NLM IRP Seminar Schedule



Scheduled Seminars on Feb. 8, 2024

Joey Thole
3 p.m.
Presentation Title
Deep learning-assisted protein design to characterize the alpha-helix <-> beta-sheet transition of a universally conserved transcription factor

Contact with questions about this seminar.


Although most proteins adopt a single energetically favorable fold, some proteins have been evolutionarily selected to reversibly interconvert between distinct folds that regulate biological processes or perform different functions. One such fold-switching protein is Escherichia coli RfaH, a member of the only known family of universally conserved transcription factors. RfaH is composed of an N-terminal NGN domain and a C-terminal KOW domain expected to fold into a beta-roll structure. Strikingly, RfaH’s KOW domain adopts an alpha-helical fold bound to the NGN domain in its apo form, but upon binding its target DNA sequence and RNA polymerase, RfaH KOW dissociates from the NGN domain and reversibly switches to the expected beta-roll topology. Previous biophysical measurements indicate that RfaH’s KOW domain is marginally stable and interconverts with a sparsely populated unfolded state with alpha-helical propensity. Although these factors may poise RfaH’s KOW domain to switch folds, the transition between the two distinctly folded states has not been observed. One possible explanation is RfaH’s poor solubility, particularly of its NGN domain, which aggregates at concentrations above 2 uM and hampers biophysical characterization. To circumvent this problem, we used the deep learning tool ProteinMPNN, to design an RfaH-like protein sequence with a soluble NGN domain. We then fine-tuned this sequence to switch folds. In this work, we use circular dichroism and nuclear magnetic resonance to characterize the unique folds assumed by the designed KOW domain. This structural characterization paves the way to biophysically characterize the fold-switching behavior of an RfaH-like protein.