Refolding out of guanidine hydrochloride is an effective approach for high-throughput structural studies of small proteins

Abstract

Low in vivo solubility of recombinant proteins expressed in Escherichia coli can seriously hinder the purification of structural samples for large-scale proteomic NMR and X-ray crystallography studies. Previous results from our laboratory have shown that up to one half of all bacterial and archaeal proteins are insoluble when overexpressed in E. coli. Although a number of strategies may be used to increase in vivo protein solubility, there are no generally applicable methods, and the expression of each insoluble recombinant protein must be individually optimized. For this reason, we have tested a generic denaturation/refolding protein purification procedure to assess the number of structural samples that could be generated by using this methodology. Our results show that a denaturation/refolding protocol is appropriate for many small proteins (≤18 kD) that are normally soluble in vivo. In addition, refolding the purified proteins by using dialysis against a single buffer allowed us to obtain soluble protein samples of 58% of small proteins that were found in the insoluble fraction in vivo, and 10% of the initial number of proteins provided good heteronuclear single quantum coherence (HSQC) NMR spectra. We conclude that a denaturation/refolding protocol is an efficient way to generate structural samples for high-throughput studies of small proteins.

Structure determination using NMR spectroscopy and X-ray crystallography requires the generation of large amounts of soluble recombinant protein. Most often, recombinant proteins for structural biology studies are produced in Escherichia coli because the cells grow rapidly and to high density in inexpensive medium, the expression system is well characterized, and a large number of expression vector systems and mutant host strains are available. In addition, under appropriate conditions, the recombinant protein can comprise >50% of the total cellular protein. However, the use of the E. coli expression system is limited because the target proteins often segregate partially or completely into the insoluble fraction of the cell. Recent studies of protein expression on a large number of prokaryotic and eukaryotic proteins indicate that >50% of recombinant proteins may be found in the insoluble fraction of bacterial cell lysates (Christendat et al. 2000a,b; Yee et al. 2002).

Several strategies have been used to increase the probability of producing soluble recombinant proteins in bacterial cells, including co-expressing protein folding modulators, manipulating the temperature of growth and induction, and producing the protein as a fusion with another soluble protein. However, not one of these techniques has been uniformly successful. Recently, the effect of altering fusion protein partners was evaluated systematically. Two groups monitored the solubility of different proteins expressed as fusions with six (Hammarstrom et al. 2002) or eight (Shih et al. 2002) different soluble proteins or affinity tags. By testing a variety of constructs, the percentage of recombinant proteins that could be detected in the soluble fraction of the cell increased from ∼50% to 80% to 85% overall.

An alternative approach to generate recombinant proteins for functional and structural studies is to purify proteins from inclusion bodies. In this approach, the insoluble proteins must first be solubilized with a denaturant and then refolded into a soluble native conformation to be useful for structural studies. Currently, there is no generally applicable method to refold insoluble proteins. One of the emerging trends is to use an array of refolding conditions to screen for a single condition that is compatible with a given protein. However, there is no body of evidence that describes the success rates of these procedures. The challenges associated with developing generic refolding procedures, as well as the perception that refolding methods, are usually unsuccessful and have limited the widespread use of refolding as a first-tier protein purification strategy. If such approaches could be developed, the use of insoluble protein as a starting material has several advantages. First, it would not be necessary to use widespread screens of different protein constructs and expression vectors in an attempt to produce protein that is soluble in vivo. Second, the use of chaperones to assist protein folding in vivo would not be required. Third, insoluble proteins are less susceptible to proteolytic degradation compared with soluble proteins. Finally, it would be possible to express toxic proteins, which inhibit cell growth if they are expressed in the soluble fraction.

The refolding of proteins from the insoluble cellular fraction is commonly accomplished by solubilizing the protein in a chaotropic agent, such as guanidine hydrochloride (GuHCl) or urea, and then removing the denaturant by dialysis or rapid dilution. The efficiency of protein renaturation depends on the competition between correct folding and aggregation, and there is evidence that the presence of contaminants in the refolding buffer can significantly decrease the yield of refolded protein (Maachupalli-Reddy et al. 1997). In a study of lysozyme refolding, aggregation increased when plasmid DNA, lipopolysaccharides, or proteins that aggregate upon refolding were added to the renaturation mixture. Other studies clearly show that the removal of contaminants before preparative refolding increases the yield (Babbitt et al. 1990; Wong et al. 1996; Tran-Moseman et al. 1999).

Although refolding strategies have been successful on a case-by-case basis, the proportion of proteins that can be denatured and refolded is unknown, and the success rate of refolding methods can only be inferred from anecdotal information. We set out to estimate the success rate of refolding methods by performing denaturation/refolding experiments on a large number of proteins chosen from four test organisms under study in our laboratory. Proteins from E. coli, a Gram-negative bacterium (Blattner et al. 1997); Thermotoga maritima, a thermophilic eubacterium with an optimal growth temperature of 80°C (Nelson et al. 1999); Methanobacterium thermoautotrophicum, a lithoautotrophic thermophilic archaeon that grows optimally at 65°C (Smith et al. 1997); and the unicellular eukaryotic budding yeast Saccharomyces cerivisiae (Goffeau et al. 1996) were cloned into overexpression vectors and purified from E. coli. Our aim was to determine if the success rate for refolding methods was high enough to warrant their inclusion in the standard protein purification arsenal, particularly in a structural proteomics setting in which a large number of proteins are expressed and purified at once.

Results

Target selection

Two sets of protein targets were selected for denaturation/refolding analysis. The first set comprised 70 proteins, the "insoluble proteins", which were known to partition to >90% in the insoluble fraction of an E. coli cell lysate. These 70 proteins, which were chosen from three organisms (S. cerivisiae, T. maritima, and M. thermoautotrophicum) did not have predicted transmembrane regions or known structural homologs. The second set comprised 25 proteins from four organisms (E. coli, S. cerivisiae, T. maritima, and M. thermoautotrophicum) that were produced in the soluble fraction of an E. coli cell lysate, and had been previously characterized by using NMR spectroscopy. This set was included in our analysis to compare the suitability of denaturation/refolding strategies for proteins that could otherwise be produced in soluble form using a native protein purification strategy.

All proteins in both sets ranged between 6 and 18 kD, so that the quality of the refolded sample could be analyzed by NMR spectroscopy. Each of the 25 soluble proteins (labeled with ¹⁵N) had previously been purified from the soluble fraction and analyzed by two-dimensional ¹⁵N-edited HSQC. On the basis of the HSQC spectrum, which provides a "signature" pattern of amide ¹H-¹⁵N resonances, the proteins were classified as good, promising, dilute, or poor candidates for NMR structure determination. The good spectra showed well-dispersed peaks of approximately equal intensity and of the number expected for the sequence of the protein. Promising spectra showed well-dispersed peaks that were either too few or too many, or of unequal intensities, indicating conformational heterogeneity or the presence of dynamic processes on an intermediate timescale that broaden or obscure NMR signals. The quality of these protein samples may be improved by changing the solution conditions or the size of the protein construct. Dilute protein samples were those that precipitated out of solution and, thus, gave extremely weak or no NMR signal. The classification of poor spectra was used for both unfolded and aggregated protein samples. Unfolded proteins show many sharp intense peaks with chemical shifts consistent with random coil conformation. Aggregated proteins are characterized by too few peaks, which are broadened and clustered in the center of the spectrum. Proteins that form large stable oligomers will generate spectra that will be classified as poor. Most of the 25 proteins generated either good or promising HSQC spectra or produced crystals after native protein purification. The classifications of the NMR spectra of these 25 soluble proteins were as follows: 17 good, 4 promising, 2 dilute, and 2 poor (Table 2).

Table 2.

Comparison of the HSQC results for the protein samples that were expressed in the soluble fraction in vivo

gi#a	No. of amino acids	% Refolded	Denaturing HSQC	Native HSQC
Escherichia coli
1789047	61	100	good	good
1790579	112	80	poor	promising
1786899	148	100	good	promising
1788196	167	0		good
1786341	179	50	good	good
1787927	82	80	good	good
1786351	114	50	good	good
1789689	127	100	good	dilute
Thermotoga maritima
4980762	83	100	good	good
4981224	151	50	good	good
4981243	152	0		dilute
4981256	149	80	good	poor
4981409	158	95	good	good
4981518	87	80	good	good
4981520	118	5		good
4981522	79	100	good	good
4981537	135	50	promising	good
4981552	158	95	good	good
4981598	101	100	good	promising
4981628	124	100	good	good
4981728	137	100	good	poor
4982075	150	100	good	good
Saccharomyces cerivisiae
6322902	133	80	good	good
Methanobacterium thermoautotrophicum
2621511	145	90	good	promising
2621893	62	100	good	good

^aNational Center for Biotechnology Information (NCBI) protein identification number (PID).

Purification and refolding of well-expressed insoluble proteins from E. coli

The 70 insoluble proteins were expressed in E. coli, and the whole-cell pellets were solubilized in a solution containing 6.9 M GuHCl. Purification of the His-tagged protein was performed in the same solution by using Ni-affinity chromatography, and the samples were refolded by dialysis against buffer containing 25 mM phosphate (pH 6.8) and 250 mM NaCl. A total of 41 of the 70 (58%) samples remained soluble after dialysis with at least 50% recovery as estimated by SDS-PAGE (Fig. 1 ▶, Table 1). The three-dimensional conformation of each of the 41 soluble protein samples was probed by using far-UV circular dichroism (CD) spectroscopy. In the far-UV region (<250 nm), the spectral characteristics of a protein are primarily determined by the conformation of its polypeptide backbone, especially its secondary structure. Each protein was classified according to its spectrum into one of four groups: α-helical, β-sheet, unusual, or random coil (Table 1). Representative CD spectra are shown in Figure 2 ▶. Of the 41 protein samples that remained soluble after dialysis, 21 showed significant α-helical content, 5 showed β-strand character, 5 were unusual, and 7 showed spectra consistent with random coil. Three samples could not be concentrated to yield a sufficient CD signal, due to precipitation. In summary, by using a simple refolding strategy, 31 of the 70 insoluble proteins could be purified and refolded, resulting in a protein sample that demonstrated some secondary structure and remained in solution at low concentration.

Figure 1.

SDS-polyacrylamide gels showing the fractionation of seven refolded yeast proteins into the pellet (P) or supernatant (S) after centrifugation of the sample at 15,000 rpm for 15 min. Protein marker molecular weights (kD) are indicated at right.

Table 1.

Refolding and biophysical analysis of the 70 protein samples that are insoluble in vivo

gi#a	No. of Residues	% Refolded	CD result	HSQC
Saccharomyces cerivisiae
6322622	116	100	α-helical	poor
6325453	109	100	β-sheet	poor
6320900	107	100	β-sheet	unfolded
6324270	110	0
6322568	123	0
6325316	128	0
6321730	73	0
6319465	145	0
6321928	148	50	α-helical	poor
6321669	137	100	random coil	unfolded
14318509	148	0
6324454	102	100	dilute	poor
6321245	103	100	dilute	poor
6319472	104	80	α-helical	poor
10383788	136	100	random coil	unfolded
6321445	84	100	unusual	poor
6322784	147	0
6320815	87	100	α-helical	poor
14318500	117	0
6321306	111	100	α-helical	poor
6319745	105	80	random coil	unfolded
6323397	110	60	random coil	poor
6321317	126	0
6322326	105	100	no signal	poor
6321850	130	100	unusual	poor
6323850	74	0
6321037	122	100	unusual	poor
6323326	124	100	random coil	unfolded
6324523	106	80	α-helical	dilute
6324941	117	0
83189	146	5
6324001	146	100	α-helical	poor
6325210	99	50	α-helical	unfolded
6322681	137	80	uunusual	good
6320760	122	5
6323190	114	10
6320458	149	5
6319609	133	100	α-helical	poor
6325115	105	50	α-helical	good
10383803	146	50	β-sheet	dilute
7839146	126	0
6322262	108	5
Thermotoga maritima
4981121	151	100	α-helical	promising
4981364	101	0
4981437	101	100	unusual	poor
4981475	118	100	α-helical	2.8-Åb
4981497	135	50	α-helical	poor
4981533	101	0
4981568	111	0
4981614	114	0
4981889	111	100	random coil	unfolded
4981890	144	0
4982100	92	80	β-sheet	good
4982192	121	0
4982245	117	0
Methanobacterium thermoautotrophicum
2621100	114	80	α-helical	dilute
2621252	127	100	random coil	unfolded
2621282	125	5
2621294	138	30
2621458	126	50	α-helical	promising
2621459	148	0
2622053	106	50	α-helical	poor
2622351	148	100	β-sheet	poor
2622414	63	50	α-helical	poor
2622420	136	60	α-helical	poor
2622488	132	100	α-helical	good
2622555	169	90	α-helical	dilute
2622607	148	0
2622899	135	100	α-helical	good
2621045	50	100	random coil	unfolded

^aNational Center for Biotechnology Information (NCBI) protein identification number (PID).

^bThis sample produced a poor HSQC, but formed native crystals that diffract to 2.8 Å.

Figure 2.

Representative circular dichroism spectra used for protein secondary structure classification. Proteins were classified as either α-helical (gi 2621893; diamonds), β-sheet (gi 4981224; empty circles), unusual (gi 4981537; solid circles), or unfolded (gi 10383788; squares).

To assess the suitability of proteins for NMR spectroscopy or for crystal trials, it is necessary to concentrate the proteins to >0.3 mM. Of the 31 proteins that could be solubilized and refolded, 24 could be concentrated at >0.3 mM. These samples were uniformly labeled with ¹⁵N and concentrated by ultrafiltration for NMR data collection and for crystal trials. Seven of the 24 samples yielded ¹⁵N-HSQC spectra that could be considered good or promising (Table 1, Fig. 3 ▶), and one sample formed a crystal that diffracted to 2.8 Å. Therefore, by using a simple refolding strategy, we were able to rapidly generate purified protein for 58% of the 70 small insoluble proteins and to generate structural samples for ∼10%.

Figure 3.

¹⁵N-HSQC spectra of the seven proteins expressed in the insoluble fraction of E. coli that provided good or promising structural samples.

Comparison of native and denaturing protein purification protocols

We performed denaturation/refolding studies on 25 well-characterized proteins that are soluble in vivo for two reasons. First, we wanted to compare the success rates of the native and denaturing protein purification methods on a common set of proteins. Second, if some soluble proteins could also be purified by using the denaturation/refolding approach, we wanted to ensure that the renatured proteins adopted the same three-dimensional conformation as those purified from the soluble fraction by using a native purification procedure.

The set of 25 well-expressed soluble proteins was purified by using both methods. Of the 25 samples, 22 (88%) refolded. The 88% refolding rate for this set of proteins is significantly higher than the rate of 55% achieved with the insoluble proteins, indicating that proteins that are soluble in vivo will be better behaved in vitro. Two of the three proteins that were unable to be refolded had been classified as good, with the remaining third as dilute when purified by using the native purification protocol. However, although two good samples were unable to be purified by denaturation/refolding, the renaturation approach actually improved the behavior of four other proteins. Two proteins classified as poor, one classified as dilute, and one classified as promising when purified by using the native protocol were classified as good when purified using the denaturation/refolding protocol. Altogether, there were 21 samples classified as good or promising for each of the denaturing and native protein purification strategies. Figure 4 ▶ compares the HSQC spectra obtained for six samples using both denaturing and native purification protocols. The nearly identical spectra obtained by using the two purification protocols illustrates that the renatured proteins are adopting the same three-dimensional conformations as the proteins purified under native conditions.

Figure 4.

Representative ¹⁵N-HSQC spectra of 6 of the 14 proteins that produced good structural samples when prepared by both native and denaturing protocols. Spectra for the proteins purified by both denaturing (d, upper panels) and native (n, lower panels) methods are shown.

Discussion

Up to 50% of cytosolic bacterial and archaeal proteins are sequestered to the insoluble fraction of the cell when overexpressed in E. coli. High-throughput structural proteomic projects will face an increasingly difficult task as the three-dimensional structures of the soluble proteins are solved, and new approaches to deal with nonideal protein samples will need to be developed. The development of a simple and efficient renaturation procedure that can be applied to the insoluble proteins provides the most straightforward strategy to produce large amounts of recombinant protein for structural and functional studies.

We have examined the refolding behavior of a group of 95 small proteins. A simple denaturing/refolding protocol provided a source of soluble folded protein for 58% of proteins that were insoluble in vivo. Slightly >10% of the proteins that were purified from the insoluble fraction generated good samples for NMR spectroscopy. In previous studies, we have shown that 33% of the small proteins expressed in the soluble fraction of E. coli provide good HSQC spectra (Yee et al. 2002). The success rate with which we recovered good samples from the insoluble fraction is significantly lower, although only one set of buffer refolding conditions was examined. A larger percentage of structural samples might be recovered from the insoluble fraction by exploring a wider array of refolding protocols, such as rapid dilution or refolding while immobilized on a column, which may help prevent aggregation of folding intermediates.

It may also be possible to increase the fraction of refolded proteins by exploring a number of different solution conditions, especially for those proteins that contain common metal ions and/or cofactors. There are commercially available kits that can be used for this purpose. Rapid dilution or column-based refolding could be easily automated by using 96-well plates. The extent of refolding in this format could be monitored by using UV spectrophotometry, NMR, or CD spectroscopy. There are several advantages to using CD spectroscopy as a monitor for protein structure, including the speed of data collection, the relatively simple interpretation of the spectra, the ability to collect spectra under a wide variety of conditions, the small amount of sample required, and the ability to recover the sample. The major disadvantage is that the technique gives only a global "average" view of the protein. Although deconvolution of the spectra can approximate the amount of secondary structure present, it cannot be related to the exact structure of the protein. This is not a concern for screening buffer conditions, as the samples could be evaluated according to presence of identifiable secondary structure and absence of light scattering, which indicates a soluble folded protein.

Finally, the addition of a crystal screening step could also potentially increase the number of structures determined for proteins recovered from the insoluble fraction of the cell. We set up crystal trials with 12 samples that gave poor HSQC spectra and found that one of them formed a crystal that diffracted to 2.8 Å. A previous study of 46 small proteins comparing the effectiveness of NMR and crystallography in generating structural samples found that three proteins that exhibited poor HSQCs could be solved to high resolution by using X-ray crystallography (Savchenko et al. 2003). Other unpublished results from our laboratory that support this finding include the high-resolution crystal structures of three proteins from a group of 55 small proteins that exhibited poor HSQCs (A. Yee, D. Christendat, A.M. Edwards, and C.H. Arrowsmith, pers. comm.). These results indicate that we could produce good structural samples for another 5% of proteins that are insoluble in vitro by combining NMR and crystallographic studies.

Denaturing purifications may be desirable not only for totally insoluble proteins but also for proteins that are not expressed to high levels in E. coli and for those that are partially sequestered to inclusion bodies. By denaturing the total cellular protein, it would be possible to recover a larger fraction of the recombinant protein, not just the amount present in the soluble fraction. This could translate to significant cost savings when labeling proteins with ¹³C and ¹⁵N for NMR studies, or with selenomethionine for crystallographic studies. There are also a number of technical advantages to using the denaturing protein purification protocol. For example, many of the strategies used to increase protein solubility in vivo function simply by decreasing the rate of protein expression. These methods include performing the induction step at low temperature, inducing with a nonmetabolizable carbon source such as desoxyglucose, or inducing with limited amounts of inducer added to the culture. If the requirement to express the protein in the soluble fraction of E. coli was obviated, then induction of the recombinant protein could be carried out under conditions that optimize expression levels, but not solubility (i.e., induce for a few hours at 37°C).

It is important to assess the conformational differences between proteins purified by native and denaturing methods, particularly when studying proteins for which an activity assay is unavailable. No studies have systematically compared the three-dimensional structures of proteins purified by using both purification methods. In this study, we examined the ¹⁵N-HSQC NMR spectra of 22 proteins that were purified by both native and denaturing protocols, and we found that the spectra look very similar in each case. The minor differences observed between the pairs of proteins are likely due to small amounts of degradation or slight differences in the sample buffer. These results give us confidence that the three-dimensional structures of the proteins that were refolded from denaturant are the same as those purified by using a native purification protocol.

Conclusion

The recovery of recombinant protein from the insoluble fraction of E. coli cell lysates was thought to require technically diverse and often complex refolding procedures. We purified 25 proteins by using native and renaturation approaches. Each of the procedures generated 21 good or promising structural samples. A denaturation/refolding protocol is therefore appropriate for most small proteins that would normally be found in the soluble fraction. By using a simple denaturing protein purification strategy, we were able to obtain soluble protein samples for 58% of small proteins that were insoluble in vivo, with 10% providing good HSQC NMR spectra. We conclude that, particularly for high-throughput studies of small proteins, it is more efficient to perform a denaturing/refolding protocol than a native purification.

Materials and methods

Expression and solubility tests

Target proteins were PCR-amplified from genomic DNA and cloned into the expression vector pET15b (Novagen) as a fusion with an N-terminal 6-His affinity tag and a thrombin cleavage site, or a modified pET15b vector with a TEV protease cleavage site. The fusion proteins were overexpressed in the E. coli strain BL21 STAR (Novagen). Initial trials to determine protein solubility were performed by using a culture volume of 3 mL of Luria broth (LB) in 24-well polypropylene microtiter plates. Three to five colonies were picked from fresh transformations and used to inoculate the LB cultures, which were grown at 37°C to an A₆₀₀ of ∼0.6. Protein expression was induced by the addition of 175 μg/mL of IPTG, followed by incubation overnight at room temperature. The cells were harvested and lysed in BugBuster (Novagen), to release any soluble proteins. The insoluble proteins and cell debris were then removed by centrifugation, and the soluble and insoluble cell fractions were analyzed by SDS-PAGE followed by Coomassie staining.

For large-scale production of the protein samples, cells were grown at 37°C in M9 minimal medium enriched with 0.7 g/L of ¹⁵N-NH₄Cl to an A₆₀₀ of 1.0. Protein expression was induced by the addition of isopropyl-β-d-thiogalactopyranoside (final concentration, 175 μg/mL), followed by incubation for 4 h at 37°C. The cells were harvested by centrifugation and lysed in 6 M GuHCl, 100 mM NaH₂PO₄, 10 mM Tris-HCl, and 10 mM imidazole (pH 8.0) and were purified in the same buffer via a batch method by using nickel-nitrilotriacetic acid-agarose resin (Qiagen). The pure proteins were eluted with 6 M GuHCl and 0.2 M acetic acid and were refolded by dialysis into 25 mM phosphate (pH 6.8), 250 mM NaCl, and 2 mM DTT. After dialysis, any protein precipitate was collected by centrifugation at 15,000 rpm for 15 min, and was resuspended in 8 M urea, 100 mM NaH₂PO₄, and 10 mM Tris-HCl (pH 8.0). Equal amounts of the soluble fraction of the sample and the solubilized precipitate were analyzed by SDS-PAGE followed by Coomassie staining. The percentage of soluble protein present in solution versus the insoluble precipitate was then estimated.

Biophysical analysis

CD wavelength scan experiments were performed in an Aviv 62A DS CD spectrometer. The soluble fractions of the refolded protein solutions were analyzed immediately after removal from dialysis. The protein concentrations ranged from 20 to 50 μM. Data was collected at 25°C from 260 to 200 nm (1-nm increments), with a 2-s averaging time.

All ¹H-¹⁵N HSQC spectra were acquired at 25°C in a Varian INOVA 500- or 600-MHz spectrometer equipped with a pulse-field gradient unit and actively shielded z-gradient triple resonance probes. The total number of t₁ increments was 64, with the number of scans per increment ranging from eight to 64, depending on the concentration of sample being examined. The data were processed by using NMRPIPE software package (Delaglio et al. 1995).