Temperature effects on the structure and dynamics of liquid dimethyl sulfoxide: A molecular dynamics study

Temperature effects on structure and dynamicsof the psychrophilic protease subtilisin S41and its thermostable mutants in solution

Ronny Martinez 1, Ulrich Schwaneberg1,2 andDanilo Roccatano1,3

1Jacobs University Bremen, Campus Ring 1, D-28759 Bremen, Germanyand 2RWTH Aachen University, Worringerweg 1, D-52074 Aachen,Germany

3To whom correspondence should be addressed.E-mail: [email protected]

Received July 11, 2010; revised March 6, 2011;accepted March 8, 2011

Edited by Kam-Bo Wong

The psychrophilic protease subtilisin S41 from theAntarctic bacillus TA41, and two variants with twoand seven amino acid substitutions were studied usingmolecular dynamics simulation at 283 and 363 K. Theanalysis of protein dynamics revealed that the averageglobal flexibility of both variants was slightly higherthan wild type at both 283 and 363 K. Essentialdynamics analysis evidenced that the most relevant col-lective motions, especially at 363 K, differ in distri-bution and intensity for each protein variant. At hightemperature and for the thermo labile wild type, anamplification of a subset of the low-temperature largestcollective motions was observed. On the other hand,the two thermostable variants showed a rather differ-ent pattern of essential motions at 363 K from those at283 K. These results support the hypothesis that theintroduced amino acid substitutions, rather thanimproving the global stability of the variants byincreasing its rigidity, lead to a change on the princi-pal fluxional modes allowing the protein to explore adifferent subset of conformations. A better understand-ing of this process can open alternative strategies toincrease the enzyme stability in addition to increasingthe rigidity of the protein scaffold.Keywords: enzyme flexibility/essential modes/proteindynamics/protein stability/thermal inactivation

Introduction

Extremophile organisms express proteins that are designedto operate under extreme conditions, from the icing arcticsea poles to the scorching volcanic vents in the deepoceans (Thomas and Dieckmann, 2002; Jenney Jr andAdams, 2008). These examples of adaptation are of greatinterest for academic research and for biotechnologicalapplication (Herbert, 1992; Demirjian et al., 2001; van den

Burg, 2003). Among these proteins, psychrophilic enzymesare important for sustainable biotechnological processes asa replacement of the efficient, but more energy demand-ing, mesophilic or thermophilic variants (Vieille andZeikus, 2001; Gupta et al., 2002; Maurer, 2004).Unfortunately, low thermal stability is a common draw-back of cold-adapted enzymes preventing their efficientindustrial exploitation. Differences in stability amongextremophile proteins have been widely investigated(Vieille and Zeikus, 1996, 2001). In all these studies, thelack of a consensus on the mechanisms responsible forthe different stability in extremophiles suggests that thereare different design principles governing thermalresistance.

A common strategy reported to improve protein thermalstability consists in the reinforcement of the overall structuralrigidity of the protein by increasing the number of disulfidebonds, intra-molecular salt bridges and trimming the lengthof loop regions (Matthews et al., 1987; Storch et al., 1999;Zhang et al., 2006). As a complement to these structure-oriented approaches, in the last few years, new ideasemerged concerning the connection between stability andprotein dynamics. In particular, it was proposed that proteinfluctuations can provide a mechanism of thermal stability inaddition to the structural modifications (D’Amico et al.,2002; Feller, 2007). This hypothesis has been strongly sup-ported by the recent progress in experimental techniquesbased on neutron diffraction methods (Tehei and Zaccai,2007), nuclear magnetic resonance (NMR) (Boehr et al.,2006; Henzler-Wildman and Kern, 2007) and in theoreticalapproaches based on computer simulations on proteindynamics in solution (Adcock and McCammon, 2006; vanGunsteren et al., 2006). In particular, the new NMR methodscan provide detailed new information on the dynamics ofproteins at different levels of time scales (Henzler-Wildmanand Kern, 2007; Mittermaier and Kay, 2009). These NMRstudies evidenced the importance of protein dynamics on thepicosecond and nanosecond time scale for the stability andactivity of extremophilic enzymes (Boehr et al., 2006;Krishnamurthy et al., 2009). Among computational tech-niques, molecular dynamics (MD) simulation is used toprovide detailed atomic models of the protein stability anddynamics (Daggett, 2006; Scheraga et al., 2007., 2008; vanGunsteren et al., 2008). Many examples in the literatureshow that the intertwine combination of MD simulations andexperimental data provides a powerful approach to unravelthe detail of protein dynamics (see, for example, Morin andGagne, 2009).

Several MD studies have been performed to study andcompare the thermal behavior of extremophilic enzymes

# The Author 2011. Published by Oxford University Press. All rights reserved.

For Permissions, please e-mail: [email protected]

533

Protein Engineering, Design & Selection vol. 24 no. 7 pp. 533–544, 2011Published online April 6, 2011 doi:10.1093/protein/gzr014

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

(Colombo and Merz, 1999; Grottesi et al., 2002; Polyanskyet al., 2004; Melchionna et al., 2006; Spiwok et al., 2007;Papaleo et al., 2008; Kundu and Roy, 2009; Merkley et al.,2010; Priyakumar et al., 2010). Some of them report thatthermostable enzymes have larger overall flexibility thantheir mesophile homologous or cold-adapted counterparts(Colombo and Merz, 1999; Grottesi et al., 2002; Papaleoet al., 2008), others state that the difference in flexibility pro-files were not as remarkable as often supposed (Spiwoket al., 2007; Merkley et al., 2010). Most of these studies havebeen conducted using short time scale (,5 ns) simulationsand this may be one of the reasons for the different con-clusions. In fact, the long time scale MD simulations are nor-mally necessary to obtain quantitative information to becomparable with experimental data (see for example vanGunsteren et al., 2008; Bos and Pleiss, 2009). In this work,the modification induced by the temperature and amino acidsubstitutions on the structural and dynamics properties of amodel psychrophilic protein was explored using long timescale MD simulations (50 ns). The model proteins used forthis study were the Subtilisin S41 (named WT) and two var-iants with enhanced thermal stability (Wintrode and Arnold,2001). Subtilisin S41 (Fig. 1a) is a cold-adapted protease iso-lated from Antarctic bacillus TA41 (Davail et al., 1994). Thegene of the enzyme encodes for a 419 amino acidpro-enzyme but only 309 are present in the mature protease.The enzyme shares most of its properties with mesophilicsubtilisins (structure of the precursor, 52% amino acidsequence identity, alkaline pH optimum, broad substratespecificity, Ca2þ binding) but it is characterized by a higherspecific activity on macromolecular substrates, a shift of theoptimum of activity toward low temperatures and a lowthermal stability (Davail et al., 1994). Successful attempts togenerate thermostable variants of the enzyme by directedevolution have been reported by Miyazaki and Arnold(Miyazaki and Arnold, 1999; Miyazaki et al., 2000; Wintrodeand Arnold, 2001). The variant 1-14A7 (named MUT2 inthis work) has two amino acid substitutions: Lys211Pro andArg212Ala (Fig. 1b), with a half-life at 333 K 10 timeslonger (84 min) than that of the wild-type S41 (8 min). Asecond variant, 3-2G7 (named MUT7, Fig. 1b), with fiveadditional amino acid substitutions (Ser145Ile, Ser175Thr;Lys221Glu, Asn291Ile and Ser295Thr), shows a half-life at333 K 500 times longer than that of WT. In addition, none ofthe mutations in both variants were directly related to theactive site conserved residues (Asp34, His71, Asn168 andSer249) but an increase of their enzymatic activity at bothlow and high temperatures was observed (Miyazaki et al.,2000). This result suggests that improvement in proteinthermal stability does not necessarily decrease the activity atlower temperatures, as it normally occurs in naturallyevolved thermophilic enzymes. Recently, the crystal structureof wild-type Subtilisin S41 has been solved (Almog et al.,2009) giving the possibility to study the molecular foun-dations of the improved thermal stability of both the mutants.

From the crystallographic structure the main difference ofSubtilisin S41 from homologous mesophilic ones is the pres-ence of more extended loops regions. Although, experimentalstudies of this protein using NMR methods are not yetreported in literature, it is common opinion that psychrophilicenzymes have evolved toward a high conformational

flexibility that increases the catalytic efficiency at low temp-eratures but it reduces the thermal stability. In this paper, wewill verify this hypothesis by comparing the structural anddynamic properties of WT, MUT2 and MUT7. In particular,In the Results section, the structural and dynamics propertiesobtained from the simulations at 283 and 363 K of the wildtype and the two variants are reported. The dynamics proper-ties of the enzyme are dissected in their principal fluctuationmodes using essential dynamics methods. A comparison ofthese modes at the different temperatures was performed toassess their degrees of similarity. In addition, the results ofsimulations at 450 K used to induce the unfolding and quali-tatively compare the relative stability of the models are alsoreported. In the Discussion section, the results from the

Fig. 1. Cartoon representations of the crystal structure of Subtilisin S41indicating (a) the active site residues (D34, H71, N168 and S249), the ionbinding sites (Ca-1, Ca-2, Ca-3, Ca-4 and Na-5) and the extended loops.The most flexible loop groups are identified in dark grey. (b) The position ofthe amino acid substitutions in MUT2 (K211P, R212A) and the additionalsubstitutions leading to MUT7 (S145I, S175T, K221E, N291I, S295T). Theactive site residues (D34, H71 and S249) and the structural ions arerepresented in surface mode.

R.Martinez et al.

534

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

simulations are further examined by comparing with the lit-erature. Finally, in the Conclusion section a summary andoutline of the paper is presented.

Materials and methods

Starting structuresThe starting structure for the MD simulations of SubtilisinS41 was the crystal structure of the protein recently resolvedat 1.4 A resolution (Almog et al., 2009) and deposited withthe PDB code 2GKO. The four structural calcium ions (indi-cated as Ca-1, Ca-2, Ca-3 and Ca-4, respectively), thesodium ion (indicated as Na-5) and water molecules presentin the crystal structure were retained while the other ligands(phenylmethylsulfonyl fluoride) were removed. The mutantsMUT2 and MUT7 were generated from the WT structureusing the mutate tool of the program Swiss PDB Viewer(Guex and Peitsch, 1997) by selecting the most favorablerotamers for each mutated residue. The optimized potentialsfor liquid simulations (Jorgensen and Tirado-Rives, 1988)force field was used for all simulations. TIP4P was adoptedas the water model (Jorgensen et al., 1983).

The simulation boxes were built by centering each proteinmolecule in a truncated octahedral periodic box. The size ofeach box was set to have a minimal distance between theprotein atoms and any side of the box of 0.75 nm. The finalvolume of each box was equal to 320 nm3. The proteins weresolvated by stacking an equilibrated box of 216 water mol-ecules to completely fill the simulation box. All solvent mol-ecules with any atom within 0.15 nm from solute atoms wereremoved. The resulting total charge of the box was 22 forWT, 24 for MUT2 and 26 for MUT7, respectively. Sodiumcounter ions were added by replacing the water molecules atthe most negative electrostatic potential to balance the totalcharge of the box. The size of the simulated systems isreported in Table I.

Simulation protocolThe potential energy of the systems was minimized using thesteepest descendent method for at least 500 steps to relax theprotein structure and remove clashes caused by the water boxgeneration. The three proteins were studied at the three refer-ence temperatures of 283, 363 and 450 K. The large tempera-ture gaps were chosen to enhance the temperature effect onthe protein dynamics in the simulation time scale. The simu-lations at 450 K were used to induce the protein unfoldingand qualitatively study the relative stability of the differentproteins. All simulations were performed by keeping thetemperature at the reference values by weak coupling (coup-ling constant t ¼ 0.1 ps) with an external bath (Berendsenet al., 1984). The protein and the rest of the system were

coupled to two different temperature baths. The pressure ofthe system was kept constant at 1 bar by using theBerendsen’s barostat (Berendsen et al., 1984) with a couplingconstant of tp¼ 0.5 ps. The linear constraint solver algorithmwas used to constrain all bond lengths (Hess et al., 1997).For the water molecules, the SETTLE algorithm was used(Miyamoto and Kollman, 1992). A dielectric permittivity,1r¼ 1 and a time step of 2 fs were used. The non-bondedinteractions (electrostatic and Leonard-Jones) were calculatedusing the particle mesh Ewalds method (Essmann et al.,1995). For the calculation of the long-range interactions, agrid spacing of 0.12 nm combined with a fourth-orderB-spline interpolation were used to compute the potentialand forces between grid points. A non-bonded pair-list cutoffof 0.9 nm was used and the pair-list was updated every fivetime steps. The MD simulations were started by assigning toall atoms initial velocities obtained from a Maxwellian distri-bution at the given temperatures. The water distribution andthe box density were equilibrated for 100 ps using positionrestraints on the protein heavy atoms. After the equilibration,the position restraints on the protein were removed and thesystem was gradually heated from 50 K to the simulationtemperature during 50 ps of simulation. The following pro-duction runs were 50 ns long.

Analysis of the simulationsEach simulation was monitored using standard structuralanalysis. Total and per residue backbone (C, Ca and N)root-mean-square deviations (RMSD) were calculated withrespect to the reference crystal structure of WT S41Subtilisin. The secondary structure of the protein was ana-lyzed using the DSSP criteria (Kabsch and Sander, 1983).The presence in the binding site of the four calcium and onesodium ions was monitored by analyzing time series ofselected distances (for Asp and Glu the minimum distancewith the two side chain carboxylic oxygen and for Leu182the carbonyl oxygen atom were used, respectively) betweenCa-1 and Asp286, Ca-2 and Asp216, Ca-3 and Glu49, Ca-4and Asp114 and Na-5 and Leu182, respectively. Total andlocal root-mean-square fluctuations (RMSF) of heavy andbackbone atoms were used to quantify the protein flexibility.The average RMSF(i) for the ith residue was calculated byaveraging the RMSF of atoms belonging to the same residue.In addition, the local RMSF (LRMSF) of loop regions wascalculated from the RMSF(i) as:

LRMSF ¼ 1

N2 � N1

XN2

i¼N1

RMSFðiÞ

where N2 and N1 are the residue numbers (Polyansky et al.,2004). The differences of LRMSF were used to characterizethe variation of flexibility for the different variants and at283 and 363 K. The significance of the mutation- andtemperature-induced differences observed in the dynamic be-havior of loop regions was evaluated using the t-student test.The analysis of RMSD and RMSF per residue was per-formed on the last 20 ns of the trajectories

Cluster analysisThe study the protein backbone conformations sampledevery 10 ps during the MD simulations was conducted using

Table I. Summary of simulations

WT MUT2 MUT7

Box volume (nm3) 320 320 320Number of atoms 41 790 41 326 41 360Solvent molecules 9373 9262 9267Counter ions 2 4 6

Temperature effects on structure and dynamics of the psychrophilic protease subtilisin S41

535

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

the groningen molecular simulation program package clusteralgorithm (Daura et al., 1999). For each structure, aleast-square translational and rotational fit was performedusing backbone atoms, and the RMSD for the backboneatoms of the proteins were calculated.A cutoff of 0.07 at283 K and 0.10 nm at 363 K were used.

Essential dynamics analysis (EDA)This method is based on the principal component analysis ofMD trajectories and allows separating the large (nonlinear)amplitude correlated motions present in the different proteinsat different temperatures from the quasi-harmonic modeassociated with thermal noise fluctuations (Garcia, 1992;Amadei et al., 1993). The analysis was performed on proteinbackbone atoms obtained from the last 20 ns of the MDsimulation trajectories of each protein at both 283 and363 K. The details of the procedure for the analysis aredescribed elsewhere (Amadei et al., 1993). The total numberof eigenvalues and eigenvectors considered for the analysiswas 2781. The WT crystal structure was used as referencefor the analysis of all the simulations. The resulting eigen-vectors and eigenvalues define the direction and the ampli-tude, respectively, of specific orthogonal modes offluctuation of the molecular system. The comparison ofeigenvectors obtained from the different simulations was per-formed using the root-mean-square inner product (RMSIP)defined as

RMSIP ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

N

Xm

i¼1

Xm

j¼1

ðvi � ujÞ

vuut

where vi and uj are ith and jth eigenvectors of the two differ-ent m dimension essential subspaces of different systems.RMSIP gives a simple measure to assess their dynamicalsimilarity. The convergence of the essential modes was per-formed by comparing the RMSIP’s calculated from the twohalves (0–25 and 25–50 ns) of the trajectories. The valuesobtained from this analysis are reported in Table SIV of theSupplementary materials.

The statistical significance of the overlap between twoeigenvectors was estimated by comparing it with the projec-tions of a randomly oriented unit vector onto a given set(Amadei et al., 1999) and onto eigenvectors obtained fromthe simulation of a structurally equivalent but uncorrelatedbiopolymer. For the first case, the probability density r(s,M)of finding a value s of the square projection of a random unitvector onto one of the eigenvectors of the given set is(Amadei et al., 1999)

rðs;MÞ ¼ ðM � 1Þð1� sÞM�1

where M ¼ 2781 is the dimension of the space. The value ofs0 giving the 99% of confidence that a given s is similar to arandom distribution is calculated from the integral of theprobability density (Roccatano et al., 2003):

Pðs0Þ ¼ðs0

0

r ds ¼ 1� ð1� s0ÞM�2 ¼ 0:99

and hence

s0 ¼ 1�ffiffiffiffiffiffiffiffiffi0:01

M�1p

� 0:002:

Fig. 2. (I) Backbone RMSD values are plotted versus simulation time forWT (a), MUT2 (b) and MUT7 (c). (II) RMSD values per residue werecalculated for the last 20 ns of simulation trajectory for wild-type S41 (a),MUT2 (b) and MUT7 (c); (III) RMSF values per residue in last 20 ns ofsimulations for S41 (a), MUT2 (b) and MUT7 (c). The black line and thegray line represent values for the 283 and 363 K simulation, respectively.Amino acid substitutions are indicated with a dotted line. The dark bars andgray arrows at the top of RMSF graphs indicate a-helical and b-sheetsregions, respectively.

R.Martinez et al.

536

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

The square projection of one eigenvector onto a referenceset is statistically significant if s . s0 or in case s , s0 can beconsidered compatible with the random distribution. Thevalue of s0 gives a lower limit for testing the inner productmatrices (IPMs) of the eigenvectors derived from the back-bone EDA of the simulations. The upper-limit estimationwas obtained from a topological equivalent but uncorrelatedmolecular system. For this purpose, 30 ns MD simulation invacuum and at 500 K of a poly-glycine chain having thesame length and initial fold of the WT S41 Subtilisin crystalstructure backbone was performed. The first 30 eigenvectorsobtained from the analysis of the backbone atoms of the last20 ns of the simulation were compared with those from simu-lations in water. The IPMs showed distributions with valuesnot exceeding 0.2.

SoftwareMD simulations and the analysis of trajectories were per-formed using the GROMACS 4.0.7 software package (Hesset al., 2008). The program VMD (Humphrey et al., 1996)was used to produce all the pictures of the molecularstructures.

Results

Structural propertiesThe secondary structure (in particular a-helix and b-sheet)of proteins in all the simulations did not show remarkablevariations along the simulation time (see Table SI).

The average radius of gyration and total surface areas ofWT, MUT2 and MUT7 indicate that the variation of theprotein structure compactness at 363 K did not exceed 6% ofthe one at 283 K (see Table SIIa).

The total backbone RMSD values along the 50 ns simu-lations for the two different temperatures are reported inPanel I of Fig. 2. All curves reach a stable plateau within30 ns, with RMSD values not exceeding 0.3 nm. RMSDvalues calculated per residue are reported in Panel II ofFig. 2. For WT (a), the most relevant deviations includeloops 83–87, 105–110 and 237–241. The amino acid substi-tutions introduced in positions 211 and 212 of MUT2 (b)resulted in a notable deviation from the crystallographicstructure for the loop 209–222 in the simulation at 363 K.On the contrary, this deviation was not observed in the simu-lation at 283 K. In the case of MUT7 (c), the same regiondid not show remarkable variation from the crystal structureat either 283 or 363 K.

Cluster analysisThe cluster analysis performed with a 0.1 nm cutoff revealeda convergence in the number of clusters formed at 283 and363 K, with median structures at 363 K representing 63% ofthe total conformations for WT, 20% for MUT2 and 38% forMUT7. The structural variation was concentrated in theloops mentioned above, and clearly visible when the confor-mation of the crystal structure of S41 is overlapped with themedian structure calculated for each simulation at 283 and363 K (Fig. 3, Panel I).

Active siteThe conserved active site residues (Asp 34, His 71, N168and S249) show a very low backbone RMSD and RMSFvalues (Fig. 2) for all the simulations. The only large confor-mational change was observed for the His71 side chain thatreturn to its unbinding conformation on the removal of theligand PMSF from the active site (Almog et al, 2009).

Fig. 3. (I) Overlap of the crystal structure of S41(white) with the median structures calculated by cluster analysis WT (left), MUT2 (center) and MUT7 (right)at 283 K (black) and 363 K (gray) main structural differences are in loop areas. (II) Structural changes in S4-binding site in S41. Median structures of the mostabundant clusters found from the MD trajectories of S41 at 283 K (b) and 363 K (c) were analyzed and compared with the structural configuration of thecrystal structure (a). The dark grey surface area represents residues that are part of the S4-binding site; the light grey surface represents the catalytic triad. Thearrow indicates the position of residue Tyrosine 111 (in licorice).

Temperature effects on structure and dynamics of the psychrophilic protease subtilisin S41

537

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

Panel II of Fig. 3 shows the active site conformation of therepresentative (median) structures of the most populated back-bone conformation obtained from the cluster analysis of theWT trajectories at 283 K (a) and 363 K (b), respectively, com-pared with the configuration of the crystal structure. Tyrosine111 (licorice representation) is shown blocking the access toS4-binding site when the simulation is run at 363 K, mainlydue to a configuration change on loop 105–110 (left com-ponent of S4 pocket rim). There are reports that Tyr104 ofSubtilisin BPN0 (corresponding to Tyr111 in S41) might actas a flexible lid for the S4-binding site (Takeuchi et al., 1991).Similar conformational changes are also observed in MUT2and MUT7 clusters at 363 K but in less represented clusters.

Structural ionsIn all simulations, calcium ions remained in their initial pos-ition within a distance of 0.4 nm from the control coordinat-ing residues. The same occurred for the sodium ion inposition Na-5, with the exception of MUT2 at 363 K inwhich the ion left the coordination site after 30 ns of simu-lation time (see Figs S1–S3). The loose coordination of thesodium ion is a consequence of both the lower change andthe less number of coordinating residues in the positionNa-5. In fact, the Ca-4 and Na-5 position were identified inthe crystal structure as low-affinity calcium ions binding sitefrom the number and type of calcium-binding ligands(Almog et al, 2009).

Protein hydrationThe hydration of the protein during the simulations was ana-lyzed by averaging the number of water molecules within0.5 nm from the protein during the simulations. The averagenumber of water molecules at 283 and 363 K show adecrease of �6% due to the decrease in the temperature ofthe overall box density. The values did not show significantdifferences among the different proteins at both temperatures(see Table SIIb).

Dynamic propertiesRMSF values per residue are reported in Panel III of Fig. 2.The regions with the largest flexibility corresponded to theloop regions for all three variants.

As expected, the largest values are in correspondence withloop regions, accounting for 50 % of the overall backboneflexibility. The observed fluctuation pattern was different foreach variant, especially in correspondence of the amino acidsubstitutions (dotted line).

In Table II, the average RMSF values for heavy and back-bone atoms calculated over the last 20 ns of each simulation

are reported. The differences between the values were stat-istically tested using the t-student method. The increase ofthe average RMSF values ranging from 283 to 363 K is thesame for all proteins. Furthermore, the WT average fluctu-ations for both heavy and backbone atoms at the two temp-eratures were found slightly lower or equal to those ofMUT2 and MUT7 simulation, respectively. Contrary to thecommon expectation, this result suggests that the mutationsof the two thermostable variants do not increase the overallrigidity of the enzymes.

Although the overall fluctuations do not have dramaticchanges, from Fig. 2 it is evident that the changes in theprotein dynamics reflect a redistribution of the fluctuationpattern mainly in the loop regions of the proteins. Therefore,the differences among average local fluctuations (LRMSF) ofthe most flexible loops were analyzed.

Loops classified into three groups are represented in darkgrey in Fig. 1a. The first group, containing loops 15–21,39–47, 83–87 and 237–241, is located to the opposite sideof the catalytic site (surface). The second group is formed bythe extended loop 209–222 and loop 288–291. Both loopsare near to (or even part of) high-affinity calcium-bindingsites Ca-1 (loop 288–291) and Ca-2 (loop 209–222) (Almoget al., 2009). The third group is formed by loops 105–110

Table II. Average RMSF (ARMSF) values obtained from the last 20 ns of MD simulation for WT, MUT2 and MUT7. The values and their standard deviation

were calculated for all heavy and backbone atoms at 283 and 363 K. In the H0 columns, the hypothesis H0: ARMSFWT = ARMSFMUT2, MUT7 obtained is

tested at P . 0.1 confidence using the t-student test

Temperature WT RMSF (nm) MUT2 RMSF (nm) H0 MUT7 RMSF (nm) H0

Heavy atoms 283 K 0.061+0.039 0.070+0.047 1 0.062+0.044 0363 K 0.089+0.068 0.097+0.064 1 0.091+0.057 0Difference 0.028 0.027 0.029

Backbone atoms 283 K 0.047+0.023 0.055+0.027 1 0.048+0.029 0363 K 0.069+0.048 0.077+0.048 1 0.072+0.041 0Difference 0.022 0.022 0.024

Table III. The backbone LRMSF values for loop regions were calculated

for WT, MUT2 and MUT7 simulations at 283 K (a) and 363 K (b). In the

H0 columns, the hypothesis H0: LRMSFWT = LRMSFMUT2, MUT7 obtained

is tested at P . 0.1 confidence using the t-student test

Loopnumber ()

Loopgroup

WT LRMSF(nm)

MUT2LRMSF (nm)

H0 MUT7LRMSF (nm)

H0

(a) At 283 K15–21 I 0.067+0.011 0.085+0.022 1 0.070+0.011 039–47 I 0.049+0.010 0.043+0.006 0 0.040+0.006 183–87 I 0.158+0.016 0.084+0.025 1 0.071+0.013 1105–110 III 0.125+0.035 0.149+0.046 0 0.196+0.076 1138–146 III 0.062+0.013 0.117+0.031 1 0.064+0.010 0209–222 II 0.065+0.016 0.099+0.018 1 0.050+0.008 1237–241 I 0.070+0.012 0.083+0.031 0 0.057+0.012 1288–291 II 0.048+0.003 0.172+0.035 1 0.046+0.002 0(b) At 363 K15–21 I 0.095+0.017 0.161+0.038 1 0.133+0.036 139–47 I 0.063+0.012 0.063+0.015 0 0.058+0.009 083–87 I 0.196+0.035 0.195+0.039 0 0.206+0.040 0105–110 III 0.249+0.081 0.278+0.125 0 0.176+0.052 1138–146 III 0.111+0.044 0.210+0.104 1 0.141+0.049 0209–222 II 0.075+0.013 0.197+0.084 1 0.080+0.010 0237–241 I 0.168+0.058 0.126+0.059 0 0.071+0.011 1288–291 II 0.059+0.003 0.174+0.024 1 0.103+0.022 1

R.Martinez et al.

538

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

and 138–146. Both loops are near to the active site and arein proximity (or even part of) the so-called S4-bindingpocket (Siezen and Leunissen, 1997).

In Table IIIa and b, the LRMSF of loop regions of thethree groups from all the simulations are reported. The com-parison with the WT values revealed remarkable differenceson the loop flexibility for each variant and at different temp-eratures. WT shows two major flexible regions (loops 83–87and 105–110) with RMSF values significantly higher thanthe less mobile loops. The large flexibility of these two loopsis not limited to WT but it is also present in the twomutants. Since loop 105–110 is close to a low-affinitycalcium-binding site and is part of the S4-binding pocket, itshigh flexibility at low temperature might be beneficial forthe enzymatic activity. However, at high temperatures, theenhanced flexibility can alter the structure of both thecalcium-binding site and of the S4-binding pocket. Thismight eventually lead to a loss of activity even before majorstructural failure occurs by unfolding. MUT2 and MUT7 at283 K have the same number of loops (three) showingsimilar LRMSF as for the WT but none of them are similarwith each other. At the higher temperature, the number ofLRMSF of similar intensity as the WT is four with two ofthem in common (loops 39–47 and 83–87). The comparisonof LRMF between MUT2 and MUT7 indicates no similarity.Similar conformational change is also observed in the MUT2and MUT7 simulations at 363 K but with less frequentoccurrence.

The double-mutant MUT2 showed a decrease in the flexi-bility on loop 105–110 and an increase in loops 209–222and 138–146. This change on the flexibility profile mightresult in stabilization of the S4-binding pocket, allowing theenzyme to remain active at high temperatures for longer timethan WT. Loop 209–222 is localized far from the active sitebut it is still part of the Ca-2-binding site and its disruptioncan lead to a structural failure of the protein.

MUT7 evidenced a change of the most flexible areas fromloop 209–222 to loops 39–47 and 237–241. Four out of thefive introduced substitutions are in the neighborhood of theCa-2 and Ca-1 calcium-binding sites, most probablyre-stabilizing these loops from the perturbation caused by theprevious substitutions in positions 211 and 212. Finally, thefluctuation level on the loop 138–146 decreased back tothe values of WT.

Essential dynamics analysisThe analysis of the RMSF and LRMSF can only tell the vari-ation of the fluxional pattern of the protein but not how thesefluctuations are correlated among them and along the tridi-mensional degree of freedom of the molecule. In order toobtain this information, the fluctuations of the protein back-bone were analyzed by means of EDA. The number of eigen-vectors contributing to the 70% of the total motion (essentialeigenvectors) calculated for WT and the two variants at 283and 363 K are reported in Table IV (the cumulative percen-tage of the fluctuation for the first 100 eigenvectors isreported in Fig. S4 of the Supplementary materials). For WTsimulation at 363 K, the number of essential eigenvectorsdecreased from 47 to 27, suggesting a narrowing of theessential space as a consequence of a change on the principalsoft motions.

The comparison of the essential subspace using theRMSIP, represented by the first 30 eigenvectors at bothtemperatures, resulted in values higher than 0.6 (Table SIIIof Supplementary materials). These values are comparablewith those obtained by comparing the two halves of the tra-jectories (Table SIV) indicating a significant overlap amongessential subspaces of these proteins. In addition, the innerproduct (IP) matrix of the same eigenvectors (Fig. 4) showsan evident correspondence (IPs . 0.4; note that the purerandom model and pseudo random model of the analyzedprotein provide only values IP , 0.2) of different low-temperature essential modes with those obtained at a hightemperature.

The backbone RMSF calculated for the projection of thetrajectories along the first 3 eigenvectors show collectivefluctuations mainly involving loops 83–87 and 105–110(Fig. 5a). The RMSF profiles at both temperatures are verysimilar consistently with their IP (.0.6). The other twoeigenvectors at both the temperatures has less similarity andno fluctuations along the loop 83–87. In Fig. 6, the tridimen-sional representation of concerted motions along the essentialeigenvectors for WT (a) at 283 K (1) and 363 K (2) is

Table IV. Comparison of the number of eigenvectors that represent 70% of

the total motions of each S41 variant

Eigenvectors representing 70%of motion at 283 K

Eigenvectors representing 70%of motion at 383 K

WT 47 27MUT2 32 28MUT7 34 31

Fig. 4. Inner product matrices between eigenvectors representing the essential subspace of WT (right), MUT2 (center) and MUT7 (left) at 283 and 363 K.

Temperature effects on structure and dynamics of the psychrophilic protease subtilisin S41

539

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

reported. At 283 K, the only major variation occurred in loop105–110, which is related to the flexibility of the S4-bindingpocket, necessary for enzyme activity (Siezen and Leunissen,1997). At 363 K, the increased flexibility of loops 83–87 and105–110 was clearly visible (Fig. 7, a2) along with theincrease in flexibility for loop 237–241.

Differently from WT simulation, the number of essentialeigenvectors calculated for MUT2 shifted from 32 at 283 Kto 28 at 363 K. The overlap of the essential subspaces(0.515) and IP matrices also suggested a lower similarity ofthe essential modes at the two temperatures (Fig. 4). TheRMSF profile along the first 3 eigenvectors (Fig. 5b) isdifferent at the two temperatures. In fact, collective motionsat 283 K involve mainly loop 138–146 and, at 363 K, loops83–87 and 105–110. Furthermore, the fluctuations along the

loop 209–222 at 363 K occur mainly along the second andthird eigenvectors. The IP matrices calculated among thedifferent proteins show a higher essential subspaces simi-larity at 283 K rather than at 363 K (Fig. S5 ofSupplementary materials). In Fig. 6b, the tridimensional rep-resentation of the essential modes shows clearly that, at lowtemperature, the only major fluctuation occurred in loop138–146 in correspondence of the S4-binding pocket.However, at 363 K, the larger fluctuations involved the loops209–222, 83–87 and 105–110 (Fig. 6, b2).

Finally, in the case of MUT7, the number of essentialeigenvectors does not show a significant change, having 34at 283 K and 31 at 363 K. The RMSIP value (0.521) wasagain lower than WT as also the similarity between eigen-vectors at both 283 and 363 K (Fig. S5 of Supplementarymaterials), whereas the IP matrix shows that the subset ofeigenvectors representing the principal motions at 283 and363 K have a lower degree of similarity. The RMSF profilealong the first and second eigenvectors at 283 K was verydifferent to that obtained at 363 K since collective motions at283 K included loop 105–110, whereas at 363 K loops83–87, 105–110 and 138–146 (Fig. 5c) were involved inthe overall motion of the protein. The concerted modes at283 K reported in Fig. 6, c1 involve mainly loop 105–110.At 363 K, these modes were described in different loopregions, in particular loops 15–21, 83–87, 105–110 and138–146. In this case, the temperature increase did notsimply enhance the modes present at lower temperature butseems to activate different ones in other regions of theprotein (Fig. 6, c2).

Unfolding simulationsIn Fig. 7a and b, the backbone RMSD and RMSF perresidue from the simulations at 450 K are reported. Thetrends of the curve show a qualitative consistency with thethermal tolerance observed experimentally for the three pro-teins. The highest deviations and fluctuations are concen-trated in the N-terminal end of the protein (residue 1–30),loops 165–180 and 237–241 suggesting that these arerelated with the unfolding process of S41, which is consistentin WT, MUT and MUT7.

While the fluctuation profile is similar, analysis of thethree-dimensional conformations at 30 ns and the increase inthe radius of gyration of the variants over simulation timesupport that, on a given simulation time, there is a differentqualitative level of unfolding, which is higher for WT, com-pared with MUT2 and MU7 (Fig. 8).

Discussion

The increased thermal stability of MUT2 with respect to theWT enzyme was considered as the effect of stabilization ofthe extended loop 209–222 by the two mutations at211–212 positions (Miyazaki and Arnold, 1999). In MUT7,four of the additional five amino acid substitutions are loca-lized on the same loop. In particular, the substitutionLys221Glu is localized in proximity of the calcium-bindingsite Ca-2, whereas Asn291Ile and Ser295Thr are close to theCa-1-binding site. The half-life at 608C of MUT7 increasesat high calcium concentrations (Miyazaki et al., 2000) inclear consistency with the possible effect of the Ca-2 binding(Almog et al. 2009). Our high-temperature simulations

Fig. 5. RMSF values per residue obtained by projecting the WT (a), MUT2(b) and MUT7 (c) trajectories on the corresponding first 3 eigenvectors. Theblack line represents values obtained at 283 K and the gray line 363 K.Dotted lines represent the position of the introduced amino acidsubstitutions.

R.Martinez et al.

540

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

provide a possible qualitative scenario of the unfolding ofthe three proteins. The total RMSD and radius of gyrationdistinguish the WT from the mutant’s trend. The latter showa similar variation with the temperature. A more detailedanalysis of the residue deviation and fluctuation shows thatthe unfolding starts from the N-terminal region of the pro-teins. However, these results are obtained by very high temp-erature that may flatten the subtle difference of thermalstabilization mechanism in the variant. The unfolding ofSubtilisin proteases is an irreversible process and it can shareat very high temperature a similar trend. Furthermore,another important effect to take into account in the interpret-ation of these modeling data is the tendency of Subtilisin(and proteases in general) to the auto-proteolysis that

normally occurs in the loop regions. This effect can be influ-enced by the mobility of these regions (Siezen andLeunissen, 1997).

In this sense, the results of our MD simulations alsosuggested that the effect of mutations on the thermal stabilityof the variants may be more complex than a structuralreinforcement and could involve changes of their internaldynamics. In particular, WT simulations showed a low mobi-lity for the loop 209–222 (probably determined by the pres-ence of Ca-3) and higher mobility of the shorter loop regions(83–87 and 105–110) at both 283 and 363 K. In MUT2, theloop 209–222 showed an increased mobility only at 363 Kwhereas mobility of the active site loop 105–110 was ratherreduced at the same temperature. Our results support the

Fig. 7. Total RMSD (a) RMSD per residue (b) and RMSF per residue (c) analysis of the trajectories obtained at 450 K for WT (black), MUT2 (dark grey) andMUT7 (light gray).

Fig. 6. Tridimensional representation of the essential modes representing 70% of the total protein from the simulations of WT (a), MUT2 (b) and MUT7 (c).The structural conformations were extracted each 600 ps and superimposed to evidence the dominant fluctuation modes on each condition. Subscripts 1 and 2refer to the 283 and 363 K simulations, respectively.

Temperature effects on structure and dynamics of the psychrophilic protease subtilisin S41

541

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

stabilization of the region where the amino acid substitutionwas introduced in MUT7, since RMSD and RMSF valuesper residue decreased with respect to those in the same pos-itions in MUT2 at 363 K (Fig. 2). Despite MUT7 shows ahigher overall flexibility than WT, a detailed inspection evi-dences that there are no loops with very high RMSF values(Table III). This, together with the change on the essentialmodes at 283 and 363 K suggests that the protein structureremains stable at higher temperatures. The idea of anenhancement of different set of essential modes uponincrease of temperature has also been proposed in otherstudies of thermophilic (Colombo and Merz, 1999; Grottesiet al., 2002; Wintrode et al., 2003; Motono et al., 2007) ormesophilic proteins (Roccatano et al., 2003) but not yetexperimentally demonstrated. However, recently experimen-tal results from NMR and Neutron scattering are showing theimportance of localized enhanced fluctuations in hingeregions of the protein for the enzyme activity (Meinholdet al., 2008; Krishnamurthy et al., 2009).

In the last years, there have been also different experimen-tal results proving the role of flexible loop region on proteinthermal stability. Jang et al. (2002) showed that removingone of the two unique long loops from thermicin subtilisin-

like protease yields either a less active variant or a lessthermostable one. The removal of the extended loops wasperformed to eliminate possible weak points on the structureto further increase thermal stability, but the results suggestedthat those loops are an important structural factor that mayplay a role on the protein ability to withstand high tempera-tures. This results support also the idea expressed before thatthe mechanism of thermal stabilization in these proteins canbe more elaborate and involve also the dynamic aspect of theprotein structure.

At high temperature, the essential modes of WT are rathersimilar but reduced in number in comparison with thoseobtained from low-temperature simulations. These modesinvolve few loop regions of the protein that show an amplifi-cation of the overall fluctuations as the temperature isincreased. On the contrary, the essential modes of MUT7 arerather different at 283 and 363 K. Furthermore, the numberof essential subspaces is constant with the increase of temp-erature resulting in a larger number of collective modesinvolving loop regions than in WT. Moreover, the loopregions involved in these essential modes are not in structuralcorrespondence of the binding or active site region of theprotein. Finally, the essential eigenvectors for both mutants

Fig. 8. (I) Comparison between the initial structure of S41 (a) and the final configuration after a 30 ns long simulation at 450 K of WT (a), MUT2 (b) andMUT7 (c). (II) The variation of the radius of gyration over time for WT (black) MUT2 (dark gray) and MUT7 (light gray) evidences the differences on thelevel of unfolding on the variants.

R.Martinez et al.

542

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

have lower resemblance with those at a higher temperature(Fig. S5).

Conclusions

In this work, we have reported a comparative study usingMD simulations of the psychrophilic Subtilisin S41 and twovariants with enhanced thermal stability and activity at 283and 363 K. The results of these studies show that the aminoacid substitutions introduced in the variants do not changesubstantially the structure and overall fluctuations withrespect to WT. A detailed study of the fluctuation profilesrevealed consistent variations on the collective modes,suggesting that the increase on temperature enhances theamplitude of some of the essential modes also present at alow temperature. Based on these results, we propose as apossible and alternative explanation of high-temperatureadaptation on these variants, based on the change of theessential modes present at low and high temperatures(mainly involving loops that are not close to the active site).This variation keeps flexibility of the variants comparablewith WT levels for residues relevant to the catalytic activity,allowing them to retain activity at both high and lowtemperatures.

Although these results are not conclusive and based on atheoretical model, they can provide argument for furtherexperimental and theoretical investigations. Understandingthe possibility to modulate the dynamic behavior of anenzyme might offer alternative strategies to improve proteinstability, in addition to the structural stabilization approaches.

Supplementary data

Supplementary data are available at PEDS online.

Funding

This work was supported by the German Governmentthrough the Bundesministerium fur Bildung und Forschung[FKZ: 0315035A to U.S.] and Henkel AG & Co. KGaA.This study was performed using the computational resourcesof the CLAMV (Computer Laboratories for Animation,Modeling and Visualization) at Jacobs University Bremen.

ReferencesAdcock,S.A. and McCammon,J.A. (2006) Chem. Rev., 106, 1589–1615.Almog,O., Gonzalez,A., Godin,N., de Leeuw,M., Mekel,M.J., Klein,D.,

Braun,S., Shoham,G. and Walter,R.L. (2009) Proteins: Struct. Funct.Bioinf., 74, 489–496.

Amadei,A., Ceruso,M.A. and Nola,A.D. (1999) Proteins: Struct. Funct.Genet., 36, 419–424.

Amadei,A., Linssen,A.B.M. and Berendsen,H.J.C. (1993) Proteins: Struct.Funct. Genet., 17, 412–425.

Berendsen,H.J.C., Postma,J.P.M., van Gunsteren,W.F., DiNola,A. andHaak,J.R. (1984) J. Chem. Phys., 81, 3684–3690.

Boehr,D.D., Dyson,H.J. and Wright,P.E. (2006) Chem. Rev., 106,3055–3079.

Bos,F. and Pleiss,J. (2009) Biophys. J., 97, 2550–2558.Colombo,G. and Merz,K.M. (1999) J. Am. Chem. Soc., 121, 6895–6903.D’Amico,S., Claverie,P., Collins,T., Georlette,D., Gratia,E., Hoyoux,A.,

Meuwis,M.A., Feller,G. and Gerday,C. (2002) Philos. Trans. R. Soc.Lond., B, Biol. Sci., 357, 917–925.

Daggett,V. (2006) Chem. Rev., 106, 1898–1916.Daura,X., Gademann,K., Bernhard,J., Dieter,S., Wilfred,F.v.G. and

Alan,E.M. (1999) Angew. Chem. Int. Ed. Engl., 38, 236–240.

Davail,S., Feller,G., Narinx,E. and Gerday,C. (1994) J. Biol. Chem., 269,17448–17453.

Demirjian,D.C., Morıs-Varas,F. and Cassidy,C.S. (2001) Curr. Opin. Chem.Biol., 5, 144–151.

Essmann,U., Perera,L., Berkowitz,M., Darden,T., Lee,H. and Pedersen,L.(1995) J. Chem. Phys., 103, 8577–8593.

Feller,G. (2007) Extremophiles, 11, 211–216.Garcia,A.E. (1992) Phys. Rev. Lett., 68, 2696–2699.Grottesi,A., Ceruso,M.A., Colosimo,A. and Nola,A.D. (2002) Proteins:

Struct. Funct. Bioinf., 46, 287–294.Guex,N. and Peitsch,M.C. (1997) Electrophoresis, 18, 2714–2723.Gupta,R., Beg,Q. and Lorenz,P. (2002) Appl. Microbiol. Biotechnol., 59,

15–32.Henzler-Wildman,K. and Kern,D. (2007) Nature, 450, 964–972.Herbert,R.A. (1992) Trends Biotechnol., 10, 395–402.Hess,B., Bekker,H., Berendsen,H.J.C. and Fraaije,J.G.E.M. (1997)

J. Comput. Chem., 18, 1463–1472.Hess,B., Kutzner,C., van der Spoel,D. and Lindahl,E. (2008) J. Chem.

Theory Comput., 4, 435–447.Humphrey,W., Dalke,A. and Schulten,K. (1996) J. Mol. Graph., 14, 33–38.Jang,H.J., Lee,C.H., Lee,W. and Kim,Y.S. (2002) J. Biochem. Mol. Biol.,

35, 498–507.Jenney,F., Jr and Adams,M. (2008) Extremophiles, 12, 39–50.Jorgensen,W. and Tirado-Rives,J. (1988) J. Am. Chem. Soc., 110,

1657–1666.Jorgensen,W.L., Chandrasekhar,J., Madura,J.D., Impey,R.W. and Klein,M.L.

(1983) J. Chem. Phys., 79, 926–935.Kabsch,W. and Sander,C. (1983) Biopolymers, 22, 2577–2637.Krishnamurthy,H., Munro,K., Yan,H. and Vieille,C. (2009) Biochemistry,

48, 2723–2739.Kundu,S. and Roy,D. (2009) J. Mol. Graph. Model., 27, 871–880.Matthews,B.W., Nicholson,H. and Becktel,W.J. (1987) Proc. Natl Acad. Sci.

U.S.A., 84, 6663–6667.Maurer,K.H. (2004) Curr. Opin. Biotechnol., 15, 330–334.Meinhold,L., Clement,D., Tehei,M., Daniel,R., Finney,J.L. and Smith,J.C.

(2008) Biophys. J., 94, 4812–4818.Melchionna,S., Sinibaldi,R. and Briganti,G. (2006) Biophys. J., 90,

4204–4212.Merkley,E.D., Parson,W.W. and Daggett,V. (2010) Protein Eng. Des. Sel.,

23, 327–36.Mittermaier,A.K. and Kay,L.E. (2009) Trends Biochem. Sci., 34, 601–611.Miyamoto,S. and Kollman,P.A. (1992) J. Comput. Chem., 13, 952–962.Miyazaki,K. and Arnold,F.H. (1999) J. Mol. Evo., 49, 716–720.Miyazaki,K., Wintrode,P.L., Grayling,R.A., Rubingh,D.N. and Arnold,F.H.

(2000) J. Mol. Biol., 297, 1015–1026.Morin,S. and Gagne,S.M. (2009) Biophys. J., 96, 4681–4691.Motono,C., Gromiha,M.M. and Kumar,S. (2007) Proteins: Struct. Funct.

Bioinf., 71, 655–669.Papaleo,E., Pasi,M., Riccardi,L., Sambi,I., Fantucci,P. and Gioia,L.D. (2008)

FEBS Lett., 582, 1008–1018.Polyansky,A.A., Kosinsky,Y.A. and Efremov,R.G. (2004) Russ. J. Bioorg.

Chem., 30, 421–430.Priyakumar,U.D., Ramakrishna,S., Nagarjuna,K.R. and Reddy,S.K. (2010)

J. Phys. Chem. B, 114, 1707–1718.Roccatano,D., Daidone,I., Ceruso,M.A., Bossa,C. and Nola,A.D. (2003)

Biophys. J., 84, 1876–1883.Schaeffer,R.D., Fersht,A. and Daggett,V. (2008) Curr. Opin. Struct. Biol.,

18, 4–9.Scheraga,H.A., Khalili,M. and Liwo,A. (2007) Annu. Rev. Phys. Chem., 58,

57–83.Siezen,R.J. and Leunissen,J.A.M. (1997) Protein Sci., 6, 501–523.Spiwok,V., Lipovova,P., Skalova,T., Duskova,J., Dohnalek,J., Hasek,J.,

Russell,N. and Kralova,B. (2007) J. Mol. Model., 13, 485–497.Storch,E.M., Daggett,V. and Atkins,W.M. (1999) Biochemistry, 38,

5054–5064.Takeuchi,Y., Noguchi,S., Satow,Y., Kojima,S., Kumagai,I., Miura,K.-i.,

Nakamura,K.T. and Mitsui,Y. (1991) Protein Eng., 4, 501–508.Tehei,M. and Zaccai,G. (2007) FEBS J., 274, 4034–4043.Thomas,D.N. and Dieckmann,G.S. (2002) Science, 295, 641–644.van den Burg,B. (2003) Curr. Opin. Microbiol., 6, 213–218.van Gunsteren,W.F., Bakowies,D., Baron,R., Chandrasekhar,I., Christen,M.,

Daura,X., Gee,P., Geerke,D.P., Glattli,A., Hunenberger,P.H., et al. (2006)Angew. Chem. Int. Ed. Engl., 45, 4064–4092.

van Gunsteren,W.F., Dolenc,J. and Mark,A.E. (2008) Curr. Opin. Struc.Biol., 18, 149–153.

Vieille,C. and Zeikus,G.J. (1996) Trends Biotechnol., 14, 183–190.Vieille,C. and Zeikus,G.J. (2001) Microbiol. Mol. Biol. Rev., 65, 1–43.

Temperature effects on structure and dynamics of the psychrophilic protease subtilisin S41

543

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from

Wintrode,P.L. and Arnold,F.H. (2001) In Frederic,M.R., David,S.E.,Peter,S.K. and Frances,H.A. (eds), Advances in Protein Chemistry.Academic Press, San Diego, California, USA and London, UK, pp.161–225.

Wintrode,P.L., Zhang,D., Vaidehi,N., Arnold,F.H. and Goddard,W.A., III(2003) J. Mol. Biol., 327, 745–757.

Zhang,Y., Porcelli,M., Cacciapuoti,G. and Ealick,S.E. (2006) J. Mol. Biol.,357, 252–262.

R.Martinez et al.

544

by guest on August 24, 2016

http://peds.oxfordjournals.org/D

ownloaded from