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

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 dual selection for directed evolution of SpCas9 PAM specificity.

The widely used Streptococcus pyogenes Cas9 (SpCas9) nuclease derives its DNA targeting specificity from protein-DNA contacts with protospacer adjacent motif (PAM) sequences, in addition to base-pairing interactions between its guide RNA and target DNA. Previous reports have established that the PAM specificity of SpCas9 can be altered via positive selection procedures for directed evolution or other protein engineering strategies. Here we exploit in vivo directed evolution systems that incorporate simultaneous positive and negative selection to evolve SpCas9 variants with commensurate or improved activity on NAG PAMs relative to wild type and reduced activity on NGG PAMs, particularly YGG PAMs. We also show that the PAM preferences of available evolutionary intermediates effectively determine whether similar counterselection PAMs elicit different selection stringencies, and demonstrate that negative selection can be specifically increased in a yeast selection system through the fusion of compensatory zinc fingers to SpCas9.

The geometric influence on the Cys2His2 zinc finger domain and functional plasticity.

The Cys2His2 zinc finger is the most common DNA-binding domain expanding in metazoans since the fungi human split. A proposed catalyst for this expansion is an arms race to silence transposable elements yet it remains poorly understood how this domain is able to evolve the required specificities. Likewise, models of its DNA binding specificity remain error prone due to a lack of understanding of how adjacent fingers influence each other's binding specificity. Here, we use a synthetic approach to exhaustively investigate binding geometry, one of the dominant influences on adjacent finger function. By screening over 28 billion protein-DNA interactions in various geometric contexts we find the plasticity of the most common natural geometry enables more functional amino acid combinations across all targets. Further, residues that define this geometry are enriched in genomes where zinc fingers are prevalent and specificity transitions would be limited in alternative geometries. Finally, these results demonstrate an exhaustive synthetic screen can produce an accurate model of domain function while providing mechanistic insight that may have assisted in the domains expansion.

GLUT12 promotes prostate cancer cell growth and is regulated by androgens and CaMKK2 signaling.

Despite altered metabolism being an accepted hallmark of cancer, it is still not completely understood which signaling pathways regulate these processes. Given the central role of androgen receptor (AR) signaling in prostate cancer, we hypothesized that AR could promote prostate cancer cell growth in part through increasing glucose uptake via the expression of distinct glucose transporters. Here, we determined that AR directly increased the expression of SLC2A12, the gene that encodes the glucose transporter GLUT12. In support of these findings, gene signatures of AR activity correlated with SLC2A12 expression in multiple clinical cohorts. Functionally, GLUT12 was required for maximal androgen-mediated glucose uptake and cell growth in LNCaP and VCaP cells. Knockdown of GLUT12 also decreased the growth of C4-2, 22Rv1 and AR-negative PC-3 cells. This latter observation corresponded with a significant reduction in glucose uptake, indicating that additional signaling mechanisms could augment GLUT12 function in an AR-independent manner. Interestingly, GLUT12 trafficking to the plasma membrane was modulated by calcium/calmodulin-dependent protein kinase kinase 2 (CaMKK2)-5'-AMP-activated protein kinase (AMPK) signaling, a pathway we previously demonstrated to be a downstream effector of AR. Inhibition of CaMKK2-AMPK signaling decreased GLUT12 translocation to the plasma membrane by inhibiting the phosphorylation of TBC1D4, a known regulator of glucose transport. Further, AR increased TBC1D4 expression. Correspondingly, expression of TBC1D4 correlated with AR activity in prostate cancer patient samples. Taken together, these data demonstrate that prostate cancer cells can increase the functional levels of GLUT12 through multiple mechanisms to promote glucose uptake and subsequent cell growth.

Deep mutational scanning of S. pyogenes Cas9 reveals important functional domains.

RNA-guided endonucleases (RGENs) have invigorated the field of site-specific nucleases. The success of Streptococcus pyogenes Cas9 (SpCas9) has led to the discovery of several other CRISPR-associated RGENs. As more RGENs become available, it will be necessary to refine their activity before they can be translated into the clinic. With this in mind, we sought to demonstrate how deep mutational scanning (DMS) could provide details about important functional regions in SpCas9 and speed engineering efforts. Consequently, we developed a nuclease screening platform which could distinguish active Cas9 mutants. We screened a library of 1.9 × 107 with over 8500 possible non-synonymous mutations and inferred the effects of each mutation using DMS. We demonstrate that the RuvC and HNH domains are the least tolerant regions to mutation. In contrast, the Rec2 and PI domains tolerate mutation better than other regions. The mutation information defined in this work provides a foundation for further SpCas9 engineering. Together, our results demonstrate how DMS can be a powerful tool to uncover features important to RGEN function. Application of this approach to emerging RGENs should enhance their engineering and optimization for therapeutic and other applications.