each subunit binds one biotin molecule, shown in stick representation. ChiAVD is a hybrid of avidin and AVR4, in which the segment highlighted in orange in avidin (residues 38-60, 23 residues) is replaced by the sequentially related segment found in AVR4 (residues 38-58, 21 residues). Also, Ile 117 of avidin, shown in red, is replaced by a tyrosine found in AVR4. doi:10.1371/journal.pone.0100564.g001 kindly provided by M. Blackledge, Institut de Biologie Structurale (Grenoble, France).
H/D exchange experiments were carried out by first lyophilizing a sample in D 2 O and then dissolving it to H 2 O. Increase in cross peak intensity was followed by measuring 1 H, 15 N HSQC spectra at 58uC, 800 MHz. The first time point was at approximately 15 minutes after dissolution. The last time point for biotin-free ChiAVD sample was at 96 hours, and for the bound form weeks after dissolution. Peak intensities were fitted to a three-parameter equation of the form I(t) = I(')+I(0)*(12exp (2k ex 6t)). Residue specific protection factors [24,25] were derived from the H/D exchange rates k ex using the spread sheet available from the Englander lab's website, http://hx2.med.upenn.edu/ download.html

RDC measurements
As ChiAVD is positively charged at the sample pH, we used bicelles, composed of 5% (w/V) DMPC/DHPC phospholipids at a molar ratio of 3:1, as the liquid crystal medium. Due to the instability of this liquid crystal medium at elevated temperatures we measured 1 H-15 N RDCs at 40uC. As no deuterium labelling was utilized, we employed a modified version of the MQ-HNCO-TROSY experiment that has been successfully used for measuring 1 H-15 N RDCs in the 558-residue Filamin A 16-21 fragment [26].
RDCs were applied as constraints in all four subunits.

Molecular dynamics simulations
The X-ray crystallographic structure of chimeric avidin [8] (PDB identifier 3MM0) was completed for the missing residues in L6,7 of chains E, F, G, and H (1-3 residues each) and L3,4 of chain G (Asn43) using Modeller 9v5 [27]. Biotin was placed into the binding pockets of ChiAVD with the help of wild-type avidin [4], PDB identifier 2AVI). Cocrystallized water molecules from within a 5 Å radius were included, and waters clashing with biotin were removed. Hydrogens were added using PDB2PQR 1.3.0. [28,29]. GAFF parameters were assigned for the biotin using the antechamber module, and Amber_99SB parameters [30] were assigned for the protein in the tleap module in Amber 10 [20]. The tetrameric protein with or without biotin was placed in a 75 Å686 Å690 Å box filled with TIP3P water molecules. 11 and 15 Cl 2 ions were added, resulting in a total of 52844 and 52973 atoms in the bound and ligand-free systems, respectively. Energy minimizations and molecular dynamics simulations were carried out in NAMD 2.6 [31]. Three 4000-step conjugate gradient minimizations were carried out for the ligand-bound complex: first with protein and ligand frozen, second with the ligand and Ca atoms frozen, and third without restraints. For the ligand-free system the second minimization was omitted. The systems were heated from 0 to 310 K in 31 ps and equilibrated for 2 ns in 310 K. The simulations were continued for 10 ns at three temperatures: 310, 333, and 523 K. The simulations were carried out in NPT conditions (1 atm) using the Berendsen thermostat and barostat. A 1-fs timestep was used in all simulations. The trajectory was superimposed by the Ca atoms using the RMSD Visualizer Tool plugin, and root mean square fluctuation (RMSF) was calculated using the ''measure rmsf'' command and a 1-ps step in VMD 1.9.1 [32].
The dissociation rate constant (k diss ) of fluorescently labelled ArcDiaTM BF560 (ArcDia, Turku, Finland) biotin was determined by fluorescence spectrometry essentially as described in [33]. The assay was performed at 50uC using a QuantaMasterTM Spectrofluorometer (Photon Technology International, Inc., Lawrenceville, NJ, USA) equipped with circulating water bath thermostat. The fluorescence probe was excited at 560 nm and emission was measured at 578 nm.
Determination of hydrodynamic radius was performed by dynamic light scattering (DLS) using Zetasizer Nano ZS (Malvern Instruments Ltd) in 50 mM NaH 2 PO 4 /Na 2 HPO 4 , 100 mM NaCl, pH 7 at 25uC. Six measurements were performed each consisting of 10 6 10 s measurement. Data was analysed using Zetasizer software v7.01 (Malvern Instruments Ltd) using ''General purpose'' model and volume distribution.
The transition midpoint (T m ) of the ChiAVD forms was measured by differential scanning calorimetry (DSC) using VPcapillary DSC (GE Healthcare, MicroCal) in 50 mM NaH 2 PO 4 / Na 2 HPO 4 , 100 mM NaCl, pH 7. The scanning was carried out from 20uC to 140uC at a rate of 120uC/h, using a 5 s filter period and a low feedback mode. Measurements were done using 13.8 mM protein in the presence or absence of 36 mM D-biotin (Fluka prod. no. 14400). Data analysis was made using the Origin 7 software (GE Healthcare, MicroCal). The T m s were determined using a Non-2-state fitting model.

Structure of biotin-bound ChiAVD
We have solved the first solution structure of a member of the avidin-family of proteins. The structure of this 56 kDa protein was solved without resort to the laborious, yet relatively expensive method in which perdeuteration is combined with selective methyl protonation [34]. The deciding factor was the thermostability of the protein -a raise of the measurement temperature to 58uC reduces the protein's overall rotational correlational time (,25 ns at room temperature) to half the value. Amide proton exchange rate with solvent significantly increases, however, along with temperature. To address this problem, a set of methyl proton detection experiments with high sensitivity and resolution [14] were employed in the assignment of methyl containing residues [16].
The homotetrameric symmetrical structure simplifies the NMR spectra of ChiAVD at the expense of losing potential NOE distance restraints at the very center of the protein structure (Arg114, Val115) where intra-and intersubunit correlations are indistinguishable. Intermolecular NOE cross peaks, here referring to those between ChiAVD subunits, were manually assigned from heteronuclear-edited NOE spectra. Identification of these peaks was based on an analysis of subunit interfaces of existing avidin crystal structures, mainly that of biotin-free ChiAVD [8]. To overcome the symmetry issue it would have been possible to create a tetrameric ChiAVD with differentially labelled subunits by using the methodology utilized earlier to produce monomeric avidin which tetramerizes upon addition of biotin [35]. Alternatively, dual chain avidin technology could have been applied [36].
Several NOE cross peaks from ChiAVD to biotin were observed in the NOE spectra. Despite numerous attempts with different concepts, we were unfortunately unable to obtain unambiguous chemical shift assignments for the bound biotin. We thus excluded these cross peaks from structure calculations. Additional restraints for structure calculation were obtained from chemical shifts, Jcoupling constants, H/D exchange experiments, and 1 D NH RDCs.
RDC measurements. To measure 1 D NH RDCs in ChiAVD, the newly modified MQ-HNCO-TROSY scheme (MQ-HNCO-TROSY+, Figure 2) was devised. The pulse scheme reduces losses associated to exchange broadening due to solvent exchange or J couplings. This experiment enables the determination of 1 H-15 N RDCs by measuring 1 J NH (in water) and 1 (J+D) NH (in liquid crystal medium) splittings in the 15 N dimension between two anti-TROSY components whose relative position and effective linewidth, with respect to the TROSY component, can be fine-tuned with two parameters k (0,k,1) and l (l.0). In the case of ChiAVD, the 1 (J+D) NH splittings were obtained by recording two MQ-HNCO-TROSY+ spectra in an interleaved manner with k = 0.5 and l = 0.5 i.e. the apparent splitting measured in 15 Figure 3.
In the case of ChiAVD, we were able to measure 93 RDCs. In structure calculation we used 58 RDC restraints omitting RDCs from the termini, more than half of RDCs originating from loops, as well as some RDC in secondary structure regions constantly giving large violations. Experimental RDCs, description of the alignment tensor and correlation plots are given in Figure S1 and Table S1. RDCs slightly improved the precision of the ensemble of ChiAVD structures, on average 0.06 Å for the backbone atoms of the ordered part of the sequence, residues 5-123. The RMSD to the biotin-free structure also marginally improved (0.03 Å for the same atoms).
An ensemble of biotin-bound ChiAVD structures with good precision was achieved ( Figure 4 and Table 1). The backbone and heavy atom RMSDs to the mean monomer structure are 0.27 and 0.69 Å , respectively, and in the tetramer 0.32 and 0.71 Å . The RMSDs for the monomer and the tetramer are of the same order indicating that the relative orientation of the monomers and the monomer itself are structurally equally well defined.

Comparison of the structures of biotin-bound and free ChiAVD
Despite the different experimental method and data collection conditions, namely the temperature, close structural similarity is evident for the two forms of ChiAVD. It is obvious that at 58uC the biotin-bound ChiAVD structure is still intact showing no indication of unfolding or dissociation into monomers. The b sheet regions of the biotin-bound ChiAVD solution structure superimpose well with those of the biotin-free form crystal structure [8] ( Figure 5A In loops the free versus bound RMSD values range from 0.8160.10 Å for the four-residue loop connecting b strands 1 and 2 (L1,2, residues 13-16 of the tetramer) to 1.6860.28 Å found for L4,5 (residues 54-62). The higher RMSDs in loops are the result of increased mobility as compared to the structured parts (see below the 15 N relaxation analysis). Notably, this is true also for L3,4 (residues [35][36][37][38][39][40][41][42][43][44][45][46] in which the backbone atom RMSD to the free form is 1.2960.13 Å . In the biotin-free form residues Pro41-Gly42 of this loop fold to a helix-like conformation (Q/y angles on average 264u/29u and 269u/222u) whereas in the biotin-bound form these residues are found in multiple conformations. None of the conformations precludes the formation of the stabilising intramonomeric salt bridge between side chains of Asp39 and Arg114 from b8 observed in the structures of biotin-free ChiAVD and AVR4 (Asp39-Arg112, [8,9]). This salt bridge is present in half of the structures of the ensemble.
Although NOEs to biotin were excluded from the structure calculations, residues in the biotin binding site are well defined ( Figure 5B). Some of the side chains of polar residues have, however, mutually different orientations. Besides the lack of biotin resonance assignments, here also the fact that no attempts were made to assign the side chain hydroxyl and amide group protons contributes to the differences observed.

Biotin-free ChiAVD in solution
The biotin-free ChiAVD sample deteriorates substantially faster than that of the bound form at the high temperature needed for sufficient NMR experiment sensitivity. Cross peaks become wider and their shape gets distorted although remaining at same positions with no additional peaks appearing over time. We suspect that over time in the prevalent solution conditions the protein molecules transiently interact to form higher molecular weight states. Assignment of the backbone resonances was however successfully conducted at 70uC [16]. Methyl group chemical shifts of free ChiAVD are listed in Table S2.
The structural similarity of the bound and free forms is evident from chemical shift comparison. The Dd Ca, Cb, N, H N (freebound) persuasively show that the chemical environments within monomers differ uniquely at residues located in the biotin binding site (see Figure S2 for Dd Cb). The largest shift differences, up to 3-5 ppm, are observed in L3,4 for residues Val37 and Ala38. The 1 H, 13 C chemical shifts of methyl groups located at the interfaces also match. A comparison of the 1 H, 13 C HSQC spectra of free (at 80uC) and bound (58uC) ChiAVD reveals that the methylcontaining residues at the 1-4 dimer interface exhibit comparable side chain chemical shifts (Dd( 13 C) ,0.27 ppm, Dd( 1 H) ,0.04 ppm) in the two forms. No significant methyl chemical shift changes are observed for Met96, Thr113, or Val115 at the 1-2 and 1-3 interfaces either. The N e1 -H e1 pair of Trp110 bridging monomers 1 and 2 shifts notably. This is, however, caused by direct interaction with biotin.
Because of the different acquisition temperatures, the 1 H line widths in the 1 H, 13 C HSQC spectra of free and bound ChiAVD cannot be directly compared. However, the measured line widths in both spectra can be divided in to three categories depending on their magnitude. We observe that each methyl resides in the same line width category in both protein states. We deduce that in the two forms the methyl groups at the interfaces have similar dynamical characteristics arising from similar chemical environments.
In all, the chemical shift and line width data indicate a close tertiary and quaternary structure similarity between the free and the bound form. It is thus justified to use the structure of the bound form in solution at 58uC in the analysis of the 15 N relaxation data of both forms. When extracting diffusion tensor parameters from the relaxation data this structure also gave lower x 2 target function values as compared to the crystal structure of the free form. Details of the diffusion tensor parameters and their derivation are given in Figure S3.
Model-free analysis R 1 , R 2 and heteronuclear Overhauser (hetNOE) 15 N relaxation data were recorded on 600 and 800 MHz spectrometers at 58uC for free and bound ChiAVD. For the bound form also relaxation data at 40uC on 800 MHz were recorded. These data as well as the average values are presented in Figure S4 and Table S3. The relaxation data were interpreted using the Model-free approach [40,41] and are presented for the 800 MHz data. Similar results were obtained from the analysis of the 600 MHz data.
Assuming isotropic diffusion, the overall rotational correlation times (t c ) are 13.060.4 ns for the biotin-bound ChiAVD, and 13.260.4 for the free form at 58uC. At 40uC, a t c of 18.060.6 ns is found for the bound form. By assuming a linear correlation between the ratio of solvent viscosity and temperature (g/T) and t c , at 25uC ChiAVD has an overall rotational correlation time of 25.4 ns. It is interesting to note, that this t c differs markedly from that estimated by the empirical formula, t c = 0.59986MW+ 0.1674, giving 33.8 ns. The Stokes-Einstein relation for the reorientation of a hard sphere gives an estimate of 21.0 ns for t c assuming a hydration radius of 3.2 Å . The observed rigidity of ChiAVD (see below) might lower the rotational correlation time towards the value predicted by the latter relation. This is consistent with the notably low t c s found for the highly rigid b-lactamases TEM-1 [42] and PSE-4 [43].
The Model-free parameters S 2 , t e and R ex are presented in Figure 6. The squared order parameter, S 2 , provides information  13 Ca regions. The 13 C carrier is placed in the middle of 13 C9 region (175 ppm) and rectangular 180u pulses are applied off-resonance for 13 Ca with phase modulation by V. Removal of 13 C9-13 Ca and 15 N-13 Ca coupling interactions during t 1 and t 2 , respectively, can be accomplished using either the SEDUCE-1 decoupling sequence [37] or three 180u 13 Ca rectangular pulses applied off-resonance with phase modulation by V. The delays used for coherence transfer are: D = 1/(4J NH ); TN = 1/(4J NC 9) = 12.5-16.6 ms; e = duration of gradient + recovery delay. Inset (A9) shows implementation to select the anti-TROSY component, which is downscaled by a factor of k (0, k,1) with respect to the TROSY component (see panel B). The phase cycling used is: Q 1 = x, 2x; Q 2 = x; Q 3 = 2x; Q 4 = 2x; y = 2x; Q rec. = x, 2x. Inset (A") shows pulse sequence implementation to select the anti-TROSY component which is scaled up by a factor of l (l.0) with respect to the TROSY component (see panel C). The phase cycling used is: Q 1 = x, 2x; Q 2 = x; Q 3 = 2x; Q 4 = 2x; y = x; Q rec. = x, 2x. Hence, for measuring 1 J NH (and 1 (J+D) NH ) couplings, the k and l values can be selected independently, for instance using k = 0; l = 1 yields two subspectra whose resonance frequencies differ by 2pJ NH i.e. 1 J NH couplings can be obtained directly from the frequency separation. Quadrature detection in the indirect 15 N (t 2 ) dimension, the 90u( 15 N) with the phase y is inverted simultaneously with the gradient G N to obtain echo/antiecho selection. The data processing is according to the sensitivity enhanced method [38]. The axial peaks are shifted to the edge of the spectrum by inverting Q 2 together with Q rec. in every second t 2 increment. Quadrature detection in the 13 C9 dimension is obtained by States-TPPI protocol applied to Q 1 [39]. Corresponding data are available for 66 residues. Large reduction (DS 2 .0.1) in the amplitude of motion upon biotin binding is observed for residues Thr19 (b2), Gly31 (b3), Ala38, Asn40 and Ile44 (L3,4), Gln53 (b4) and Arg100 (b7). The most impressive change is observed in L3,4 with DS 2 values of 0.13-0.19. Mobility in L3,4 is however still present in the biotin2bound form as Asn40 and Ile44 have S 2 values of ,0.7 and Gly42 as low as 0.2. Two residues, Ser73 and Phe120, show significant increase in mobility upon biotin binding. Two residues exhibit modest values of R ex (1.6#R ex #3.1 s 21 ) in the bound form, and seven residues in the free form, with slightly larger values (2.0#R ex #5.1 s 21 ). No correlation of R ex with H/D exchange data (see below) is detected. Instead, the presence of R ex for Val37, Arg100, Asp109 and Ala112 in the free form could possibly be associated with missing stabilizing sidechain interactions with biotin present in the bound form. The first of these is located in the lid-making loop whereas the three others, located in the protrusion of the b-barrel, are in contact with the 1-2-related monomer via Trp110.
In both formst e is relaxation active for several residues in loops and in a few loop-flanking residues. Largest contributions of t e to relaxation are observed for residues in L3,4 and L5,6, with the bound form values outweighing those of the free form.
Comparison of bound ChiAVD at two different temperatures. The rigidity of the protein is almost completely retained when increasing the temperature from 40 to 58uC. The average S 2 of 0.8660.03 observed at 58uC for all residues excluding the flexible residues at the termini has only slightly decreased from the 0.8860.03 observed at 40uC. If only residues in secondary structures are considered the same average S 2 value is found for the two temperatures. Interestingly, a per residue analysis shows that it is the biotin-binding region that becomes more mobile when temperature is raised. Largest decrease in S 2 are observed for residues Gly15, Ala36, Gly42, Ile44, Thr45, Ser73 and Arg114, all at the biotin-binding end of the b-barrel. Three of these are located in the lid-making loop L3,4. At the lower temperature the number of residues with a R ex contribution is larger. Unexpectedly, the majority of the residues exhibiting slow exchange are located in secondary structure regions.
The Model-free parameters derived from the relaxation data imply that ChiAVD is a remarkably rigid protein. A search of entries including order parameters deposited in the BioMagRes-Bank (http://www.bmrb.wisc.edu/) reveals that in the thirty data sets with an experimental temperature below 34uC the calculated average S 2 (omitting possible flexible terminal residues) ranges from 0.71 to 0.92. Only a few deposited data sets with an experimental temperature above 40uC are available. In these the average S 2 are 0.90 at 45uC for Trp repressor (BMRB entry 17041), 0.86 at 50uC for CtCBM11 (18389), 0.82-0.84 at 47-73uC for calmodulin-peptide complex (4970), and 0.81 at 44uC for azurin (6243). From the literature we find, in addition, average S 2 of 0.54 (50uC) for the B1 domain of Streptococcal protein G [44], 0.82 (45uC) for cardiac troponin C [45], 0.81-0.76 at 15-47uC for ubiquitin [46], and 0.88 (b strands only, 45uC) for OspA [47]. Considering that for most proteins studied to date the temperature dependency of S 2 is negative, ChiAVD with an average S 2 of 0.88 (40uC) and 0.86 (58uC) ranks among the most rigid proteins studied. The current understanding that among factors potentially increasing chemical and thermal stability is the reduction of conformational flexibility [48] is nicely in line with the fact that ChiAVD is extremely stable towards harsh conditions. It is, however important to note that the 15 N-1 H vector motions represent only a subset of the backbone dynamics, and acquisition of motional data of the 13 C9-13 C a vector would result in a more comprehensive perception of the overall backbone motions [49].

Conformational entropy
The contribution of conformational entropy changes to binding free energy can be derived from the S 2 values [50,51]. For the 66 residues considered the net loss in conformational entropy is 2122.9 J6mol 21 K 21 . This figure includes only the fast ps-ns time-scale motion of amide bond vectors of a subset of residues of the protein. It is however close to the experimentally determined DS of 2115.7 J6mol 21 K 21 found for the structurally related protein AVR4/5(C122S) [7] meaning that this type of motion makes a significant contribution to the entropic term of the Gibbs free energy of binding biotin to ChiAVD.

H/D exchange
H/D exchange studies were performed at 58uC for the free and biotin-bound ChiAVD. Data (partly qualitative) were obtained for 89 (free) and 104 (bound) residues (see Figure 7 and Figure S5 showing the curve fitting to the data). Twenty-five (free) and thirty-seven (bound) backbone amide hydrogens as well as Trp10 (free) and Trp10, 70, 97, and 110 (bound) side chain H e1 hydrogens are very efficiently buried and/or hydrogen bonded. Their 1 H, 15 N HSQC cross peak remain unperturbed for days after solvent exchange. Most of these are located at subunit interfaces: residues in b4-b7 at the 1-4 interface, residue 114 and residues 115-116 in b8 at the 1-2 and 1-3 interfaces, respectively. Trp70, 97 and 110 are essential in providing hydrophobic interactions to biotin. Trp10 is located at the opposite end of the b-barrel and forms, in both free and bound ChiAVD, a hydrogen-bond with its H e1 to Leu6 carbonyl oxygen. An estimate for the lower limit of the highest rate constant was calculated from peak intensities in the first spectrum after solvent addition [52]. Residues which had exchanged before the first time point have a rate constant of .2.9610 23 s 21 (represented with a protection factor P, expressed as log P, of 2.5 in Figure 7). Forty and thirty-nine residues belong to this group in the free and bound ChiAVD, respectively. For residues which had exchanged before the second time point the approximate rate constant is 1.7610 23 ,k ex ,2.9610 23 s 21 (log P 2.8). Six (free) and one (bound) residues belong to this group. Backbone amide protons of fifteen (free) and twenty-five (bound) residues exchanged within the experimental time. On average, the rates for free ChiAVD are 0.28610 23 s 21 faster than for the bound form.
It is evident from these data that the bound form is considerably more stable in terms of H N protection arising from hydrogen bonding and/or burial. The free and bound ChiAVD differ in H N protection at the side facing solvent (residues in b1-b3) as well as in L7,8 and the following 3 10 helix ( Figure 7B). Similar results have been obtained for streptavidin by H/D exchange and mass spectrometric studies [53]. In both proteins b1-b3 at the solventexposed face of the structures, as well as the 3 10 helix at the 1-2 interface show a reduction in exchange upon biotin binding. At the 1-4 interface a larger number of amide hydrogens are labile in the biotin-free form of streptavidin as compared to that of ChiAVD: In addition to residues in b5-b6 and b8 protected in both proteins, in ChiAVD solvent protection covers also residues in b7. H/D exchange and infrared spectroscopic studies with avidin [54] also indicated a reduction in the proportion of exchangeable hydrogens and reduction in the fast exchange kinetic constant upon biotin binding. ChiAVD has a biotin binding affinity similar to that of AVR4 [7], which is higher than that observed for streptavidin [55]. This mutual difference might partially be accounted for by the differences observed in hydrogen protection. In ChiAVD the more extensive hydrogen bond network in the free form would imply a smaller loss in entropy upon biotin binding which would have a favourable effect on the thermodynamics of the reaction and thus increase affinity. A further entropic benefit for ChiAVD might result from L3,4 retaining at least partially its mobility (again a smaller loss in entropy) as opposed to streptavidin in which a reduction of exchange is observed for the entire loop.

MD analysis
Molecular dynamics simulations were carried out at three temperatures for 10 ns. Movement of the protein chain is visualized by plotting the root mean square fluctuation over time (RMSF; Figure 8). A clear correlation between secondary structure and RMSF is observed: all eight b strands are found to show much lower fluctuation as compared to the loops connecting them. L3,4 and L6,7 are the most mobile. Biotin stabilizes most loops in all the temperatures tested, but the stabilizing effect is rather modest. Using an average of five residues to calculate the DRMSF, the highest degree of stabilization is observed in sequence stretches Thr35-Asp39 for 310 K (DRMSF: 232%) and 333 K (DRMSF: 232%) and Trp70-Phe74 for 523 K (DRMSF: 231%). RMSF is found to correlate with the temperature used. Movement of Ala22 in L2,3 increases in the presence of biotin.
The results of the molecular dynamics simulation agree remarkably well with the order parameter data, with the exception of L6,7. The simulation nicely exposes the small differences observed in the experimental data between the free and the bound protein form, namely the larger number of mobile residues in L3,4 of the free form and the loop's overall higher amplitude of motion. Also the lower stability of the 3 10 helix in the free form is evident. However, the order parameters show no indication of high mobility for L6,7. In fact the simulation data indicates higher mobility for twelve residues, encompassing half of the b strands flanking the three-residue L6,7.

ChiAVD mutants G42A and G42F
Apart from the termini, three regions stand out from the average motional regime in ChiAVD: L3,4, L4,5 and L5,6. According to the S 2 and t e values residue Gly42 in L3,4 is the most mobile residue. In quest of further stability, we mutated Gly42 to alanine and phenylalanine. The GlyRAla mutation represents the simplest attempt to create rigidity with a bulkier side chain whereas the objective of the GlyRPhe mutation was to create a stabilizing p-p aromatic interaction with Phe72 in the structurally opposite L5,6. (Strept)avidins have extensively been modified by mutagenesis [56], but to our knowledge the outcome of mutating Gly42 has not yet been described. Intriguingly, Chivers et al. [57] applied point mutations S52G and R53D to streptavidin (streptavidin Ser52 equals to ChiAVD Gly42 in sequence) and obtained streptavidin with both decreased dissociation and association rates, but also with increased thermal stability.
The mutants expressed efficiently (data not shown), folded properly to tetramers and had physical properties very similar to ChiAVD. A hydrodynamic radius of 3.0460.75 nm was measured by DLS for ChiAVD-G42A, 3.0960.70 nm for ChiAVD-G42F, and 3.1860.69 nm for ChiAVD. The main peak consisted of 99.7-100% of the volume-adjusted intensity. The obtained hydrodynamic size corresponds to globular protein with molecular weights of 45.4 kDa (ChiAVD-G42A), 47.2 kDa (ChiAVD-G42F) and 50.4 kDa (ChiAVD), which is very close to the theoretical size of the tetramer.
Mutation of Gly42 had no notable influence on the thermostability. However, we observed a slightly lower biotin dissociation rate for the mutants. At 50uC ChiAVD-G42A showed a slightly lower fluorescently labelled biotin dissociation rate than ChiAVD-G42F or ChiAVD: (4.6460.33)610 25 s 21 as compared to (5.8060.40)

Conclusions
We have shown that at 58uC biotin-bound ChiAVD maintains its compact, well-defined structure. With an average order   parameter S 2 of 0.86, it is one of the most rigid proteins studied to date. Rigidity plays a significant role in promoting stability. The extent of the subunit interfaces, reflected in the high number of very slowly exchanging amide hydrogens also contributes to the stability. The biotin-free forms of the avidin family of proteins are less stable than the bound forms. For ChiAVD the difference in stability can be (at least partially) ascribed to the smaller extent of H-bonding/burial of amide hydrogens and higher nano-picosecond flexibility (S 2 ) at the biotin-binding end of the b-barrel as compared to the bound form. for the oblate tensor (1 st minimun) D | = D xx and D H = D zz = D yy , for the prolate tensor (2 nd minimum) D | = D zz and D H = D xx = D yy . The angles a, b and c describe the orientatation of the diffusion tensor in the structure frame. For axially symmetric tensor, the polar angles Q and h are given. (TIF) Figure S4 15 N R 1 , R 2 , and hetNOE data of biotin-bound and free ChiAVD. The 800 MHz data are represented with red (58uC) and blue (40uC) dots, and those of 600 MHz (58uC) with black dots. Trp N e1 data at 800 MHz, 58uC are shown with green dots. Secondary structure regions are highlighted with b strands in blue and 3 10 helices in red. The depicted secondary structure of the free form is that present in the crystal structure [1]. Average errors are 0.04 (800 MHz, 58uC), 0.03 (800 MHz, 40uC), and 0.03 (600 MHz, 58uC) s 21 for R 1 , 0.40/0.70/0.31 s 21 for R 2 and 0.04/0.04/0.06 for hetNOE in the bound form and 0.03/0.03 (800/600 MHz, R 1 ), 0.31/0.48 (R 2 ) and 0.05/0.04 (hetNOE) for the free form. The amino acid sequence has a gap (highlighted in grey): His54 is followed by Lys57. The first three residues Gln(23), Thr (22) and Val (21)