ÌÇÐÄÊÓƵ

Understanding human biology requires more than mapping our genes—we must also understand how gene expression is regulated to guide healthy development, growth, and maintenance of our body systems over a lifetime.

Scientists at Lawrence Berkeley National Laboratory (Berkeley Lab) and Stanford University have revealed new insights into how regulatory sequences called enhancers drive gene expression during embryonic development. Enhancers are sections of DNA that orchestrate the expression of a gene despite being located far away from the actual coding sequence.

Their work, today in Nature, shows how multiple short, modular sequences within an enhancer are needed to properly guide expression, and that even a single nucleotide mutation in one of these regions can change how and where a gene is activated. The team also used their experimental results to develop and assess a machine learning model.

"These findings show that even the smallest changes can have huge impacts," said first author Michael Kosicki, a postdoctoral researcher at Berkeley Lab. "Our approach will give scientists a powerful tool to investigate normal gene regulation and unravel the increasingly strong link between disease and variations in the non-coding genome."

The Berkeley Lab team members used a mouse model to study seven human enhancers known to govern the development of the brain, heart, limbs, and face. They created a huge variety of different mutations in those enhancers, then looked for changes to developing tissues across the whole body. In one striking example, alterations to an enhancer associated with building structures in the face and limbs caused it to activate in the heart and nervous system tissues instead.

"Seeing that just a single base pair mutation can change where in the body an enhancer activates, and potentially change how the body develops, has profound implications for studying human disorders and designing gene therapies," said co-lead author Len Pennacchio, a senior scientist in Berkeley Lab's Environmental Genomics and Systems Biology (EGSB) division. "It also means scientists need to be cautious when designing tissue-specific gene therapies, to avoid unintended effects."

Investigating enhancers has always been challenging, because each of these sequences contains multiple binding sites for transcription factors, the molecules that switch DNA transcription on or off. The effects of mutations depend on the specific combination and location of sites that are altered, and can only be revealed through systematic experiments. This complexity, and the lack of sufficient data to train machine learning algorithms, makes it difficult to build accurate predictive models.

Using the large experimental dataset the Berkeley Lab team created, the Stanford collaborators developed a new model and tested whether it could identify the same important sequences revealed in the experiments.

"We wanted to explore how far AI can take us in understanding enhancer biology right now," said Axel Visel, senior scientist at Berkeley Lab and a co-corresponding author.

They found that although the model could identify many functionally important regions of enhancers by searching for sequence patterns known to indicate binding sites, it missed other sequences that are clearly critical based on the team's experimental evidence.

"Currently available models tell the truth, but not the whole truth," said Kosicki. "In other words, the predictions we have are typically correct, but they sometimes miss functional regions we identified experimentally. Identifying these modeling blind spots will help us improve them in the future."

For now, the findings serve as both a resource and a reminder: Even the best predictive models need grounding in experimental biology. As researchers continue to refine AI tools, studies like this one will be essential for revealing what those models get right and where they still fall short.