This article was originally published online as an accepted preprint. The “Published Online” date corresponds to the preprint version. You can request a copy of the preprint by emailing the Biopolymers editorial office at email@example.com
The vast literature on bovine pancreatic ribonuclease A (EC 184.108.40.206, RNase) was reviewed in 19981 by Ronald T. Raines of the University of Wisconsin in which he concluded, “RNase has been the most extensively studied enzyme of the 20th century.” While Raines' review emphasized the mechanistic enzymology of RNase, the use of RNase as a model system for protein chemistry and biophysics was also highlighted. Park and Raines2 have shown that catalysis by RNase is limited by the rate of substrate association, placing RNase in a very select group of highly evolved biological catalysts. Chris Anfinsen of NIH used RNase to demonstrate that its amino acid sequence determined its unique three-dimensional structure.3 Four Nobel Prizes have been awarded (Anfinsen, Moore, Stein, and Merrifield) for work associated with studies on RNase. Privalov used calorimetric studies of RNase and four other small globular proteins to characterize the highly cooperative nature and thermodynamics of protein folding.4 Using disulfide-bond chemistry, Scheraga and coworkers5–9 studied the refolding of RNase. Ribó et al. illustrated the use of pressure to study protein folding/unfolding with RNase and its mutants as a model system.10 As RNase contains four disulfide bonds (Figure 1) as well as two cis-amide bonds, its folding has been a topic of considerable interest.
In this more personal review, the emphasis is on the historical relation between RNase, the Rockefeller Institute/University, and its influence on R. Bruce Merrifield and his students. In addition, both current and potential future uses of RNase to address fundamental problems in the thermodynamics driving molecular recognition and the dynamics of enzyme catalysis will be highlighted.
RIBONUCLEASE AS MODEL SYSTEM
Bovine Pancreatic Ribonuclease A—Best-Studied Enzyme of the 20th Century and Possibly of the 21st Century as Well
RNase was isolated by René Dubos in 1938,12 crystallized by Moses Kunitz in 1940,13 sequenced (Figure 1) by Stanford Moore and William Stein in 196314 (with considerable assistance by the Anfinsen group at NIH15 as well16), and synthesized in the Merrifield lab in 1969,17, 18 all at the Rockefeller Institute/University. While GRM was a graduate student in the Merrifield lab (1962–1966), Professors Dubos, Kunitz, Moore, and Stein were still daily participants in lunches at long tables held in the Faculty Lunchroom at Rockefeller, where graduate students could mingle with these legends of protein chemistry. Bruce devoted significant effort to using RNase and its fragments as a testbed for solid phase synthesis, as well as a means of understanding the binding and recognition of peptides by proteins in well-characterized systems. He discussed his fascination with RNase as a model system for examining protein structure and enzymatic activity in his autobiography.19 Two of the best-characterized protein assembly systems that generate enzymatic activity on complex formation are the S-protein/S-peptide and C-terminal peptide/RNase complexes, the latter being discovered in the Merrifield and Moore labs.
Ribonuclease S and S-Peptide
Fred Richards discovered that RNase (124 residues) could be cleaved by the proteolytic enzyme subtilisin between residues 20 and 21, to yield a fully active complex of RNase S (S-protein, residues 21–124 with S-peptide, residues 1–20), as illustrated in Figure 2.23, 24 Dissociation of the complex eliminated ribonuclease activity, and activity was fully restored when the two components were remixed. Wyckoff and Richards25, 26 determined the three-dimensional structure of RNase S to compare with that of intact RNase A. S-peptide/S-protein (Figure 2) has proven to be an ideal model system for exploring the thermodynamics and molecular recognition requirements for a peptide by a protein-binding site.27 Based on syntheses and structure–activity studies of S-peptide analogs, Finn and Hofmann28 suggested that RNase S may be the prototype for peptide–hormone activation of receptors by completion of an active site of an enzyme. The structure of a dimer of RNase in which two N-terminal helical domains are swapped in analogy with the S-peptide interaction has also been determined.29 A hexapeptide antagonist, YNFEVL, of S-peptide, which has comparable affinity for RNase S-protein, has been discovered utilizing phage display and panning with S-protein.30
Further studies have shown that the full-length S-peptide is not required when recombined with S-protein, as the first 14 residues RNase (1–14) alone were able to restore full enzymatic activity.28 Additionally, nuclear magnetic resonance (NMR) studies have shown the S-peptide to adopt a predominantly disordered state in solution and a helical state when bound to S-protein,31–33 as shown in the crystal structure of the complex. Simplified analogs of 1–15, AEAAAAKFARAHMAA (as compared to the native 1–15 sequence (S15) KETAAAKFERQHMDS), in which residues not contacting S-protein are converted to alanine, led to active complexes with RNase.34–36 This fact lends further support to the idea that an α-helical arrangement of residues has functional significance35 and provides a roadmap to possible introduction of constrained amino acids such as aminoisobutyric acid (Aib). The φ,ψ backbone torsion angles from crystal structures of the S-peptide when complexed with S-protein are plotted in Figure 3. While residues 4–10 have precisely α-helical torsion angles, residues Gln11 and His12 do not.
Variation in binding affinities suggests that understanding and separation of variables in binding thermodynamics is not trivial. Calorimetric studies of the RNase S/S-peptide system are extensive.40–43 Ratnaparkhi et al.44 introduced an Aib residue to replace Ala4 in shortened S-peptide (Ala4Aib S16) and compared its binding thermodynamics with S16 (S15 with a C-terminal glycine replacing the carboxamide) using isothermal titration calorimetry (ITC). The more rigid Aib-containing peptide has slightly less affinity than S16 at all temperatures investigated and did not impact the melting temperature of the complex with S-protein. This suggests that initial melting may involve another subdomain of RNase, rather than the S-peptide. The crystal structure of the Aib-containing S16 and enzymatic activity of the complex were “essentially” (authors' comment) identical to that of the S15 complex. At room temperature, neither Ala4Aib nor S16 contained detectable amounts of helix by CD, suggesting that the single Ala4Aib did not induce any helical preference under the conditions examined. While the ΔS of binding is more positive for Ala4Aib, the ΔH for binding is also more positive, and, thus, the system displays entropy/enthalpy compensation. The positive value of ΔΔH was unexpected and possibly due to favorable interactions formed in uncomplexed Ala4Aib. Obviously, these data challenged the basic assumption being tested: that preorganization of the helical backbone will reduce the change in entropy of binding, and that this change will be reflected in enhanced affinity. This makes the RNase S system particularly attractive as a model system for further investigation of the details of binding thermodynamics to resolve this conflict. Nevertheless, examples exist of helical peptides containing Aib residues with nanomolar binding affinities and relatively small sizes in other less well-defined systems. What factors are necessary for preorganization to translate the anticipated loss in entropy of binding to affinity? How complementary do the interacting surfaces need to be? A minor mismatch in steric complementarity can obviously be catastrophic.
Complex of RNase With C-Terminal Segments
Merrifield described a tea conversation he had with T. P. King and Bill Konigsberg, two contemporary protein chemists, at Rockefeller in 1965 in his autobiography.19 The discussion centered on generating a construct similar to RNase S, but which explored the role of the C-terminal segment of RNase contributing His119 to the catalytic site. Ultimately, the semisynthetic ribonuclease, RNase-(1–118)-(111–124), was evolved to study the role of the C-terminal segment in recognition and catalysis.45–48 A three-component system, combining equimolar amounts of RNase fragments 1–20, 21–118, and 111–124, retained 30% of the enzymatic activity of intact RNase.19 Possible utilization of this three-component system in studies of chimeric RNase and preorganization are attractive.
RNase and RNase S Domain-Swapped Dimers
Domain swapping occurs when two monomers exchange structural motifs, α-helix, β-strands or even larger subdomains, to form dimers or higher oligomers and retain the structure of the swapped domains. RNase was the first protein that demonstrated the three-dimensional domain swapping, as shown in the Moore and Stein laboratory.49 Subsequent work has shown that RNase can form a variety of dimers, trimers, and higher oligomers by swapping the N-terminal α-helix, the C-terminal β-strand, or both, as recently reviewed.50 The amyloid-like structure seen in the C-terminal dimer of RNase led Eisenberg and coworkers to a general proposal for amyloid formation via domain swapping.51 Lopez-Alonso et al.52 have shown that RNase S can also form a dimer analogous to the C-dimer of RNase whose dissociation is strongly slowed by excess S-peptide, suggesting that S-peptide release is rate limiting.
The RNase A Superfamily
A number of enzymes with homology, both in sequence and function to pancreatic RNase A, have been characterized as reviewed by Dyer and Rosenberg.53 All eight human ribonuclease A-like genes are located on chromosome 14, each encoding a secretory signal sequence with an invariant catalytic triad of two histidines located at each end of the protein and one lysine within a conserved motif (CKXXNTF). Only one, angiogenin/RNase 5, has three disulfide bonds, perhaps representing the ancestral gene of what is thought to be the sole vertebrate-specific enzyme54; the others contain the four disulfides seen in RNase itself (Figure 1), although the location of disulfide 65–72 in RNase has been shifted in onconase. Modifications of their respective sequences are thought to be involved with a diversity of functions in host defense against viral and bacterial pathogens.
From a phage library displaying hexapeptides, clones that bound S-protein were selected. Bound peptides showed a sequence motif, (F/Y)NF(E/V)(I/V)(L/V), that bore little resemblance to S-peptide. YNFEVL was synthesized and shown by ITC to bind S-protein with a Kd of 5.5 mM at 25°C, comparable to those reported for some S-peptide variants. The YNFEVL peptide was an antagonist of S-peptide. It did not generate RNAse activity upon binding to S-protein, but blocked the ability of S-peptide to restore enzyme activity by competition. The hexapeptide is considerably smaller than S15 or S-peptide and could provide insight into different binding interactions, as well as possible interaction of the S-peptide with the RNA substrate. To our knowledge, the structure of the complex of YNFELV with S-protein has not been determined.
Intracellular Inhibitors of RNase
Ribonuclease has demonstrated potential in oncology. Within the cell, RNase is complexed with an inhibitory protein (RI).56 When the surface of RNase is mutated to eliminate recognition by the cytosolic inhibitor (RI), the modified RNase is toxic to cancer cells by degradation of intracellular RNA.57 These observations provide another opportunity for development of helix mimetics of S-peptide as potential therapeutics. The interaction between RNase and RI is very tight (femtomolar Kd, estimated binding energy = 73.3 kJ mol−1; Ref.58), and the residue/residue basis of interaction has been mapped by Rutkoski et al.57 The crystal structure of RNase complexed with RI has been determined (Figure 4) and the role of individual residues of the S-peptide domain to binding affinity examined.58 Two basic residues of the S-peptide, Lys1 and Lys7, dominate the interaction. Replacement of Lys1 by alanine decreased affinity by 4 kJ mol−1, whereas replacement of Lys7 with alanine reduced affinity by 10 kJ mol−1. Loss in binding energy (16 kJ mol−1) by a double mutation was comparable to RI binding to S-protein alone.58 Smith and Raines59 have developed a creative expression system in an E. coli strain in which the reductive intracellular environment has been eliminated to allow nascent RNase to form its disulfide bonds and generate enzymatic activity. Mutations within this E. coli strain allowed the interface between RNase and RI to be mapped.
HELIX MIMETICS IN MOLECULAR RECOGNITION OF RNASE S
Design and syntheses of a variety of scaffolds, primarily polyaromatic, to orient amino acid side chains and mimic the recognition surface of peptide helices has been reviewed by Che et al.60 S-peptide is an excellent model for exploring surface complementarity and the thermodynamics of S-protein recognition. Because the S-peptide contributes His12 to complete the active site of RNase S, helix mimetics may be designed to bind to S-protein and regenerate enzymatic activity. Binding of a helix mimetic to S-protein could be detected by a number of techniques, including competition with a fluorescently labeled S-peptide, ITC, etc. Yet, to determine that the correct (i.e., “designed”) alignment of the two components in the RNase S complex occurred would require structural studies by crystallography or NMR. The requirement for catalysis of precise positioning of the catalytic His12 residue, however, effectively couples a thermodynamically driven binding event to a simple enzymatic assay. Deciding what exact variations in helix geometry to target with a helix mimetic of S-peptide is nontrivial as discussed below.
The prevailing viewpoint is that the classical α-helix (φ = −57°, ψ = −47°) with a 13-member hydrogen-bonding scheme predominates in protein crystal structures. Further investigation challenges this tacit assumption and the frequently incorrect assignment of classic α-helicity for many helical subdomains in the PDB, including the S-peptide. A Ramachandran (φ,ψ) plot of atomic resolution crystal data for RNase S-peptide (two molecules per asymmetric unit) in Figure 3 shows that not all helical residues are precisely α-helical. Forcing the side chains of residues 11 and 12, however, to orient on a “perfect” α-helical scaffold would misalign the histidine side chain and disrupt catalytic activity. Further, this would also misalign the methionine-13 side chain that has been shown to be the “hot spot” in the S-peptide for binding affinity.20, 42 A recent examination of over 900 helical residues in high-resolution crystal structures of proteins indicated that the experimentally observed distribution (Figure 5) is skewed from the classic α-helix, toward the region of the 310 helix, suggesting a predominance of bifurcated hydrogen-bonding schemes.61 This view is supported by molecular dynamics (MD) simulations of crambin (58 residues) in explicit solvent using the next-generation AMOEBA force field,62 followed by analysis of helical length (Figure 6). Most helical-mimetic scaffolds proposed to date do not possess backbone flexibility that would allow for induced fit upon binding a protein. Rather most scaffolds have been designed against ideal or single-case α-helices and, thus, will not generically complement other helix-recognition sites.
α-Methyl Amino Acids
Marshall and Bosshard63 first published the dramatic effect on helical propensity of replacing the α-proton of an amino acid with a methyl group in 1972. Examination of the reproduced Ramachandran plots (Figure 7) from this paper showed the differences in allowed regions of the two variables, φ and ψ, for the capped amino acid, acetyl-L-alanine N-methylamide (Figure 7A) and acetyl-D-alanine N-methylamide (Figure 7E). Replacement of a hydrogen with a methyl group on the peptide backbone results in significant loss in possible values for φ and ψ; for example, the impact of a methyl replacement for the amide hydrogen is shown in two plots, the first for acetyl-N-methyl-L-alanine N-methylamide, which restricts φ (Figure 7B), and the second for acetyl-D-alanine N-methylamide which restricts ψ (Figure 7C). These plots indicate why proline, or any other N-methyl amino acid, has difficulties being accommodated within the helical segments of proteins. The most dramatic change was seen, however, when the α-hydrogen of the alanine was replaced by a methyl group, acetyl-L-aminoisobutyric N-methylamide (Figure 7D), simultaneously restricting both φ and ψ. This prediction that α-methyl amino acids would restrict the flexibility of the peptide backbone to only those regions of φ,ψ space common to both acetyl-L-alanine N-methylamide (Figure 7A) and acetyl-D-alanine N-methylamide (Figure 7E) has been confirmed subsequently by numerous experimental/theoretical studies by the Marshall,65–69 Karle,70, 71 Toniolo,72, 73 Balaram,70 and Millhauser74 groups. By focusing the minimal energy states of the peptide backbone to φ,ψ space associated with right- or left-handed helices, substitutions with α-methyl amino acids preorganize the peptide to turn or helical conformations at those residues and reduce the entropy of folding/binding. α-Methyl amino acids have become a standard tool in the peptide chemists' repertoire for inducing turns and stabilizing helices.75 The only negative aspect of their use is the ambiguity regarding which type (α- helix, 310-helix, or a mixture of the two) of helix is being recognized upon binding.65 The ability to form rigidified complementary helical surfaces by subtle changes in backbone torsional angles may, in fact, turn out to be a compelling argument for the use of oligomers of α-methyl amino acids as generalized helical mimetics.
Replacing two residues in a sequence with α,α-disubstituted amino acid containing both an α-methyl group and an α-vinyl group followed by olefin metastasis allows rigidification of the helical structure through cyclization of the side chains by “hydrocarbon stapling.” A stapled BH3 helix that activated apoptosis was reported by Walensky et al.76 Alternatively, a similar cyclic stabilization is obtained with side-chain lactam77 or disulfide formation.78 Shepherd et al.79 reported synthesis of a 13-residue helical peptide stabilized by two side-chain lactam bridges that inhibited RSV fusion with an IC50 of 36 nM. Multiple strategies for stabilizing helices have been developed and are illustrated in Figure 8. These approaches should be explored as additional ways to stabilize α-methyl oligomers in either the α- or 310-helical conformations, or perhaps, trap a peptide in the intermediate state with bifurcated hydrogen bonds.
Enthalpy/entropy compensation remains one of the most difficult aspects of molecular recognition to quantify. It seems intuitively obvious that a tightly binding ligand with high surface complementarity and enhanced enthalpy of binding should bind tighter and be more constrained at the binding site, thus reducing its entropy. It is common to observe a change in the relative contribution of enthalpy and entropy of binding as a set of ligands are examined. As the simplest example of preorganization, incorporation of aminoisobutyric acid (Aib, α-methylalanine) into peptides/proteins restricts the φ and ψ backbone angles adjacent to Aib to those associated with helix formation (Figure 7).63, 65 The Marshall lab utilized this observation to investigate the receptor-bound conformation of several peptides binding to GPCRs.80–83 By restriction of the adjacent dihedral torsions, insertion of an Aib should also lower the entropic penalty of turn/helix formation upon protein folding, due to preorganization. In a study of the thermodynamics of S-peptide/S- protein complex formation in which α-methylalanine, or aminoisobutyric acid (Aib), had been substituted for Ala4 in the S-peptide by Ratnaparkhi et al. using ITC,84 the following conclusion was drawn,44 “The surprising lack of temperature dependence of ΔGo for Ala4Aib is likely to be due, in part, to the reduced conformational entropy of the uncomplexed peptide as a result of introduction of the Aib residue.” Also, it was noted that while the Ala4Aib substitution reduces conformational entropy, TΔS compensation is difficult to dissect into its component parts. “Although Aib reduces conformational entropy of the unbound state, there is no increase in protein stability, because of significant enthalpy-entropy compensation.”44 This is exactly the sort of observation that needs detailed analysis by both modern experimental and computational approaches to dissect the fundamentals of these observations.85
Combinatorial Library of Helix Mimetics
Bourne et al. (Phenyl-pyridine-based a-helical mimetics: Targeting the CheY-FliM binding site, In preparation) have developed a protocol for synthesis of substituted phenyldipyridyl libraries to serve as helix mimetics (Figure 9). The choice of the phenyldipyridyl scaffold over the diphenyl, terphenyl, and other scaffolds suggested by Jacoby,86 Hamilton and coworkers,87–90 and others91 was based on extensive Density Functional Theory analyses of candidate scaffolds (Figure 10) by Che et al.60 This study provided an understanding for the design of helix mimetics, and why the simple concept of moving the side chains from the interacting surface of a peptide helix to the aromatic scaffold is not straightforward (Figure 11).
REVERSE TURNS IN RIBONUCLEASE AND BEYOND
Expressed Protein Ligation
Expressed protein ligation (EPL) depends on the use of inteins, bacterial cysteine–protease domains, which can catalyze their own excision from a protein sequence while splicing the N- and C-terminal segments together.97 By using a mutated intein, the enzyme mechanism can be trapped with the N-terminal sequence as the intein-bound thioester. This activates the unprotected N-terminal segment for chemical ligation to an N-terminal cysteine C-terminal segment. Generation of hybrid proteins with expressed protein domains ligated to synthetic peptide or organic component (Figure 12) is readily accomplished. This approach has provided an ability to combine segments of protein labeled differentially to facilitate biophysical studies.98–100 Three examples of EPL to generate hybrid constructs of RNase have appeared; two of which probe the role of preorganization on thermal stability and enzyme catalysis. In 1998, Evans et al.101 produced RNase A by expressing an inactive truncated form, the first 109 amino acids of RNase A, as a fusion protein consisting of RNase 1–109, an intein, and a chitin-binding domain. Thiol-induced cleavage of the precursor fusion protein led to the liberation of RNase 1–109 with a C-terminal thioester. Addition of synthetic peptides representing residues 110–124 led to the generation of full-length products with catalytic activity indicative of wild-type RNase. The turnover numbers and Km for ligated and renatured RNase A were 8.2 s−1 and 1.5 mM, which are in good agreement with reported values of 8.3 s−1 and 1.2 mM.48 Arnold et al.102 also used EPL to prepare semisynthetic RNase, essentially indistinguishable from the wild-type enzyme.
Impact of Reverse-Turn Mimetics on Protein Stability
Reverse-turn mimetics103 have been constructed from such diverse structural approaches as cyclically constrained peptides,104, 105 organic bicyclic rings such as benzodiazepines,106 and metal complexes of chiral pentaazacrowns.107 Arnold et al.102 used EPL to construct RNase analogs in which synthetic peptide fragments 95–124 were ligated with expressed RNase 1–94. This allowed exploration of the role of the highly conserved Type-VI reverse turn observed in RNase; WUBLAST analysis of genomic sequences of RNAses from 50 species show 49/50 sequences contain a proline at residue 114, the one exception being a leucine found in the capybara (Hydrochoerus hydrochaeris) RNase. Residue 113 was a conserved asparagine in most cases, with a few examples of aspartic acid, serine, or lysine substitutions. The question of any selective advantage of preserving this dipeptide reverse turn during the evolution of these different species is intriguing. Residues Asn113-Pro114 located at the end of a β-hairpin were replaced with a reverse-turn peptidomimetic (Figure 13), the dipeptide (R)-nipecotic acid–(S)-nipecotic acid, shown by Chung et al.,108 to nucleate reverse-turn segments. The substitution of two exocyclic six-membered rings in the peptide backbone for the single five-membered proline ring and Asn residue was anticipated to enhance the preorganization of the unfolded state and, thereby, reduce the entropy of folding. Such a reduction in the entropy of folding should be reflected in a higher melting temperature compared with the wild-type protein.
Estimation of the entropy changes that result from conformational constraints and/or binding to a macromolecule is a topic of considerable interest when attempting to estimate affinities of ligands for receptors.109 This issue has been extensively studied experimentally by Williams et al.110–112 and forms the theoretical basis of fragment-based drug discovery,113 as exemplified in “SAR by NMR.”114 Mammen et al.115 developed a classical model for estimating the entropy of torsional motion about a single bond. The entropy associated with torsional motion ranges from 0 to 15 J mol−1 K−1 and varies with the bond length, the hybridization, the symmetry, the sizes of these atoms or groups of atoms, and the extent of conjugation to adjacent bonds. The question of interest, however, is in the actual entropic change introduced by either preorganization of the ligand and/or binding to another molecule as determined experimentally or by computation.
Table I from Arnold et al.116 is reproduced as part of Table I. Note the three different controls for the native protein (Asn-Pro) and the quantitative agreement in experimental values for both Tm and kcat/Km. It is clear that the dinipecotic acid substitution at residues 113–114 had little impact, if any, on the enzyme activity, but a measurable, small effect on the melting temperature [ΔTm by (1.2 ± 0.3)°C, or ΔΔGm = 2.0 ± 0.4 kJ mol−1].
Table I. Impact of 5,5′-Dimethylproline (dmP) Substitution116 for Pro114 and Dinipecotic Dipeptide (R-Nip-S-Nip) Turn-Mimetic Substitution116 for Asn113-Pro114 in RNase on Catalytic Properties (kcat/Km) and Melting Temperatures (Protein Stability)
kcat/Km (×107 M−1 s−1)
64.6 ± 0.2
1.6 ± 0.4
0.03 ± 0.01
63.4 ± 0.2
1.5 ± 0.2
63.5 ± 0.1
1.4 ± 0.3
63.6 ± 0.2
1.5 ± 0.2
ΔTm = (1.2 ± 0.3)°C; ΔΔG = 2.0 ± 0.2 kJ mol−1
ΔTm = −10°C
66.4 ± 0.2
1.6 ± 0.2
Folds 4X Faster
ΔTm = (2.8 ± 0.3)°C; ΔΔG = 4.6 ± 0.2 kJ mol−1
In a similar study, Arnold et al.117 introduced 5,5-dimethyl-L-proline (dmP) for Pro114 to stabilize the cis-amide bond118 between residues 113 and 114 by EPL (Figure 13). The catalytic activity kcat/Km was indistinguishable from that of the native enzyme. The midpoint of the thermal transition for dmP114 RNase was increased, ΔTm = (2.8 ± 0.3)°C or ΔΔGm = 4.6 ± 0.4 kJ mol−1. For comparison, replacing Pro114 with glycine or L-alanine causes a very significant decrease in conformational stability (ΔTm = −10°C, ΔΔGm = 3.6 kJ mol−1),119 suggesting an important role of this reverse turn and its associated β-hairpin in stabilizing RNase.
In addition, the rate of protein folding was increased approximately sixfold, as isomerization of the proline trans-amide bond was thought to be rate limiting. This is in sharp contrast to the previous study where the introduction of the two nipecotic acid six-membered rings at position 113 and 114 only increased the ΔTm by (1.2 ± 0.3)°C. RNase has been a focus of protein folding studies by the Scheraga group which have shown multiple folding pathways depending on the redox reagent (DTT versus glutathione)119, 120 or the location of the four disulfide bonds (RNase versus onconase).119, 121 It would be intriguing to analyze the pathways associated with folding of the Pro114dmP mutant. Other proline analogs impact cis–trans amide isomerization and, therefore, type-VI reverse-turn formation.118, 122 An excellent geometrical mimic of the cis-amide bond, the 1,5-disubstituted tetrazole ring, has been used by Zabrocki and coworkers123–125 to probe type-VI reverse-turn recognition. Incorporation of a series of such dipeptide analogs to replace Asn113-Pro114 in RNase by EPL with subsequent detailed biophysical studies should provide some insight into the role of this highly conserved dipeptide sequence in RNase function.
MD Simulations of Chimeric RNase
To quantitatively evaluate the relative propensity of reverse-turn mimetics to stabilize β-hairpins, Takeuchi and Marshall126 monitored various parameters during a MD simulation. For example, the relative time that the distance between the two α-carbons of the first and fourth residue of a capped tetrapeptide containing the mimetic was less than 7 Å was monitored. To investigate the proclivity of the newer reverse-turn mimetics, the native tetrapeptide sequence Gly112-Asn113-Pro114-Tyr115 was capped with acetyl at the N-terminal and with N-methyl amide at the C-terminal. Starting with the native sequence, nine mutants with potential reverse-turn mimetics were generated in silico. MacroModel 9.1 was used to run 10-ns MD simulation of the modified peptides in implicit solvent (GB/SA) at 300 K using the OPLS 2005 force field. The distance between the C-alphas of Gly112 and Tyr115 and the distance between the carbonyl oxygen of Gly112 and amide hydrogen of Tyr115 were recorded (Figure 14) at each time step after initial equilibration similar to the studies by Takeuchi and Marshall126 on reverse-turn propensity. The virtual dihedral angle defined by the four C-alpha carbons of the reverse turn, as suggested by Tran et al.,127 of Gly112, Asn113, Pro114, and Tyr115 was also monitored as shown in Figure 14.
The results of the tetrapeptide simulations were quite revealing. In the top two panels of Figure 14, the impact of the nine different dipeptide substitutions on frequency of observation versus distance between the glycine carbonyl oxygen and the tyrosine amide nitrogen (prevalence of a classic hydrogen bond between the i and i + 3 residue) are plotted. The middle two graphs of Figure 14 show the frequency versus distance between the α-carbons of glycine and tyrosine, another measure of the propensity to form conformations resembling reverse turns. In native RNase, the distance between the two α-carbons was less than 7 Å over 80% of the simulation; in the R-Nip-S-Nip chimeric protein, the distance was less than 7 Å for only 1% of the time. While this difference does not directly estimate the amount of preorganization in the unfolded RNase versus the chimeric protein, it does indicate that the introduction of two additional methylenes in the backbone of the hairpin loop by R-Nip-S-Nip dramatically increases its inherent flexibility and compromises any impact of preorganization of the entropy of folding. In contrast, the use of the reverse-turn nucleators, Pro-D-Pro or D-Pro-Pro, enhanced the reverse-turn potential to equal or greater than native Asn113-Pro114 in the simulations, in accordance with previous estimates of reverse-turn nucleation by Takeuchi and Marshall.126 It is clear from these graphs that the R-Nip-S-Nip or S-Nip-R-Nip dipeptides do not dynamically stabilize the reverse turn seen with Asn-Pro (red line in all graphs) while Asn-dmP, D-Pro-Pro, and Pro-D-Pro mimic and stabilize the reverse-turn as well or better then Asn-Pro itself; in fact, the two bottom graphs of Figure 14, where the virtual dihedral angle between the four α-carbons is shown versus frequency, further confirms stabilization of the reverse turn by these three dipeptides. These graphs can be used to give a crude estimate of the entropic consequences of these dipeptide substitutions. From these results, one would predict that the thermal stability of a chimeric RNase with either Pro-D-Pro or D-Pro-Pro replacing Asn113-Pro114 would be greater than that of the chimeric RNase with R-Nip-S-Nip. More sophisticated computations using replica exchange are underway to estimate the changes in melting temperature seen for small chimeric proteins in a model system described later.
In further preliminary simulation studies128 of intact RNase analogs, R-Nip-S-Nip of Arnold et al.116 and other potential replacements of Asn113-Pro114 were compared with native RNase. To investigate the minimal effect on the melting temperature [ΔTm = (1.2 ± 0.3)°C] observed, the crystal structure of RNase was minimized, the turn mimetic R-Nip-S-Nip inserted for Asn113-Pro114 and the chimeric structure reminimized. The two additional methylenes of the two β-amino acids were readily incorporated into the structure by simply extending the hairpin loop with nearly identical torsion angles of the rest of the peptide backbone. Thus, little, if any, difference was found between the minimum-energy structures of the two hairpins, suggesting that R-Nip-S-Nip for Asn-Pro substitution did not disrupt the extended β-sheets of the hairpin, and enthalpic stabilization of the hairpin loop was maintained.
MD Simulations of RNase and RNase S
Nadig et al. did 100-ps simulations of both RNase and RNase S at 300 and 330 K129 using AMBER 4.0 in TIP3P water. No significant differences in the behavior of the two enzymes was observed, despite the cleavage of the bond between residues 20 and 21 in RNase S that was supposedly responsible for enhanced hydrogen-deuterium exchange and sensitivity to trypsin in the case of RNase S that was previously reported. Differences in sensitivity to trypsin digestion were attributed to the enhanced cleavage of S-protein that had dissociated from RNase S during the experiments, rather than to differences in inherent susceptibility to enzymatic cleavage of RNase and RNase S. A similar argument was applied to hydrogen–deuterium exchange, due to the measured dissociation rate of RNase S of 1.45 × 10−8 M at 15°C.42 Thus, the presence of small amounts of S-protein and S-peptide during the experimental studies on RNase S can explain the reported enhanced dynamics of RNAse S compared with RNase. It is difficult to imagine that cleavage of the peptide bond between residues 20 and 21 with the resulting changes in the structure of RNase S versus RNase did not impact normal modes and other dynamic properties of the enzyme. One plausible explanation is that the changes in low frequency modes are not sufficiently sampled in the 100-ps simulations.
Dynamics and Enzyme Mechanism
The effect of protein motion on different aspects of enzymatic function and protein complex formation in a variety of systems has become accessible through sophisticated NMR relaxation experiments.130–138 Alexandrescu et al.139 determined the 15N backbone dynamics of S-peptide in both its free and bound form. The order parameters derived were used to calculate 1H-15N bond vector entropies that averaged −12.6 ± 1.4 J mol−1 residue K for Thr3-Ser15, the S-peptide residues that became ordered upon binding to S-protein. Further studies140 of structure/disorder in S-peptide based on NMR residual dipolar couplings indicated that the presence of a native-like α-helix (residues 3–13) undergoing dynamic fraying at either end under experimental conditions that traditionally stabilize helical conformations. These results were entirely consistent with previous studies by Brown and Klee,141 Kim and Baldwin,142 and Nelson and Kallenbach143 that showed that S-peptide had a nascent α-helical segment that could be stabilized by experimental conditions. The dynamics of RNase A itself were later investigated by Cole and Loria33 by NMR spin-relaxation experiments. RNase A was conformationally rigid on time scales faster than overall rotational tumbling (picoseconds to nanoseconds). The average order parameter, S(2), for RNase A was 0.910 ± 0.051. Twenty-eight of the 124 amino acid residues in RNase underwent chemical exchange on the microsecond-to-millisecond time scale. The microscopic chemical exchange rates, kex, were quantified by the relaxation-compensated Carr-Purcell-Meiboom-Gill (rcCPMG) experiment for 16 of the 28 residues. The value of kex was identical for all 16 residues with an average of 1640 s−1 and very similar to the RNase kcat value of 1900 s−1. Many of these mobile residues localize to the active site in RNase, including the catalytically crucial amino acids His119 and Asp121. Additional motion was found in the B1, B2, and P0 binding subsites, suggesting possible coupling of motion between the binding and catalytic sites. The activation energy of the motion was measured by applying the rcCPMG experiment at temperatures of 283, 293, and 298 K and was determined to vary between 3.6 and 7.4 kcal mol−1. This activation-energy barrier to conformational motion is very similar to that of the RNase-catalyzed reaction, and thus would not be thermodynamically limiting for catalysis. Correlations of motion and activation energy with enzymatic parameters suggest a possible role for protein flexibility in the molecular enzymatic mechanism of RNase. The data, however, from two modifications of the β-hairpin at residues 113–114 in RNase by EPL yielded identical values of catalytic activity kcat/Km116, 117 with those of the native enzyme. This is more in concert with the experimental data of Park and Raines2 that substrate association is the rate-limiting step for RNase. Marcus rate theory suggests that efficient catalysis minimizes the energetics of reorganization,144 and the correlations in protein dynamics in RNase with enzymatic rates may be misleading. Alternatively, the dynamics of the β-hairpin centered at residues 113–114 may not couple to the active site in RNase. The role of the Asn-Pro reverse turn and β-hairpin centered at residues 113–114 in RNase remains a mystery, as it is highly conserved in different species. As additional motion was observed by NMR in the phosphate (PO)-binding site by Cole and Loria,33 similar experiments on the K7H/R10H/H12K/H119Q mutant of Moussaoui et al.145 (discussed later) where the catalytic site now resides at the PO-binding site could be instructive.
ENGINEERING OF CHIMERIC PROTEINS
Bruce was intrigued with the idea of an all-D enantiomer of RNase. “Will a protein of unnatural configuration bind and react with a ligand of unnatural enantiomeric structure in exactly the same way as the natural all-L enzyme reacts with its normal substrate?” was a rhetorical question he asked on page 139 of his autobiography.19 Bruce started the synthesis of all-D RNase and its enantiomeric substrate, cytosine-1-β-L-ribofuranosyl-1′,3′ cyclic phosphate in his group, but the project was never completed to our knowledge. His influence inspired his former technician, Angela Corigliano, to synthesize all-D S-peptide while in the lab of Irwin Chaiken and to show it would neither inhibit nor activate S-protein.146 Steve Kent would address this question with the synthesis of all-D HIV-1 protease and an all-D substrate.147 The results were as anticipated: a mirror-image biological universe is feasible; the mirror-image enzyme was specific for the mirror-image peptide substrate.
Toward Protein Engineering
Active Site Design
By mutation, it has been possible to reengineer proteins,148, 149 i.e. move an active site from one protein to a new scaffold, change the enzymatic mechanism, etc. Recently, Moussaoui et al.145 changed the phosphate-binding site of RNase into a catalytic site for cleavage of phosphodiester bonds by the double mutant K7H/R10H. Suppression of the normal active site in K7H/R10H/H12K/H119Q shows a clear increase in exonucleolytic activity which can only be attributed to the new active site. K7H/R10H retained 17% relative activity while the activity of K7H/R10H/H12K/H119Q was reduced to 9%. The kcat/Km ratio was reduced approximately fourfold for each of the two mutant enzymes compared with native RNase.
Incorporation of nonnatural amino acids and semirigid peptidomimetics provide unique possibilities for designing proteins that adopt a stable predetermined fold, allowing rational protein engineering to become a reality. Bruce Erickson, a Merrifield lab alumnus, pioneered the incorporation of D-proline to nucleate β-turns150 in the de novo design of proteins,151 particularly his seminal work on betabellins152 started as a collaboration with Richardson et al.153 Limiting segmental dynamics may be a useful probe of enzyme mechanism and/or specificity, a topic of increased interest. The ability to design proteins with enhanced stability is of commercial interest in the production of super stabile biocatalysts for green chemistry. For example, multiple tons of the proteolytic enzyme subtilisin, engineered to be stable in detergents at alkaline pH and elevated temperatures, are consumed annually in laundry detergents.154 To be able to experimentally explore such topics with chimeric proteins is a direct outcome of the invention of solid-phase peptide chemistry and a stated objective (see discussion on pages 125–141 on ribonuclease A in his autobiography) of its inventor, R. Bruce Merrifield,19 as well as a goal of Emil Fisher whose 100th anniversary of the synthesis of a peptide amide bond was recently celebrated in 2006.
By the same principle of preorganization, incorporating semirigid mimetics of α-helices, β-sheets, and reverse turns into a protein would minimize the entropy ΔSfolding lost on folding through preorganization, while retaining the interactive surface features that optimize the favorable enthalpic interactions ΔHfolding in the folded state. Thus, the stability (ΔGfolding) of the mutant folded protein was enhanced compared to the wild-type protein. In effect, one could select a particular folded geometry from the manifold of potential folds for a given sequence by forcing the protein to adopt particular secondary-structure elements at the desired locations in the amino-acid sequence. The first examples of incorporation of reverse-turn mimetics with measurement of enhanced stabilities were into the enzymes HIV protease155 and RNase A.116 Chimeric proteins should be thermodynamically more stable because their fold space is limited by semirigid mimetics that reduce the entropic penalty upon folding into the desired 3D structure. In addition, semirigid mimetics should promote the rate of protein folding by nucleation.117 Modular secondary structure mimetics can serve as building blocks in the design of ultrastable, catalytically active chimeric proteins that resist both proteolytic degradation and denaturation by heat, pH, detergents, etc.
MD Simulations of Chimeric Protein, FSD-1
Designing a protein sequence that folds into a desired three-dimensional shape is known as the inverse protein-folding problem. In nature, protein sequences are limited to combinations of the 20 naturally occurring amino acids plus any posttranslational modifications. Computational protein design methods can identify de novo amino acid sequences that can be folded into predefined topologies.156, 157 As a representative example, the 28-residue full sequence design (FSD-1) ββα-protein was computationally designed by Dahiyat and Mayo to form a stable zinc-finger ββα-fold independent of zinc binding.156 Starting with the backbone coordinates of the zinc-finger protein Zif268, they selected side-chain rotamers to optimize side-chain/side-chain and backbone/side-chain interactions. The designed protein was synthesized and its structure solved by NMR; the resulting structure's overall backbone RMSD was 1.98 Å relative to the computationally designed target. Residues 3–12 were assigned as a β-hairpin, and residues 15–26 were assigned as α-helical, illustrated in Figure 15. FSD-1 was used as a model system for MD simulations of folding and stability for chimeric proteins incorporating helical mimetics.128
FSD-1 was chosen as a prototype for chimeric protein engineering because of its small size and because its α-helical segment could be replaced with a semirigid, organic helix mimetic. It should be possible to predict enhanced stability by computational methods due to its small size (28 residues versus 124 residues for RNase). Its short β-hairpin peptide segment could be readily synthesized by solid-phase synthesis and chemically ligated to the organic helical mimetic.128 Experimental observations and theoretical calculations suggested that helical mimetics based on a triphenyl, tripyridyl, or phenyldipyridyl scaffolds could correctly orient the i, i + 3, i + 4, and i + 7 side chains that formed one side of the surface of an α-helix.89, 92, 103 To further investigate the stabilities of helix mimetics, chimeric proteins were designed in silico by ligating four different helix mimetic designs with the β-hairpin subdomain of FSD-1 (Figure 16). A 100-ns MD simulation with GB/SA implicit solvation showed that one of these chimeric proteins was more stable than the native structure and maintained the expected fold during simulations.128 More detailed simulations using replica exchange at several temperatures are underway to determine if the melting temperatures of the chimeric proteins can be predicted.
Future Studies of Chimeric RNase Proteins
The extensive experimental database available in the literature1 on RNase A, RNase S, and the three-component RNase system described in Merrifield's autobiography19 provides a unique opportunity to dissect many aspects of the thermodynamics of helix recognition, the contributions of entropy to the binding free energy, and the impact of preorganization on protein stability. As these topics require further clarification, several approaches with RNase model systems are of interest. These include measurement of the dynamics of the RNase S chimeric proteins by modern NMR relaxation measurements to determine the impact of preorganization on dynamics and enzyme activity. A direct spectroscopic method to measure the binding entropy of a ligand has recently been developed.158 In studies on an 11-residue peptide fragment from the C-terminus of the G-alpha subunit of transducin that binds to the photoactivated state of rhodopsin (R*), Kisselev et al. determined the R*-bound conformation of the peptide by transfer NOE experiments.159 Utilizing an analog of the peptide in which two spin labels had been introduced, Van Eps et al.158 used pulse EPR to measure the distance between the two nitroxides in the presence of rhodopsin, both dark-adapted and light-activated. In the first case, a distribution of distances between 16 and 36 Å was observed. After irradiation with light, the distribution collapsed to a sharp peak centered at 19 Å, the anticipated distance between the two spin labels based on the R*-bound conformation. The entropy of the peptide in solution and when R* bound can be estimated (see discussion of the configuration partition function, Chapter III in Flory64) from the distribution of distances between the two nitroxide labels to give the change in entropy upon peptide binding. The overall change in entropy of the S-peptide/S-protein binding measured by ITC includes the change in entropy of both S-peptide and S-protein, as well as changes in the overall entropy of solvation. Similar studies on spin-labeled S-peptide analogs when free or bound to S-protein would provide an independent experimental estimate of changes in entropy of the peptide analog upon complex formation.
RNase has proven to be a worthy enzyme of study, as extensive prior investigations have laid the groundwork for more precise experiments to address many questions of current interest. The historical association of this enzyme with the Rockefeller Institute/University had a profound influence on both Bruce Merrifield and his first graduate student (G.R.M.). There is still much to be learnt about the detailed thermodynamics of protein recognition and the impact of preorganization on binding affinity. The role of protein dynamics in enzymatic catalysis by RNase is still an open question that is amenable to detailed analysis by studies on chimeric RNase constructs. Many of the techniques necessary to analyze these issues depended on the development of solid-phase synthesis and the influence that Bruce Merrifield had on the next generation of protein chemists.
The authors acknowledge the overwhelming abundance of literature focused on various aspects of RNase and offer sincere apologies to those investigators whose seminal work was not cited. Computational resources were supported in part by TeraGrid. Prof. Ronald T. Raines of the University of Wisconsin kindly recommended some additional papers to be included after reading an early draft; Profs. Carl Frieden and Jay Ponder, as well as Dr. Christy Taylor of Washington University, kindly read and made suggestions to improve the manuscript. We must finally acknowledge the enormous influence that Bruce Merrifield and the talented colleagues whom he trained at Rockefeller University have had on modern protein chemistry and enzymology. It is indeed a pleasure to stand on the shoulders of such giants of protein biochemistry to glimpse into the future.