Data-driven prediction and design of bZIP coiled-coil interactions

Area of Science:

• Molecular Biology
• Protein Interactions
• Bioinformatics

Background:

• Basic-region leucine-zipper (bZIP) transcription factors exhibit specific dimerization crucial for their function.
• Predicting these interactions from sequence data is challenging due to the structural similarity of bZIP proteins.
• The coiled-coil motif is central to bZIP homo- and hetero-dimerization.

Purpose of the Study:

• To develop a predictive model for bZIP coiled-coil interactions using conserved residue interactions.
• To design synthetic peptides capable of selective binding to target bZIP proteins.
• To explore applications in protein inhibition and synthetic biology.

Main Methods:

• Utilized a dataset of 4,549 bZIP coiled-coil interactions to train a predictive model.
• The model incorporates residue-pair and triplet interaction energies.
• Employed an integer linear programming strategy to assemble native heptad modules for peptide design.

Main Results:

• The developed model achieved a correlation of R = 0.68 with experimental binding free energies, outperforming existing scoring functions.
• Designed synthetic peptides demonstrated tight and selective heterodimer formation with target bZIP domains (JUN, XBP1, ATF4, ATF5).
• Over 132 candidate complexes were tested using fluorescence resonance energy transfer (FRET) assays.

Conclusions:

• The predictive model accurately captures bZIP coiled-coil interactions.
• De novo peptide design strategy enables selective targeting of specific bZIP proteins.
• This approach offers potential for creating protein inhibitors and novel peptides for synthetic biology and nanotechnology.