Evaluating Protein Engineering Thermostability Prediction Tools Using an Independently Generated Dataset

Abstract

Engineering proteins to enhance thermal stability is a widely utilized approach for creating industrially relevant biocatalysts. The development of new experimental datasets and computational tools to guide these engineering efforts remains an active area of research. Thus, to complement the previously reported measures of T ₅₀ and kinetic constants, we are reporting an expansion of our previously published dataset of mutants for β-glucosidase to include both measures of T _M and ΔΔG. For a set of 51 mutants, we found that T ₅₀ and T _M are moderately correlated, with a Pearson correlation coefficient and Spearman's rank coefficient of 0.58 and 0.47, respectively, indicating that the two methods capture different physical features. The performance of predicted stability using nine computational tools was also evaluated on the dataset of 51 mutants, none of which are found to be strong predictors of the observed changes in T ₅₀, T _M, or ΔΔG. Furthermore, the ability of the nine algorithms to predict the production of isolatable soluble protein was examined, which revealed that Rosetta ΔΔG, FoldX, DeepDDG, PoPMuSiC, and SDM were capable of predicting if a mutant could be produced and isolated as a soluble protein. These results further highlight the need for new algorithms for predicting modest, yet important, changes in thermal stability as well as a new utility for current algorithms for prescreening designs for the production of mutants that maintain fold and soluble production properties.

Figure 1

Structure of BglB (PDB ID: 2JIE) from the bacterium Paenibacillus polymyxa. PyMOL rendering of BglB showing the 28 sequence-positions (teal spheres) of the 51 mutants chosen out of the original 92 previously expressed proteins for the T_M analysis. The reaction scheme of the hydrolysis of 4-nitrophenyl β-d-glucopyranoside by BglB used in the T₅₀ study.

Figure 2

Comparison of two different experimental thermal stability datasets and experimentally derived ΔΔG. (A) Relationship for each mutant between T₅₀ and T_M. The PCC of 0.58 illustrates that the two methods are modestly positively correlated with mutations that are in the extreme ends of the temperature range (±5 °C). (B) Evaluation of ΔT_M with experimentally derived ΔΔG shows the two qualities are highly correlated (PCC = −0.76), unlike (C) where the relationship between ΔT₅₀ and experimentally derived ΔΔG has a PCC of −0.35.

Figure 3

Evaluation of the algorithms ΔTSE versus the experimentally derived ΔΔG and the T_M and T₅₀ datasets. The Pearson correlation coefficient and Spearman’s rank correlation for each performance against three types of experimental data were determined. Five representative comparisons are illustrated above, with four additional algorithms, SDM, AUTO-MUTE (DDG), AUTO-MUTE (ΔT_M), and HoTMuSiC provided in Figure SI 1-3. No algorithm resulted in a significant correlation between the calculated energies and the observed T_M, T₅₀, or ΔΔG for this dataset.

Figure 4

Computational prediction for the effect on mutant soluble protein production using nine different algorithms. From left to right: Rosetta ΔΔG, FoldX PSSM, ELASPIC, DeepDDG, PoPMuSiC, SDM, AUTO-MUTE (DDG), AUTO-MUTE (ΔT_M), and HoTMuSiC of soluble (green) and nonisolated protein (blue). In this case, mutants that resulted in a significant (>10-fold) decrease in yield of purified soluble protein are considered nonisolatable. Significance in population differences was determined using a Student’s t-test. The units (ΔTSE and ΔT_M) of all algorithms are individually normalized between 1 to −1. For visualization purposes, outliers were omitted after normalization. Each graph without normalization and with outliers can be found in Figure SI 1-4 and all raw values in Figure SI 4.