New Zinc-Finger Technology Makes For Faster, Better Genetic Editing
By Deborah Borfitz
February 14, 2023 | CRISPR-based approaches that rely on the enzyme Cas9 to cut DNA at precise locations has been the darling of genetics research over the past few years. The revolutionary gene-editing system is great at suppressing or knocking out unwanted genes, but it can also produce a lot of off-target activity. And when delivered via adeno-associated virus (AAV) vectors, Cas9 occupies so much space that little room is left for any other therapeutic cargo that might be needed.
So says Marcus Noyes, Ph.D., assistant professor in the Institute for Systems Genetics at NYU Langone Medical Center, who has spent most of his career designing zinc fingers, an abundant if complicated group of proteins that dominate DNA binding in the real world. In the 1990s, zinc fingers were receiving as much press as CRISPR is today, but their potential wasn’t realized until recently because engineering them to bind to new sequences was astronomically challenging.
Zinc fingers could potentially have a greater impact on medicine than CRISPR-Cas9 because they’re smaller and comprised of human rather than bacterial proteins, Noyes says. Importantly, they could be used to deliver multiple artificial transcription factors simultaneously to treat complex diseases—including central nervous system conditions, heart disease, diabetes, and obesity—where most of the associated mutations are in the non-coding sequence of the genome controlling gene expression.
New technology, ZFDesign, has overcome the difficulty of designing zinc fingers for specific tasks using a transformer model of deep learning that interprets the meaning of a protein sequence in the context of what neighboring zinc fingers are doing, explains Noyes. It can quickly generate zinc fingers that manipulate transcription factors in the most natural possible way, by “replacing zinc fingers with zinc fingers within a natural protein coding sequence.”
This technology is being spun out into a company called TBG Therapeutics and Noyes is a cofounder, along with University of Toronto molecular geneticists Philip Kim, Ph.D., and Mikko Taipale, Ph.D. A report on the tool recently published in Nature Biotechnology (DOI: 10.1038/s41587-022-01624-4).
For the paper, Noyes and his team used a customized zinc finger to disrupt the coding sequence of a gene in human cells. They also built several zinc fingers that successfully reprogrammed transcription factors to bind near a target gene sequence and turn up or down its expression, demonstrating that ZFDesign can be used for epigenetic changes.
Most of what scientists are now doing with CRISPR, including both epigenetic and genomic editing, were originally done with zinc fingers, says Noyes. But only four or five labs around the world ever had the expertise to design them at all—let alone in large, library-scale quantities.
Noyes came from one of those labs, having worked alongside Scot Wolfe at the University of Massachusetts Medical School where they published one of the very first papers that demonstrated how nucleases, such as zinc finger nucleases or CRISPR, could be used to create model organisms in the lab (Nature Biotechnology, DOI: 10.1038/nbt1398). Relative to making zinc fingers, CRISPR is comparatively easy since it just involves designing a piece of RNA that is complementary to the sequence of the target DNA. “Zinc fingers are a little more complicated in that you have to consider how they interact with the DNA... [and] with the protein sequence that surrounds them.”
Each zinc finger is a tiny protein domain that binds to three bases of DNA that gets linked into a zinc finger array to extend the target sequence, he explains. The main hurdle for many years was understanding which zinc fingers in their array would “work nicely with their neighbors” so as not to inhibit the intended DNA binding.
The transformer model of artificial intelligence (AI), used primarily in the fields of natural language processing and computer vision, solved for this problem, continues Noyes. Just as a word in English can mean many different things, depending on other words in the same sentence, zinc fingers work in the context of adjacent protein sequences.
Before AI, it was necessary to build a library of zinc fingers for every single DNA target binding site and their surrounding protein sequences, continues Noyes. The process of library building and selection of zinc fingers that would function on their target would take many months.
“The scale is really important,” he says, drawing a comparison to looking under the lamppost for a set of keys on the other side of a dark parking lot. In the early days, scientists were limited to a small set of zinc fingers “as the assays used to uncover them were biased toward high-affinity interactions” that led to a limited understanding of zinc finger function, restricted the ability to engineer the proteins to bind to new sequences, and produced a lot of unintended side effects.
Over the past decade, starting with his lab at Princeton University and now at NYU, all the zinc finger selection work has been done, says Noyes. Those libraries represent nearly 50 billion possible zinc finger-DNA interactions, data feeding an AI model that knows how to make DNA bind to the desired sequence and compatibly within a zinc finger array.
High-throughput sequencing technology wasn’t even invented until 2000, points out Noyes. Before then, “we were basing our understanding of zinc fingers on 10 or 20 pieces of data because there just wasn’t the ability to sequence a million pieces of DNA at the same time.” As a result, the scale of the zinc finger screens carried out in the latest (Ichikawa et al.) study enabled investigation of the complexity required to understand zinc fingers, but also at a time when the sequencing technologies allowed researchers to capture and understand that data.
ZFDesign promises to shorten the front end of the lengthy development process for gene therapies by allowing scientists to ask more comprehensive questions, says Noyes. In the past, they would start by targeting a sequence known to regulate a gene and then spend six to 18 months evolving a bunch of zinc fingers that will bind to that sequence. “Now, we can say we know this one sequence works, but we can cover a 500-base region.... We are designing zinc fingers that cover every single position over those 500 bases and we can design them instantaneously and can have them tested within a matter of weeks.”
Anything that can be done with CRISPR-Cas9 are potential applications for zinc fingers, which have the advantage of being the most common way that nature has selected to bind to DNA and thus provide the potential for a much safter therapeutic option, says Noyes. Cas9 being a bacterial protein, the body will have an immune response for applications that require long-term expression.
“Interestingly enough,” he adds, “half of the transcription factors in humans are using zinc fingers to determine which DNA sequence they engage with, and this is true for most organisms.” Cas9 is a “very specialized way of binding DNA when we think of... how most organisms turn on and off genes.”
By speeding up the creation of zinc fingers, coupled with their smaller size, the ZFDesign system will enable use of these proteins to control multiple genes at the same time and thus potentially correct a large number of human diseases, says Noyes. Currently, the most productive and regulator-approved way of delivering a gene therapy to patients is through AVV vectors whose cargo space is limited to about 4,500 bases of DNA. Cas9 itself is 4,200 bases in size, versus about 1,000 bases for zinc finger regulators, meaning three or four artificial transcription factors could be efficiently packed into a single viral vector.
The safety concerns with both Cas9 and zinc fingers are off-target activity whereby they bind to the intended sequence but also to many other unintended ones, says Noyes. The Nature Biotechnology paper discusses a few ways of optimizing the specificity of zinc fingers, such as minimizing binding to G-rich DNA targets as these interactions tend to be the most promiscuous, and manipulating the contacts between zinc fingers and the phosphate backbone that can provide binding-affinity at the wrong targets.
In terms of optimization, he adds, “we’re at the 2012 stage [when Cas9 was invented] with zinc fingers.” The first versions of Cas9 did not work well in mammals and had a lot of off-target activity. “Now that we have a zinc finger model, we can start to probe for the influences that will improve specificity.”
Noyes and Kim are now working on optimizing the specificity of ZFDesign. The updated model will be the subject of their next published paper that could be out within a year.
In the interim, the research team is discussing partnership deals with various pharmaceutical companies where ZFDesign would be used to repress proteins being overexpressed in the central nervous system in neurodegenerative diseases such as ALS, Alzheimer’s, and Parkinson’s, he reports. The technology will soon be demonstrated in animal models to see if it improves neurodegeneration via repression of TDP-43, he offers as an example.
Other strong candidates for zinc fingers are any of the roughly 600 so-called “haploinsufficiency” diseases caused by deletion or inactivation of a copied gene, he adds. Increased expression of the healthy copy could be an approach to address any haploinsufficiency. Among these are epilepsy and Dravet syndrome, where upregulating the expression of SCN1A could lead to a cure.