Background

During the development of organisms, genes operate under precise temporal and spatial regulation. A key element responsible for this regulation is the enhancer, which exists in non-coding regions of the DNA sequence that do not encode proteins. These enhancers control gene expression in specific tissues, forming the foundation for each cell to perform unique functions. In the past, genomic research teams focused primarily on discovering enhancers that exist in nature. However, identifying enhancers that act only in specific tissues among numerous DNA sequence combinations is extremely challenging.

Recent advancements in artificial intelligence (AI) technology have spurred efforts to predict and redesign DNA sequences. Efforts to decode the grammar of gene regulatory sequences have so far been limited to model organisms such as fruit flies or cultured cell lines. The environment within mammalian organisms is far more complex and tightly controlled, making it nearly impossible to design enhancers that function in actual animal tissues. This is because the signaling systems and base composition within the fetal environment can interfere with the normal operation of artificial enhancers. As a result, the demand for custom-designed regulatory sequences that function accurately in vivo has been steadily growing.

Key Discovery

A joint research team led by Dr. Shendy Chen and Professor Alexander Stark at the Institute of Molecular Pathology (IMP) in Vienna developed DeepSTARR-Mouse, an AI system for designing artificial enhancers that specifically target tissues in mammals. This model predicts chromatin accessibility and enhancer activity by inputting DNA sequences of 1,001 base pairs (bp) in length. The research team first trained the model's foundational recognition ability using whole-genome chromatin accessibility data (ATAC-seq) from mouse embryo tissues. They then performed transfer learning to fine-tune tissue-specific activity using validated data from the VISTA Enhancer Browser. Based on these results, the team successfully derived 15 artificial enhancer sequences that selectively activate only in the central nervous system (CNS), heart, and limb regions of mouse embryos. To verify the efficacy of the designed sequences, the team conducted reporter gene analysis using transgenic mouse embryos at developmental day 11.5 (E11.5). The analysis revealed that all 15 designed sequences achieved 100% success in activating only in the target tissues. This marks the first case in which gene expression in mammalian embryos has been fully controlled using artificial DNA sequences.

Significance and Outlook

This research elevates genome engineering from the stage of replicating natural sequences to the level of directly designing sequences according to desired purposes. It is widely regarded as a breakthrough that significantly enhances the safety of gene therapy. Conventional gene therapy delivery vectors, such as adeno-associated virus (AAV) vectors, have had issues with unwanted expression in non-target organs, including the liver, when administered systemically, leading to toxicity. Using AI-designed enhancers can block expression outside target cells, thereby effectively reducing the risk of side effects.

However, there remain practical obstacles to overcome before clinical application. Since this validation was conducted at the embryonic developmental stage, it is still uncertain whether the same level of accuracy will be guaranteed in adult or diseased tissue environments. Additionally, a method to compress the relatively long 1,001 bp artificial enhancer sequences into virus vectors with limited carrying capacity is required. Ultimately, the technology must undergo precision validation in human cell lines and clinical studies to progress toward practical application.