A universal deep-learning model for zinc finger design enables transcription factor reprogramming

Abstract

Cys2His2 zinc finger (ZF) domains engineered to bind specific target sequences in the genome provide an effective strategy for programmable regulation of gene expression, with many potential therapeutic applications. However, the structurally intricate engagement of ZF domains with DNA has made their design challenging. Here we describe the screening of 49 billion protein–DNA interactions and the development of a deep-learning model, ZFDesign, that solves ZF design for any genomic target. ZFDesign is a modern machine learning method that models global and target-specific differences induced by a range of library environments and specifically takes into account compatibility of neighboring fingers using a novel hierarchical transformer architecture. We demonstrate the versatility of designed ZFs as nucleases as well as activators and repressors by seamless reprogramming of human transcription factors. These factors could be used to upregulate an allele of haploinsufficiency, downregulate a gain-of-function mutation or test the consequence of regulation of a single gene as opposed to the many genes that a transcription factor would normally influence.

Engineered transcription-associated Cas9 targeting in eukaryotic cells

“…It was furthermore shown, in the same work 39 , that ZF-DNA interactions can compensate for suboptimal activity on NGG PAMs when Cas9’s PAM-interacting domain 40 (PID) was genetically attenuated with a single substitution known as ‘MT3’ (Cas9 MT3 ). Although ZFs can now be designed to recognize nearly any DNA sequence 41 , they do not confer temporal control over Cas9’s activity. We therefore initially tested a single tetO2 DNA element, the target substrate for wild-type (Wt) TetR and reverse TetR (RevTetR) DNA binding proteins that can be dynamically induced with doxycycline (Dox) to dissociate from or bind tetO2 , respectively 42 .…”

High-throughput thermodynamic and kinetic measurements of transcription factor/DNA mutations reveal how conformational heterogeneity can shape motif selectivity

“…Obtaining these measurements will be essential for improved design of selective binders. While algorithms to design synthetic TF-like binders with user-specified sequence specificity 9,10 are increasingly successful, attempts to improve selectivity – such as by mutating contacts involved in “non-specific” contacts like charged interactions with the phosphate backbone – yield scaffold-dependent success 9 ( Fig 4G, 6A ). Our work suggests that prediction and design of selective binders (beyond TF-DNA interactions) will necessitate consideration of energy landscapes that govern both folding and specific recognition.…”

“…Many prior efforts have uncovered the “sequence specificity” (the motif bound with the highest affinity) 1,2 of thousands of TFs across multiple organisms, with preferences often represented as mononucleotide or dinucleotide models 3–8 . This information, combined with structural models, has fueled progress in designing 9,10 and predicting 11–15 sequence-specific nucleic acid binders. However, while motif representations effectively represent “sequence specificity”, they often fail to capture other important features of TF-DNA binding, including absolute binding affinities and preferences for sequences that fall outside of the highest affinity set 16 .…”

Synthetic Gene Circuits for Regulation of Next‐Generation Cell‐Based Therapeutics

“…Computational approaches can be used to simplify the design process. [205] Artificial TALEs are customized DNA-binding domains, consisting of a series of tandem repeats, typically 33-35 amino acids long, where each repeat targets a single DNA base pair. [206] The specificity of the DNA binding is determined by two variable amino acid residues within each repeat (typically at positions 12 and 13), known as the repeat variable diresidues (RVDs).…”