Ipa

ZI

a

ARRAA

KAAMS

1

t(attmasbftTwrih

o

0d

Journal of Biotechnology 154 (2011) 248– 254

Contents lists available at ScienceDirect

Journal of Biotechnology

j ourna l ho me pag e: www.elsev ier .com/ locate / jb io tec

ntegrating molecular dynamics and co-evolutionary analysis for reliable targetrediction and deregulation of the allosteric inhibition of aspartokinase formino acid production

hen Chen, Sugima Rappert, Jibin Sun1, An-Ping Zeng ∗

nstitute of Bioprocess and Biosystems Engineering, Hamburg University of Technology, Denickestrasse 15, D-21073 Hamburg, Germany

r t i c l e i n f o

rticle history:eceived 14 February 2011eceived in revised form 2 May 2011ccepted 9 May 2011vailable online 14 May 2011

eywords:spartokinase

a b s t r a c t

Deregulation of allosteric inhibition of enzymes is a challenge for strain engineering and has beenachieved so far primarily by random mutation and trial-and-error. In this work, we used aspartokinase, animportant allosteric enzyme for industrial amino acids production, to demonstrate a predictive approachthat combines protein dynamics and evolution for a rational reengineering of enzyme allostery. Molec-ular dynamic simulation of aspartokinase III (AK3) from Escherichia coli and statistical coupling analysisof protein sequences of the aspartokinase family allowed to identify a cluster of residues which arecorrelated during protein motion and coupled during the evolution. This cluster of residues forms an

llosteric regulationolecular dynamics simulation

tatistical coupling analysis

interconnected network mediating the allosteric regulation, including most of the previously reportedpositions mutated in feedback insensitive AK3 mutants. Beyond these mutation positions, we have suc-cessfully constructed another twelve targeted mutations of AK3 desensitized toward lysine inhibition.Six threonine-insensitive mutants of aspartokinase I–homoserine dehydrogenase I (AK1–HD1) were alsocreated based on the predictions. The proposed approach can be widely applied for the deregulation ofother allosteric enzymes.

. Introduction

Allosteric regulation is one of the fundamental mechanismshat control almost all cellular metabolisms and gene regulationTsai et al., 2009). Deregulation of allsoteric inhibition has been

challenge in designing and optimizing metabolic pathways forhe production of target metabolites such as amino acids. So far,his is achieved almost exclusively by multiple rounds of random

utation and selection. Despite the successful application of thesepproaches for the development of amino acid producers, they haveeveral disadvantages. For example, undesirable mutations woulde introduced which may cause growth retardation and by-productormation. Furthermore, well selectable phenotypes such as resis-ance to analogs of inhibitors are prerequisite for these processes.hus, these approaches cannot be used for some allosteric enzymeshich lack corresponding selectable phenotypes for the mutants. A

ational approach that could be used to guide targeted reengineer-ng of allosteric enzymes without screening or selection process isighly desired.

∗ Corresponding author. Tel.: +49 40 42878 4183; fax: +49 40 42878 2909.E-mail address: aze@tu-harburg.de (A.-P. Zeng).

1 Present address: Tianjin Institute of Industrial Biotechnology, Chinese Academyf Sciences, Tianjing, 300308, PR China.

168-1656/$ – see front matter © 2011 Elsevier B.V. All rights reserved.oi:10.1016/j.jbiotec.2011.05.005

© 2011 Elsevier B.V. All rights reserved.

Recent advances in structural biology together with computa-tional analysis are opening a new avenue toward understandingand rational reengineering of allosteric enzymes (Chen et al., 2010).Two approaches, molecular dynamic simulation (MD) and statisti-cal coupling analysis (SCA) are especially useful for such purpose(Estabrook et al., 2005). Allosteric regulation is a dynamic processand thus MD can provide valuable information for correlated oranti-correlated motions among different structural elements relat-ing dynamics to allostery (Smock and Gierasch, 2009). On the otherhand, SCA can reveal correlated mutations of protein family andhelp to identify coupled residues contributing to the allosteric com-munication (Lockless and Ranganathan, 1999; Suel et al., 2003).Estabrook et al. (2005) demonstrated the usefulness of the com-bined approach of SCA and MD for identification of amino acid pairsessential for catalysis. In this work, we further show that such anintegrated approach is efficient to define a cluster of residues thatare essential for allosteric regulation and can be used for rationalderegulation of allosteric inhibition.

Aspartokinase was chosen in this work as a model enzyme. Itcatalyzes the phosphorylation of aspartate and controls the biosyn-thesis of several industrially important amino acids such as lysine,

threonine and methionine (Yoshida et al., 2007). In Escherichia coli,there exist three aspartokinase isozymes. Two of them, aspartok-inase I–homoserine dehydrogenase I (AK1–HD1) encoded by thrAgene and aspartokinase III (AK3) encoded by lysC gene are allosteric

Z. Chen et al. / Journal of Biotechn

Fig. 1. Conformational transition of aspartokinase III (AK3) by allosteric regulationof lysine. The regulatory domain is shown in red. The N-lobe and C-lobe of the cat-alytic domain are denoted by blue and cyan, respectively. Residues 352–362 thatshowed the largest conformational change are colored in purple. The substrates(ADP, aspartate) and the inhibitor (lysine) are represented by CPK models. (For inter-pw

eds12tdSe(m

obclasiarouorid

2

2a

w

on both evolutional and dynamical contribution. It was created by

retation of the references to color in this figure legend, the reader is referred to theeb version of the article.)

nzymes and especially important for lysine and threonine pro-uction. AKI–HDI is allosterically inhibited by threonine and itsynthesis is repressed by threonine plus leucine (Bearer and Neet,978) while AK3 is inhibited and repressed by lysine (Kotaka et al.,006). In the last fifty years, considerable efforts have been madeo deregulate these two enzymes from allosteric inhibition by ran-om mutation and selection of mutants resistant to lysine analogue-(2-aminoethyl)-L-cysteine (AEC) (Kikuchi et al., 1999; Miyatat al., 2001), or threonine analogue �-amino-�-hydroxyvaleric acidAHV) (Lee et al., 2003). However, only a limit number of positive

utations have been identified so far, especially for AK1–HD1.The crystal structures of AK3 complex with substrates (R-state)

r lysine (T-state) have been solved (Kotaka et al., 2006). Lysineinding induces a large conformational change of AK3 (Fig. 1). AK3onsists of an N-terminal catalysis domain and a C-terminal regu-atory domain. The regulatory domain possesses two motifs calleds ACT domains (Chipman and Shaanan, 2001) which are respon-ible for the lysine binding. Mapping of the reported mutationsnto the three-dimensional structure of AK3 enables us to evalu-te their roles for allosteric regulation. Interestingly, these mutatedesidues are located not only within lysine binding sites but also inther regions of the protein (Table 1 and Fig. 4A). This motivatess to carry out a more systematic analysis of the whole structuref AK3 to identify residues which may form an interacting networkesponsible for the allosteric regulaltion. Specific residues of thisnteracting network should be evaluated as potential targets foreregulation of the allostery.

. Materials and methods

.1. Molecular dynamic simulation (MD) and cross-correlation

nalysis

The starting structure for the MD simulation of AK3 with lysineas based on the crystal structure of T-state AK3 (PDB code

ology 154 (2011) 248– 254 249

2J0X). Aspartate and other ligands were removed from the struc-ture and the missing residues were repaired using MODELLER9v5 (http://salilab.org/modeller/). For the MD simulation of AK3without lysine, lysine was also removed from the previous struc-ture. Dynamic trajectories were computed using AMBER 10.0 withparm99SB force field (Duan et al., 2003). Protein was solvated in abox of TIP3P water molecules (Jorgensen, 1981) with the minimaldistance of 1.5 nm from the protein to the box wall. Na+ ions wereadded to neutralize the systems. 1500 steps of steepest-descentenergy minimization and 2500 steps Newton–Raphson minimiza-tion were performed before the MD simulation. The systems werethen heated to 300 K, followed by 500 ps equilibration and 10 ns MDsimulations. The particle mesh Ewald method (Darden et al., 1993)was used to calculate the long-range electrostatics interactions.Non-bonded interactions were cutoff at 12.0 A and updated every25 steps. The SHAKE method (Ryckaert et al., 1977) was appliedto constrain all covalent bonds involving H atoms. Each simulationwas coupled to a 300 K thermal bath at 1.0 atm of pressure by apply-ing the algorithm of Berendsen et al. (1984). The temperature andpressure coupling parameters were set as 0.2 and 0.05 ps, respec-tively. The integration step was set to 2 fs and the coordinates weresaved every 0.1 ps, giving a total number of 100,000 structures foreach trajectory.

Cross-correlation analysis of the trajectories from 2 ns to 10 nswas performed to evaluate the dynamical correlation between anytwo residues. The cross-correlation coefficient is defined as C(ij) =(⟨

�ri • �rj

⟩)/(⟨

�ri2⟩1/2⟨

�rj2⟩1/2

), where �ri and �rj denotes

the displacement vectors of residue i and j and the angle bracketsdenote ensemble average. The coordinate sets of 2 ns were used asthe references. C(ij) = 1 indicates that the motions of two residuesare completely correlated (same phase) while C(ij) = − 1 indicatesthat the motions of two residues are completely anti-correlated(opposite phase). The extent of correlation was calculated usingAMBER10.0.

2.2. Multiple sequence alignment (MSA) and statistical couplinganalysis (SCA)

Sequences of the aspartokinase family proteins were col-lected from the UniRef90 database in UniProt Knowledgebase(http://www.uniprot.org/). Any sequence sharing >90% similarityto another sequence was removed in order to get a diverse distribu-tion of samples. The sequences were aligned with MUSCLE (Edgar,2004) followed by structure-guided manual adjustment (Doolittle,1996). The sequence positions with gap frequency higher than 20%were deleted. The final alignment consisted of 340 sequences and424 positions.

SCA measures the evolutional correlation between any tworesidue positions Cab

ij= �i�j(f ab

ij− f a

if bj

), where f ai

and f bj

denote the

observed frequency of amino acid a and b at position i and j. f abij

rep-resents the joint frequency of having a at position i and b at positionj, and �i and �j are the positional conservation-based weights. Thedetailed procedure for the SCA calculation has been described else-where (Halabi et al., 2009; Suel et al., 2003). The SCA matrix wascalculated using the METLAB script derived from the publication ofHalabi et al. (2009).

2.3. The SCA·MD matrix

SCA·MD matrix is to measure the positional correlation based

multiplying the individual elements of the SCA matrix with the cor-responding elements of the truncated cross-correlation matrix ofMD trajectory of AK3 with lysine.

250 Z. Chen et al. / Journal of Biotechnology 154 (2011) 248– 254

Table 1The 30 predicted positions in this study essential for allosteric regulation of AK3 and AK1–HD1 and point mutations.

No. Enzymes and corresponding point mutations

AK3 Referencesa AK1–HD1 Referencesa

1b M251P This study L257 –2 T253R This study Y259 –3 R305A This study L312 –4 S315A This study G322 –5 M318I Miyata et al. (2001) M325 –6 H320A This study G327 –7 G323D Kikuchi et al. (1999), Miyata et al. (2001) G3308 F324 –* M331 –9 L325F Miyata et al. (2001) A332 –

10 F329R This study F336 –11 I337P This study I344P This study12 S338L This study S345F Lee et al. (2003)13 V339A This study V346 –14 T344M Kikuchi et al. (1999) Q351A This study15 S345L Kikuchi et al. (1999), Miyata et al. (2001) S35216 E346R This study S353 –17 V347M Miyata et al. (2001) S354 –18 V349M Miyata et al. (2001) Y356 –19 T352I Kikuchi et al. (1999), Miyata et al. (2001) S357 –20 T355 – S359 –21 C378 – E388 –22 I392 – V402 –23 F407 – F417 –24 N414 – N424A This study25 R416A This study N426A This study26 M417I Miyata et al. (2001) I427P This study27 S423 – S435 –28 S424 – E436A This study29 N426 – S438 –30 C428R This study S440 –

a The corresponding mutations from other references were originally derived from random mutation and selection and the single mutation desensitizes aspartokinasef

ences–

2

wTcSstwpc

2

bm1Am5mp

3

3

cd

rom feedback inhibition.b The residues of the same row correspond to the same position in multiple sequ

*, Mutations of the corresponding positions have not yet been reported.

.4. Molecular cloning and enzyme purification

The lysC gene and the thrA gene encoding for AK3 and AK1–HD1ere amplified by PCR from the genomic DNA of E. coli K12.

he DNA fragments were double-digested by NdeI and XhoI andloned into corresponding sites of plasmid pET-22b (+) (Novagen).ite-directed mutageneses were carried out using Quick-changeite-directed mutagenesis kits (Stratagene) according to the pro-ocol. The constructed proteins have (His)6-tags at the C-terminushich could be used for enzyme purification. Overproduction andurification of AK3 and AK1–HD1 as well as their mutants werearried out following the protocol of Yoshida et al. (2007).

.5. Enzyme assay

The aspartokinase activities of AK3 and AK1–HD1 were assayedy hydroxamate method (Black and Wright, 1954). 1 ml reactionixture contained 200 mM Tris–HCl (pH 7.5), 10 mM MgSO4·6H2O,

0 mM aspartate, 10 mM ATP, 160 mM NH2OH·HCl and the enzyme.fter incubation at 303 K for 30 min, the reaction was stopped byixing with 1 ml 5% (w/v) FeCl3 solution and the absorbance at

40 nm was monitored. The specific activity is expressed as micro-oles aspartyl hydroxamate produced per minute per milligram

rotein.

. Results

.1. MD simulation of aspartokinase III

As mentioned above, allosteric regulation is a dynamic pro-ess involving the correlated or anti-correlated motions betweenifferent structural elements. Thus, two 10 ns MD simulations

alignment.

of AK3 with lysine and without lysine were firstly carried outindependently. Cross-correlation analyses were performed onthe trajectories of atomic motions from 2 ns to 10 ns, respec-tively. As shown in Fig. 2A and C, lysine binding induces bigchanges of dynamic motions of AK3, especially for the anti-correlated motions between domains (negative values). We definedthree sectors according to the correlated motions: A, B and C,which correspond to the main parts of the N-lobe and C-lobeof the catalytic domain and the regulatory domain, respectively(Fig. 2B). For the dynamic trajectory of AK3 with lysine, corre-lated motions dominate the internal motions within each sector.A strong anti-correlated motion is found between the sector Aand sector C. This extensive anti-correlated motion is relatedto the allosteric regulation of AK3 identified by the compari-son of crystal structures of R-state and T-state AK3 (Fig. 1A).Interestingly, the anti-correlated motion between the sectors Aand C did not appear in the dynamic trajectory of AK3 with-out lysine, indicating that this motion is induced by lysinebinding.

A detailed comparison of Fig. 2A and C also highlighted theimportant role of sector B for a cross-sector communication. Inthe trajectory of AK3 with lysine, the residues 249–262 (light-blue spheres), 227–238 (marine spheres), and 201–220 (purplebluespheres) of sector B show obvious correlated motions with sector C,peptide 434–449 (orange) and sector A, respectively (Fig. 2B). Thecross-sector correlated motions are mainly mediated by van derWaals interactions. The cross-sector correlated motions betweenresidues 249–262 and sector C are especially important as they did

not appear in the trajectory of AK3 without lysine (Fig. 2C). Wetherefore supposed that the residues 249–262 have a high proba-bility to be involved in the allosteric communication between theregulatory domain and the catalytic domain. Indeed, the muta-

Z. Chen et al. / Journal of Biotechnology 154 (2011) 248– 254 251

Fig. 2. Correlated and anti-correlated motions of AK3 in molecular dynamic simulation. (A) Cross-correlation matrix of AK3 with lysine. Correlated motions are shown bypositive value (red) and anti-correlated motions are shown by negative value (blue). (B) Mapping of the correlated motions derived from (A) into AK3. Only the motions inchain A are indicated. Chain B is represented by a ribbon. Three independent sectors are defined according to the correlated motions, corresponding to the main parts of theN-lobe (sector A) and the C-lobe (sector B) of the catalytic domain and the regulatory domain (sector C). Sector B connects sector A and sector C through the colored spheres.R correS K3 wr

t(

3f

cpicdtSrcttri

ett

esidues 249–262 (light blue), 227–238 (marine) and 201–220 (purple blue), showector A and sector C show anti-correlated motion. (C) Cross-correlation matrix of Aeader is referred to the web version of the article.)

ion of E250K transferred lysine from an inhibitor to an activatorKikuchi et al., 1999).

.2. Combination of SCA and MD to identify the residues essentialor the allosteric regulation of aspartokinase III

Although the MD simulation quantifies the correlated and anti-orrelated motions between different structural elements and thusrovides clues for their contributions for allosteric regulation, it

s difficult to directly identify a cluster of key residues from theross-correlation matrix such as Fig. 2A. To this end, a pronouncediagonal matrix by giving a weight of evolutionary conservation tohe cross-correlation matrix can be helpful. We thus performed aCA analysis of aspartokinase family proteins to evaluate the cor-elations between any two residue positions by examining theirorrelated mutations. The SCA matrix of aspartokinase family pro-eins is shown in Fig. 3A. A high correlation value of SCA indicateshat the two residues are co-evolved and the mutation of oneesidue would cause the change of the other residue correspond-ngly in order to keep a specific function.

The combined SCA·MD matrix was constructed by a multiply oflements of the SCA matrix with the corresponding elements of theruncated cross-correlation matrix derived from the MD simula-ion of AK3 with lysine (Fig. 3B). Differently from the results shown

lated motions with sector C, peptide 434–449 (orange) and sector A, respectively.ithout lysine. (For interpretation of the references to color in this figure legend, the

in Figs. 2A and 3A, the combined SCA·MD matrix (Fig. 3B) showsmuch sharper peaks and valleys. This is due to the fact that theabove process of combined analysis eliminates either evolution-arily significant interactions that do not show dynamic couplingsor MD correlations that are not evolutionarily conserved. Thus,the residues that are both dynamically and evolutionarily corre-lated are considered as the most important residues in mediatingthe allosteric regulation. The residues that are directly involved insubstrates binding are also not considered as targets to avoid affect-ing enzyme activity. 30 such highly ranked residues within sectorB and sector C were finally selected from the SCA·MD matrix foraspartokinase.

3.3. Distribution of the identified residues and site-directedmutagenesis

The 30 top-ranked residues within sector B and sector C of theSCA·MD matrix are mapped into the crystal structure of AK3 asdepicted in Fig. 4A. These residues are sparsely distributed through-out the protein structure and form an interconnected network

mainly by van der Waals contact. Most of the previously reportedmutation points of feedback-resistance AK3 are among these iden-tified positions (Table 1). The residues can be divided into threegroups. The first group consists of two pairs of residues, M251–I392

252 Z. Chen et al. / Journal of Biotechnology 154 (2011) 248– 254

Fig. 3. SCA and SCA·MD matrix. (A) SCA matrix of the aspartokinase family. The color scale linearly maps the correlated value from 0 (uncorrelated) to 3 (correlated).Numbering is according to the sequence of AK3. (B) SCA·MD matrix. The cross-correlation matrix of AK3 with lysine from Fig. 2A was truncated and multiplied by the SCAmatrix.

Fig. 4. Distribution of the 30 dynamically and evolutionally correlated positions and site-directed mutagenesis. (A) Mapping of the 30 highly correlated positions on AK3.Lysine is represented by CPK models. The residues involved in lysine binding are colored in blue. The residues in the catalytic domain are shown in cyan. Other residues inregulatory domain are colored in red. (B) Inhibition profiles of AK3 and its mutants by lysine. The activities were displayed as relative activities normalized by the specificactivities without inhibition. The specific activities without inhibition for wildtype, M251P, T253R, R305A, S315A, H320A, F329R, I337P, S338L, V339A, E346R, R416A, C428Rare: 352 ± 32, 104 ± 12, 334 ± 23, 348 ± 26, 484 ± 34, 339 ± 24, 302 ± 14, 335 ± 26, 338 ± 24, 358 ± 22, 349 ± 23, 478 ± 36, 98 ± 12 mU mg−1 protein, respectively. Data representmean values and standard deviation from three assays. (C) Inhibition profiles of AK1–HD1 and its mutants by threonine. The specific activities without inhibition for wildtype,I344P, Q351A, N424A, N426A, I427P, E436A are: 335 ± 13, 218 ± 22, 209 ± 28, 242 ± 23, 212 ± 21, 248 ± 17, 331 ± 24 mU mg−1 protein, respectively. (For interpretation of thereferences to color in this figure legend, the reader is referred to the web version of the article.)

techn

acttetbia

l(2MmtV(torl

bmtsoaMtAsCto

itaeotn

3a

tttsomftIsosptvod

Z. Chen et al. / Journal of Bio

nd T253–R305. They are especially important for the allostericommunication between sector B and sector C as discussed in Sec-ion 3.1 because their dynamic correlations only exist in the MDrajectory of AK3 with lysine. Destroying the pair interactions, forxample, by mutation of Met251 to Pro or Arg305 to Ala to reducehe van der Waals interaction significantly, release the lysine inhi-ition (Fig. 4B). This is also verified by the mutation of T253R which

ntroduces a repulse interaction with Arg305 and thus destroys thellosteric regulation by lysine (Fig. 4B).

The second group contains the highly correlated residues withinysine binding sites. Mutations of these residues such as M318IMiyata et al., 2001), G323D (Kikuchi et al., 1999; Miyata et al.,001), L325F (Miyata et al., 2001), S345L (Kikuchi et al., 1999;iyata et al., 2001) have been previously identified by randomutation and selection. In this work, we created several addi-

ional mutations belonging to this group of residues such as S338L,339A, E346R, all of which desensitize AK3 from lysine inhibition

Fig. 4B). Mutation of this group of residues, either by destroyinghe hydrogen bonds with lysine (e.g. M318I, S345L, S338L, V339A)r introducing steric hindrance (e.g. G323D, L325F) or electrostaticepulsion (e.g. E346R), can hinder lysine binding thus reduce theysine inhibition.

The third group of residues is not directly involved in lysineinding. They are sparsely located within regulatory domain, anday participate in the allosteric communication within regula-

ory domain. Several mutations belonging to this group of residuesuch as T344M, V347M, V349M, T352I and M417I have been previ-usly reported to be able to deregulate AK3 (Table 1) through thepproach of random mutation and selection (Kikuchi et al., 1999;iyata et al., 2001). We have constructed several additional muta-

ions including S315A, H320A, F329R, I337P, R416A and C428R.ll of the mutations greatly reduced lysine inhibition (Fig. 4B). Ithould be mentioned that several residues such as R416, M417 and428 are located far from lysine binding sites and the identifica-ion of these residues cannot be achieved without application ofur integrated approach.

The enzyme activities of most our constructed mutants are sim-lar to that of wildtype except M251P and C428R indicating thathese two residues also related to the catalysis (the specific enzymectivities of the wildtype and mutants are shown in the figure leg-nd of Fig. 4). Mutations of other residues close to but not belong tour predicted ones such as Asn308 to Ala, Ser321 to Ala, or Glu411o Ala, however, do not change the inhibition profiles of AK3 (dataot shown) showing that our prediction is highly efficient.

.4. Applicability of the identified network to otherspartokinases

To verify whether the identified residue network is also func-ional in other members of the aspartokinase family, we mappedhe corresponding positions of AK3 into the sequence of AK1–HD1hrough sequence alignment (Table 1). AK1–HD1 shows a lowequence identity with AK3 and is allosterically inhibited by thre-nine. The crystal structure of AK1–HD1 is not available. The onlyutation (S345F) for threonine-insensitive AK1–HD1 from E. coli

ound in the literature (Lee et al., 2003) is located within our iden-ified network. Other six mutations, I344P, Q351A, N424A, N426A,427P and E346A were created in this study. All of these mutationsuccessfully reduced the feedback inhibition of AK1–HD1 by thre-nine without significant change of enzyme activity (Fig. 4C, thepecific enzyme activities are shown in the figure legend). We sup-ose that the predicted positions in this study are also applicable

o other member of aspartokinase family as evolutionary conser-ation is imposed in our integrated approach. Actually, applicationf the identified residue interaction network has enabled us toeregulate the allosteric inhibition of aspartokinase by lysine and

ology 154 (2011) 248– 254 253

threonine in Corynebacterium glutamicum for high lysine produc-tion (Chen et al., 2011).

4. Discussion

Allosteric regulation is a complex process which has not beencompletely clarified presently. It has been proposed that only a fewresidues play the essential roles during the allostery process (Tsaiand Nussinov, 2009). Identify these key residues represent the firststep toward the rational engineering of enzyme allostery for indus-trial strain development. This study intends to introduce a fast andefficient approach integrating co-evolutionary analysis of familyproteins and molecular dynamic simulation of one selected pro-tein member to identify a key and well-conserved residue–residueinteraction network which is responsible for the allosteric regula-tion in this protein family. The combination of SCA and MD has beenshown to be especially useful for identification of amino acid pairsessential for catalysis (Estabrook et al., 2005). However, so far thereis no application of this approach to solve the problem of enzymeallostery. Here, we demonstrated the successful implementation ofthis method to deregulate the allosteric inhibition of an importantindustrial enzyme for amino acid production – aspartokinase.

Because of the importance of allosteric regulation for cellularmetabolism, large efforts have been made to reengineer allostericenzymes in industrial strain development (Chen et al., 2010;Fastrez, 2009). The traditional random mutation and selectionmethods or more recently the approach of directed evolution havebeen routinely used for such purposes. However, these methodsneed to construct libraries of tens of thousands up to millionsof mutants to find an effective mutation; hence its application islimited by the availability of efficient screening methods andquite often only limited points were identified (Kikuchi et al.,1999; Miyata et al., 2001). Some recent studies tried to rationallyreengineer protein allostery based on the structure of crystal-lized proteins (Zhang et al., 2009). Nevertheless, these methodsare strongly limited by the availability of protein structures. Evenmore, the previous paradigm of rational design often excludesresidues that are far away from the effector binding sites whichmight, however, significantly influence the properties of pro-tein. This is due to the difficulty to link the static proteinstructure to its function or regulation which are dynamic pro-cesses. The approach described herein uses MD simulation tocorrelate protein motions to its regulation and co-evolutionalinformation to further differentiate the most important residues.It thus systematically analyzes the residues’ interaction net-work within the whole sequence space. As indicated in theresults, not only the residues directly involved in lysine bindingbut also others could affect the allostery process. The latter mayplay important roles for the transduction of allosteric signal. Muta-tion of these residues that are far from the effector binding sites,such as M251 and T253 which are even not located within the regu-latory domain of AK3, can also strongly change the inhibition profileof AK3. It thus provides more solutions for rational design whichis especially important from an application point of view when thefocused residues have already been patented.

With the fast development of structure biology, we can nowaccess to large number of protein structures. However, most casesthere are only limited protein structures within the same proteinfamily. In other hand, the amino acid sequences and the inhibi-tion profiles could be varied significantly during the evolution.For example, there exist many different isozymes of aspartokinase

which may be inhibited by lysine or threonine alone or their con-certed action. Thus, traditionally it is difficult to directly transferthe dynamic information of allsotery from one protein member toanother. Since co-evolution weight is imposed in this approach to

2 techn

efiosacwhidc

R

B

B

B

C

C

C

D

D

D

54 Z. Chen et al. / Journal of Bio

xtract the most conserved residue–residue correlation, the identi-ed network has been shown to be able to guide the reengineeringf other members of protein family, for example, AK1–HD1 of thistudy which has no crystal structure available and the sequencend the inhibition profiles is different to AK3. The same approachan also be applied to other allosteric protein families which areidespread especially in pathways of amino acid synthesis. Weave used the same method to successfully deregulate the allosteric

nhibition of phosphoenolpyruvate carboxylase and homoserineehydrogenase from C. glutamicum, which still have no availablerystal structure (Chen et al., in preparation).

eferences

earer, C.F., Neet, K.E., 1978. Threonine inhibition of the aspartokinase–homoserinedehydrogenase I of Escherichia coli. A slow transient and cooperativity of inhibi-tion of the aspartokinase activity. Biochemistry 17, 3523–3530.

erendsen, H.J.C., Postma, J.P.M., van Gunsteren, W.F., DiNola, A., Haak, J.R., 1984.Molecular dynamics with coupling to an external bath. J. Chem. Phys. 81,3684–3690.

lack, S., Wright, N.G., 1954. (-Aspartate kinase and (-aspartyl phosphate. J. Biol.Chem. 213, 27–38.

hen, Z., Meyer, W.Q., Rappert, S., Sun, J.B., Zeng, A.P., 2011. Co-evolutionaryanalysis enables rational deregulation of allosteric enzyme inhibition inCorynebacterium glutamicum for lysine production. Appl. Environ. Microbiol.,doi:10.1128/AEM.02912-10.

hen, Z., Wilmanns, M., Zeng, A.P., 2010. Structural synthetic biotechnology: frommolecular structure to predictable design for industrial strain development.Trends Biotechnol. 28, 534–542.

hipman, D.M., Shaanan, B., 2001. The ACT domain family. Curr. Opin. Struct. Biol.11, 694–700.

arden, T., York, D., Pedersen, L., 1993. Particle mesh Ewald: An N·log (N) methodfor Ewald sums in large systems. J. Chem. Phys. 98, 10089–10092.

oolittle, R.F., 1996. Computer Methods for Macromolecular Sequence Analysis, vol.

266. Academic Press, San Diego, CA, USA.

uan, Y., Wu, C., Chowdhury, S., Lee, M.C., Xiong, G., Zhang, W., Yang, R., Cieplak, P.,Luo, R., Lee, T., Caldwell, J., Wang, J., Kollman, P., 2003. A point-charge force fieldfor molecular mechanics simulations of proteins based on condensed-phasequantum mechanical calculations. J. Comput. Chem. 24, 1999–2012.

ology 154 (2011) 248– 254

Edgar, R.C., 2004. MUSCLE: multiple sequence alignment with high accuracy andhigh throughput. Nucleic Acids Res. 32, 1792–1797.

Estabrook, R.A., Luo, J., Purdy, M.M., Sharma, V., Weakliem, P., Bruice, T.C., Reich, N.O.,2005. Statistical coevolution analysis and molecular dynamics: identification ofamino acid pairs essential for catalysis. Proc. Natl. Acad. Sci. U.S.A. 102, 994–999.

Fastrez, J., 2009. Engineering allosteric regulation into biological catalysts. Chem-biochem 10, 2824–2835.

Halabi, N., Rivoire, O., Leibler, S., Ranganathan, R., 2009. Protein sectors: evolutionaryunits of three-dimensional structure. Cell 138, 774–786.

Jorgensen, W., 1981. Transferable intermolecular potential functions for water,alcohols, and ethers, application to liquid water. J. Am. Chem. Soc. 103, 335–340.

Kikuchi, Y., Kojima, H., Tanaka, T., 1999. Mutational analysis of the feedback sitesof lysine-sensitive aspartokinase of Esherichia coli. FEMS Microbiol. Lett. 173,211–215.

Kotaka, M., Ren, J., Lockyer, M., Hawkins, A.R., Stammers, D.K., 2006. Structures ofR- and T-state Esherichia coli aspartokinase III. Mechanisms of the allosterictransition and inhibition by lysine. J. Biol. Chem. 281, 31544–31552.

Lee, J.H., Lee, D.E., Lee, B.U., Kim, H.S., 2003. Global analyses of transcriptomes andproteomes of a parent strain and an L-threonine-overproducing mutant strain.J. Bacteriol. 185, 5442–5451.

Lockless, S.W., Ranganathan, R., 1999. Evolutionarily conserved pathways of ener-getic connectivity in protein families. Science 286, 295–299.

Miyata, Y.O., Kojima, H., Sano, K., 2001. Mutation analysis of the feedback inhibi-tion site of aspartokinase III of Esherichia coli K-12 and its use in L-threonineproduction. Biosci. Biotechnol. Biochem. 65, 1149–1154.

Ryckaert, J.P., Ciccotti, G., Berendsen, J.C., 1977. Numerical integration of the carte-sian equations of motion of a system with constraints: molecular dynamics ofn-alkanes. J. Comput. Phys. 23, 327–341.

Smock, R.G., Gierasch, L.M., 2009. Sending signals dynamically. Science 324,198–203.

Suel, G.M., Lockless, S.W., Wall, M.A., Ranganathan, R., 2003. Evolutionarily con-served networks of residues mediate allosteric communication in proteins. Nat.Struct. Biol. 10, 59–69.

Tsai, C.J., Sol, A., Nussinov, R., 2009. Protein allostery, signal transmission and dynam-ics: a classification scheme of allosteric mechanisms. Mol. Biosyst. 5, 207–216.

Yoshida, A., Tomita, T., Kurihara, T., Fushinobu, S., Kuzuyama, T., Nishiyama, M., 2007.

Structural Insight into concerted inhibition of alpha 2 beta 2-type aspartatekinase from Corynebacterium glutamicum. J. Mol. Biol. 368, 521–536.

Zhang, G., Chen, J.A., Tanaka, T., 2009. Deregulation of allosteric response of lacto-coccus lactis prolidase and its effects on enzyme activity. Biochim. Biophys. Acta,968–975.