AI tools speed development of antibody probes to see activity inside living cells  

Researchers at Colorado State University have determined how to use artificial intelligence to modify antibodies so they act as lightbulbs, enabling scientists to better see inside living cells to track errors in gene expression that can lead to cancer and other disorders.  

The findings, published in Science Advances, outline an approach that is significantly faster than existing manual testing and development methods to address an ongoing challenge to see activity in tiny cells continuously and clearly. That has limited basic understanding of a variety of cellular functions and reactions related to DNA transcription and protein creation. Those processes – known as gene expression – are key to many outstanding questions about human health and disease.  

A major reason for this limitation is that DNA inside cells is tightly wrapped around proteins called histones, which carry small chemical tags that help determine whether and when genes are turned on or off. These histone modifications act like tiny traffic stoplights for genes and can constantly change over time. 

Researchers have tried to use antibodies to bind to targets inside cells to better track those changes and the conditions that bring them about. However, antibodies evolved to be found in the bloodstream and often fall apart when inserted in the difficult conditions found inside living cells. It can also take years of manual screening to identify potential antibody proteins that could work for that purpose. The CSU team overcame that barrier by using AI to rapidly re-design potentially useful antibodies into intrabodies. Intrabodies are engineered antibody fragments that are specifically designed to fold correctly, stay soluble, and remain stable inside a cell or the bloodstream. When tagged with special fluorescent markers, they help visualize specific histone modifications and their chemical relationships in real time.  

Associate Professor Tim Stasevich, in CSU’s Department of Biochemistry and Molecular Biology, is the primary author on the interdisciplinary paper. He said you can think of these intrabody probes as lightbulbs that help to better see the activity that gives cells their identity and informs their function.  

“These probes could eventually inform cancer research where the wrong genes are activated – the tumor suppressor genes are off in a cell when they should be on – for example,” he said. “With these probes we can see if it is a small subset of cells that are malfunctioning or if it is every cell in a larger pattern. This work develops a new tool that allows us to see interactions and cell activation in a way that was not previously possible.”  

He added that the probes allow researchers to see movies of these tiny processes rather than a set of still images, which may miss key moments.  

AI-assisted protein design to convert antibody sequences to intrabodies  

This research is part of a growing movement to apply AI techniques to better understand how living cells operate. 

For this project, the CSU team used Google DeepMind’s AlphaFold2 and ProteinMPNN software tools to quickly identify and then re-design potentially useful antibody protein sequences – provided through a partnership with researchers at the Institute of Science, Tokyo – into stable intrabodies for use. 

The headstart from the AI programs identifying targets that were “well-behaved” for this use meant the team only needed to test around five potential designs to get one that worked as a probe. Previously, a Ph.D. student may have had to work for years on a single option with no guarantee of success. 

The team’s approach builds on existing work to accelerate protein engineering with AI by Professor Chris Snow in the School of Biomedical and Chemical Engineering, who also served as an author on this paper. 

In all, the team was able to create 19 new antibody-based probes that still function at high temperatures and are easier to generate for testing. Notably, 18 of these 19 sequences had previously failed before being refined by the AI-driven process, demonstrating how effective the approach is at developing leads. The team also noted that their failures with the pipeline were just as informative as their successes because they emphasized which patterns and substitutions in the protein structures proved effective and worthy of further pursuit. 

First author and Ph.D. student Gabriel Galindo said the team found an overall success rate of about 70% when converting conventional antibodies into intrabodies during the project.  

“That is obviously much better than the 5% to 10% success rate we had observed before and is very encouraging. I look forward to seeing how other newly developed AI tools such as AbMPNN, IgMPNN and AntiFold could further improve antibody structure prediction and design by incorporating antibody-specific data,” he said. 

Stasevich said one ongoing goal is to help develop a large database that can be used as a training set for further refining the design algorithms used by the AI software to identify sequences of interest. 

“The potential for this work is staggering because there are already more than 2,000 solved antibody structures and 147,000 sequences available publicly – each of which could potentially be converted into a useful intrabody with our methods,” Stasevich said. “And that number will continue to grow over time.” 

Gretchen Fixen is an undergraduate in the Department of Biochemistry and Molecular Biology and an author on the study within Stasevich’s lab. She helped design several potential intrabodies and validate their functionality and efficiency under the mentorship of Galindo. 

She said the project improved her academic performance by helping to connect material in the classroom to actual work in the lab. 

“The most important lesson I took from this publication was learning how to integrate diverse scientific concepts and approaches to gain deeper biological insight and push the boundaries of what is technically possible,” said Fixen, who is now a graduate student at the University of Heidelberg. 

Future work to create intrabodies against viruses like West Nile  

Professor Brian Geiss is a co-author on the findings and sees a lot of potential in using this approach to address similar problems in the field of virology. His team in the Department of Microbiology, Immunology, and Pathology studies how viruses like West Nile replicate and affect the cells they infect. 

By re-engineering virus-specific antibodies as probes, he hopes to track groups of natural viral proteins over time to understand infection in unprecedented detail. 

“Traditional fixed-cell microscopy using virus-specific antibodies has long been used to visualize viral proteins, but that approach only takes snapshots in time and doesn't show how individual proteins move over the course of an infection,” said Geiss. “This approach will hopefully help us understand how these proteins influence replication and alter the cellular environment during infection from the very start all the way through.”  

He added that the team is now in an excellent position to monitor West Nile virus assembly and maturation in real-time and “see” virus particles being formed. 

Stasevich said the durability of the new probes presented other opportunities as well.  

“Their toughness could make them great for diagnostic work, for example. And because they remain stable even at high temperatures, they could be easily transported across a variety of conditions,” he said. “Our approach is showing a lot of potential avenues, which will be exciting to pursue going forward.” 

---

---