Influence of circular permutations on the structure and stability of a six‐fold circular symmetric designer protein

Abstract

The β‐propeller fold is adopted by a sequentially diverse family of repeat proteins with apparent rotational symmetry. While the structure is mostly stabilized by hydrophobic interactions, an additional stabilization is provided by hydrogen bonds between the N‐and C‐termini, which are almost invariably part of the same β‐sheet. This feature is often referred to as the “Velcro” closure. The positioning of the termini within a blade is variable and depends on the protein family. In order to investigate the influence of this location on protein structure, folding and stability, we created different circular permutants, and a circularized version, of the designer propeller protein named Pizza. This protein is perfectly symmetrical, possessing six identical repeats. While all mutants adopt the same structure, the proteins lacking the “Velcro” closure were found to be significantly less resistant to thermal and chemical denaturation. This could explain why such proteins are rarely observed in nature. Interestingly the most common “Velcro” configuration for this protein family was not the most stable among the Pizza variants tested. The circularized version shows dramatically improved stability, which could have implications for future applications.

Short abstract

PDB Code(s): 6I37, 6I38, 6I39, 6I3A and 6I3B;

1. INTRODUCTION

For over 30 years, the goal of protein design has been to put our understanding of protein stability and folding to the test. The ability to create at will a desired protein to chosen specifications would demonstrate irrefutably the depth of our understanding of the forces governing protein structures. ¹ While tests of fundamental physico‐chemical principles drove the first designs of α‐helical peptides, ² the dynamic between utilizing and gathering knowledge has now changed. In recent years, protein design has been used as a tool to broaden our comprehension of protein evolution and folding, as well as creating molecules with practical applications. In such studies, repeat proteins were often the target because of their low sequence complexity relative to their size. Current evolutionary theory holds that repeat proteins evolved from small peptide fragments which were duplicated and then fused together. ³ However, while the symmetry in the gene sequences and structures of these proteins strongly indicates that such events happened, the process requires perfectly symmetrical evolutionary intermediates, ⁴ and protein design first showed that tandem linking of identical repeats can produce a stable, folded protein molecule.

In general, repeat proteins can be grouped into one of two classes, linear or cyclic. The proteins that arrange in a linear fashion have repeated domains that only contact their immediate neighbors. These proteins vary widely in the number of repeats, and include for example the Ankyrin repeats, the Leucine‐rich repeats and the HEAT repeats. ⁵ , ⁶ Designer proteins based on these include the DARPin proteins among others. ⁷ Utilizing the same principles as these natural proteins, the group of Baker created a de novo protein with a sequence unrelated to any found in nature ⁸ demonstrating that such proteins can even be designed without the aid of a natural template.

Cyclic repeat proteins show a structure built from similar domains arranged around a central symmetry axis. These proteins often possess a hydrophobic core and have a fixed number of repeats. This group includes the β‐trefoil fold and the widespread TIM‐barrel family. ⁹ The internal symmetry of these proteins has been exploited to reduce the sequence design space, resulting in the first designed TIM‐barrel. ¹⁰ This study showed that polar interactions, as well as the hydrophobic core, are required to stabilize the barrel structure. Trefoil proteins are built from three tandem copies of a sub‐domain with roughly 50 residues. The Blaber group designed a perfectly three‐fold symmetrical trefoil called Symfoil, with three identical copies of the same sequence, ¹¹ and then used this as a model system to investigate the folding nucleus of the trefoil fold. ¹²

The β‐propeller family is atypical of the cyclic repeat proteins because propeller proteins are found to vary unusually widely in the number of repeats. Natural propeller proteins are known with as few as four repeats, and as many as 10. Each repeat (also called a blade) consists of four anti‐parallel β‐strands. ¹³ β‐propellers are common and have variety of functions in nature, ranging from enzymatic to protein–protein interactions. ¹⁴ The propeller itself is held together by hydrophobic interactions between the blades, but frequently some additional stabilization is present, such as the disulfide bond often found between the first and last blade of four‐bladed propellers.

In the larger propellers extra stabilization is provided by a “Velcro” closure, an arrangement in which the N and C termini are not found between repeats, but within one blade. When the C‐terminal part of the split blade consists only of the innermost β‐strand, so that the N‐terminus is buried within the protein, the protein is said to possess a “3 + 1 Velcro.” Alternatively, in a protein with “1 + 3 Velcro,” the C‐terminal part of the split blade consists of the three inner strands. Both these conformations are common in nature. ¹³ The “2 + 2 Velcro,” with one blade split evenly, has only been observed in RCC1‐like domains, as far as we are aware. ¹⁵ RCC1 (Regulator of Chromosome Condensation 1) was the first protein shown to adopt a seven‐fold β‐propeller structure with a “2 + 2 Velcro.” ¹⁶ Since then homologues have been discovered, in both prokaryotes and eukaryotes, possessing the same repeating motif. ¹⁵ Similarly, to date only one sub‐family of propeller proteins is known with no “Velcro” strap or other closing mechanism present, the prolyl‐oligopeptidases. ¹⁷ The lack of the usual stabilization mechanisms found in propellers was predicted to enhance catalysis by the protein, which was confirmed by showing the introduction of a cystine bond between the blades decreased enzyme activity. ¹⁸

We previously designed a β‐propeller protein with six identical blades by attempting to reverse evolutionary drift and recreate a theoretical evolutionary precursor. This protein, called Pizza6, can fold readily and is extremely stable. ¹⁹ Crystal structures showed that each identical 42‐residue blade within the protein closely matched the predicted model. It was found that a protein consisting only of either two or three of these blades could also fold, giving a trimer or dimer respectively. These proteins, called Pizza2 and Pizza3 were found to be roughly as stable as Pizza6, implying that the entropic cost of oligomerization played little role in the overall protein stability. Both Pizza2 and Pizza3 were initially designed with a seven residue “1 + 3 Velcro” closure, but “non‐Velcro” variants were also found to form stable complexes with a total of six blades. The group of Tawfik also illustrated the gene duplication and fusion pathway with the creation of a symmetric form of Tachylectin‐2, although they used a protein engineering method based on phage display instead of computational protein design. ²⁰

In order to investigate the stability and unfolding of the Pizza protein, mutants with tryptophan residues were designed by virtual screening, testing each residue in the original blade sequence. One mutant called Pizza6‐AYW, with a tyrosine replaced by tryptophan in each blade, was found to express well and fold with a stability similar to PknD propeller domain, the naturally occurring template of the Pizza protein. ²¹ In this study, we have used this Pizza variant to test the influence of different “Velcro” arrangements on the protein's structure, stability and folding.

2. RESULTS

Pizza6‐AYW, like the template protein from which it was created (RCSB: 1RWL), possesses a “1 + 3 Velcro” closure, so that the N‐terminus forms the outermost strand of the split blade. In order to study the influence of the “Velcro” position, we made cyclic permutations of the sequence to change the location of N‐ and C‐termini of the protein. We created four mutants in all, two of which, nv1Pizza6‐AYW and nv2Pizza6‐AYW, lack a “Velcro” closure (“nv” stand for “non‐Velcro”). The third protein (v31Pizza6‐AYW) has a “Velcro” closure formed by a single strand at the C‐terminus of the protein, a common arrangement in natural proteins. The last mutant (v22Pizza6‐AYW) has a rarely observed feature with the split domain shared equally between the two ends of the protein. The positions of the N‐ and C‐termini are shown for each mutant in Figure 1. In addition to creating mutants with permuted termini, we also created a protein lacking any termini by utilizing the split‐intein gp41‐1 to create a cyclic polypeptide. We refer to this protein as circular Pizza6‐AYW or cPizza6‐AYW.

FIGURE 1.

Model structures of all proteins used in this study. A beads and ribbon representation of the first blade is shown on the left with the residues colored from N‐terminus (blue) to C‐terminus (red). As indicated by the black lines, the N‐terminus changes position within the β‐sheet depending on the location of the “Velcro.” Each line is labeled with a colored letter, linking it to the cartoon representation of this mutant on the right. In this cartoon representation the first repeat of each protein is colored blue to red from N‐terminus to C‐terminus, this again illustrates the shifting position of the N‐terminus within a blade. The peptide bond formed after the split‐intein is self‐cleaved is shown in purple in the structure of cPizza6‐AYW

DNA sequences encoding the proteins were purchased and cloned into a suitable vector for protein expression (Table S1). Variants with two (Pizza2) and three (Pizza3) blades were also cloned, in order to test whether the mutants retained the self‐assembling properties of the original Pizza protein. cPizza6‐AYW was created using a plasmid developed specifically for the gp41‐1 split‐intein system. ²²

All proteins were expressed in E. coli BL21, and the histidine‐tagged proteins could be purified with IMAC. Protein purity was verified with SDS‐PAGE and analytical size exclusion chromatography (see Figure 2). Both indicate that the proteins were monodisperse and had the expected molecular weight. cPizza6‐AYW appeared slightly smaller. The circular dichroism (CD) spectra of all the proteins (measured at room temperature) are nearly identical, with a negative peak at 212 nm, indicating the proteins have the same fold. To confirm this, we crystallized all the different variants. Each protein crystallized under different conditions (Table S2), and diffraction data could be collected to resolution limits between 1.0 and 1.58 Å (Table S3). The structure could be phased using the original Pizza6‐AYW (PDB: 6F0Q) as a model for molecular replacement. All the loops of cPizza6‐AYW were clearly observable in the electron density map, indicating the peptide bond was successfully formed (Figure 3). We could observe seven bromine ions in the crystal structure of v22Pizza6‐AYW and verify their identity with anomalous diffraction (Figure S1). The central cavity of v31Pizza6‐AYW contains a magnesium ion in a highly coordinated water cluster (Figure S2).

FIGURE 2.

Purification data and protein structure. (a) SDS‐PAGE analysis of all six proteins, (b) result of analytical size exclusion performed on a Superdex200 increase column, and (c) circular dichroism spectra for all the proteins at room temperature

FIGURE 3.

Protein structure of cPizza6‐AYW and Pizza6‐AYW illustrating the circularized peptide bond. On top, the protein is depicted in ribbon, with lines and electron density at sigma level at the regions where the termini could be located. A zoom in on the black square region is depicted at the bottom. A clear connection can be observed in cPizza6‐AYW while one residue is clearly absent in the case of Pizza6‐AYW

None of the two‐bladed fragments expressed sufficiently for the protein to be analyzed. The three‐bladed fragments could be expressed, but nv2Pizza3‐AYW aggregated during purification. The soluble Pizza3 variants behaved like the Pizza6 parent protein under size‐exclusion chromatography, apparently combining as a dimer. However, only Pizza3‐AYW, with the native “Velcro” position, possesses an identical CD‐spectrum to the six‐bladed proteins. The spectra of the other three‐bladed mutants have shifted towards 200 nm more closely resembling the spectra in the unfolded state (Figure S3).

To determine whether the “Velcro” position affects the protein stability, we exposed the mutants to high temperatures and different denaturants. Thermal unfolding was monitored with CD spectroscopy at 218 nm and with differential scanning fluorimetry (DSF) (see Figure 4). The curves were fitted with a Boltzmann sigmoid (Figure S4), and the results can be seen in Table 1. CD spectroscopy showed that the protein unfolds reversibly (Figure S5), although a prolonged stay at high temperature will cause the proteins to aggregate causing irreversible changes. Since these techniques evaluate different unfolding processes and different buffers were used, the melting temperature differed slightly but the relative difference is the same for each protein. The variants lacking the “Velcro” closure are the least stable, and of the proteins with “Velcro,” v22Pizza6‐AYW is the least stable. From the fitting of the CD data, the difference in free energy between the mutants and Pizza6‐AYW was calculated. ²³ As a negative value indicates an increase in stability. The circular construct is the most stable protein of all. For this last protein, although the trough in CD at 218 nm becomes smaller above 65°C, it does not completely disappear as is the case with the other proteins.

FIGURE 4.

The thermal and chemical stability of the “Velcro” mutants. (a) CD signal measured at 218 nm as a function of temperature in a phosphate buffer at pH 7.6, (b) fraction of denatured protein as observed with DSF in the proprietary Protein Thermal Shift™ buffer, (c) fraction of denatured protein plotted against increasing concentration of guanidine hydrochloride. (d) The Chevron plot, combined folding and unfolding rates of the four least stable proteins

TABLE 1.

Thermodynamic stability parameters

	Tm (°C) (CD)	ΔHm (kJ/mol) (CD)	ΔΔGD − N (kJ/mol) (CD)	Tm (°C) (DSF)
nv1Pizza6‐AYW	41.78 ± 0.01	880 ± 10	24.9 ± 0.3	46.6 ± 0.4
nv2Pizza6‐AYW	45.32 ± 0.02	970 ± 14	16.7 ± 0.2	50.6 ± 0.6
v31Pizza6‐AYW	50.52 ± 0.02	430 ± 4	4.56 ± 0.06	60.5 ± 0.4
v22‐Pizza6‐AYW	44.13 ± 0.01	720 ± 5	19.4 ± 0.3	50.88 ± 0.17
Pizza6‐AYW	52.48 ± 0.02	758 ± 9	—	57.2 ± 0.8
cPizza6‐AYW	64.61 ± 0.05	490 ± 10	−28.2 ± 0.4	68.1 ± 0.9

Unfolding with chemical denaturants was monitored by tryptophan fluorescence, measuring the emission at 330 nm. Experiments were performed with guanidine hydrochloride (GdnHCl), ethanol, hydrochloric acid and sodium hydroxide. The emission was plotted from zero to 6 M GdnHCl for each construct. A much larger difference was observed than with thermal melting, but the order remained the same. nv1Pizza6‐AYW is the least stable followed by nv2Pizza6‐AYW, v22Pizza6‐AYW and Pizza6‐AYW have similar stability with 50% of the protein unfolded around 1 M GdnHCl, v31Pizza6‐AYW remained highly stable with only cPizza6‐AYW having a higher resistance to the denaturants. When possible, the data were fitted with a Boltzmann sigmoid, fitting parameter are provided in Table 2, the difference in free energy between mutants and Pizza6‐AYW was calculated again. The same hierarchy as for thermal measurements is observed. To distinguish differences in stability between the four least stable proteins, an additional experiment was carried out with GdnHCl concentrations between 0 and 2 M with smaller intervals. This indicated that the original Pizza6‐AYW is the most stable of these. With possible applications in protein nanotechnology in mind, we also tested the stability at different pH and ethanol concentrations. The “non‐Velcro” variants unfolded at low pH while v31Pizza6‐AYW remained stable above pH ~ 3, and unfolded substantially at pH ~ 2, whereas cPizza6‐AYW remained stable even at pH ~ 2. In contrast, all proteins remained stable at high pH. Ethanol showed changes in tryptophan fluorescence between 10% and 30% (vol/vol)

TABLE 2.

Isothermal denaturation parameters

	$\mathrm{\Delta }{G}_{D-N}^{{\mathrm{H}}_2\mathrm{O}}$ (kJ/mol) a	m value (kJ/mol M−1)	[D]50% (M)	$\mathrm{\Delta \Delta }{G}_{D-N}^{D_{50\%}}$ (kJ/mol)
nv1Pizza6‐AYW	25.1 ± 1.6	38 ± 2	0.660 ± 0.007	9.2 ± 1.1
nv2Pizza6‐AYW	31 ± 4	40 ± 5	0.782 ± 0.006	5.5 ± 1.2
v22Pizza6‐AYW	21.7 ± 1.3	29 ± 1.5	0.749 ± 0.006	5.4 ± 1.0
Pizza6‐AYW	24.6 ± 1.0	25.9 ± 0.7	0.949 ± 0.014	/

^a $\mathrm{\Delta }{G}_{D-N}^{{\mathrm{H}}_2\mathrm{O}}$ ₌ m ∙ [D]_50%.

We measured the kinetics of unfolding and folding for the four least stable proteins by monitoring tryptophan fluorescence emission at 330 nm with a plate reader. v31Pizza6‐AYW and cPizza6‐AYW could not be fully unfolded so the kinetics could not be measured. The refolding of proteins was measured by first unfolding the proteins completely in 8 M GdnHCl and then introducing them to a buffer with a low GdnHCl concentration, to give a final denaturant concentration between 0.12 and 0.72 M. Unfolding was measured between 1.5 and 5.9 M GdnHCl. The time traces were fitted with an exponential to extract the observed rate constant (Table S4, Figure S7), these were used to create a chevron plot (Figure 4, Table S5). While the folding rate is comparable among all four proteins, the constructs lacking a “Velcro” unfolded significantly faster than the other two. v22Pizza6‐AYW showed the slowest unfolding rate, with a rate constant of unfolding in the absence of denaturant ($k_{\mathrm{u}}^{{\mathrm{H}}_2\mathrm{O}})$ of 5 e − 6 s⁻¹.

Further analysis was carried out with NMR on the proteins to discover if they showed any significant structural differences in solution. Isotopic labeling was not used. The 1D proton and 2D NOESY spectra of all the proteins were found to have different line broadening but chemical shifts of signals are highly similar. Differences could be monitored on the isolated peak in the downfield region that has a lower intensity in the less stable proteins (Figure S8). This peak could be assigned to exchangeable HN in the imidazole ring of His23 (in nv1Pizza6‐AYW) in the loop region connecting the two middle blades. A closer view of the spectrum clearly shows a single peak for the stable proteins with “Velcro strap,” splitting into multiple peaks for proteins lacking this closure. The peak splits into two for v22Pizza6‐AYW probably because the histidine is close to the chain termini (Figure S9). Since the NMR spectra were all highly similar, we did not proceed to experiments with labeled protein samples.

3. DISCUSSION

The β‐propeller family is very large and diverse. Not only are the repeat motifs themselves highly variable, but the number of repeats is notably unrestricted over a considerable range. The location of the termini also differs, giving different “Velcro” conformations. It is impossible to estimate the influence of the “Velcro” on the stability of natural propellers as each repeat has a unique sequence. Recently however we were able to design the completely symmetrical Pizza protein with the same blade sequence repeated six times. This symmetry makes it possible to compare the effects of different “Velcro” positions and help our understanding in the evolution of the β‐propeller fold, as well as how to increase the stability of artificial propeller proteins. We compared proteins built from the same repeat sequence, but differing in the “Velcro” position, for their stability, folding kinetics and ability to self‐assemble.

The arrangement of the “Velcro” closure does not strongly influence the structure of the protein. Comparing the crystal structures, the C_α RMSD between the mutants and Pizza6‐AYW is below 1 Å, on a par with experimental error. It can also be seen that most of the deviation among these models occurs in the flexible loop between the two outermost strands. This might be due to crystal contacts or simple surface flexibility as these regions lie on the outside of the protein. Even the propellers with no form of “Velcro” remained stable. The identical chemical shifts in NMR spectra confirm that the same structure is adopted by each protein, but differences in line broadening indicate that parts of it might become more flexible without the “Velcro” strap. His23 of nv1Pizza6‐AYW, located in the middle connecting loop region, becomes more disordered. This retention of structure is consistent with a previous study on the propeller domain of the prolyl‐oligopeptidase family, which showed that other features of the structure, in particular the hydrophobic interactions, are strong enough to maintain the fold even without a “Velcro” strap. ²⁴

In contrast to the results for the Pizza6 variants, the “Velcro” strap does affect the self‐assembly of Pizza3 mutants. Only the protein with the original “Velcro” arrangement showed an unchanged CD spectrum compared to the Pizza6 proteins. In a previous study it was noted that “non‐Velcro” Pizza2 had a slight tendency to remain a monomer, but the self‐assembling properties could be restored by introduction of a metal coordinating site. ²⁵ The presence of the “Velcro” closure provides a stronger link between the folding of the individual subunits of Pizza2 and the protein as a whole, as the isolated subunit cannot fold completely without close interaction with its neighbors. This provides a powerful mechanism to ensure complete assembly of the protein into the final functional form and may explain why it is almost always retained in β‐propellers.

Contrary to these findings, the group of Tawfik previously used ancestral library design of the β‐propeller Tachylectin‐2 to find a functional ancestor that would self‐assemble from single repeats. Interesting they could only identify functional, self‐assemblies when there was no “Velcro” present. Suggesting that self‐assembly evolved before the circular permutations that led to the “Velcro” closure. ²⁶ , ²⁷ As Pizza can only self‐assemble from more than two repeats, it represents a further step in evolution then the single repeats of Tachylectin‐2. This might explain why the “Velcro” closure is needed for Pizza to self‐assemble.

In addition to β‐propellers, Trefoil repeats also possess a feature similar to a “Velcro” closure. In a recent study, the effects of this closure mechanism were investigated with the designed trefoil protein Symfoil. Of all circular permutants, only the original sequence was able to assemble correctly, indicating not only the importance of the closure mechanism in the trefoil fold, but suggesting it is not added as an evolutionary afterthought. ²⁸

In the case of Pizza we found that it is possible for the different circular permutations of the protein to fold, but the stability of the protein dependent significantly on the location of the “Velcro” closure. Unexpectedly, although the natural template possesses a “1 + 3 Velcro,” the 3 + 1 arrangement proved to be more resistant to denaturation, as can be seen from the ΔΔG of chemical denaturation experiments and melting temperature measured with DSF. From this study alone it is impossible to conclude whether this is a general feature of β‐propellers or specific for the Pizza protein. Different families of repeat proteins show different preferences for certain permutations. The Kelch and YWTD families tend to show “3 + 1 Velcro,” while the NHL and WD repeat families display the 1 + 3 variety. Since only Pizza variants with “1 + 3 Velcro” proved able to fold from smaller fragments (with 2 or 3 blades per polypeptide), it may be that the different propeller families have adopted the closure mechanism that gives them the best ability to assemble as complexes. Alternatively, they may have just retained the closure present in the ancestral motif prior to duplication and fusion. Future experiments using different recently designed proteins, such as Tako8 and Cake9, belonging to different propeller families may shed further light on the matter. ²⁹ , ³⁰ Nevertheless, the propeller fold may be highly successful because of an intrinsic adaptability that allows individual proteins to adopt different permutations to regulate their flexibility.

While the absolute value of the melting temperature between CD spectroscopy and DSF differs, the trend remains the same. In CD spectroscopy a characteristic absorption is used to observe secondary structure directly, while in DSF a fluorophore binds to hydrophobic residues of the protein exposed when it loses its tertiary structure. The propeller architecture consists almost entirely of β‐strands and is highly compact, so in this case fluorophore binding can begin only after substantial loss of secondary structure. The indirect nature of the measurement may explain the higher melting temperatures observed by DSF compared to the CD method. In addition, the measurements are also carried out in a different buffer (phosphate buffer at pH 7.6 for CD, and a proprietary Protein Thermal Shift™ buffer for DSF). Lastly, while in CD the folding is reversible, the DSF experiment is not reversible due to binding of the fluorophore, which may also contribute to the difference in melting temperatures.

When comparing the thermodynamic and kinetic measurements from chemical denaturation experiments the same trends can be observed. The unfolding transition point is found between 0.7 M and 0.8 M GdnHCl. Fitting the data to a Boltzmann sigmoid model yields a value for m, the proportionality constant that relates change in ΔG to change in denaturant concentration. nv1Pizza6‐AYW and nv2Pizza6AYW have the highest m‐value, and this sensitivity to denaturant may be because proteins without a “Velcro” have more flexible termini, less constrained by hydrogen bonds and more accessible to the solvent. On the other hand, restricted solvent access probably increases the resistance to chemical denaturation of v31Pizza6‐AYW and cPizza6‐AYW. This effect could also explain why these proteins are significantly more resistant to GdnHCl than to thermal denaturation, as heat energy naturally penetrates the protein regardless of its structure, allowing it to disrupt bonds that are required to maintain the fold. The rarely observed “2 + 2 Velcro,” placing both chain termini in the middle of the blade, slows down the rate of unfolding drastically, giving a rate of unfolding ($k_u^{{\mathrm{H}}_2\mathrm{O}})$ less than a quarter of that of Pizza6‐AYW. This slow unfolding does however not result in higher stability. The slow folding, with no compensating advantage in stability, may explain why this permutation is rarely observed in natural propeller proteins. It is also possible that it is evolutionarily harder for a variant to arise with precisely trimmed N‐ and C‐termini that fit the fold. In contrast, the 1 + 3 and “3 + 1 Velcro” allow extra residues at one end of the polypeptide chain, which extend into the bulk solvent. The “2 + 2 Velcro” position seems to be unique to the RCC1‐like domains. A bioinformatics study of the evolution of the β‐propeller fold showed that these domains are only distantly related to other propellers, which could also explain its different closing mechanism. ³¹

The effect of the Velcro position on function could not be investigated using Pizza, since it has no ligand binding or enzymic activity, but previous studies have investigated the effect of circular permutations on the function of β‐propellers. In a study on strictosidine synthase it was found that such mutants could be made but the catalytic activity of the proteins decreased. ³² However, instead of focusing on the “Velcro” closure, the study targeted two flexible regions near the active site to introduce the new termini.

Several recently designed symmetrical proteins have been reported, but none are perfectly symmetric due to the chain termini. Utilizing a split‐intein to form a peptide bond between the N‐ and C‐termini, we created a circular Pizza variant, the first perfectly symmetric protein. This has a dramatic effect on the stability, raising the T _m over 10°C and lowering the free energy compared to Pizza6‐AYW by 28.2 kJ/mol. This increased thermal resistance was previously observed with other circularized proteins. ²² , ³³ Unlike the other variants, the CD signal of cPizza6‐AYW does not disappear completely at 218 nm, suggesting this protein retains considerable native structure even after the thermal transition. The difference in stability is even greater when considering chemical denaturants. While most of the constructs showed essentially complete unfolding below 1.5 M GdnHCl, cPizza6‐AYW is barely unchanged by 4 M GdnHCl at room temperature.

One of the first steps in the development of new protein nanotechnology is the creation of stable and adaptable protein frameworks onto which new functions can be grafted. Pizza was designed as a symmetrical protein with such applications in mind. Variants of the Pizza protein have recently been shown to be capable of enzymatic activity ³⁴ and further work towards nanoreactors and cargo‐carrying complexes is underway. Since stability of the protein framework is a necessary requisite for many biotechnological applications, the use of split‐intein based vectors to produce novel Pizza‐based complexes could prove highly advantageous. Here we have shown that the protein stability can be controlled to some extent by circular permutation, and that evolution has generally selected a form of “Velcro” closure that is not the most stable but that allows for self‐assembly of propeller repeats. In addition, Pizza can readily be generated as a cyclic protein, which greatly enhances its stability against chemical denaturation, and thus results in a stable, perfectly symmetrical framework highly suited for catalysis.

4. MATERIALS AND METHODS

4.1. Protein preparation

Synthetic genes encoding Pizza6‐AYW and variants were ordered as synthetic DNA fragments (gBlocks™, Integrated DNA Technologies, Coralville, Iowa). They were cloned into the pET28 (Novagen®, Merck, Darmstadt, Germany) vector using the NdeI and XhoI restriction sites. Pizza2 and Pizza3 variants, encoding proteins with 2 or 3 blades per protein, were produced by digesting the plasmid with BamHI or HindIII respectively, and subsequent re‐ligation. Cyclic Pizza6‐AYW (cPizza6‐AYW) was created with golden gate cloning. A synthetic DNA fragment (gBlock™, Integrated DNA Technologies, Coralville, Iowa) was designed to have BsaI restrictions sites and overhangs matching the gp41‐1 vector (Addgene, Watertown, Massachusetts), seamlessly linking the protein to the split‐intein. This vector was created specifically for the purpose of circularizing proteins. ²²

All plasmids were transformed into E. coli BL21(DE3) cells, Pizza6‐AYW and its variants were expressed and purified following the protocol as described by Noguchi et al. ²¹ Protein expression was induced by adding IPTG to a final concentration of 0.5 mM, and subsequently cells were kept at 20°C for 20 hr. The pellets of proteins containing a hexa‐histidine tag were suspended in 50 mM NaH₂PO4, 200 mM NaCl and 10 mM imidazole. The cells were lysed by sonication. The supernatant solution was loaded onto a 10 ml volume nickel‐sepharose column (GE Healthcare, Chicago, Illinois, US) equilibrated with the same buffer and washed with a similar buffer containing 20 mM imidazole. The protein was finally eluted with buffer with 300 mM imidazole and digested by thrombin overnight at 4°C during dialysis into 50 mM NaH₂PO4, 200 mM NaCl, 10 mM imidazole. The protease:Pizza ratio was around 1:200. The protein was re‐loaded onto the washed nickel‐sepharose column and the same steps were repeated. The protein eluted at low imidazole concentrations as the hexa‐histidine tag was removed. The fractions containing Pizza protein were pooled before loading onto a Superdex‐200 column (GE Healthcare, Chicago, Illinois) equilibrated with 20 mM HEPES and 200 mM NaCl buffer at pH ~ 8.0.

The circularized protein lacks a hexa‐histidine tag and cannot be purified by immobilized metal ion chromatography. In this case, after protein induction, the cells were lysed in 50 mM sodium acetate pH ~ 4.5. After centrifugation the lysate was loaded onto a 5 ml HiTrap‐SP HP cation exchange column (GE Healthcare, Chicago, Illinois) and the protein was eluted by applying a salt gradient. Pooled fractions were concentrated with Vivaspin 15R (Sartorius, Goettingen, Germany) and applied on a Superdex 75 pg 16/600 size exclusion column (GE Healthcare, Chicago, Illinois) equilibrated with a 20 mM HEPES and 200 mM NaCl buffer at pH ~ 8.0.

The purified proteins were concentrated to 10 mg/mL and shown to be 95% pure by SDS‐PAGE. All purified proteins were analyzed by size‐exclusion chromatography (SEC). The SEC analysis was performed using a Superdex 200 increase 10/300 GL column (GE Healthcare, Chicago, Illinois) with the same buffer as Superdex 75 PEG 16/600 column.

4.2. Tryptophan fluorescence

Denaturation of the protein samples was measured by observing intrinsic tryptophan fluorescence with a Sapphire ² 96‐well plate reader (Tecan, Männedorf, Switzerland). 2 μL of protein sample (OD₂₈₀ of 10) were pipetted into 98 μL of the denaturant solution to give a final protein concentration of 0.5 mg/mL.

The denaturants used are guanidine hydrochloride (GdnHCl), sodium hydroxide (NaOH), sodium chloride (NaCl), ethanol (EtOH) and hydrochloric acid (HCl). GdnHCl concentration was tested from zero to 6 M, in steps of 0.25 M and between zero and 2 M with varying step size to increase the data points in the transition region. HCl and NaOH were tested from 0.05 M to 1 M, and NaCl from 1.5 M to 5 M. Ethanol concentrations were varied from 5% and 90%. Tryptophan fluorescence was measured immediately, and after 7‐ and 14‐days storage at 20°C, by observing the emission intensity at 330 nm after excitation at 280 nm. In case of ethanol, the emission ratio 340/330 was chosen because of global shift in intensity with increasing ethanol concentration. Samples were prepared and measured in triplicate. ³⁵ Fluorescence measurements from blank samples (containing no protein) were subtracted from the measured values. The data points were fitted to the following Boltzmann‐sigmoid equation with a custom Python script utilizing the SciPy library. ³⁶

$$F\left(\left[\text{Denaturant}\right]\right)=\frac{{\alpha }_N+{\beta }_N\left[\text{Denaturant}\right]+\left({\alpha }_D+{\beta }_D\left[\text{Denaturant}\right]\right)e\frac{-m\left({\left[D\right]}_{50\%}-\left[\text{Denaturant}\right]\right)}{\mathit{RT}}}{1+e\frac{-m\left({\left[D\right]}_{50\%}-\left[\text{Denaturant}\right]\right)}{\mathit{RT}}}$$

F is the measured signal in this case abs330, α _N is the signal of the native state at 0 M denaturant, β _N is the slope of this signal, α _D and β _D are the corresponding values for the completely denatured state. These parameters are introduced because the denatured and native states change linearly with the denaturant's concentration. [D]_50% is the concentration at which 50% of the protein is denatured, m is the constant of proportionality ($\frac{-\partial G_{D-N}}{\partial \left[\text{Denaturant}\right]}$) and has the dimensions of cal mol⁻¹ M⁻¹. ³⁷ From these values the free energy difference between the folded and denatured state in water can be extrapolated as $\mathrm{\Delta }{G}_{D-N}^{{\mathrm{H}}_2\mathrm{O}}$ ₌ m ∙ [D]_50%. The difference in free energy between the mutants and Pizza6‐AYW at the mean value of [D]_50%was calculated with this equation ²³ : $\mathrm{\Delta \Delta }{G}_{\mathit{D}\mathit{-}\mathit{N}}^{D_{50\%}}=0.5\left(m+m^{\prime }\right)\mathrm{\Delta }{\left[D\right]}_{50\%}.$ In which m′ is the m‐value of the mutant and Δ[D]_50% is the difference in the concentration point at which 50% is denaturated.

4.3. Isothermal kinetics of denaturation

To measure unfolding kinetics, 2 μL protein samples (OD₂₈₀ of 10) were added to 98 μL of guanidine hydrochloride solutions between 1.5 and 5.9 M. Refolding was measured by adding proteins samples denatured in 6 M guanidine hydrochloride to buffer with low concentration of denaturant, giving a final concentration between 120 mM and 720 mM. In both cases the fluorescence emission at 330 nm after excitation with 280 nm was measured over 10 s intervals for 2.5 hr (or longer) on a Safire ² plate reader (Tecan, Männedorf, Switzerland). ³⁵ The emission at 330 nm was fitted to the exponential equation: $F\left(t\right)=F\left(\infty \right)+B{e}^{-k_{\mathit{obs}}t}$ using a custom Python script utilizing the SciPy library. ³⁶ This exponential equation corresponds two a two‐state kinetics model. F is the measured signal in this case abs330, F(∞) is the offset value, B is the change in signal and k _obs is the apparent rate of folding/unfolding. The experiments were repeated with different GdnHCl concentrations, this allows for the construction of a Chevron plot. The natural logarithm of the rate constants is plotted in function of GdnHCl concentration. Both the unfolding and folding behave linear this following equation:

$$\ln \left(k_{\mathit{obs}}\right)=\ln \left(k_f^{{\mathrm{H}}_2\mathrm{O}}{e}^{-m_{k_f}\left[\text{Denaturant}\right]}+{\mathrm{k}}_{\mathrm{u}}^{{\mathrm{H}}_2\mathrm{O}}{e}^{-m_{k_u}\left[\text{Denaturant}\right]}\right)$$

where k _obs is the rate constant from the exponential fitting equation to the time traces, $k_f^{{\mathrm{H}}_2\mathrm{O}}$ and $k_u^{{\mathrm{H}}_2\mathrm{O}}$ are the values extrapolated to the absence of denaturant and $m_{k_f}$, $m_{k_u}$ are factors of proportionality of folding and unfolding, respectively. ²³

4.4. CD spectroscopy

All circular dichroism measurements were performed on a JASCO J‐1500 instrument (JASCO, Easton, Maryland). The spectra were taken at 20°C using a 1 mm path quartz cuvette, with protein diluted to 0.1 mg/mL in 20 mM phosphate pH ~ 7.6. For thermal denaturation measurements, 0.05 mg/mL protein samples were placed in a 2 mm path quartz cuvette. The samples were heated from 0°C to 95°C in steps of 0.2°C while monitoring the CD signal at a wavelength of 218 nm.

The CD signal was fitted to a Boltzmann‐sigmoid equation with a custom python‐script utilizing the SciPy library. ³⁶

$$F\left(T\right)=\frac{{\alpha }_N+{\beta }_{N}T+\left({\alpha }_D+{\beta }_{D}T\right)e^{\frac{-\mathrm{\Delta }{H}_m\left(1-\frac{T}{T_m}\right)}{\mathit{RT}}}}{1+e^{\frac{-\mathrm{\Delta }{H}_m\left(1-\frac{T}{T_m}\right)}{\mathit{RT}}}}$$

F is the measured signal in this case the measured CD signal at 218 nm, α_N is the signal of the native state β_N is the slope of the signal in the native state, α_D and β_D are the corresponding values for the completely denatured state. These parameters are introduced because the denatured state change linearly with temperature. T _m is temperature at which 50% of the protein is denatured, ΔH_m is the change of enthalpy upon denaturation at the melting point. ³⁸ The changes in free energy between the mutants and Pizza6‐AYW were calculated with the following equation $\mathrm{\Delta \Delta }{G}_{D-N}=\mathrm{\Delta }{H}_m\frac{\mathrm{\Delta }{T}_m}{T_m}$. ²³ In which ΔH_m and T_m are the enthalpy and melting temperature of Pizza6‐AYW and ΔT_m is the difference in melting temperature between the mutant and Pizza6‐AYW.

4.5. Differential scanning fluorimetry

Differential scanning fluorimetry was performed using a Quick Studio3 (Thermo‐Scientific, Waltham, Massachusetts). The stability was determined by adding Protein Thermal Shift Dye™ (Thermo‐Scientific, Waltham, Massachusetts) and Protein Thermal Shift™ buffer to protein samples at 1.0 mg/mL and subsequent incubation from 25°C to 95°C with a gradient of 3°C per min. All reactions were performed in quadruple and T_m was determined using the manufacturer's software.

4.6. NMR

NMR spectra were recorded with a 0.2 mM protein solution in phosphate buffer (pH 8.0) in H₂O:D₂O (9:1) mixture at 25°C on a 600 MHz Bruker Avance II spectrometer equipped with a 5 mm TCI HCN Z‐gradient cryoprobe. Spectra were processed using Topspin 3.2 (Bruker Biospin, Evere, West‐Vlaanderen, Belgium) and analyzed by using CARA software (version 1.9.1.7, Vaughan, Ontario, ON, Canada) In the one‐dimensional and two‐dimensional spectra, the water signal was suppressed by using excitation sculpting with gradients. The two‐dimensional NOESY spectra with mixing time 200 ms were recorded with a sweep width of 8,400 Hz in both dimensions, 32 scans, 4,096 data points in t2, and 1,024 free induction decays (FIDs) in t1. In the processing of two‐dimensional spectra, the data were apodized with a shifted sine‐bell square function in both dimensions. Proton using external DSS signal as reference (0.000 ppm). The signal assignment of the unlabeled proteins was obtained from homonuclear 2D COSY, TOCSY and NOESY spectra as described by Kurt Wuthrich. ³⁹

4.7. Crystallization and structure determination

Prior to crystallization experiments all proteins were dialyzed against 20 mM HEPES pH ~ 8.0, and the concentration adjusted to 10.0 mg/mL. Crystallization experiments were performed at 20°C using the sitting‐drop vapor diffusion method with a Gryphon robot (Art Robbins, Sunnyvale, California, US). Crystals grew with a variety of precipitants, and for some proteins, conditions were optimized by hanging‐drop vapor diffusion method. The crystals were washed in mother liquor containing a percentage of glycerol as cryoprotectant before being stored in liquid nitrogen. Data sets were collected on X06DA at the Swiss Light Source (SLS) (Villigen, Switzerland) for the “Velcro” constructs or on I03 at the Diamond Light Source (DLS) (Oxfordshire, UK) for the circular Pizza. At SLS total of 1800 images were taken with a 0.1° oscillation, the used X‐ray wavelength was 1.0000 Å. At DLS a wavelength of 0.7000 Å was used, and 1,800 images were taken with a 0.2° oscillation, in both cases a Pilatus 6 M detector was used. The data was processed with XDS ⁴⁰ and Aimless. ⁴¹ The model of Pizza6‐AYW (PDB: 6F0Q) was used for molecular replacement with PHASER. ⁴² Refinement was carried out with PHENIX.REFINE ⁴³ with manual modifications in COOT. ⁴⁴ The CCP4 suite was used to handle the data. ⁴⁵ An overview of data collection and refinement statistics is given in Table S2. The structures of the new “Velcro” mutants: nv1Pizza6‐AYW, nv2Pizza6‐AYW, v31Pizza6‐AYW and v22Pizza6‐AYW were deposited in the PDB with respective accession codes: 6I37, 6I38, 6I39, and 6I3A. The circular construct cPizza6‐AYW was deposited with accession code, 6I3B.

AUTHOR CONTRIBUTIONS

Bram Mylemans: Conceptualization; data curation; formal analysis; investigation; visualization; writing‐original draft; writing‐review and editing. Els De Ridder: Investigation; resources; writing‐review and editing. Hiroki Noguchi: Data curation; investigation; methodology; writing‐review and editing. Eveline Lescrinier: Data curation; investigation; methodology; writing‐review and editing. Jeremy Tame: Resources; supervision; writing‐original draft; writing‐review and editing. arnout voet: Conceptualization; formal analysis; funding acquisition; investigation; methodology; project administration; resources; supervision; writing‐original draft; writing‐review and editing.

SIGNIFICANCE STATEMENT

Different “Velcro” strap positions exist in the β‐propeller fold. In this study, we use a symmetric designer protein as a model system, showing that the most common position in nature does not agree with the most stable one. These proteins can therefore be stabilized by a simple circular permutation. In order to maximize the stability, the proteins should be circularized using split‐inteins.

Supporting information

Appendix S1: Supplementary materials: The DNA and protein sequences are shown in Table S1. Table S2 contains crystallization conditions of the diffracted crystals and Table S3 the diffraction statistics of these crystals. Table S4 and S5 contain the fitting parameters for the kinetic experiments. Figures S1 and S2 show the interesting features in the crystal structures. Figure S3 shows purification data of the 3‐bladed mutants. Figure S4 shows the thermal fits and S5 shows the thermal reversibility. Figure S6 gives the additional denaturation curves. The kinetic fits are shown in Figure S7. Figures S8 and S9 show the results of 1D proton and 2D NOE NMR scans.

Mylemans B, Noguchi H, Deridder E, Lescrinier E, Tame JRH, Voet ARD. Influence of circular permutations on the structure and stability of a six‐fold circular symmetric designer protein. Protein Science. 2020;29:2375–2386. 10.1002/pro.3961