www.elsevier.com/locate/yprep

Protein Expression and Purification 55 (2007) 166–174

Expression in Escherichia coli, refolding and crystallizationof Aspergillus niger feruloyl esterase A using a serial factorial approach

Isabelle Benoit a, Bruno Coutard b, Rachid Oubelaid b,Marcel Asther a, Christophe Bignon b,*

a UMR 1163 INRA de Biotechnologie des Champignons Filamenteux, IFR86-BAIM, Universite de Provence et de la Mediterranee, ESIL,

163 avenue de Luminy CP 925, 13288 Marseille cedex 09, Franceb Architecture et Fonction des Macromolecules Biologiques, UMR 6098, CNRS et universite d’Aix-Marseille I et II, 163 avenue de Luminy,

13288 Marseille cedex 09, France

Received 20 February 2007, and in revised form 23 March 2007Available online 10 April 2007

Abstract

Hydrolysis of plant biomass is achieved by the combined action of enzymes secreted by microorganisms and directed against the back-bone and the side chains of plant cell wall polysaccharides. Among side chains degrading enzymes, the feruloyl esterase A (FAEA) spe-cifically removes feruloyl residues. Thus, FAEA has potential applications in a wide range of industrial processes such as paper bleachingor bio-ethanol production. To gain insight into FAEA hydrolysis activity, we solved its crystal structure. In this paper, we report how theuse of four consecutive factorial approaches (two incomplete factorials, one sparse matrix, and one full factorial) allowed expressing inEscherichia coli, refolding and then crystallizing Aspergillus niger FAEA in 6 weeks. Culture conditions providing the highest expressionlevel were determined using an incomplete factorial approach made of 12 combinations of four E. coli strains, three culture media andthree temperatures (full factorial: 36 combinations). Aspergillus niger FAEA was expressed in the form of inclusion bodies. These weredissolved using a chaotropic agent, and the protein was purified by affinity chromatography on Ni column under denaturing conditions.A suitable buffer for refolding the protein eluted from the Ni column was found using a second incomplete factorial approach made of 96buffers (full factorial: 3840 combinations). After refolding, the enzyme was further purified by gel filtration, and then crystallized follow-ing a standard protocol: initial crystallization conditions were found using commercial crystallization screens based on a sparse matrix.Crystals were then optimized using a full factorial screen.� 2007 Elsevier Inc. All rights reserved.

Keywords: Full factorial; Incomplete factorial; Sparse matrix; Refolding screening; Protein expression; Crystallization

Introduction

In the pre-genomics era, protein producers had no otherchoice but to use the trial and error method. This methodin addition to being generally performed by hand, had themajor drawback that only one parameter could be alteredat a time which was both time-consuming and labor-intensive.

1046-5928/$ - see front matter � 2007 Elsevier Inc. All rights reserved.

doi:10.1016/j.pep.2007.04.001

* Corresponding author. Fax: +33 491 26 67 20.E-mail address: bignon@afmb.univ-mrs.fr (C. Bignon).

The urgent and quantitative demand of protein struc-tures by structural genomics (SG)1 created optimalconditions for a technological jump consisting of twocomplementary halves: the use of robots for processingnumerous samples in parallel, and the set up of screeningtests for assessing the expression [1,2], solubility [3,4] andcrystallization [5,6] of recombinant proteins (for recent

1 Abbreviations used: SG, Structural Genomics; FAEA, feruloyl esteraseA; FF, full factorial; IF, incomplete factorial; SM, sparse matrix; E. coli,

Escherichia coli; A. niger, Aspergillus niger; LB, Luria-Bertani; TB, turbobroth; SB, superior broth.

I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174 167

reviews on these topics, see [7,8, whole issues]). Althoughthese tests were set up independently from each other forsolving unrelated problems, they have in common the com-bined use of several variables.

One way to combine different states of different vari-ables is to use a factorial approach. The factorialapproach of an experiment is defined by the combinationof different states of different variables used in that exper-iment. A full factorial (FF) is made of all the combina-tions of all the states of all the variables. Since thenumber of experimental points of a FF increases rapidlywith the number of variables and states, mathematical(incomplete factorial (IF)) and empirical (sparse matrix(SM)) bias have been found to reduce the number ofexperimental points while retaining the statistical signifi-cance of the full factorial.

The fraction of FF combinations constituting an IF ful-fills two conditions: (i) the selected combinations are evenlysampled throughout the combination space; (ii) all two-variable interactions are sampled as uniformly as possible.In a SM, the factorial incompleteness is obtained via anintentional bias towards experimental conditions that haveworked before. Therefore, in contrast with IF the statisticalspace of a SM is not uniform. Rather, the space density(the number of experimental points) is higher in the vicinityof the ideal solution. An illustration of these three kinds offactorials can be found at: http://www-structmed.cimr.cam.ac.uk/Course/Crystals/Theory/matrix_screens.html#sparse.

Optimization of crystallization conditions has alwaysfollowed a FF. In the late 70s, Carter and Carter [9] pro-posed using an IF to sample the crystallization space. Lateron, Jancarick and Kim [10] reached the same goal bymeans of a SM. Since then, commercial crystallization kitsintended for defining initial crystallization conditions andbased on a SM approach and on published crystallizationdata have been launched. In line with the work of Carterand Carter, Audic et al. [11] proposed an on line freeware(SAmBA) to help sampling crystallization conditions fol-lowing an IF approach. The use of SM [12] and of IF[13,14] was also reported for screening the refolding ofinclusion bodies. More recently, Abergel et al. usedSAmBA to set up an IF made of 12 experimental points(out of the 36 of the corresponding FF) for finding the bestconditions for expressing recombinant proteins in Esche-

richia coli (E. coli) [1].In early SG programs, we used in a trial and error

approach the same variables as those used in IF by Abergelet al. [1] for screening expression conditions of recombinantproteins in E. coli [6]. Shortly after switching to the facto-rial approach, we compared the time spent and the numberof positive results obtained with either method. Althoughthe same variables and states were used, the comparisonwas unambiguously in favor of the factorial approach.

Since then, every step performed in our laboratory fromprotein expression to crystallization is based on a full fac-torial, an incomplete factorial or a sparse matrix. There-fore, we consider the factorial approach as a powerful

tool not only for crystallizing but also for producingrecombinant proteins.

Type A feruloyl esterase (FAEA) is a protein secreted bycellulolytic microorganisms. It participates in the degrada-tion of plant cell wall polysaccharides by removing feruloylresidues from hemicellulose and pectin. This enzyme hastherefore a wide range of potential industrial applications.Enzymes are generally used in industry under non physio-logical conditions, particularly in terms of incubation tem-peratures. In a previous paper, we have evaluated therespective importance of protein folding and glycosylationin the thermal stability of Aspergillus niger (A. niger)FAEA. In an attempt to find structural properties support-ing the observed differences, we solved the 3D structure ofglycosylated and non glycosylated forms of A. niger FAEA[15]. To solve the 3D structure of the non glycosylated pro-tein, A. niger FAEA was expressed in E. coli and therecombinant protein was used to produce crystals.

In the present paper, we report how the use of four consec-utive factorials (two incomplete factorials, one sparsematrix, and one full factorial) allowed expressing in E. coli,refolding and then crystallizing A. niger FAEA in 6 weeks.

Materials and methods

FAEA expression and purification

PCR amplification of A. niger FAEA coding sequence,and sub-cloning of the PCR product into a prokaryoticexpression vector encoding a 6His-tag at the N terminusof the recombinant protein have been reported already [15].

FAEA expression was assessed at analytical scale (4 mlcultures) using an IF [1] combining four E. coli strains(BL21(DE3)pLysS, Rosetta(DE3)pLysS, Origami(DE3)pLysS, and C41(DE3)pRos [Novagen]), three temperatures(17, 20, and 37 �C), and three culture media (superior broth(SB), turbo broth (TB) [AthenaES], and 2TY, [DIFCO]).The 12 IF combinations were obtained out of the 36 com-binations of the corresponding full factorial (4 strains · 3temperatures · 3 culture media) using SAmBA software(http://igs-server.cnrs-mrs.fr/samba/). The detaileddescription of each combination is provided in Fig. 2b,lines 1–3). C41(DE3)pRos is C41(DE3) strain transformedwith Rosetta(DE3)pLysS plasmid.

Transformed cells were selected on agar plates contain-ing 100 lg/ml ampicillin and 34 lg/ml chloramphenicol(AC). Luria-Bertani (LB)-AC medium (4 ml/well in 24-welldeepwells) was seeded with a single colony of each trans-formant, and pre-cultures were grown overnight at 37 �Cin a shaking incubator. The next day, 4 ml culture med-ium-AC (SB, TB, or 2YT) corresponding to the 12 IF con-ditions listed in Fig. 1b (lines 1–3) were individually seededwith 150 ll of each overnight pre-culture, and grown at37 �C. At 0.5 < OD600 < 1, expression was induced by addi-tion of 0.5 mM IPTG and allowed to proceed for 3 h at37 �C or overnight at 17 and 20 �C. When expression timehad elapsed, cells were spun at 4000g for 5 min, and the cell

IF1: 12 expression conditions

IF2: 96 refolding conditions

SM: 240 crystallization conditions

FF: 36 optimisation conditions

Fig. 1. Flowchart of the different factorials. IF, incomplete factorial; SM,sparse matrix; FF, full factorial.

168 I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174

pellet was re-suspended in 700 ll of lysis buffer (50 mMTris pH 8, 300 mM NaCl; 1 mM PMSF, and 0.25 mg/mllysozyme) per well. The cell suspension was frozen at�80 �C overnight. After thawing, MgSO4 (20 mM) andDNAse 1 (10 lg/ml) were added and the cell lysate wasincubated at 37 �C for 30 min (or until the lysate was notany more viscous). Total protein expression was assayedas previously described using DNAse-treated lysates andanti-6His antibodies in an automated dot–blot procedure[2]. A description of the trend analysis of the results is pro-vided in the results section.

The expression condition providing the highest expres-sion level was scaled up (2 L culture), and cells were pro-cessed as described above. After DNAse treatment, thecell lysate was spun for 30 min at 10000g. The supernatantwas discarded and the pellet containing FAEA inclusionbodies was washed three times with water. The washed pel-let was solubilized by shaking in 100 ml of 50 mM Tris pH8, 300 mM NaCl, 6 M guanidium hydrochloride, and therecombinant protein was purified by affinity chromatogra-phy on Ni column under denaturing conditions. For wash-ing (50 mM imidazole) and elution (250 mM imidazole)steps, guanidium hydrochloride was substituted with 8 Murea. FAEA eluted from the Ni column was dialyzedagainst 50 mM Tris, pH 8, 300 mM NaCl, 8 M urea toremove imidazole, and concentrated to 5 mg/ml. The con-centrated protein was used in a 96-well refolding screeningas described [3] except that protein precipitates weredetected by measuring the OD at 405 nm instead of350 nm. Preparative refolding was performed using theremaining material of the 2 L culture (12 ml at 5 mg/ml),and the best refolding buffer deduced by trend analysis(see results section) from the results of the refolding screen-ing. In practice, denatured FAEA was drop by drop dilutedinto 200 ml of pre-chilled refolding buffer (50 mM CHES,pH 9, 10 mM b-mercaptoethanol) with constant stirring.After concentration to 6 ml, the refolded FAEA was fur-ther purified by size exclusion chromatography on Super-dex 200 (Amersham) using 50 mM CHES pH 9 asrunning buffer. Eluted fractions were pooled, and the pro-tein concentrated to 10 mg/ml.

Enzymatic activity assay

The esterase activity was assayed by a continuous spec-trophotometric method as previously described [16] using

methyl ferulate (MFA) (Apin chemical Ltd., Oxon, UK)as substrate, and a microplate reader (Tecan Sunrise).Activities were expressed in nanokatal/ml. One nanokatalwas defined as the amount of enzyme catalyzing the releaseof 1 nmol of ferulic acid per second.

Crystallization

Initial crystallization conditions were obtained using anautomated sitting drop technique and 96-well plates (Gre-iner). In addition to the crystallization buffer contained inthe tank, each well of these plates can accommodate threedrops, and so three protein concentrations were used perwell. A nano-drop 8-needle robot (Cartesian Inc.) was usedfor initial crystallization trials [5]. Three commercialscreens (240 crystallization buffers) were used: Stura Foot-print screens, Structure screens 1 and 2 (Molecular Dimen-sions Limited), and SM1 (Nextal). In practice, the robotloaded 0.1 ll of each crystallization buffer, and then 0.1,0.2, or 0.3 ll of a 10 mg/ml protein solution (in 10 mMCHES pH 9). After sealing, plates were stored at 20 �Cand visually inspected weekly.

Optimization of crystallization conditions was per-formed manually using the hanging micro-drop technique.Crystals obtained in initial crystallization trials wereimproved using initial conditions as a starting point, anda FF made of 36 combinations obtained by combiningsix pH values (from 9.5 to 11 using CAPS buffer) withsix precipitant concentrations (from 1 to 2 M (NH4)2SO4).

Results

FAEA expression and purification

The general strategy is summarized as a flowchart inFig. 1.

FAEA expression was evaluated under different cultureconditions following an IF approach (IF1) made of 12experiments (the full factorial consisted of 36 experiments).When Coomassie blue staining was used, polyacryamidegel electrophoresis (SDS–PAGE) analysis revealed FAEAexpression in the insoluble fraction of cell lysates only(Fig. 2a). As previously observed [15], non glycosylatedHis-tagged FAEA migrated as a molecular species slightlylarger (�35 kDa) than predicted by in silico translation ofthe coding sequence (28.5 kDa). Since the actual molecularmass was determined to be 28.55 kDa by mass spectrome-try analysis [15], this discrepancy is likely to be an electro-phoresis artifact. To quantify expression levels moreaccurately, a dot–blot analysis was performed using wholecell lysates and an anti-His antibody. Raw data are pre-sented in Fig. 2b, lines 4 and 5. The highest expression level(29679 light units, Fig. 2b, line 5) was obtained in an exper-iment using Rosetta(DE3)pLysS cells at 17 �C in TB med-ium, and was 11.3 times higher than the lowest expressionlevel (2629 light units) obtained in an experiment usingBL21(DE3)pLysS cells at 37 �C in 2YT medium.

B R O C B R O M C B R O C M

< 20°C > < 37°C > < 17°C >

Sol

uble

Inso

lubl

e

a

M B R O C B R O C M B R O C

2YT 2YT SB TB TB SB TB 2YT SB TB 2YT SB

37 20 17

2629

9877

3997

6739

5037

1983

4

1990

3

7032

1107

5

2967

913

697

3104

b123

4

5

B R O C B R O C B R O C

Volume (ml)

O

D28

0

0

0. 25

0 50 100 150 200 250 300 350 400

f GF M

Washing

Elution

OD

280

Time

2.5

0

e M Ni

0

2

4

6

Ligh

t int

ensi

ty (x

104 )

B R O C 2YT TB SB 37 20 17

c

strain medium temperature

d

H12

Microplate w

ells

OD405 nanokatal / ml0 0.5 1

A1

(1.96)

pI (4.66)

pH

4

5

6

7

8

9

8*

Fig. 2. FAEA expression and purification. (a) Ten microliters of soluble or insoluble fraction of cell lysate from each of the 12 combinations of IF1 were run ona 12% polyacrylamide gel under reducing and denaturing conditions. The expression pattern was revealed by Coomassie blue staining. M, molecular weighmarkers (116, 66.2, 45, 35, 25 kDa). B, R, O, C refer to the four E. coli strains (for a detailed description of each combination, see (b) lines 1–3). (b) Dot–bloanalysis of FAEA expression under the 12 conditions of IF1 using anti-6His antibodies. Line 1, expression temperatures; line 2, E. coli strains: BBL21(DE3)pLysS; R, Rosetta(DE3)pLysS; O, Origami(DE3)pLysS; C, C41pRos; line 3, culture media: SB, superior broth; TB, turbo broth; line 4, raw data(grey scale); line 5, same as line 4 except that raw data are provided as the amount of light signal (in arbitrary units) emitted by the dot–blot and recorded byCCD camera [2]). (c) Statistical analysis of the raw data. For each variable (E. coli strain, culture medium, and expression temperature), the best state isindicated by a black triangle at the top of the corresponding histogram. (d) Refolding screening. The 96 refolding buffers of the refolding plate are ordered fromtop (well A1) to bottom (well H12). Protein precipitate formation (OD405) and enzymatic activity (nanokatal/ml) are reported as grey and black histogramsrespectively. Refolding buffer pH are indicated on the right; *contains minichaperones and redox components (see [3] for details). The arrow locates FAEA pI

on the pH scale. (e) Washing and elution profiles of Ni affinity column. The dotted line is the imidazole step gradient. Inset: Coomassie blue stainedpolyacrylamide gel in which an aliquot of the Ni elution fraction pool (Ni) was run along with molecular weight markers (M: 97, 66, 45, 30, 20.1 kDa from topto bottom). (f) Size exclusion chromatography purification of refolded FAEA. Inset: Coomassie blue stained polyacrylamide gel in which an aliquot of thefraction pool of the peak eluting at 200 ml (GF) was run along with molecular weight markers (M: 116, 66.2, 45, 35, 25, 18.4 kDa from top to bottom).

I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174 169

tt,

,

170 I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174

The variable states combination providing the highestexpression was determined by means of a trend analysis(Fig. 2c) [1], using the amount of light emitted by eachexperimental point (Fig. 2b, line 5). In practice, the beststate for each of the 3 variables (strain, culture medium,and temperature) was determined by summing the lightvalues of all the experiments using the considered variablestate. For example, the score for state ‘‘B’’ (BL21(DE3)-pLysS strain) of the variable ‘‘strain’’, was 18741. Thisnumber resulted from the addition of light values 2629,5037, and 11075 (Fig. 2b, line 5), which were the three ofthe 12 IF1 experiments using state ‘‘B’’. The same calcula-tion was done for the 10 variable states (B, R, O, C, 2YT,TB, SB, 37, 20, and 17). The result of this calculation isreported in the form of histograms in Fig. 2c. The beststate for each variable is that having the highest score.Rosetta(DE3)pLysS strain, TB medium and 17 �C werethe best states for strain, culture medium and temperaturevariables, respectively (indicated by a black triangle at thetop of the corresponding histogram). Since individuallydetermined benefit of each of these states has the virtueof being cumulative with that of the other variables, thesethree best states also constituted the combination the trendanalysis predicted to provide the highest expression level ofthe 36 possible combinations of the FF. Interestingly, thiscombination happened to be one of the 12 experiments ofthe incomplete factorial, and effectively had the highestIF1 score (Fig. 2b, lines 4 and 5). This coincidenceprovided an immediate validation of the trend analysisprediction.

A large scale production was performed under the abovedefined expression conditions. Out of 2 L culture, �60 mgFAEA were purified from inclusion bodies by Ni affinitychromatography under denaturing conditions (Table 1).Fig. 2e shows the chromatogram of the Ni affinity column(washing and elution steps), and a SDS–PAGE analysis ofthe pooled elution fractions. This single affinity purificationstep provided pure enough FAEA to be directly used inrefolding screening.

Table 1FAEA purification scheme

Purification step Proteine (mg) Total activity (nanokata

Inclusion bodiesa 105 0IMACb 60 0Refoldingc 55 1045SECd 30 651

The values are for a 2 L culture.a Protein amount and activity were assayed before dissolving inclusion bodib Immobilized-metal affinity chromatography; protein amount and activity w

concentration.c Protein amount and activity were assayed just after refolding before proted Size exclusion chromatography; protein amount and activity were ass

concentration.e Estimated by SDS–PAGE by comparison with known quantities of pro

substrate.

The best buffer for refolding Ni-purified urea-denaturedFAEA was found using an automated refolding screening[3]. The screen is composed of 96 buffers that were definedby IF approach (the full factorial was made of 3840 com-binations). The refolding test is based on the measure at405 nm of the light scattered by precipitated proteins. Ifthe denatured protein refolds upon dilution in a givenrefolding buffer, it remains soluble and OD405 does notincrease. Conversely, if the protein does not fold upon dilu-tion in the refolding buffer, it precipitates and OD405

increases. The result of FAEA refolding screening isreported in Fig. 2d. After dilution in the 96 refolding buf-fers, the protein precipitated in acidic buffers (pH 6 6) andremained soluble in buffers with pH P 6, a result in linewith its acidic pI (4.66). The enzymatic activity paralleledthe solubility and increased with pH. To find the most suit-able refolding buffer for both solubility/activity and com-patibility with downstream crystallization steps, a trendanalysis of the effect of the different variables of the refold-ing screen on FAEA solubility and activity was performed(see Supplementary material online). This analysis indi-cated that pH was by far the most important variable, withpH 9 providing both the highest solubility and activity.Other variables such as ionic strength and detergent didhave some influence on either solubility or activity, butnone had a positive effect on both and some (glycerol,detergents) could interfere with crystallization. FAEA hasseven cysteins and three disulfide bonds, so one cystein res-idue remains free after disulfide bridges have formed. Toavoid inter molecular disulfide bond formation duringrefolding, a low concentration of b-mercaptoethanol(10 mM) was used although the reducing agent had noinfluence on FAEA solubility (see Supplementary materialonline).

Ideally, proteins to crystallize should be dissolved in thesimplest possible buffer to avoid interfering with crystalli-zation buffers. Based on the above considerations, a con-sensus refolding buffer made of 50 mM CHES pH 9 and10 mM b-mercaptoethanol was considered to provide the

l) Specific activity (nanokatal/mg) Activity yield (%)

0 00 019 10021.7 62

es in guanidinium buffer.ere assayed in the pooled fractions eluted from the column before protein

in concentration.ayed in the pooled fractions eluted from the column, before protein

teins of known size. The enzymatic activity was measured using MFA

I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174 171

best balance between solubility/activity and compatibilitywith crystallization.

Ni-purified urea-denatured FAEA from two liter culturewas refolded as described in Materials and methods usingthat buffer. After refolding, the protein was further purifiedby size exclusion chromatography. We took advantage ofthis step to get rid of the reducing agent. As predicted bythe trend analysis (see Supplementary material online), thisdid not result in protein precipitation. The refolded proteinmigrated as two peaks (Fig. 2e). Linear regression analysisusing retention volumes of molecular weight markers ofknown size indicated aggregation states of 5 (elution vol-ume = 200 ml) and 32 (elution volume = 120 ml) mole-cules, respectively. Although both peaks containedidentical amounts of enzymatic activity (not illustrated),only fractions of the peak containing the pentamericenzyme were used in subsequent steps because high molec-ular weight polymers could hinder crystallogenesis. Out of�60 mg FAEA eluted from the Ni column, �30 mg pureFAEA were recovered after gel filtration (table 1). Pooledfractions were concentrated to 10 mg/ml, and protein pur-ity was checked by SDS–PAGE (Fig. 2f). This batch ofFAEA was used to measure kinetic parameters before pro-ceeding with crystallization trials. The enzymatic propertiesof this refolded non glycosylated FAEA were identical tothose of glycosylated FAEA secreted by its natural host(A. niger) [15].

FAEA crystallization

Pure, soluble and functional FAEA was used in crystal-lization trials. Three commercial kits based on a SM designand amounting to a total of 240 crystallization buffers wereused for finding initial crystallization conditions.

As reported in Fig. 3a, different kinds of results wereobtained. Phase separations and protein precipitates wereobserved in a wide range of pH (4.5–9.5). By contrast, crys-tals were produced at very alkaline pH only (10.5). The twoconditions providing crystals shared the same ionicstrength (0.2 M LiSO4), but used different precipitants([2 M (NH4)2SO4] or [1.2 M NaH2PO4/0.8 M K2HPO4]).Although morphologically identical (not illustrated), crys-tals grew in 1 week in the presence of [2 M (NH4)2SO4],and in more than 6 months in the presence of [1.2 MNaH2PO4/0.8 M K2HPO4]. Interestingly, 0.2 M LiSO4

was also found in conditions providing phase separationsand protein precipitates at lower pH. Similarly, 2 M(NH4)2SO4 produced protein precipitates at pH 7 and8.5. Thus, 0.1 M CAPS pH 10.5 was necessary for obtain-ing FAEA crystals. This was strengthened by the observa-tion that the same combination of ionic strength (0.2 MLiSO4) and precipitant (2 M (NH4)2SO4), produced a pro-tein precipitate at pH 7 and crystals at pH 10.5 (Fig. 3a,lines 8 and 13). On the other hand, since no crystal wasobserved in conditions made of 0.1 M CAPS pH 10.5and either [30%PEG 400] (SM1 screen condition 42) or[0.2 M NaCl/20%PEG 8000] (SM1 screen condition 77),

it could be said that 0.1 M CAPS pH 10.5 was not sufficientand that 0.2 M LiSO4 and 2 M (NH4)2SO4 were alsorequired for obtaining FAEA crystals. In conclusion,SM-based crystallization screening kits allowed defininginitial crystallization conditions in a single experimentmade of 240 experimental points.

Optimization of crystallization conditions was per-formed by combining six states for each of two variables.The variables were pH and precipitant. A pH scale from9.5 to 11 (0.3 pH unit increments), and precipitant concen-trations ranging from 1 to 2 M (NH4)2SO4 (0.2 M incre-ments) were tested simultaneously in a single FFexperiment (36 conditions). The result of this optimization(Fig. 3b) definitely highlighted the absolute dependency ofFAEA crystallization on high concentrations of precipitant(2 M (NH4)2SO4) whereas pH, which proved critical in ini-tial crystallization steps (Fig. 3a), influenced only crystalquality at this later stage. In practice, pH 11 was used inthe final crystallization condition because it provided bettercrystals than lower pH (see pictures of the crystals at thebottom of Fig. 3b). In conclusion, FAEA crystallizationconditions were optimized by tuning the pH and the preci-pitant concentration in a FF made of 36 experimentalpoints.

Crystals diffracting at 1.7 A were obtained in the opti-mized condition, which allowed solving FAEA 3D struc-ture [15].

Discussion

When over produced in its natural host, A. niger FAEAwas secreted as a functional protein in the culture medium[17] presumably because protein folding and glycosylationwere correctly carried out by the fungus. By contrast, whenE. coli was used as expression host post-translation (fold-ing, glycosylation) was not properly performed. Conse-quently, illegitimate inter-protein hydrophobicinteractions occurred, leading to the formation of inclusionbodies [15]. In addition, A. niger FAEA was expressed inE. coli at one tenth the level reached in A. niger [17]. Thus,producing a soluble form of A. niger FAEA in E. coli waschallenging and required using additional methods forscreening the expression and the refolding of the protein.This was conveniently achieved in the present study bymeans of two incomplete factorial approaches. Afterrefolding, the protein was further purified by size exclusionchromatography, and FAEA crystals were obtained usingsparse matrix-based commercial kits. Initial crystals wereoptimized using a full factorial, and optimized crystals wereused to solve the non glycosylated/refolded FAEA 3Dstructure [15].

As described in Results, the combination of strain/cul-ture medium/temperature that led to the highest expressionyields ([R–TB–17]) was correctly identified using IFscreening. This analysis can also be used to identify thatcombination of variable states that is predicted to havethe smallest expression yield. By doing so, we found

pH Ionic strength Precipitant1 4.5 (0.1M NaAc) 0.2M LiSO4 30% PEG 8000 2 6.5 (0.1M MES) 0.2M (NH4)2SO4 30% MME PEG 5000 3 8.5 (0.1M Tris) 0.2M LiSO4 30% PEG 4000 Ph

ase

sepa

ratio

n

4 9.5 (0.1M CHES) 0.2M NaCl 1.26M (NH4)2SO4

5 5.5 (0.1M NaAc) 12% MPEG 5000 6 5.5 (0.1M NaAc) 24% MPEG 5000 7 7 (0.1M Tris) 20% PEG 1000 8 7 (0.1M Tris) 0.2M LiSO4 2M (NH4)2SO4

9 8 (0.1M imidazole) 0.2M Zn(OAc)2 20% 1,4-butanediol 10 8 (0.1M imidazole) 0.2M Zn(OAc)2 2.5M NaCl

118.20 (0.1M

HEPES)40% MPEG 550

Prec

ipita

te

12 8.5 (0.1M Tris) 2M (NH4)2SO4

13 10.5 (0.1M CAPS) 0.2M LiSO4 2M (NH4)2SO4

Cry

stal

14 10.5 (0.1M CAPS) 0.2M LiSO41.2M NaH2PO4

0.8M K2HPO4

a The use of a sparse matrix for finding initial crystallization conditions provided the following results

b Use of a full factorial for finding optimal crystallization conditions

Final crystallization condition: 0.1M CAPS pH 11, 0.2M LiSO4, 2M (NH4)2SO4

1.7 Å diffracting crystals

Influence of pH on crystal quality

(NH

4)2S

O4

2M

1M

9.5 pH 11

X X X X X X

~ 24 hours*

Less than 48 hours*

1 week*

6 months*

Fig. 3. The consecutive use of a sparse matrix (a) and of a full factorial (b) led to diffraction-grade crystals. (a) Out of the 240 crystallization conditions ofthe sparse matrix screen, 14 provided results such as phase separations, protein precipitates or protein crystals (crystallization drops remained clear underthe remaining 226 crystallization conditions). The composition (pH, ionic strength, precipitant) of these 14 responsive conditions is reported here as atable. *Phase separation, protein precipitate, or crystals appeared under the indicated condition after this length of time has elapsed. (b) X denotes thepresence of crystals.

172 I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174

that expression conditions routinely used in many labora-tories for expressing recombinant proteins in E. coli

(BL21(DE3)pLysS strain , LB medium, 37 �C) would haveprobably provided expression results lower than thoseobtained using the worst IF1 condition. At the bottomof IF1 expression scale, [C–2YT–37] was predicted to bethe combination providing the lowest expression level(Fig. 2c). [C–2YT–37] was one of the 36 combinations ofthe full factorial, but was not one of the 12 combinationsof IF1. However, IF1 condition 1 ([B–2YT–37]) bestapproximated this worst combination in terms of bothcomposition (same medium, same temperature) andexpression results ([B–2YT–37] did provide the lowestFAEA expression level of IF1), which indirectly validatedthe trend analysis prediction. Using the best variable statescombination, 30 mg pure FAEA were obtained out of 2 Lculture. Since the amount of FAEA produced by [B–2YT–37]

was 11.3 times lower than the combination we have used,less than 2.7 mg pure FAEA would have been obtainedif [B–2YT–37] had been used. Interestingly, this lowestexpression level was obtained under expression conditionsused in many laboratories. BL21(DE3)pLysS strain is themost widely used E. coli strain for expressing recombinantcoding sequences borne by T7-based plasmids, and 37 �Cis the by default used temperature for expressing recombi-nant proteins in E. coli. Only the culture medium was oddin that LB is more often used than 2YT. As discussedbelow, LB is not a rich medium and does not allow celldensities as high as those obtained with 2YT to be reached.Therefore, although we did not perform the experiment,we speculate that the canonical [BL21(DE3)pLysS cells/LB medium/37 �C] combination would have providedan even lower expression level than the lowest obtainedwith IF1.

I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174 173

Expression temperatures, strains, and media used in IF1were essentially those used by Abergel et al. [1]. We slightlymodified the original protocol in light of our experience inSG prior to switching to factorial sampling [6], because SGprograms allowed reliable statistics to be made thanks tothe large number of recombinant proteins featuring theseprograms.

Temperatures

In the original paper [1], 25, 37, and 42 �C were used.Although not specified in the paper, we assume that 25and 42 �C were chosen for improving protein solubilityby respectively reducing E. coli metabolism and inducingheat shock proteins [18]. Escherichia coli can express pro-teins at temperatures as low as 15 �C, and protein solubilitycan be significantly improved by reducing the expressiontemperature from 25 to 17 �C. In some cases however, thistemperature drop resulted in a complete loss of expression.In those cases limiting the temperature reduction to 20 �Crestored expression while still improving recombinant pro-tein solubility. On the basis of these observations, we chose20 and 17 �C instead of 25 and 42 �C. Thirty-seven degreeCelsius was kept because it is the optimal E. coli growthtemperature.

Media

We used the same culture media as those of the originalpaper. The three media have the following features. 2YTresembles LB (Miller formulation) in that it contains thesame amount of sodium chloride and is not buffered. It ishowever richer than LB in casein digest (16 instead of10 g/L) and yeast extract (10 instead of 5 g/L). Thus,2YT can be described as ‘‘enriched LB’’. SB and TB arealso rich media, but in contrast with LB and 2YT theyare buffered (with potassium phosphate for TB, which alsoprovides a source of phosphate) to maintain the pH around7.2 throughout the culture. They also contain glucose (SB)or glycerol (TB) as carbon source. In some instances, wetried substituting these with other media, but the latterdid not provide any expression improvement over the ori-ginal set.

Strains

In contrast with Abergel et al. [1], we only use pLysSstrains in order to limit expression leakage which mayresult in poor expression when expressing toxic proteins.In addition, pLysS is borne by a plasmid conferringresistance to chloramphenicol. This allows a single ampi-cillin and chloramphenicol cocktail to be used in the 12experimental points of IF1. BL21(DE3)pLysS is theoriginal strain from which the other three derive [19].Rosetta(DE3)pLysS cells overproduce tRNAs that are inlimiting supply in E. coli. A. niger FAEA coding sequencecontains 3 CUA (Leu), 7 GGA (Gly) and 1 CCC (Pro)

codons. The corresponding tRNAs, which are scarce inE. coli, are supplied by Rosetta but not by the otherthree strains. This could explain the higher expressionlevel obtained with Rosetta (Fig. 2c). The cytosol of Ori-gami(DE3)pLysS was made more oxidative by geneticengineering than that of BL21(DE3)pLysS, allowingdisulfide bonds to be made without the need foraddressing recombinant proteins to the periplasm.C41(DE3)pLysS were selected upon their higher resistancethan BL21(DE3)pLysS to toxic proteins, particularlymembrane proteins [20]. We tried other commerciallyavailable E. coli strains, but these proved less effectiveand were not included in the final set.

The use of factorial approaches has been individuallyreported for expressing [1], refolding [6,12–14], or crystal-lizing [9,10] proteins. However, as far as we know thepresent study is the first example of a systematic use offactorial approaches at every step of a protein produc-tion and crystallization process. Because of their combi-natorial nature factorials allow the results of all theexperiments they are composed of to be obtained simul-taneously. This has two consequences. First, experimen-tal points serve as reciprocal internal negative andpositive controls. Second, reliable trend analysis can beperformed, and can be performed immediately. Thus,in contrast with the classical trial and error approach[6] there is no need to wait for the result of the firstexperiment to design the next one which saves a signifi-cant amount of time. For instance, only 6 weeks wererequired to go from PCR amplification of FAEA codingsequence to diffracting crystals.

Despite their combinatorial nature (as opposed toserial), this report suggests that organizing factorials as asuite might further increase their efficiency by cumulatingthe benefits of the time saved individually by each factorial.If this holds true, then it will be worth serially linking thehighest possible number of factorials. For instance, in theprocess of recombinant protein production factorialdesigns of tests for cell lysis [21,22] and for protein solubil-ity and stability [22] could valuably increment the currentflowchart (Fig. 1).

Acknowledgments

This research was supported by grant from the ConseilRegional Provence Cote-d’Azur, France and TembecS.A., France and by the French Research National Agency(National Research Program on Bioenergies). We thankGerlind Sulzenbacher for helpful advices and MarseilleNice Genopole for the high throughput crystallogenesisplateform.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.pep.2007.04.001.

174 I. Benoit et al. / Protein Expression and Purification 55 (2007) 166–174

References

[1] C. Abergel, B. Coutard, D. Byrne, S. Chenivesse, J.-B. Claude, C.Deregnaucourt, T. Fricaux, C. Gianesini-Boutreux, S. Jeudy, R.Lebrun, C. Maza, C. Notredame, O. Poirot, K. Suhre, M. Varagnol,J.M. Claverie, Structural genomics of highly conserved microbialgenes of unknown function in search of new antibacterial targets, J.Struct. Funct. Genomics 4 (2003) 141–157.

[2] R. Vincentelli, S. Canaan, J. Offant, C. Cambillau, C. Bignon,Automated expression and solubility screening of His-tagged proteinsin 96-well format, Anal. Biochem. 346 (2005) 77–84.

[3] R. Vincentelli, S. Canaan, V. Campanacci, C. Valencia, D. Maurin,F. Frassinetti, L. Scappucini-Calvo, Y. Bourne, C. Cambillau, C.Bignon, High-throughput automated refolding screening of inclusionbodies, Protein Sci. 13 (2004) 2782–2792.

[4] A. Geerlof, J. Brown, B. Coutard, M.-P. Egloff, F.J. Enguita, M.J.Fogg, R.J.C. Gilbert, M.R. Groves, A. Haouz, J.E. Nettleship, P.Nordlund, R.J. Owens, M. Ruff, S. Sainsbury, D.I. Sverguna, M.Wilmanns, The impact of protein characterization in structuralproteomics, Acta Cryst. D 62 (2006) 1125–1136.

[5] G. Sulzenbacher, A. Gruez, V. Roig-Zamboni, S. Spinelli, C.Valencia, F. Pagot, R. Vincentelli, C. Bignon, A. Salomoni, S. Grisel,D. Maurin, C. Huyghe, K. Johansson, A. Grassick, A. Roussel, Y.Bourne, S. Perrier, L. Miallau, P. Cantau, E. Blanc, M. Genevois, A.Grossi, A. Zenatti, V. Campanacci, C. Cambillau, A medium-throughput crystallization approach, Acta Crystallogr. D Biol.Crystallogr. 58 (2002) 2109–2115.

[6] R. Vincentelli, C. Bignon, A. Gruez, S. Canaan, G. Sulzenbacher, M.Tegoni, V. Campanacci, C. Cambillau, Medium-scale structuralgenomics: strategies for protein expression and crystallization, Acc.Chem. Res. 36 (2003) 165–172.

[7] Acta Crys. D 62 (2006) 1103–1285 (whole issue).[8] Methods Mol. Biol. 363 (2006) 1–276 (whole issue).[9] C.W. Carter, C.W. Carter, Protein crystallization using incomplete

factorial experiments, J. Biol. Chem. 254 (1979) 12219–12223.[10] J. Jancarik, S.-H.J. Kim, Sparse matrix sampling: a screening method

for crystallization of proteins, Appl. Cryst. 24 (1991) 409–411.[11] S. Audic, F. Lopez, J.M. Claverie, O. Poirot, C. Abergel, SAmBA: an

interactive software for optimizing the design of biological macro-molecules crystallization experiments, Proteins 29 (1997) 252–257.

[12] A. Hoffman, M. Tai, W. Wong, C.G. Glabe, A sparse matrix screento establish initial conditions for protein renaturation, Anal. Bio-chem. 230 (1995) 8–15.

[13] G. Chen, E. Gouaux, Overexpression of a glutamate receptor(GluR2) ligand binding domain in Escherichia coli: application of anovel protein folding screen, Proc. Natl. Acad. Sci. USA 94 (1997)13431–13436.

[14] M.S. Willis, J.K. Hogan, P. Prabhakar, X. Liu, K. Tsai, Y. Wei, T.Fox, Investigation of protein refolding using a fractional factorialscreen: a study of reagent effects and interactions, Protein Sci. 14(2005) 1818–1826.

[15] I. Benoit, M. Asther, G. Sulzenbacher, E. Record, L. Marmuse, G.Parsiegla, I. Gimbert, M. Asther, C. Bignon, Respective importanceof protein folding and glycosylation in the thermal stability ofrecombinant feruloyl esterase A, FEBS Lett. 580 (2006) 5815–5821.

[16] M.C. Ralet, C.B. Faulds, G. Williamson, J.F. Thibault, Degradationof feruloylated oligosaccharides from sugar-beet pulp and wheat branby ferulic acid esterases from Aspergillus niger, Carbohydr. Res. 263(1994) 257–269.

[17] E. Record, M. Asther, C. Sigoillot, S. Pages, P.J. Punt, M. Delattre,M. Haon, C.A.M.J.J. van den Hondel, J.-C. Sigoillot, L. Lesage-Meessen, M. Asther, Overproduction of the Aspergillus niger feruloylesterase for pulp bleaching application, Appl. Microbiol. Biotechnol.62 (2003) 349–355.

[18] F. Baneyx, M. Mujacic, Recombinant protein folding and misfoldingin Escherichia coli, Nat. Biotechnol. 22 (2004) 1399–1408.

[19] F.W. Studier, A.H. Rosenberg, J.J. Dunn, J.W. Dubendorff, Use ofT7 RNA polymerase to direct expression of cloned genes, MethodsEnzymol. 185 (1990) 60–89.

[20] B. Miroux, J.E. Walker, Over-production of proteins in Escherichia

coli: mutant hosts that allow synthesis of some membrane proteinsand globular proteins at high levels, J. Mol. Biol. 260 (1996) 289–298.

[21] N. Moreland, R. Ashton, H.M. Baker, I. Ivanovic, S. Patterson, V.L.Arcus, E.N. Baker, J.S. Lott, A flexible and economical medium-throughput strategy for protein production and crystallization, ActaCrystallogr. D 61 (2005) 1378–1385.

[22] T. Ducat, N. Declerck, T. Gostan, M. Kochoyan, H. Demene, Rapiddetermination of protein solubility and stability conditions for NMRstudies using incomplete factorial design, J. Biomol. NMR 34 (2006)137–151.