Computational design of serine hydrolases

Abstract

The design of enzymes with complex active sites that mediate multistep reactions remains an outstanding challenge. With serine hydrolases as a model system, we combined the generative capabilities of RFdiffusion with an ensemble generation method for assessing active site preorganization at each step in the reaction to design enzymes starting from minimal active site descriptions. Experimental characterization revealed catalytic efficiencies (k_cat/K_m) up to 2.2 × 10⁵ M^-1 s^-1 and crystal structures that closely match the design models (Cα root mean square deviations <1 angstrom). Selection for structural compatibility across the reaction coordinate enabled identification of new catalysts remove with five different folds distinct from those of natural serine hydrolases. Our de novo approach provides insight into the geometric basis of catalysis and a roadmap for designing enzymes that catalyze multistep transformations.

Figure 1.. Design methods.

(A) Active site specific backbone generation with RFdiffusion. Given the geometry of a possible active site configuration, RFdiffusion denoising trajectories generate backbone coordinates which scaffold the site. (B) Generation of active site ensembles with PLACER. The coordinates of the sidechains around the active site and any bound small molecule for the step in the reaction being considered are randomized, and n samples are carried out to generate an ensemble of predictions. (C) Mechanism of ester hydrolysis by serine hydrolases. (D) PLACER ensembles for distinct states along the reaction coordinate for hydrolysis of 4MU-Ac for a native serine hydrolase (top, PDB: 1IVY) and a designed serine hydrolase (bottom, r3-josie).

Figure 2.. Functional characterization of designed serine hydrolases.

(A) Chemical schematic of a serine hydrolase active site. (B) Summary of design method and experimental success rate for probe labeling, single turnover acylation, and catalytic turnover for each design round. (C) Chemical schematic depicting probe labeling, acylation, and catalytic turnover. (D) Fold (left) and active site (right) of serine hydrolase design models. (E) Reaction progress curves for the parent design and catalytic residue knockouts. Dashed line represents the enzyme concentration and shaded areas represent standard deviation of three technical replicates. (F) Michaelis-Menten plots derived from initial (r1, r2) or steady state velocities (r3). Error bars represent standard deviation of three technical replicates.

Figure 3.. Structural characterization of designed serine hydrolases.

(A, D) Structural superposition of design models (gray) and crystal structures (rainbow) for r3-super (A) and r3-win (D). (B and E) Active site overlays of design models (gray) and crystal structures (rainbow) of r3-super (B) and r3-win (E) with 2Fo-Fc map shown at 1σ (blue mesh). (C and F) Superposition of substrate binding sites of the design models (gray) and crystal structures (rainbow) of r3-super (C) and r3-win (F) with 2Fo-Fc map shown at 1 σ (blue mesh). Distances shown in Å.

Figure 4.. Computational redesign and more complex folds improve catalysis.

(A) Computational pipeline for redesign of r3-win. (B,C,D) k_cat (B), K_m (C), and k_cat/K_m (D) of parent r3-win compared to computational redesigns. (E,F,G) Structural superposition of design model and crystal structure of r4-win1 (E), r4-win31 (F), and (G) r4-dadt1 with 2Fo-Fc map shown at 1σ. (H,I,J,K) Design models (H,J) and Michaelis-Menten plots (I,K) for active designs with distinct folds. (L) Chemical and structural comparison of n and n+1 oxyanion hole motifs. (M) Design model of r6-momi that utilizes two backbone amide oxyanion hole contacts, one from an n+1 backbone amide. (N) Michaelis-Menten plot for r6-momi with 4MU-PhAc. Error bars represent standard deviation of three technical replicates.

Figure 5.. PLACER ensembles reveal geometric determinants of catalysis.

(A) Frequencies of catalytic Ser-His H-bond formation in PLACER ensembles for each reaction step, grouped by experimental outcome. (B) Apo PLACER ensembles of representative inactive (top) and acylating (bottom) designs. (C) Median angle (α) between serine Oγ, histidine Nϵ and Cϵ across PLACER ensembles of inactive and acylating designs. (D) Apo PLACER ensembles of representative inactive (top) and acylating (bottom) designs, angle indicates median α. (E) AEI PLACER ensemble H-bond frequencies for designs that undergo acylation or full turnover. (F) PLACER ensembles of the apo state for an acylating (top) and multiple turnover design (bottom). (G) PLACER ensembles of the AEI state for a representative design that undergoes acylation (top) and a design that catalyzes turnover (bottom). Measurements shown represent median distances (Å) of key H-bonds indicated for each ensemble and percentages represent frequency of H-bond formation across all PLACER trajectories. (H) Newman projections of serine g+ and g− rotameric states (left). (I) PLACER ensembles of an acylating design (top) and a design that catalyzes turnover (bottom). (J) Median serine χ₁ angle across TI1 and AEI state PLACER ensembles for designs that catalyze acylation or turnover (left) and for the same designs grouped by number of oxyanion hole H-bonds. (K) AEI state PLACER ensembles for r3-win, r4-win1, r4-win31, and r4-dadt1, with percent of frames with correct oxyanion hole rotamer, Ser χ₁ angle, and catalytic Ser-His H-bond distance shown. Boxplots represent median, upper and lower quartiles; whiskers extend 1.5×IQR above and below the upper and lower quartiles (respectively). Observations falling outside these ranges plotted as outliers.