Computationally designed libraries for rapid enzyme stabilization

Abstract

The ability to engineer enzymes and other proteins to any desired stability would have wide-ranging applications. Here, we demonstrate that computational design of a library with chemically diverse stabilizing mutations allows the engineering of drastically stabilized and fully functional variants of the mesostable enzyme limonene epoxide hydrolase. First, point mutations were selected if they significantly improved the predicted free energy of protein folding. Disulfide bonds were designed using sampling of backbone conformational space, which tripled the number of experimentally stabilizing disulfide bridges. Next, orthogonal in silico screening steps were used to remove chemically unreasonable mutations and mutations that are predicted to increase protein flexibility. The resulting library of 64 variants was experimentally screened, which revealed 21 (pairs of) stabilizing mutations located both in relatively rigid and in flexible areas of the enzyme. Finally, combining 10-12 of these confirmed mutations resulted in multi-site mutants with an increase in apparent melting temperature from 50 to 85°C, enhanced catalytic activity, preserved regioselectivity and a >250-fold longer half-life. The developed Framework for Rapid Enzyme Stabilization by Computational libraries (FRESCO) requires far less screening than conventional directed evolution.

Keywords: enzyme stability; in silico design; in silico screening; protein stability engineering; thermostability.

Scheme 1.

FRESCO. In Step 1, stabilizing mutations are generated with multiple algorithms. The in silico screening Steps 2 and 3 remove false positives. In Step 2, variants are eliminated which have properties that are known to typically decrease thermostability, such as increased hydrophobic surface exposure to the water phase or an increased number of unsatisfied H-bond donors and acceptors (for details, see the Materials section and the Results section). Step 3 eliminates variants in increased flexibility (an example is shown in Supplementary Fig. S3). An experimental screening (Step 4) is used before combining the most stabilizing mutations in Step 5. Details regarding Step 5 are described in the Results section.

Fig. 1.

Experimentally characterized point mutants of LEH. Protein variants that are significantly more thermostable are labeled (e.g. T85V). The abbreviation NE stands for no soluble expression. The gray background is used to distinguish the mutations with a predicted ΔΔG^Fold that does not significantly differ from zero (−5 to +5 kJ/mol). The variants that would not survive Steps 2 or 3 are plotted with different symbols as indicated in the inset. (A) The point mutations that were predicted to be stabilizing using Rosettaddg and also survived Steps 2 and 3 in FRESCO (Scheme 1). (B) Idem for the point mutations predicted to be stabilizing using FoldX; (C) Library characterized for a control experiment, in which the effects of omitting Steps 2 and 3 of the FRESCO protocol (Scheme 1) were tested. The best FoldX variants were selected, with maximally one mutation per position (thus, only T85I with a ΔΔG^Fold of −20 kJ/mol, not T85V with ΔΔG^Fold of −14 kJ/mol).

Fig. 2.

Overview of the stabilized variants.

Fig. 3.

Example of the predicted structure for a stabilizing mutation. The substitution T85V (ΔT_m^app = +7°C) removes a hydroxyl group in an apolar environment, and replaces it with a more hydrophobic methyl group. The polar side-chain atoms of Thr97 and Arg99 are >5 Å from the hydroxyl oxygen of Thr85, which excludes hydrogen bonding of Thr85 with those residues, indicating that Thr85 has an unsatisfied H-bond donor or acceptor.

Fig. 4.

Distribution of stabilizing mutations over the protein (crystal structure 1NWW). (A) B-factors of the C_α atoms of 1NWW (thickest traces with red color correspond to the highest B-factors). (B) Location of all the point mutations shown with spheres for which the color reveals the level of stabilization.

Fig. 5.

Overview of the improved thermal stability of the LEH mutants F1 and F2. The melting temperatures of F1 (blue) and F2 (red) compared with WT (black) as determined by DSC (A) and thermofluor assay (B). (C) The remaining catalytic activity measured at 30°C after pre-incubation for 15 min at the indicated temperatures. (D) The 30°C increase in optimum temperature for catalytic activity of these variants and (E) the slower enzyme inactivation by incubation at 55°C. (F) Retained regioselectivity of the final variants as determined by chiral GC.The elution profiles are those of the produced limonene diols. A reference sample contained both (1R,2R,4R)-limonene diol at 27.5 min and (1S,2S,4R)-limonene diol at 29.6 min (shown in gray).

Fig. 6.

Rate of (4R)-limonene 1,2-epoxide conversion versus its concentration. Wild-type LEH (black circles), variant F1 (blue triangles) and variant F2 (red squares) are indicated with different symbols. The fit is according to k_t = k_cat × [S]/([S] + K_M), in which k_t is the catalytic turnover rate per enzyme active site and [S] is the substrate concentration. The turnover rate is plotted at (A) 30°C and (B) at the optimum temperature for catalytic activity (for WT 50°C; for variant F1 80°C; for variant F2 70°C). At 30°C, the mutants have a lower catalytic activity compared with the WT, whereas the mutants clearly outperform the WT at the optimum temperature.