Computational design of ligand-binding proteins with high affinity and selectivity

Abstract

The ability to design proteins with high affinity and selectivity for any given small molecule is a rigorous test of our understanding of the physiochemical principles that govern molecular recognition. Attempts to rationally design ligand-binding proteins have met with little success, however, and the computational design of protein-small-molecule interfaces remains an unsolved problem. Current approaches for designing ligand-binding proteins for medical and biotechnological uses rely on raising antibodies against a target antigen in immunized animals and/or performing laboratory-directed evolution of proteins with an existing low affinity for the desired ligand, neither of which allows complete control over the interactions involved in binding. Here we describe a general computational method for designing pre-organized and shape complementary small-molecule-binding sites, and use it to generate protein binders to the steroid digoxigenin (DIG). Of seventeen experimentally characterized designs, two bind DIG; the model of the higher affinity binder has the most energetically favourable and pre-organized interface in the design set. A comprehensive binding-fitness landscape of this design, generated by library selections and deep sequencing, was used to optimize its binding affinity to a picomolar level, and X-ray co-crystal structures of two variants show atomic-level agreement with the corresponding computational models. The optimized binder is selective for DIG over the related steroids digitoxigenin, progesterone and β-oestradiol, and this steroid binding preference can be reprogrammed by manipulation of explicitly designed hydrogen-bonding interactions. The computational design method presented here should enable the development of a new generation of biosensors, therapeutics and diagnostics.

Figure 1. Computational Design Methodology and Experimental Validation

a, Overview of the computational design procedure. First, the geometric positions of a set of pre-chosen interaction side chains are defined with respect to the ligand (left panel), and rotamers for each interaction side chain are enumerated (left panel, inset). Second, a set of scaffolds is searched for backbones that can accommodate all of the desired interactions. For cases in which all chosen interaction residues can be placed in the scaffold protein and orient the ligand in the native binding cavity with no backbone clashes, the binding site sequence is optimized for binding affinity (center panel). Designs having native-like properties, such as high shape complementarity and binding site pre-organization, are chosen for experimental characterization (right panel). b, Ranking of the 17 experimentally characterized DIG designs by ligand interaction energy (Rosetta energy units, Reu) and the average (geometric mean) Boltzmann weight of the conformations of the side chains that hydrogen bond to the ligand. DIG10, depicted in red, scores the best by both metrics. c, Flow cytometric analysis of yeast cells expressing computationally designed proteins as part of a surface-targeted fusion protein with a C-terminal c-Myc tag. Yeast surface expression and DIG binding were confirmed by labeling the cells with a fluorescein (FITC)-conjugated anti-c-Myc antibody and a pre-incubated mixture of 2.7 μM biotinylated DIG-functionalized BSA (~10 DIG/BSA) and phycoerythrin (PE)-conjugated streptavidin, respectively. Cell populations shown are a negative control for binding (ZZ(-)), an anti-DIG antibody serving as a positive control for binding (ZZ(+)), DIG10, DIG10 in the presence of excess (730 μM) unlabeled DIG competitor, and scaffold 1z1s. DIG10 labeled with 2.7 μM biotinylated DIG-functionalized RNase (~6 DIG/RNase) is also shown. d, On-yeast substitutions of DIG10 designed hydrogen-bonding residues Tyr34, Tyr101, and Tyr115 to Phe and binding cavity residue Val117 to Arg reduces expressing-population compensated mean binding (PE) signals to background nonbinding (ZZ(-)) levels.

Figure 2. Affinity Maturation

a, Equilibrium fluorescence polarization measurements of DIG-PEG₃-Alexa488 treated with increasing amounts of DIG10 (blue), DIG5 (cyan), scaffold 1z1s (black), and negative control bovine serum albumin (red). Solid lines represent fits to the data to obtain dissociation constants (K_d values). Error bars represent standard deviations for at least three independent measurements. b, K_d values of relevant designs and affinity matured DIG10 variants. FP and ITC were used to measure the affinities for DIG-PEG₃-Alexa488 and unlabeled DIG, respectively. The high affinity of DIG10.3 for DIG precluded measurement of a reliable K_d value by ITC. The differences in the K_d values observed by the two methods may be due to interactions with the PEG linker present in the DIG-PEG₃-Alexa488 conjugate; DIG10.2, which shows the largest discrepancy, has two mutations at the cavity entrance. c, Mutations identified during affinity maturation to generate DIG10.1 (blue), DIG10.2 (orange), and DIG10.3 (green) mapped on to the computational model of DIG10.3. d, ITC thermodynamic parameters of DIG10 (cyan), DIG10.1 (blue), DIG10.2 (orange), and DIG10.3 (green) variants binding to DIG. The ΔG used to determine TΔS for DIG10.3 was calculated from the FP K_d value. e, Binding fitness landscape of DIG10.1 probed by deep sequencing. The effect of each amino acid substitution at 39 binding site residues on binding (${\Delta }_i^x$) was assessed by determining the log₂ ratios of the frequencies of substitutions to each amino acid at each position after selection for DIG binding to the frequencies of the substitutions in the unselected population. Colored grids represent single point mutations having ≥7 counts in the unselected N-terminal (fragment 1) and C-terminal (fragment 2) libraries, where red and blue indicate high enrichment and depletion, respectively. White grids show mutations for which there were not enough sequences in the unselected library to make a statistically significant conclusion about function. The initial DIG10.1 amino acid at each position is indicated in bold using its one-letter amino acid code. f, The optimality of each DIG10.1 input residue type mapped onto the computational model of DIG10.1. Optimality is defined as the positional Z-score: $Z=\frac{x-\mu }{\sigma }$ where x is the sum of enrichment values at position i, μ is the mean sum of enrichment values for all interrogated positions within the fragment library, and, σ is the standard deviation of the sums of enrichment values for all interrogated positions within the fragment library. Blue is very optimal (mutations to all other amino acids are disfavored) and red is suboptimal (mutations are preferred).

Figure 3. Crystal Structures of the DIG10.2- and DIG10.3-DIG Complexes

a, Surface representation of the DIG10.2-DIG complex showing the high overall shape complementarity of the interface. DIG is depicted in magenta spheres. DIG10.2 is a dimer and crystallized with four copies in the asymmetric unit. b, 2F_o - F_c omit map electron density of DIG interacting with Tyr34, Tyr101, and Tyr115 contoured at 1.0 sigma. c, Backbone superposition of the crystal structure of the DIG10.2-DIG complex (magenta) with the computational model (gray) shows close agreement between the two. d, Binding site backbone superposition shows that the ligand and the three programmed Tyr hydroxyl groups are in their designed conformations. e, Configurational side chain entropy between the four crystallographic copies of the DIG10.2-DIG (left panel) complex and chains A, B, C, H, and I of the DIG10.3-DIG (right panel) complex. The side chains of DIG10.3 at positions 103, 105, and 115 each adopt only a single rotamer. DIG10.2 Tyr115 conformation A features a more canonical hydrogen-bonding geometry than that of conformation B.

Figure 4. Steroid Binding Selectivity

a, The x-ray crystal structure of the DIG10.3-DIG complex (left panel) and the chemical structures of steroids interrogated in equilibrium competition fluorescence polarization assays (right panel). b, Steroid selectivity profile of DIG10.3. Solid lines represent fits to the data to obtain half-maximal inhibitory concentrations (IC₅₀ values) and error bars indicate standard deviations for at least three independent measurements. c, Steroid selectivity profile of DIG10.3 Tyr101Phe. Dashed lines show qualitative assessments of the inhibitory effects for cases in which the data could not be fit due to experimental limitations (see Supplementary Methods). d, Steroid selectivity profile of DIG10.3 Tyr34Phe. e, Steroid selectivity profile of DIG10.3 Tyr34Phe/Tyr99Phe/Tyr101F.