Design of strictly orthogonal biosensors for maximizing renewable biofuel overproduction

Transcription factors (TFs) activate transcriptional initiation by binding specific signal molecules (SMs), yet designing TFs to precisely target non-natural SMs remains challenging.

Using transcriptional activator BmoR as an example, a machine-learning based model named BT to predict three crucial residue regions (CRRs) was generated. This study aimed to achieve BmoR with strict SM orthogonality (SSO).

Random Forest Algorithm was used to generate a model BT that pinpointed CRRs. The BmoR-SM complexes in the prediction dataset of Model BT were batch-simulated using a computational pipeline via Discovery Studio. Semi-rational engineering of the residues in the CRRs generated BmoR mutants with SSO, validated through MicroScale Thermophoresis (MST) affinity assays. The SSO-enabled BmoR-based biosensor was used to screen microbial overproducers for 3-L fed-batch fermentation.

The transcription activation effects of 245 TF-SM complexes were experimentally verified, providing the training and test dataset to generate a machine-learning based Model BT with 88.5 % accuracy. The binding between 5,700 BmoR mutants and four SMs was simulated by Discovery Studio, generating 22,800 complexes to output BmoR-SM hydrogen bond (BSH) counts. BSH counts combined with supplementary parameters to form a prediction dataset. The CRRs containing totally 36 residues were successfully predicted by Model BT. The CRRs were semi-rational modified to obtain BmoR mutants with SSO. The SSO-enabled BmoR-based biosensor effectively screened a strain yielding 12.6 g/L isopentanol.

By demonstrating the dominant role of the BSH counts in TF-SM interactions and establishing a machine learning-guided framework for TF evolution, this work advances rational design principles for engineering TFs with precise molecular recognition, offering broad applications in synthetic biology and metabolic engineering.