All of the mutations are located in the catalytic cleft on the inside of the concave jelly-roll protein fold, in one of three regions that directly interact with the ligand. A Cartoon representation of the model and X-ray structure showing the 1. B Detailed comparison of the residues comprising the ligand interface. Most of the residue side chains are superimposable, while several are out of position due to the altered backbone conformation.
Only two side-chain rotamers assume substantially different conformations from prediction. C Residues identified as directly responsible for binding pocket opening. During the design process, many of the residues in the catalytic site of the 1m4w enzyme were altered in favor of the new peptide binding function, thus eliminating the proteins' native catalytic functionality. The wide and deep catalytic cleft of the protein was transformed by the design process into a tightly fitting binding pocket, closely contacting the target d -ala- d -ala or d -ala- d -lac dipeptide ligands on all sides except the N-termini, thus allowing for egress of the un-modeled remainder of the glycopeptide Fig.
Previous studies by Meiler and Baker found that R osetta energy units correspond to experimentally determined binding energies with a correlation of 0. Additionally, good hydrophobic packing of both ligand methyl groups and strong binding of the carboxyl terminus were common features in each of the nine protein designs.
Detailed schematic of ligand interface. Darker yellow, thicker lines indicate low exposed surface area; lighter, thinner lines indicate more solvent exposure. Gray dashed line denotes the path of the unmodeled portion of glycopeptide ligand.
Note the decrease in number of H-bonds and increase in degree of solvent exposure. Expression of the R osetta L igand designed proteins proceeded as outlined in the Materials and methods section. Dynamic light scattering and size-exclusion chromatography of each of the expressed proteins indicated that the 1m4w designs existed in solution as homogeneous, monomeric species. Far-UV CD spectra of the 1m4w designed proteins indicated secondary structure composition similar or identical to wild-type Supplementary data, Fig.
One-dimensional NMR results confirmed that all of the 1m4w proteins were well folded and stable. Additionally, the 1m4w designed proteins exhibited a high degree of stability and resistance to proteolysis. Samples left at room temperature for several weeks following purification showed no signs of degradation. Following computational design and expression of the chosen interface designs, binding assays were performed to validate the predicted affinities. Unfortunately, none of the designed proteins tested in this study yielded evidence of specific, high-affinity binding to their target peptide.
We thus conclude that the R osetta L igand interface designs were not successful. To determine a cause for the lack of observed binding among the designed proteins, a high-resolution X-ray diffraction structure of 1m4w6 was determined. The optimal crystallization buffer contained 0. The final conditions differed significantly from that used for the wild-type 1m4w structure Hakulinen et al. This expansion occurs through a 1.
Moreover, the solvent accessible SA surface area of the pocket increases 2. The all-atom root mean square deviation RMSD for the whole protein is 0. Notably, interface residues that contribute most to RMSD are also those possessing the highest crystallographic temperature factors B-factors. The expansion of the binding pocket disrupts interactions observed in the computational model. When the ligand is re-docked into the crystallographic structure, only 8 of 11 predicted hydrogen bond interactions are able to assume correct bonding geometry, while the ratio of ligand surface area in VDW contact with protein decreased from 0.
Thus, we hypothesized that the lack of observed ligand binding affinity was due to the expansion of the binding pocket and resulting disruption of predicted binding contacts. C Chart showing the relative degree of binding pocket expansion for each sequence substitution. In comparing the two structures, R osetta L igand calculations showed a modest but clear loss of binding affinity as pocket backbone opening increased, as indicated by several of the contributing energy terms Supplementary data, Table SI.
For example, the total number of residues involved in the hydrogen bonding network between ligand and protein decreased from 8 to 6, while the number of total hydrogen bonds dropped from 11 to 8. Similarly, solvation and electrostatic interaction energies worsened as pocket expansion increased. From this analysis, we concluded that R osetta L igand can discriminate between the binding energies of a wild-type backbone configuration and that of an enlarged binding pocket, and that this energy differential could potentially explain the lack of experimentally observed ligand binding.
Evolutionary evidence for the crucial function of these residues can be seen from a sequence alignment of 1m4w with its nearest homologues. In all , the Trp20, Pro and Val48 are either strictly or highly conserved.
Reverting these mutations, it was hoped, would restore the binding pocket to the predicted wild-type geometry, thus conferring the originally predicted ligand binding affinity. Multiple single crystals formed in several buffers centered around wells containing 0. Complete data sets down to 1. Attempts to obtain ligand bound co-crystals were unsuccessful.
All protein structures obtained were in the apo configuration. The intent of this study was to explore computational methods for designing de novo high-affinity protein—peptide interfaces. The protein designs described above did not achieve our goal of high-affinity binding to their target peptides. Our hypothesis at the outset of this study was that R osetta L igand was capable of de novo design of a high-affinity protein—peptide interface to a non-standard dipeptide ligand. Experimental testing of our original nine protein—peptide interface designs yielded negative results for high-affinity ligand binding, thus failing to prove this hypothesis.
Testing the second hypothesis by expression and assay of three re-designed proteins yielded similar negative results for ligand binding. Structure determination and analysis of the three proteins yielded further important insights. However, part of the second hypothesis was shown to be true.
We speculate that changes in the configurational dynamics of the protein as seen in crystallographic B-factors may be partly responsible for the lack of high-affinity ligand binding.
- Logical Aspects of Computational Linguistics: 5th International Conference, LACL 2005, Bordeaux, France, April 28-30, 2005. Proceedings;
- Related Articles.
- Main content!
- The First 90 Days: Critical Success Strategies for New Leaders at All Levels.
However, confirmation of this hypothesis remains outside the scope of our experimental data. An equally likely contributor to failure may be shortcomings in the R osetta energy function, in particular its solvation energy function or treatment of water molecules. As described in the Results section, we tested R osetta L igand 's ability to predict the backbone changes observed in the mutant proteins. The original protocol intentionally prevented the protein backbone from adapting in response to mutations introduced during design.
The decision to use a fixed-backbone protocol initially was made to increase speed of the calculations and was based on the erroneous assumption that a thermophilic protein scaffold such as 1m4w would be unlikely to experience significant conformational change from the mutation of a small number of residues in the enzymatic cleft. When subsequently using protocols able to accommodate backbone flexibility, R osetta L igand is quantitatively able to predict the shift in backbone configuration when the destabilizing Trp20 and Val48 mutations are alternately included or removed.
R osetta L igand flexible-backbone protocols can recapitulate backbone conformational shift. We thus conclude that R osetta L igand is able to predict the structure of the 1m4w designs to near atomic resolution of both the binding interface and the protein as a whole, and that the modeling of backbone conformational changes is important when designing protein—peptide interfaces.
Computational design of ligand-binding proteins with high affinity and selectivity.
Although R osetta L igand can accurately predict large-scale changes in backbone configuration observed in the designed protein structures, the computational protocols employed in this study are significantly limited at addressing complex protein dynamics and potential entropic factors of ligand binding. R osetta scoring and binding energy calculations are performed using a single, static, atomic representation of protein and ligand. Although recent advances in flexible backbone and relaxation functionality within R osetta have expanded its ability to address structural fluctuation during design Davis and Baker, , the ability to fully predict the effects of dynamics at a protein—ligand interface remains limited.
These elevated B-factors suggest a fundamental alteration in the dynamics of the protein as a whole Rueda et al. Visualization of crystallographic B-factors for wild-type and four 1m4w mutant proteins. A—E Backbone and residue side chains colored and sized by B-factor values for wild-type 1m4w and X-ray determined structures.
This suggests a fundamental shift in the overall dynamics of the protein. These B-factors are 1. Beyond the implications of altered proteins dynamics, standard R osetta L igand design protocols rely on a bulk, non-explicit solvation term Lazaridis and Karplus, to represent water molecules in and around the binding interface.
Entropic factors of binding-pocket desolvation are not well addressed by an implicit solvation term Gilson and Zhou, Examination of the four X-ray structures reveal 9—11 ordered water molecules within the binding pocket. Due to the increased importance of predicting individual atomic interactions in the design of high-affinity interfaces, the explicit modeling of water molecules is desirable for successful design of protein—ligand interfaces Amadasi et al.
Although recent extensions to R osetta now allow explicit interfacial waters to be modeled, this functionality did not exist at the time this study commenced. A dipeptide ligand composed of small, non-polar amino acids is a difficult target for a proof-of-concept experiment and was intended to push the boundaries of R osetta L igand technology. This, however, may have been overly ambitious. A larger, more apolar ligand possessing greater VDW surface area and opportunity for charge—charge interactions would be preferred in future work. More important to the potential success of protein—ligand interface design are the dynamics and conformational stability of a design scaffold protein.
This dynamic propensity is an undesirable trait in a protein scaffold when attempting to design a well-defined, stable, high-affinity interface. Deliberate care is advisable when choosing a de novo design scaffold, and particular attention should be given to protein dynamic modes. In this respect, scaffolds that have been extensively classified by NMR, small-angle X-ray scattering molecular dynamic simulations or other methods that yield information on protein dynamics are preferred. This interaction is crucial to maintaining a closed geometry under crystallization conditions.
That this Trp—Pro interaction is highly conserved across multiple species indicates that it is likely a key structural, dynamic and kinetic determinant common to family 11 xylanases. While the hydrophobic Trp20—Pro interaction is necessary, it is not sufficient to allow stable closing of the binding pocket.
Thus, although reversion of position 20 to the wild-type Trp is necessary for binding pocket closing, it is not in and of itself sufficient. It is intriguing that the effects of a single, conservative substitution at a spatially distal amino acid position can have such a pronounced effect on the stability of a thermophilic protein at relatively low temperature, i. This suggests that the amino acid sequence of the 1m4w protein, even in the protein core palm region , is finely tuned to accommodate this dynamic mobility.
These mutations might therefore be thought of as having enabled high-temperature, native-like dynamics at low temperatures. While the lack of success experienced in the course of this particular study may or may not be attributable to factors such as scaffold selection, unanticipated protein dynamics or the lack of explicitly modeled interfacial waters, it is important to note that progress in the field of de novo ligand interface design as a whole has lagged significantly behind other areas of de novo protein design. Not long ago, it was considered by some to be a solved problem, but retractions in several key papers Check Hayden, have led to the conclusion that the design of high-affinity protein—ligand interfaces is one of the fundamental areas of basic protein function that remains an open problem Schreier et al.
R osetta has proven adept at such challenging tasks as design of novel protein folds Kuhlman et al. Protein—protein interfaces have been re-designed for altered and multiple specificity Joachimiak et al. What is it that makes de novo design of protein—ligand interfaces so difficult, and why would de novo interface design be significantly more challenging than the re-design of a protein—peptide interface, or the design of a novel enzyme?
Computational Design of the Tiam1 PDZ Domain and Its Ligand Binding
While a completely satisfactory answer to these questions has yet to be established, one contributing factor could be protein dynamics. The requirement to design and manipulate dynamics may set a higher bar for the de novo design of ligand binding. Unfortunately, protein dynamics is also one of the most difficult and least tractable problems for current protein design programs. De novo protein design by definition entails establishing entirely new functionality in a protein. It requires an ability to recreate and manipulate all properties of a protein necessary for a given function.
Conversely, re-design, where basic protein functionality is retained but relies on conserved intrinsic properties of the protein important to its function. Such conserved intrinsic properties could include protein dynamic modes conducive to ligand binding. Similarly, re-design of protein—protein specificity may benefit from conserved functionality and dynamics, as well as having the added advantage of a larger interface surface area and number of potential interactions to offset small errors in the design algorithms.
Such small errors may have a larger impact in ligand interface design where each of a small number of interactions must be optimal for tight interaction. Yet surely the creation of novel catalytic function in the de novo design of enzymes Jiang et al. What has allowed these efforts to succeed where interface design has yet to? A partial answer may lay in the nature of enzyme function. In this case, the precise geometry of the catalytic mechanism is critical, and facilitating this geometry can be thought of as binding the chemical transition state.
Furthermore, recent studies suggest that the chemical reaction in enzyme catalysis is insensitive to global protein dynamics, which instead affect only enzyme kinetics Pisliakov et al. In this light, it is notable that all of the successful enzyme designs cited above were performed using a naturally occurring enzyme as a design scaffold some even used 1m4w and that all of these designed enzymes possess relatively poor kinetic properties, even after undergoing multiple rounds of directed evolution to improve efficiency Jiang et al.
While these speculations are far from conclusive with the small amount of evidence presented here, it is an intriguing line of thought that may warrant further attention in future studies. Our attempts at using the R osetta L igand program to design in silico a high-affinity protein—peptide interface to a bacterial dipeptide target were unsuccessful. Twelve proteins using 1m4w as a design scaffold were assayed for binding to their intended target. No high-affinity binding was detected for any of these 12 designs.
We have proposed several potential contributors to this apparent lack of success, including overambitious target peptide selection and the lack of explicitly modeled interfacial water molecules. However, possibly the most significant negative contributor to the study outcome may be the unappreciated nature and extent of dynamics inherent to the design scaffold protein. We have shown that R osetta L igand is able to predict the structure of designed interfaces to near-atomic resolution, and of large-scale protein conformational changes due to mutations introduced during the design process.
However, accurate structure prediction did not translate into high-affinity ligand binding. We therefore conclude that the computational design of proteins that tightly bind small molecules remains possibly a greater challenge than the design of enzymes. While computational enzyme design requires accurate structural prediction of catalytic residues, tight substrate binding is not needed for success. These xylanase structures may also serve as benchmark systems for future computational design protocols that model protein—peptide or protein—small molecule interfaces.
Supplementary data are available at PEDS online. Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide. Sign In or Create an Account. Sign In. Advanced Search.
Article Navigation. Close mobile search navigation Article Navigation. Barry Stoddard Basic Sciences Division and David Baker University of Washington as well as additional collaborators at UW and Rutgers University, published in Nature describes the first successful computational design and in vitro characterization of two ligand-binding proteins using a novel approach that may someday "open up a new strategy to create systems that combine the tumor targeting properties of antibodies with the therapeutic activity of traditional chemotherapy agents," says Dr.
The authors set out to design proteins with high affinity and selectivity for the steroid digoxigenin DIG , a derivative of the cardiac glycoside digoxin, which is used in the treatment of heart disease. Using these criteria, the authors designed 17 DIG-binding proteins that were then experimentally characterized. Designed proteins were expressed on the surface of yeast cells yeast surface display and tested for DIG binding using fluorescence-activated cell sorting FACS after incubation with DIG-biotin derivatives and a streptavidin-conjugated fluorophore.
To further enhance the affinity of DIG10 for DIG, the authors performed saturation mutagenesis, testing the effects of every possible amino acid substitution on this interaction.
Timeless Secrets of Health and Rejuvenation
The authors next generated a binding fitness map of 39 residues designed to interface with DIG by testing a library of variant proteins with amino acid substitutions at each of these positions for DIG affinity. Further selection led to DIG Notably, the affinity of DIG Strikingly, the steroid specificity of DIG These results indicate that the high specificity of DIG The researchers are now turning their computational approach to the design of proteins for cancer therapy.
Says Dr. Stoddard, "We are now working hard on studying the structure and mechanism of chimeric antigen receptors protein molecules that are expressed on the surface of engineered T-cells for immunotherapy and hope to one day be able to use protein engineering to create new types of antitumor targeting receptors that act to direct reprogrammed T-cells and accompanying chemotherapeutic agents directly to individual tumor cells.