To explore the mechanistic origin that determines the binding affinity of SARS-CoV-2 spike receptor binding domain (RBD) to human angiotensin converting enzyme 2 (ACE2), we constructed the homology models of RBD-ACE2 complexes of four Omicron subvariants (BA.1, BA.2, BA.3 and BA.4/5), and compared them with wild type complex (RBD_WT-ACE2) in terms of various structural dynamic properties by molecular dynamics (MD) simulations and binding free energy (BFE) calculations. The results of MD simulations suggest that the RBDs of all the Omicron subvariants (RBD_OMIs) feature increased global structural fluctuations when compared with RBD_WT. Detailed comparison of BFE components reveals that the enhanced electrostatic attractive interactions are the main determinant of the higher ACE2-binding affinity of RBD_OMIs than RBD_WT, while the weakened electrostatic attractive interactions determine RBD of BA.4/5 subvariant (RBD_BA.4/5) lowest ACE2-binding affinity among all Omicron subvariants. The per-residue BFE decompositions and the hydrogen bond (HB) networks analyses indicate that the enhanced electrostatic attractive interactions are mainly through gain/loss of the positively/negatively charged residues, and the formation or destruction of the interfacial HBs and salt bridges can also largely affect the ACE2-binding affinity of RBD. It is worth pointing out that since Q493R plays the most important positive contribution in enhancing binding affinity, the absence of this mutation in RBD_BA.4/5 results in a significantly weaker binding affinity to ACE2 than other Omicron subvariants. Our results provide insight into the role of electrostatic interactions in determining of the binding affinity of SARS-CoV-2 RBD to human ACE2.

Thermophilic β-1,4-endoglucanases (Cel5A) have garnered significant interest due to their potential applications in various industries, particularly in biofuel production and biorefineries. However, despite inherent stability, thermophilic Cel5A still face challenges in terms of further enhancing their catalytic efficiency and thermostability. In this study, a novel B-factor analysis method was used to predict beneficial amino acid substitutions within a 4 Å radius of the catalytic site in the tunnel of thermophilic Cel5A from Acidomyces richmondensis (ArCel5A). A combined strategy involving site-saturation mutagenesis and high-throughput screening was employed to identify the variants with the highest endoglucanase activity. Genomic sequencing revealed a mutation at position 299 in the starting strain T. reesei A2H, where the nucleotide sequence changed from TAC to TGC, resulting in a corresponding amino acid substitution from Tyrosine(Y) to Cystine(C). The endoglucanase activity of the mutant ArCel5A reached 3251 IU/mL, representing an 85.2% increase compared to wild-type ArCel5A at the fermentation time of 94 h. Significantly, the ArCel5A-Y299C mutant showed superior thermostability, retaining 93.8% of its initial activity after 30 min at 70 °C, and 91.5% after 10 min at 80 °C. Various computational simulation methods confirmed that the Y299C mutation enhanced the stability of the catalytic pocket, thereby improving the overall stability and catalytic efficiency of ArCel5A. This study offers an effective strategy for mining target sites for rational mutagenesis based on highly conserved sequences, which simultaneously improves both the thermostability and catalytic efficiency of thermophilic Cel5A.

Identifying the catalytic regioselectivity of enzymes remains a challenge. Compared to experimental trial-and-error approaches, computational methods like molecular dynamics simulations provide valuable insights into enzyme characteristics. However, the massive data generated by these simulations hinder the extraction of knowledge about enzyme catalytic mechanisms without adequate modeling techniques. Here, we propose a computational framework utilizing graph-based active learning from molecular dynamics to identify the regioselectivity of ginsenoside hydrolases (GHs), which selectively catalyze C6 or C20 positions to obtain rare deglycosylated bioactive compounds from Panax plants. Experimental results reveal that the dynamic-aware graph model can excellently distinguish GH regioselectivity with accuracy as high as 96–98% even when different enzyme–substrate systems exhibit similar dynamic behaviors. The active learning strategy equips our model to work robustly while reducing the reliance on dynamic data, indicating its capacity to mine sufficient knowledge from short multi-replica simulations. Moreover, the model’s interpretability identified crucial residues and features associated with regioselectivity. Our findings contribute to the understanding of GH catalytic mechanisms and provide direct assistance for rational design to improve regioselectivity. We presented a general computational framework for modeling enzyme catalytic specificity from simulation data, paving the way for further integration of experimental and computational approaches in enzyme optimization and design.