Crystallization of lysozyme: From vapor diffusion experimentsto batch crystallization in agitated ml-scale vessels
D. Hekmat*, D. Hebel, H. Schmid, D. Weuster-Botz
 Institute of Biochemical Engineering, Technische Universita¨ t Mu¨ nchen, Boltzmannstr. 15, 85748 Garching, Germany
Received 4 June 2007; received in revised form 13 July 2007; accepted 8 October 2007
Abstract
The crystallization of lysozyme was monitored in 20-
m
l sitting-drop vapor diffusion experiments and a quantitative phase diagram wasobtained. Then, batch crystallization of lysozyme in shaked 200-
m
l microtiter plates was investigated. It was observed that with rising agitationrates, the area of the nucleation zone was significantly reduced. Further batch crystallization experiments were performed (i) in 0.2–2-mlEppendorf tubes in a laboratory rotator, (ii) in 5-ml unbaffled shake flasks, and (iii) in 4-ml stirred baffled and unbaffled vessels. The crystal areadensitydistributionsofthestirred vesselswereclearly morenarrowcomparedtothe rotated Eppendorftubes.The crystalareadensitydistributionsoftheshakeflasksweresignificantlywider.Theuseofthebiocompatible,water-solubleionicliquidethanolammoniumformateasacrystallizationadditive in unbaffled stirred vessels resulted in larger, sturdy crystals and reduced formation of crystal aggregates. The experiments indicate thatml-scale batch crystallization of lysozyme in stirred vessels can be performed fast, up-scaleable, reasonably reproducible, and precipitation can beavoided reliably.
#
2007 Elsevier Ltd. All rights reserved.
Keywords:
 Agitated ml-scale vessels; Batch crystallization; Ethanolammonium formate; Ionic liquid; Lysozyme; Vapor diffusion experiments
1. Introduction
Purification of biomolecules via crystallization as analternative to chromatography or other methods may offersignificant advantages due to an expected reduction of costlydownstream processing steps and increases of product qualityand yield. The crystallization process of small biomolecules(e.g. citric acid) is rather well understood. Here, basicparameters such as suitable levels of supersaturation or desiredaverage residence times for optimum crystal growth can beadjusted and controlled by appropriate measures. Hence, large-scale crystallizers have been established and are operatingsuccessfully. Extensive R&D work with regard to the processcrystallization of polypeptides led to the implementation of thefirst approved crystalline pharmaceutical polypeptide biosynthetic human insulin – being successfully marketed ona large-scale basis [1]. This process was based on the earlyworks ofSchlichtkrull whowas the first toutilize stirredvessels[2] and Randall who used 5-ml shaked centrifuge tubes [3].The large-scale process was later on developed by E. Lilly & Co.taking place in 500-l stirred batch crystallizers from a purifiedinsulin solution [4].In contrast to the above mentioned, the crystallizationprocess of larger biomolecules such as proteins is not wellunderstood. So far, a significant lack of systematic knowledgeexists, especially regarding technical-scale crystallizationprocesses. This is because the experience from vapor diffusionexperiments cannot easily be transferred to the technical scalein many cases. Early research work with respect to technical-scale protein crystallization was performed by Judge et al. [5].Here, ovalbumin was crystallized from a crude solutioncontaining a smaller protein (lysozyme) and a larger protein(conalbumin). This process took place in a 1-l unbaffled batchcrystallizer stirred at 35 min
1
using an axial turbine impellerby employing a homogeneous seeding strategy. It was shownthat ovalbumin crystals with a protein purity greater than 99%were obtained after a single crystallization step and one crystalwash. However, little details were given with regard to the flowconditions.Jacobsenetal.reportedthebatchcrystallizationofafungal lipase from clarified concentrated fermentation broths ina225-mlstirred vessel[6].Thevesselwas stirredby a six-blade45
8
 pitched axial flow impeller at 670 min
1
. The ow
www.elsevier.com/locate/procbioProcess Biochemistry 42 (2007) 1649–1654* Corresponding author. Tel.: +49 89 289 15770; fax: +49 89 289 15714.
E-mail address:
 hekmat@lrz.tum.de (D. Hekmat).1359-5113/$ – see front matter
#
2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.procbio.2007.10.001
 
condition was found to be sufficient to fully suspend thecrystals having a typical size of 20–30
m
m. In this work, crystalsize distributions were measured. However, rather widedistributions exhibiting multiple peaks (up to four) wereobserved indicating that the process was not yet ready for anindustrial-scale application. Lee et al. investigated the batchcrystallization of a fungal lipase in a 500-ml Duran bottle on avibrating platform (200 min
1
) using seed crystals [7].Operational windows for crystallization without changes inmorphology and yield were presented. However, these resultsfrom shaked Duran bottles cannot easily be scaled-up to atechnical system. Shenoy et al. showed in a patent from AltusBiologics Inc. that the batch crystallization of rather complexproteins like monoclonal antibodies (mAbs) and antibodyfragments was feasible on a scale of 16 ml [8]. The seededbatch crystallization took place in a hematology/chemistrymixer tumbled at 50 min
1
. It was shown that the providedmAb crystals were more stable than their soluble counterpartsand allowed the advantageous subcutaneous delivery of a smallvolume of highly concentrated mAbs [9]. Again, however,these results from a tumbling mixer cannot easily be scaled-upto a technical system. Furthermore, it was frequently observedthat mAb crystals had a needle-like shape which is detrimentalto further downstream processing. With regard to proteincrystal morphology, a patent was published describing thecontrol of the shape of protease crystals by applyingtemperature shifts [10]. Klyushnichenko described the condi-tions for a large-scale protein crystallization process as a newopportunity for a robust, scaleable, reproducible industrial unitoperation by combining high-throughput screening with anengineering approach [11]. In fact, Schmidt et al. publishedsuch a technical approach using stirred batch crystallizers withtwo different impeller power inputs of 12 and 640 W m
3
[12].However, in this work, the protein was not specified, thevolumes were not given, and the flow conditions were notdescribed. Two isoforms of the protein could be separatedsuccessfully by crystallization using two different crystal-lization agents (not specified). Crystal polymorphism wasobserved in some cases depending upon the chosen region of the crystallization window even when the crystallization agentwas identical. However, no quantitative phase diagram waspresented. The final mean protein crystal size (equilibrium wasreached after approximately 24 h) was about 26
m
m byapplying the low impeller power input 12 W m
3
. Anapproximately 23% smaller mean crystal size was reachedwith the higher power input of 640 W m
3
. Since thecrystallization kinetics were found to be nearly independentof the power input, the lower power input being accompaniedby significantly smaller shear forces was favoured. Crystal sizedistributions were not presented. Carbone and Etzel reportedthe seeded batch crystallization of lysozyme in a 15-mlunbaffled glass vial stirred gently by a magnetic stir bar (stirrergeometry and stirring speed not given) [13]. This set-up wasmainly used to investigate crystallization kinetics and to test amathematical model. Recent research results on the large-scalecrystallization of a relatively largepharmaceutical protein werereported by Takakura et al. [14]. Here, a 100-l batch crystallizerstirred at 130 min
1
was used to crystallize the recombinantpharmaceutical enzyme
 L
-methionine-
g
-lyase (398 amino acidresidues) from a crude solution. A three-step bulk crystal-lization strategy was applied and almost monomorphic crystalswith a size of about 10
m
m were obtained. Again, no details onthe flow conditions were given. Further recent work on ascaleable protein crystallization process was reported byMatthews and Bean [15]. Here, a not specified recombinanttherapeutic protein (60 kDa, p
 I 
 = 9.1) expressed in
 Escherichiacoli
wascrystallizedina1-lstirredvessel.Theprocesswasthenscaled-up to a 800-l stirred crystallizer. The morphology of theprotein crystals was controlled by variation of pH and agitationconditions, and by applying a temperature ramp. It was shownthat the crystallization process was at equilibrium after alreadyabout 8.5 h using the temperature ramp.It is obvious from the above mentioned that in most cases,the flow conditions of the agitated batch crystallization systemswere not exactly defined and hence, not easily comparable.Still,alackofknowledgeregardingdesignandoperationexists,especially with regard to efforts in order to achieve a robustcrystallization process. Therefore, the aims of the presentinvestigations were (i) to obtain as quantitative as possible dataof sitting-drop vapor diffusion experiments, (ii) to determinethe experimental conditions for a quantitative and reproduciblebatch crystallization process in agitated ml-scale vessels, and(iii) to investigate the application of a promising new kind of crystallization additives – biocompatible, water-soluble ionicliquids (ILs) – as an innovative means to increase crystal size.The exemplary protein used in the experiments was lysozyme.
2. Materials and methods
2.1. Vapor diffusion experiments
Lysozyme from chicken egg white was purchased from Sigma–Aldrich,Taufkirchen, Germany (No. 62971). The other chemicals were analytical gradepurchased from Merck, Darmstadt, Germany. The sitting-drop vapor diffusioncrystallization experiments were performed at 21
 8
C in 24-well Chryschemplates type HR3-158 (Hampton Research Corp., Aliso Viejo, CA, USA) sealedwith Crystal Clear sealing tape (Manco Inc., Avon, OH, USA). The sitting-droptray had a diameter of about 6 mm and a maximum filling volume of about40
m
l. For each tray, 10
m
l of lysozyme solution was joined with 10
m
l of therespective reservoir solution. The reservoir solution contained 40–120 g l
1
NaCl as the crystallization agent in a 50 mM acetate buffer at pH 4.0. Thelysozyme start concentration in the trays was altered between 10 and 50 g l
1
.Microphotographsofthesitting-dropsweremadeusingamicroscopetypeZeissAxioplan (Carl Zeiss MicroImaging GmbH, Go¨ttingen, Germany) and a digitalcamera type DSC-S75 (Sony Deutschland GmbH, Ko¨ln). The experimental set-up was designed in order to prevent shocks and/or vibrations acting upon thecrystallization plates.
2.2. Batch experiments in agitated vessels
All batch crystallization experiments of lysozymewere performed at 21
 8
C.At first, 96-well microtiter plates (MTPs) type Rotilab (Carl Roth GmbH,Karlsruhe, Germany) together with a minishaker type MTS 2/4 D (IKA GmbH& Co. KG, Staufen, Germany) were used. Each well was filled with 200
m
l of solution containing 20–60 g l
1
lysozyme and 30–60 g l
1
NaCl in a 25 mMacetate buffer at pH 4.0. The shaking frequencies were varied from 0 to1100 min
1
at an excentricity of 3 mm. Further batch crystallization experi-ments were performed in 0.2–2-ml Eppendorf tubes in a laboratory rotator at
 D. Hekmat et al./Process Biochemistry 42 (2007) 1649–1654
1650
 
3 min
1
(neoLab GmbH, Heidelberg, Germany) with solutions containing50 g l
1
lysozyme and 40 g l
1
NaCl in a 25 mM acetate buffer at pH 4.0.More batch crystallization experiments were performed in unbaffled shakeflasks. The shaking frequency was 100 min
1
at an excentricity of 25 mm(shaking apparatus type 3022, GFL mbH, Burgwedel, Germany). The shakeflasks were made of glass and had a total volume of 25 ml. The fill volume was5 ml and the composition of the solutions was identical to the laboratory rotatorexperiments. Yet other batch crystallization experiments were performed inbaffled vessels witha fill volume of 4 ml stirredby a conicalmagneticstir barat300 min
1
. The polystyrene vessels with four baffles had a total volume of 20 ml, the inner diameter was 20 mm, and the ratio of baffles to inner vesseldiameter was 0.1 [16]. These dimensions were chosen in such a way that scale-up of the crystallization process was expected to be easily accomplished. Theconical magnetic stir bar had a length of 10 mm and a diameter of 4 mm atthe center. The tip diameter was approximately 3 mm. The composition of thesolutions was again identical to the laboratory rotator experiments. Finally,batch crystallization experiments in unbaffled stirred polypropylene vesselswith a total volume of 20 ml and an inner diameter of 20 mm were performed.The fill volume was again 4 ml. The vessels were stirred from above using apaddel impeller (height = width = 19.5 mm). The stirrer speed was 10 min
1
for the first 4 h and 50 min
1
for the rest of the operation time. The compositionof the solutions was again identical to the laboratory rotator experiments.During these experiments, the impact of the addition of 100 g l
1
of thebiocompatible, non-halogenated, substituted, water-soluble IL ethanolammo-nium formate (IoLiTec, Denzlingen, Germany) to the solutions was investi-gated. This IL has a hydroxylated primary alkylammonium cation and acarboxylate anion. The molecular weight of the IL is 107 g mol
1
, the meltingpoint is
 
82
 8
C, and the pH of the pure substance is 8.5.Forthedeterminationofthecrystalareadensitydistributions,30
m
lsampleswere placed in a Neubauercountingchamberand microphotographs were madeas described above. The samples were diluted when the crystal density was toohigh in order to perform crystal counting.The microphotographs were analyzedusinga publicdomainimageprocessingsoftware(ImageJ,version1.37v,http:// www.rsb.info.nih.gov/ij). Since nucleation and onset of crystallization were notstudied, the use of microscopy was sufficient and elaborate dynamic lightscattering methods or others were not required. The image processing softwaredetermined the distribution of the cross-sectional crystal areas. The crystal areadensity distribution
 q
2
(
¯  x
i
) was defined as
q
2
ð
¯  x
i
Þ¼
 n
ð
D
 x
i
Þ
¯  x
2
i
D
 x
i
P
i
n
ð
D
 x
i
Þ
¯  x
2
i
;
were
 n
(
D
 x
i
) was the amount of crystals in the particle size increment
D
 x
i
 = (
 x
i
 x
i
1
) and
 ¯  x
i
 = (
 x
i
 +
 x
i
1
)/2. This equation is valid assumingparticleswith similar rectangular shape [17]. In order to determine the time required toreach crystallization equilibrium in the stirred vessels, the dissolved proteinconcentration was measured from 200
m
l sample solutions. These solutionswere centrifuged at 13,000 min
1
for 3 min in a centrifuge type Biofuge pico(HeraeusInstrumentsGmbH,Du¨sseldorf,Germany). Then,the supernatantwasdilutedwith demineralized waterby factor 50 andthe UVabsorbanceat 280 nmwas measured in a spectrophotometer type BioMate 3 (Thermo Fisher Scien-tific, Schwerte, Germany) and compared to a calibration standard.
3. Results and discussion
3.1. Vapor diffusion crystallization experiments
A set of 48 experiments was made in order to obtain asufficiently quantitative crystallization phase diagram similarto the qualitative one described by Rie`s-Kautt [18]. The time toreach equilibrium was 48–72 h. The results are presented inFig. 1. The solubility curvewas derived from data published byHoward et al. [19]. From this diagram, the courses of thesupersolubility curve and of the precipitation curve wereestimated. As a result, the respectivewindows of the metastablezone, the nucleation zone, and the precipitation zone wereobtained. However, as expected with all vapor diffusionexperiments in general, reproducibility was relatively low. Atlarger concentrations of NaCl, spontaneous precipitation wasobserved in most cases. In some cases, however, the clearsitting-drop solution turned into opaque white spontaneouslyafter adding the protein solution. This observation disappearedwithin seconds and crystals were formed later on (see
 
symbols in the precipitation zone of  Fig. 1). This phenomenonwas most probably linked to a kind of reversible liquid–liquidphase separation process as described by Muschol andRosenberger and was not directly associated with theirreversible formation of precipitate [20]. On the other hand,spontaneous precipitation was observed in few cases even atlower NaCl concentrations well within the nucleation window(see
*
symbols in the nucleation zone of  Fig. 1). Based on theabove results, the experimental conditions of the batchexperiments were determined.
3.2. Batch crystallization experiments in agitated vessels3.2.1. Determination of quantitative phase diagram
In the batch crystallization experiments, the degree of nucleation can be controlled by the amount of crystallizationagent added. Hence, in contrast to the vapor diffusionexperiments, the mechanisms of nucleation and crystal growthcan be governed to a certain extent [21]. In all of the following145 experiments, the whole amount of the crystallization agentsolution was added at the start. The experiments in the MTPsrevealed that a fairly strong influence of the shaking frequencyupon the size of the crystallization window existed. The resultsare presented in Fig. 2. The distinctive measurement points forthe formation of crystals or precipitate were not shown forreasons of clarity. The precipitation curves of the experimentswith shaking frequencies from 0 up to 450 min
1
were nearlyidentical and the position was clearly left of the sitting-dropprecipitation curve. With rising shaking frequencies, theprecipitation curves moved further to the left. Hence, the
Fig. 1. Quantitative phase diagram of sitting-drop vapor diffusion crystal-lization of lysozyme (20
m
l scale). (
&
) Solubility data; (
) formation of crystals; (
*
) formation of precipitate. The error bar cross is depicted on thesupersolubility curve. The absolute measurement error for lysozyme was about
2.5 and
 
5 g l
1
for NaCl.
 D. Hekmat et al./Process Biochemistry 42 (2007) 1649–1654
 1651
 
nucleation zone was significantly reduced. The precipitationcurves for shaking frequencies from 750 to 1100 min
1
werealmost identical again. Three different durations of agitation (3,24, and 48 h) were examined, however, almost no influenceupon the position of the precipitation curves was found. Theposition of the supersolubility curvewas almost independent of the agitation intensity. Experiments with slightly elevatedcrystallization temperatures revealed that the precipitationboundary moved downwards, thus, reducing the nucleationzone further (data not shown). When precipitation took place, itwas observed within the first 3 h in all experiments. Theprecipitation curve of the experiments in 0.2-ml Eppendorf tubes positioned in a laboratory rotator operating at 3 min
1
isincluded in Fig. 2. As can be seen, the position of the curvewascomparable to the one of a MTP experiment at medium shakingfrequency.
3.2.2. Determination of crystal area density distributions
Above results were used to chose proper experimentalconditions for the agitated systems in order to preventprecipitation reliably. These conditions were 50 g l
1
lysozymeand 40 g l
1
NaCl in 25 mM acetate buffer at pH 4.0. From theexperiments in 5-ml shake flasks, the crystal area densitydistributions of lysozyme batch crystallization as a function of time were determined as presented in Fig. 3. These crystal areadensity distributions were pronounced wide representingcrystals and crystal aggregates with sizes up to 200
m
m. Inthe early hours, more smaller particles and less crystalaggregates were observed. With advancing time, less smallercrystals and more crystal aggregates were formed. Theobserved high level of polydispersity is disadvantageous forfurther downstream processing steps like filtration and crystalwashing.The crystal area density distributions of lysozyme batchcrystallization experiments in 4-ml stirred baffled vessels as afunction of time are presented in Fig. 4. In contrast to the shakeflask experiments, significantly narrower crystal area densitydistributions were obtained. In addition, the variation of thecrystal area density distributions with time was different. In thebeginning (after 10 min), almost no particles larger than 20
m
mwere observed. After 50 min, more larger crystals and crystalaggregates were found and at the same time less smallerparticles. This process was reversed after 3 h leading to lesslarger particles and more smaller particles. The crystal areadensity distribution after 24 h was similar to the one after10 min with even less larger particles and more smallerparticles than after 3 h. Hence, the mechanical shear forces of the stirrer bar led to the reduced particle sizes with time. Theprogress of the batch crystallization was tracked by measuringthe time-course of dissolved lysozyme concentration in thestirred vessels. Data from seven independent experiments withidentical conditions indicated that batch crystallization instirred vessels was considerably more reproducible than thevapor diffusion experiments (data not shown). The dissolvedlysozyme concentration dropped exponentially from 50 to9 g l
1
. Thus, about 82% of the dissolved lysozyme was movedinto the crystalline phase. After about 7–8 h, the proteinconcentration in the mother liquor kept being constant andequilibrium was reached. Examination of the crystal slurryrevealed that the stirrer speed of 300 min
1
was sufficient tofully suspend the crystals at all times.A comparison of the crystal area density distributions of lysozyme batch crystallization experiments in rotated Eppen-dorf tubes and in stirred baffled vessels after 24 h are presented
Fig. 2. Quantitative phase diagram of crystallization of lysozyme in 200
m
lshaked microtiter plates (MTPs) and in 200
m
l rotated Eppendorf tubes.Measurement points of crystal and precipitate formation not shown for reasonsof clarity. The error bar cross is depicted on the supersolubility curve. Theabsolute measurement error was about
 
2.5 g l
1
.Fig. 3. Crystal area density distributions of lysozyme batch crystallization inshake flasks at different times.
 c
lysozyme
 = 50 g l
1
,
 c
NaCl
 = 40 g l
1
, 25 mMacetate buffer at pH 4.0, liquid volume 5 ml, shaking frequency 100 min
1
. Theerror bars are equivalent to a measurement error of approximately 28%.Fig. 4. Crystal area density distributions of lysozyme batch crystallizationexperiments in 4-ml stirred baffled vessels (conical magnetic stir bar at300 min
1
) as a function of time.
 c
lysozyme
 = 50 g l
1
,
 c
NaCl
 = 40 g l
1
,25 mM acetate buffer at pH 4.0. The error bars are equivalent to a measurementerror of approximately 28%.
 D. Hekmat et al./Process Biochemistry 42 (2007) 1649–1654
1652
 
in Fig. 5. As can be seen, the crystal area density distribution of the stirred vessels was clearly narrower than the one of therotated Eppendorf tubes. It can be deduced that the agitationforces of the rotator were notedly smaller compared to thebaffled vessels stirred with the magnetic stir bar.
3.2.3. Examination of the effect of the addition of ethanolammonium formate
The results of the application of 100 g l
1
of the water-soluble IL ethanolammonium formate as a crystallizationadditive in 4-ml unbaffled tanks stirred at 10–50 min
1
with apaddle impeller after 5 h of operation compared to respectiveexperiments without the addition of the IL are presented inFig. 6. As can be seen, notedly larger, tetragonal-like shaped,sturdy crystals and less crystal aggregates were formed whenthe IL was applied.The comparison of the crystal area density distributions of lysozyme batch crystallization experiments in stirred unbaffledvessels with and without ethanolammonium formate close toequilibrium (after approximately 7 h) is given in Fig. 7. Clearly,larger crystals were obtained using the IL compared to theexperiments without addition of IL. The mechanisms of theinfluenceoftheILontheproteincrystalsizedistributionarequitecomplexandlargelyunknown.Thestrongestimpactisattributedtotheinteractionsofionsandcounterionsattheprotein–solutioninterface.Here,thedegreeofhydrationofthevariousionspeciesplaysamajorrole.TheaddedILintervenesadvantageouslywatermolecules of the protein hydration layers through excludedvolume, water activity, and interfacial effects [22]. Anothermechanism which may have affected the protein crystal sizedistribution is Ostwald ripening [23]. This mechanism is athermodynamically controlled process of formation of largercrystals at the expense of smaller crystals and takes place whenlong nucleation times occur [24]. However, at any time, nodepletion zones around growing crystals were observed whichwould indicate the occurrence of Ostwald ripening. Thecrystallization kinetics of the experiments with IL were slightlyfaster compared to the experiments without IL. However, itshouldbetakenintoaccountthatthepHvaluesdifferedfrom5.8(with 100 g l
1
IL) to 4.0 (without IL) due to the basic nature of the IL. This finding is interesting considering the fact thatprevious experimental investigations yielded considerablyslower lysozyme crystallization kinetics in aqueous solutionswithoutILandotherwisesameconditionsatapHof5.5andabove(datanotshown).ThebiocompatibilityoftheILwasexaminedbymeasuring the enzyme activity of an aqueous lysozyme solutionwithIL.Here,areductionoftheenzymeactivityofonlyabout6%was observed with 50–125 g l
1
ethanolammonium formate(monitored during a time period of 70 days, data not shown).Hence, the applicationof the sufficiently biocompatible IL as anadvantageous innovative crystallization additive in order toincrease crystal size was shown.
Fig. 5. Comparison of the crystal area density distributions of lysozyme batchcrystallization experiments in 2-ml rotated Eppendorf tubes (3 min
1
) and in 4-ml stirred baffled vessels (conical magnetic stir bar at 300 min
1
) after 24 h.
c
lysozyme
 = 50 g l
1
,
 c
NaCl
 = 40 g l
1
, 25 mM acetate buffer at pH 4.0. The errorbars are equivalent to a measurement error of approximately 28%.Fig. 6. Microphotographsoflysozymecrystalsfrombatch crystallizationin 4-ml unbaffledstirredtanksafter 5 h. Paddle impeller,stirrer speed10 min
1
forthe first4 h, subsequently 50 min
1
, 25 mM acetate buffer,
 c
lysozyme
 = 50 g l
1
,
 c
NaCl
 = 40 g l
1
. (A) No IL, pH 4.0. (B) 100 g l
1
ethanolammonium formate, pH 5.8.Fig. 7. Comparison of the crystal area density distributions of lysozyme batchcrystallization experiments in stirred unbaffled vessels with and without ILclose to equilibrium (after approximately 7 h). Paddle impeller, stirrer speed10 min
1
for the first 4 h, subsequently 50 min
1
,
 c
lysozyme
 = 50 g l
1
,
c
NaCl
 = 40 g l
1
, 25 mM acetate buffer. The error bars are equivalent to ameasurement error of approximately 28%.
 D. Hekmat et al./Process Biochemistry 42 (2007) 1649–1654
 1653
of 6