Armstrong1999

A new protein folding screen: Application to the ligand binding domains of a glutamate and kainate receptor and to lysozyme and carbonic anhydrase NEALI ARMSTRONG, ALEXANDRE DE LENCASTRE, and ERIC GOUAUX Department of Biochemistry and Molecular Biophysics, Columbia University, 650 West 168th Street, New York, New York 10032 ~Received September 15, 1998; Accepted March 19, 1999! Abstract Production of folded and biologically active protein from Escherichia coli derived inclusion bodies can only be accomplished if a scheme exists for in vitro naturation. Motivated by the need for a rapid and statistically meaningful method of determining and evaluating protein folding conditions, we have designed a new fractional factorial protein folding screen. The screen includes 12 factors shown by previous experiments to enhance protein folding and it incorporates the 12 factors into 16 different folding conditions. By examining a 10256th fraction of the full factorial, multiple folding conditions were determined for the ligand binding domains from glutamate and kainate receptors, and for lysozyme and carbonic anhydrase B. The impact of each factor on the formation of biologically active material was estimated by calculating factor main effects. Factors and corresponding levels such as pH ~8.5! and l-arginine ~0.5 M! consistently had a positive effect on protein folding, whereas detergent ~0.3 mM lauryl maltoside! and nonpolar additive ~0.4 M sucrose! were detrimental to the folding of these four proteins. One of the 16 conditions yielded the most folded material for three out of the four proteins. Our results suggest that this protein folding screen will be generally useful in determining whether other proteins will fold in vitro and, if so, what factors are important. Furthermore, fractional factorial folding screens are well suited to the evaluation of previously untested factors on protein folding. Keywords: bacterial expression; fractional factorial screen; inclusion bodies; protein folding Harvesting the fruits of whole, partial, and selective genome se- quencing projects frequently requires the heterologous expression of genes and subsequent studies of the gene products. Expression in a bacterial host, such as Escherichia coli, is an economical and time efficient method of protein production. However, in many cases expression of foreign proteins in E. coli leads to the produc- tion of insoluble inclusion bodies. Since a large number of proteins can be folded from inclusion body material and because protein production in the form of inclusion bodies has a number of merits, a method that would answer the question of whether proteins de- rived from inclusion body material can be folded into a biologi- cally relevant conformation would be useful ~Chen & Gouaux, 1997; De Bernardez Clark, 1998!. Ideally, the method would an- swer the folding question with a reasonable degree of confidence using a small number of experiments. If expression in E. coli is subsequently deemed untenable, then production of the protein in more expensive and time consuming systems can be justified. In this paper, we present an approach to search for folding conditions and apply it to the folding of four proteins. The folding buffer should favor the formation of the native state while minimizing the aggregation of folding intermediates. A wealth of experimentation has shown that polar additives ~such as argi- nine!, osmolytes, detergents, and chaotropes can minimize aggre- gation and increase the yield of biologically active material ~see Rudolph & Lilie, 1996 for a recent review!. Other factors affecting the formation and stability of the folded state are pH, redox envi- ronment, ionic strength, protein concentration, presence of ligand, and the mode by which the denaturant concentration is reduced, i.e., by dilution or dialysis, as examples. It is also known that proteinaceous chaperones can promote folding in vitro ~Cole, 1996!. However, we have devoted our attention to small molecules and polymers in the experiments described here. Reprint requests to: Eric Gouaux, Department of Biochemistry and Mo- lecular Biophysics, Columbia University, 650 West 168th Street, New York, New York 10032; e-mail: jeg52@columbia.edu. Abbreviations: GluR2, ionotropic glutamate receptor, AMPA specific, subtype 2 or B; GluR2-S1S2, ligand binding domain of GluR2; GFB-S1S2, ligand binding domain of the goldfish kainate binding protein, subtype b; CAB, carbonic anhydrase B; GuHCl, guanidine hydrochloride; GSSG, oxi- dized glutathione; GSH, reduced glutathione; FF16, 16 condition fractional factorial folding screen; AMPA, a-amino-3-hydroxy-5-methylisoxazole-4- propionic acid; KA, kainate; HPLC, high pressure liquid chromatography; PCR, polymerase chain reaction; pNAc, p-nitrophenylacetate; PEG, polyeth- ylene glycol; SEC, size-exclusion chromatography; RT, room temperature. Protein Science ~1999!, 8:1475–1483. Cambridge University Press. Printed in the USA. Copyright © 1999 The Protein Society 1475 Different proteins often require distinct conditions for folding. For example, Mycoplasma arginine deiminase folds upon dilution into only 10 mM potassium phosphate at pH 7.0 ~Misawa et al., 1994!, whereas a good condition for urokinase folding requires 50 mM Tris, pH 9.0, 1 M GuHCl, 0.2 M l-arginine, 5 mM EDTA, 0.005% Tween 80, 1.25 mM GSSG, and 0.25 mM GSH ~Winkler & Blaber, 1986!. In general, determination of folding conditions has involved a trial and error, one factor at a time approach and there is scant information available in the literature on general methods to search for folding conditions. Nevertheless, one ap- proach involved applying a crystallization screen to the search for protein folding conditions ~Hofmann et al., 1995!. However, one would predict that effective precipitation and crystallization con- ditions will almost certainly not be useful for protein folding; not surprisingly, this approach was largely unsuccessful. By contrast, optimizations of previously determined folding conditions are more widespread ~see Ahn et al., 1997!. In an effort to provide a logic to the search for protein folding conditions, we have developed fractional factorial folding screens ~Chen & Gouaux, 1997!. Here we report an improved version of the original screen and describe its application to a number of structurally diverse proteins. Since the in vitro folding of a protein may be influenced by a number of factors each bounded by two chemically reasonable levels ~i.e., the factor “protein concentration” might have the levels 0.1 and 1.0 mg0mL!, an efficient and statistically meaningful method of searching for folding conditions is desired. However, if one chooses 12 factors and each factor is assigned two levels, evalu- ation of the full factorial would require 4,096 experiments. In many cases, when one is searching for factors that impact the outcome of a particular process, evaluation of a fraction of the full factorial is sufficient to determine which factors have the greatest effects. Indeed, the fractional factorial experiment allows one to estimate main effects and multifactor interactions, depending on the resolution of the particular design ~Box et al., 1978!. Once the most significant factors have been identified, the folding condi- tions can be optimized using a subset of the original factors. Al- though some of the assumptions inherent in fractional factorial experiments, such as the assumptions that the response ~i.e., the yield of folded protein! is linearly dependent on the level of the factor and that the factors do not interact, may not be strictly valid, a screen based on a fractional factorial design nevertheless pro- vides a powerful tool by which folding conditions can be screened and factors can be evaluated. Here we describe a fractional factorial protein folding screen that includes 12 factors and a total of 16 experiments ~FF16; Table 1!. The effectiveness of the FF16 screen was tested on the ligand binding domains from the rat GluR2 receptor ~GluR2-S1S2; Chen & Gouaux, 1997; Arvola & Keinänen, 1996!, and from the goldfish kainate binding protein b ~GFB-S1S2; Wo & Oswald, 1994!, on hen egg white lysozyme, and on bovine carbonic anhy- drase B ~CAB!. GluR2-S1S2 and GFB-S1S2 have low level amino acid sequence identity, related biological activities, molecular weights of ;32 kDa, one disulfide bond, two compact domains, mixed a0b secondary structures and basic isoelectric points; prior to the work from this laboratory, folding conditions for neither GluR2-S1S2 nor GFB-S1S2 had been reported. In contrast, lyso- zyme ~Dobson et al., 1994; Hevehan & De Bernardez Clark, 1997! and CAB ~Cleland et al., 1992; Wetlaufer & Xie, 1995; Xie & Wetlaufer, 1996! are both well-studied proteins and have served as model systems for testing factors and methods for protein folding. On the one hand, lysozyme ~MW of 14.5 kDa! has four disulfide bonds, a pI of 11.35 and is primarily a-helical. On the other hand, CAB ~30 kDa! has no disulfide bonds, a catalytic metal ion center, a central b-sheet that forms the core of the protein and an isoelec- tric point of 5.9. All four proteins were successfully folded under multiple conditions from the screen. Results Statistical analysis of the data The main effects of each factor were calculated by summing the response ~i.e., @3H#-AMPA counts in the case of GluR2-S1S2! obtained when using the “1” level and then when using the “2” level of the particular factor under consideration ~Fig. 2!. The sum of the “2” experiments were then subtracted from the sum of the “1” experiments and the resulting difference was divided by 8, i.e., Main Effect 5 ( “1 level” 2 ( “2 level”08 ~Box et al., 1978!. As shown in Figure 2, polar additive ~arginine! at the “1” level, i.e., at 0.5 M, pH at the “1” level ~pH 8.5! and protein concentration at the “1” level have some of the strongest favor- able effects on the yield of active protein. To help determine if a given factor had a positive or negative effect for each of the four proteins, the main effects for each protein were scaled in the following manner. For each of the pro- teins, the values of all factors were scaled such that the main effect of the factor with the greatest effect was set to 1.0. In the case of GluR2-S1S2 the factor is pH, for example. Then, the mean main effects for each factor were calculated. This value is indicated by the unfilled bars in Figure 3. The mean main effects were then ranked in order of decreasing magnitude. As shown in Figure 3, pH, polar additive, chaotrope, and protein concentration have the largest positive mean main effects. Detergent has the largest neg- ative mean main effect. Reduction0oxidation potential has an over- all average effect of zero, although the spread in the main effects is large because lysozyme folds well in the presence of GSH0 GSSG, i.e., the “1” level of Red.0Ox. has a large positive effect ~Fig. 2! while GSH0GSSG has a smaller negative effect on the folding of the other proteins. GluR2-S1S2 As shown in Figure 1, most of the FF16 screen conditions gave significant levels of folded GluR2-S1S2 except #2, #5, #9, and #14. Condition #7 resulted in the highest @3H#-AMPA counts fol- lowed by #16 and #6. The main effects plot for GluR2-S1S2 shown in Figure 2 illustrates that the following factors had a positive effect on the folding: pH of 8.5, 0.5 M GuHCl, 0.5 mg0mL protein concentration, and 0.5 M l-arginine. The inclusion of sucrose, divalent cations, and PEG had negligible effects while dialysis, high ionic strength, GSH0GSSG, and detergent had weakly neg- ative effects. The presence of glutamate in the folding buffer had a modestly positive consequence. Based on large scale folding experiments of GluR2-S1S2, which employed folding conditions similar to #7, the yield was ;10% using OD280 measurements. GFB-S1S2 Conditions #7, #3, #4, and #11 gave the largest amount of properly folded GFB-S1S2 as judged by the @3H#-kainate binding results listed in Table 2. The yield of soluble protein was 10–15% as 1476 N. Armstrong et al. Table 1. 16 Condition fractional factorial folding screen (FF16) Buffer Patterna Modeb @Protein# ~mg0mL!c Polar add.d Detergente pHf Red.0Ox.g Chao.h Ionic str.i Dival. cat.j PEG ~%!k Ligandl NP add.m 1 222212212112 Dil. 0.1 0 0 8.5 1 mM DTT 0 250 mM EDTA 0.05 10 mM 0 2 222121121222 Dil. 0.1 0 0.3 mM 6.0 GSH0GSSG 0.5 M 10 mM Mg, Ca 0 0 0 3 221221122111 Dil. 0.1 0.5 M 0 6.0 GSH0GSSG 0.5 M 10 mM EDTA 0.05 10 mM 0.4 M 4 221112211221 Dil. 0.1 0.5 M 0.3 mM 8.5 1 mM DTT 0 250 mM Mg, Ca 0 0 0.4 M 5 212221211211 Dil. 0.5 0 0 6.0 GSH0GSSG 0 250 mM Mg, Ca 0 10 mM 0.4 M 6 212112122121 Dil. 0.5 0 0.3 mM 8.5 1 mM DTT 0.5 M 10 mM EDTA 0.05 0 0.4 M 7 211212121212 Dil. 0.5 0.5 M 0 8.5 1 mM DTT 0.5 M 10 mM Mg, Ca 0 10 mM 0 8 211121212122 Dil. 0.5 0.5 M 0.3 mM 6.0 GSH0GSSG 0 250 mM EDTA 0.05 0 0 9 122222111121 Dial. 0.1 0 0 6.0 1 mM DTT 0.5 M 250 mM Mg, Ca 0.05 0 0.4 M 10 122111222211 Dial. 0.1 0 0.3 mM 8.5 GSH0GSSG 0 10 mM EDTA 0 10 mM 0.4 M 11 121211221122 Dial. 0.1 0.5 M 0 8.5 GSH0GSSG 0 10 mM Mg, Ca 0.05 0 0 12 121122112212 Dial. 0.1 0.5 M 0.3 mM 6.0 1 mM DTT 0.5 M 250 mM EDTA 0 10 mM 0 13 112211112222 Dial. 0.5 0 0 8.5 GSH0GSSG 0.5 M 250 mM EDTA 0 0 0 14 112122221112 Dial. 0.5 0 0.3 mM 6.0 1 mM DTT 0 10 mM Mg, Ca 0.05 10 mM 0 15 111222222221 Dial. 0.5 0.5 M 0 6.0 1 mM DTT 0 10 mM EDTA 0 0 0.4 M 16 111111111111 Dial. 0.5 0.5 M 0.3 mM 8.5 GSH0GSSG 0.5 M 250 mM Mg, Ca 0.05 10 mM 0.4 M a102 factor levels are: Mode, dil 5 2, dial 5 1; @Protein#, 0.1 mg0mL 5 2, 0.5 mg0mL 5 1; Polar Additive, 0 5 2, 0.5 M 5 1; detergent, 0 5 2, 0.3 mM 5 1; pH, 6.5 5 2, 8.5 5 1; Red.0Ox., 1 mM DTT 5 2, GSH0GSSG 5 1; chaotrope, 0 5 2, 0.5 M 5 1; ionic strength, 10 mM 5 2, 250 mM 5 1; divalent cations, 1 mM EDTA 5 2, Mg, Ca 5 1; PEG, 0 5 2, 0.05% 5 1; ligand, 0 5 2, 10 mM 5 1; nonpolar additive, 0 5 2, 0.4 M 5 1. bThis factor was only included for GluR2-S1S2 and GFB-S1S2 folding experiments. cLysozyme experiments were performed using protein concentrations of 0.1 and 1.0 mg0mL. dl-arginine. eDetergent: lauryl maltoside fpH 6.0, 50 mM MES; pH 8.5, 50 mM Tris-HCl; pH was measured at 4 8C. g1 mM reduced ~GSH! and 0.1 mM oxidized ~GSSG! glutathione. hGuanidine hydrochloride. i Molar ratio of NaCl to KCl was 25:1. j 1 mM EDTA or 2 mM MgCl2, 2 mM CaCl2; except for CAB buffers that contained 1 mM ZnCl2 instead of MgCl2 and CaCl2. kPEG MWaverage 5 3,550 Da; the concentration was weight0volume. l The ligand was l-glutamate for GFB-S1S2 and GluR2-S1S2 buffers; this factor was excluded from the folding buffers for lysozyme and CAB. mSucrose. Fractionalfactorialproteinfolding screen 1477 estimated by SDS-PAGE. The fewest counts were recorded for conditions #2, #6, #8, #13, and #14. The strongest positive factors were 0.5 M l-arginine followed by dialysis, pH 8.5, and 1 mM l-glutamate ~Fig. 2!. The presence of detergent had a strong neg- ative effect while glutathione, high ionic strength, PEG, and su- crose had moderately negative consequences. Interestingly, protein concentration appeared to have no effect on GFB-S1S2 folding showing that a higher protein concentration did not yield more folded material. This in turn suggests that nonproductive folding and aggregation occurs at protein concentrations above 0.1 mg0 mL. Although most conditions in the FF16 screen were sufficient to fold GFB-S1S2, only one of the best four folding conditions were shared with its close relative GluR2-S1S2. Likewise, the chaotrope and protein concentration factors, which were important for GluR2-S1S2 folding, were relatively neutral factors for GFB- S1S2 folding. While these two proteins have many common phys- ical characteristics and functional properties, these similarities do not confer equivalence in terms of factor main effects. Lysozyme Lysozyme did not yield active material under as many conditions of the FF16 screen as compared to GluR2-S1S2 and GFB-S1S2. This result is probably due to the requirement of lysozyme for conditions that promote disulfide bond formation and rearrange- ment. In fact, none of the eight conditions that contained 1 mM DTT resulted in active lysozyme, indicating that lysozyme folding is greatly disfavored under reducing conditions. The two best con- ditions for lysozyme folding were #11 and #16, respectively. Like GFB-S1S2, protein concentration was a neutral factor in lysozyme folding. Since the data presented here measured the total amount of active protein and conditions with different protein concentrations, such as #11 ~0.1 mg0mL! and #16 ~1.0 mg0mL!, resulted in similar Fig. 1. Relative activity chart. The relative activity was calculated with respect to the condition that resulted in the greatest amount of active material for each protein; all other conditions are given as a percentage of the best condition. Condition #7 resulted in the largest amount of properly folded GluR2-S1S2, GFB-S1S2, and CAB, whereas condition #11 yielded the greatest amount of active lysozyme. Fig. 2. Plot of the factor main effects. The main effects for each factor were calculated using the equation: ~( “1 level” 2 ( “2 level”!08 ~Box et al., 1978!. The factors are plotted in order of decreasing main effects, starting from the left. A positive value on the chart indicates that the “1” level ~i.e., pH 8.5 or 0.5 M l-arginine! is estimated to enhance protein folding, whereas a negative value reflects a decrease in folding due to the “1” level of the factor. The calculations of main effects do not take into account multifactor interactions, which, if present, may alter the estimation of factor main effects; to estimate multifactor interactions, higher resolu- tion screens must be performed. The 1s uncertainty for estimation of the main effects are 502 for GluR2-S1S2, 1980 for GFB-S1S2, 0.008 for lysozyme, and 0.005 for carbonic anhydrase B. 1478 N. Armstrong et al. amounts of folded product, condition #11 gave a higher yield of active lysozyme, based on the amount of starting material. By comparing the FF16 results to an assay using native lysozyme, condition #11 had a yield of ;37% and condition #16 yielded ;4% of the activity based on the amount of original material. Carbonic anhydrase B CAB folded to yield catalytically active protein under many of the conditions in FF16. However, as was determined from the folding of GluR2-S1S2 and GFB-S1S2, condition #7 clearly gave the high- est yield per volume of protein folding solution. Both condition #7 and condition #11 resulted in a folding yield of ;60% based on the amount of starting material. It was striking that the six conditions for which the highest CAB activity was measured all had protein concentrations of 0.5 mg0mL, which made high protein concen- tration the strongest positive factor. The presence of 0.5 M l-arginine, pH 8.5, or 0.5 M GuHCl also had a positive effect on CAB folding. PEG has previously been shown to increase the yields of CAB folding ~Cleland et al., 1992!. However, in this screen the presence of PEG had a moderately negative effect. This disparity could be due to multifactor interactions, since the folding buffers used by Cleland et al. included only 1 M GuHCl, 0.5 mg0mL CAB, and PEG ~Cleland et al., 1992!. Discussion Utilization of protein produced as inclusion bodies from E. coli is a valuable strategy if the protein folds in vitro. Searching for folding conditions by trial and error, particularly in light of the substantial number of factors that may facilitate folding, is a daunt- ing and potentially fruitless task. To answer the questions of ~1! whether the protein of interest will fold and ~2! what factors are most influential, we have designed and tested a fractional factorial protein folding screen. The screen answers the question of whether the protein will fold using a small number of highly varied con- ditions and it facilitates the determination of