Introduction
Protein engineering has traditionally been dominated by directed evolution, a laboratory-intensive process of iterative mutation and selection. While powerful, this "trial and error" approach is resource-heavy and slow. Computational methods have sought to streamline this process, primarily through Ancestral Sequence Reconstruction (ASR), which identifies robust ancient variants. However, ASR is inherently retrospective. To address this, researchers have developed the Successor Sequence Predictor (SSP), an in silico tool designed to "look forward" by extrapolating a protein's evolutionary trajectory to predict beneficial future mutations.
Methodology: Mapping Evolutionary Trajectories
The core premise of SSP is that evolution often follows directional trends in physicochemical space—a concept termed "evolutionary momentum." The tool treats the history of a protein as a mathematical path.
The workflow begins by curating a diverse dataset of homologs for a target protein. Using these sequences, SSP constructs multiple phylogenetic trees and performs ASR to infer the amino acids of ancient ancestors. For every position in the protein, the algorithm identifies the "main path" from the root of the tree to the modern sequence.
Once the trajectory is established, amino acids along the path are converted into numerical values using nine manually selected AAindices (descriptors such as molecular weight, hydrophobicity, and isoelectric point). SSP then fits a linear regression model to these numerical trends and extrapolates the "next step" to predict the successor amino acid.
Physicochemical Descriptors and Prediction Reliability
The choice of physicochemical properties is critical for capturing the nuances of protein evolution. The nine indices used by SSP cover a range of structural and chemical characteristics.
To improve prediction accuracy, SSP employs two bundling strategies:
- Site Agreeing Predictions (SAPs): Occur when multiple indices agree that a mutation is likely at a specific site, regardless of the specific amino acid.
- Mutation Agreeing Predictions (MAPs): Occur when multiple indices converge on the exact same amino acid substitution. MAPs represent the highest confidence predictions.
Experimental Validation
The authors validated SSP against five experimental datasets to assess its performance across different functional traits:
- Thermostability: In the cold shock protein CspB, SSP identified 14 mutations. Eight were found to be stabilizing, with seven significantly increasing the melting temperature ($T_m$). One notable mutation provided a $+16.6^\circ C$ increase in stability.
- Enzymatic Activity: Testing against aminoglycoside 3’-phosphotransferase showed that SSP-predicted mutations significantly outperformed random mutations. While random substitutions typically decrease fitness, SSP-predicted variants achieved an average enrichment score of 1.36 (rising to 1.4 for high-confidence MAPs).
- Positive Selection: The tool successfully identified sites under positive evolutionary selection in 7 out of 8 cases for ADP-ribosylarginine hydrolase.
- Solubility: Predictions for levoglucosan kinase were largely positive or neutral, indicating that the tool effectively avoids mutations that would destabilize the protein's fold.
Conclusion and Future Directions
SSP represents a significant shift from descriptive to predictive evolutionary modeling. By leveraging evolutionary momentum, it identifies mutations that enhance stability, activity, and solubility without the need for exhaustive laboratory screening.
The primary limitation of the current pipeline is its reliance on high homolog density; a robust set of related sequences is necessary to build accurate phylogenetic trees. Future developments aim to integrate SSP into the FireProtASR web server and explore the use of Variational Autoencoders (VAEs) to further refine predictions. This tool offers a promising, cost-effective alternative for the rapid optimization of protein traits in biotechnology and medicine.
Summary updated