Researchers have harnessed artificial intelligence to create thousands of innovative DNA switches that can accurately manage the expression of genes across different cell types. This groundbreaking method opens new avenues for controlling the timing and location of gene expression in the body, offering exciting possibilities for human health and medical research that were previously unattainable.
At The Jackson Laboratory (JAX), the Broad Institute of MIT and Harvard, and Yale University, a team of researchers has utilized artificial intelligence to innovate thousands of precise DNA switches that govern gene expression in various cell types. This pioneering method paves the way for unprecedented control over when and where genes are expressed within the body, presenting significant opportunities for advancements in human health and medical science.
According to Ryan Tewhey, PhD, an associate professor at JAX and co-senior author of the study, “These specially designed synthetic elements demonstrate remarkable specificity for the targeted cell type they were intended for. This allows us to fine-tune gene expression in a particular tissue without influencing the rest of the organism.”
In recent years, advancements in genetic editing techniques and various gene therapy methods have empowered scientists to modify genes within living cells. However, selectively targeting genes in specific cell types or tissues, instead of affecting the entire organism, has proven to be quite challenging. A significant reason for this difficulty lies in the ongoing struggle to decode the DNA switches known as cis-regulatory elements (CREs), which are responsible for managing the activation and suppression of genes.
In a recent paper published on October 23 in the advanced online issue of Nature, Tewhey and his team crafted novel synthetic CREs and effectively utilized them to activate genes in brain, liver, or blood cells, all while avoiding gene activation in other types of cells.
Tissue- and time-specific instructions
While every cell in a living organism contains the same set of genes, not every gene is required in all cells, or at every moment. CREs play a crucial role in ensuring that brain-specific genes, for example, are not utilized by skin cells, or that developmental genes remain inactive in adult organisms. CREs are not part of the genes themselves but are independent regulatory DNA sequences, often located in close proximity to the genes they govern.
Researchers are aware that the human genome contains thousands of distinct CREs, each fulfilling slightly different functions. However, understanding the complex grammar of CREs has been a challenge, with “no straightforward rules governing the functionality of each CRE,” as noted by Rodrigo Castro, PhD, a computational scientist in Tewhey’s lab and co-first author of the study. “This has limited our ability to create gene therapies that effectively target specific cell types within the human body.”
“The main question we aimed to answer in this project was: ‘Can we learn to decipher and construct the code of these regulatory elements?'” explained Steven Reilly, PhD, an assistant professor of genetics at Yale and a senior author on the study. “We recognize that the grammar and syntax of these elements are still poorly understood. Therefore, we set out to build machine learning methods capable of interpreting a more complex code than what we could achieve individually.”
Utilizing advanced artificial intelligence techniques known as deep learning, the research team trained a model using extensive datasets of DNA sequences from the human genome. They analyzed the activity of these sequences in three cell types: blood, liver, and brain. This AI model empowered the researchers to predict the activity of any potential sequence from the limitless combinations possible. By examining these predictions, they unearthed new patterns within the DNA, gaining insights into how the grammar of CRE sequences influences the amount of RNA produced—an indicator of gene activation.
The team, including Pardis Sabeti, MD, DPhil, a senior author and core member at the Broad Institute and Harvard professor, subsequently developed a tool named CODA (Computational Optimization of DNA Activity). This platform leveraged their AI model to efficiently generate thousands of novel CREs with specific traits, such as activating a designated gene in human liver cells but not in blood or brain cells. Through a cycle of experimental and computational analyses, the researchers enhanced the program’s predictive capabilities regarding the biological effects of each CRE, allowing the creation of specific CREs previously unseen in nature.
“While natural CREs are abundant, they represent just a small fraction of possible genetic elements and are limited in function by evolutionary selection,” stated Sager Gosai, PhD, a postdoctoral fellow in Sabeti’s lab and co-first author of the study. “These AI-driven tools hold incredible potential for designing genetic switches that finely tune gene expression for innovative applications, including biomanufacturing and therapy, beyond the constraints of evolutionary pressures.”
Pick-and-choose your organ
Tewhey and his colleagues evaluated the newly designed synthetic CREs by incorporating them into cells and measuring their effectiveness in activating genes within the intended cell types, as well as their success in preventing gene expression in alternative cell types. They found that the synthetic CREs exhibited even greater specificity to cell types compared to naturally occurring CREs associated with those cell types.
Gosai remarked, “The synthetic CREs diverged so significantly from natural elements that initial predictions about their effectiveness seemed unrealistic. We anticipated that many of the sequences would not behave appropriately within living cells.”
Castro added, “We were thrilled to discover just how successful CODA was at generating these elements.”
The team investigated why the synthetic CREs surpassed their natural counterparts and found that the cell-specific synthetic CREs contained a mix of sequences that enabled gene expression in the intended cell types while simultaneously suppressing it in other types.
Additionally, the researchers tested several synthetic CRE sequences in zebrafish and mice, yielding promising results. For example, one CRE was able to activate a fluorescent protein specifically in the developing livers of zebrafish, without affecting other regions of the fish.
“This technology paves the way for creating new regulatory elements with predefined roles,” stated Tewhey. “Such tools will not only benefit fundamental research but also hold significant biomedical potential to manipulate gene expression in targeted cell types for therapeutic applications.”
