A master of protein origami

By the time his alarm buzzed at 6 a.m., Stephen Fried's inbox was already bursting with messages. Headlines from around the world proclaimed that an artificial intelligence system called AlphaFold had finally managed to predict the complex, multidimensional curves and folds of proteins from just a sequence of amino acid building blocks.

"DeepMind AI cracks 50-year-old problem of protein folding," proclaimed UK's The Guardian. The headline from the normally staid Science read, "'The game has changed.' AI triumphs at solving protein structures."

For those outside the chemistry cognoscenti, the announcement might have seemed little more than researchers patting each other on the back. But the question of protein folding had plagued scientists for decades, and its solution was analogous to the completion of the human genome sequence for its ability to revolutionize medicine and biotechnology. With AlphaFold, researchers could accelerate drug discovery, answer basic questions about biology and evolution, and even design new functional proteins from scratch.

For Fried, an assistant professor of chemistry who has devoted his career to dissecting the intricacies of protein folding, the November 2020 news was both exciting and somewhat disheartening.

"I was shocked," Fried says. "There was the shock, and then the fear."

In human biology, a protein is a complex molecule made up of amino acids. Proteins perform a wide array of crucial functions in the body, acting as building blocks, structural components, catalysts, and messengers. They are essential for tissue repair, immune function, regulating metabolic processes, and maintaining fluid balance like a traffic cop for the circulatory system.

"You can take any piece of living matter, mash it up, and get proteins out. It's really liberating to work on proteins because it's the common denominator of life," he says.

Understanding how individual amino acids link together and knot into their final three-dimensional state has long been a quest of biochemists. Part of the challenge was the need to scratch an intellectual itch. In theory, combining the known sequence of amino acids with a hefty dash of physics was adequate to predict any one protein shape. The sheer complexity of the calculations, however, meant that no single scientist was up to the task. Only with the second iteration of a machine learning algorithm developed by Google DeepMind and Isomorphic Labs did researchers finally have a system with the muscle to make rapid, accurate predictions on protein folding. But the main reason that researchers like Fried were obsessed with decoding protein origami was simply owing to the centrality of proteins in the evolution of life, the development of disease, and the creation of new therapeutics.

If the protein folding problem were well and truly solved, Fried would need a new calling. But the more he scrolled the seemingly endless string of news articles, the more the chemist began to realize that AlphaFold could only predict the correct folding of proteins. The program told him nothing about what happens when proteins misfold. Over the past several decades, scientists have connected a range of neurodegenerative conditions to protein misfolding, including Huntington's, Parkinson's, and Alzheimer's diseases, as well as amyotrophic lateral sclerosis (ALS) and frontotemporal dementia. Developing therapies for these brain diseases would require an intimate knowledge of why certain proteins misfold and how cells deal with the problem. Fried's own work had zeroed in on how. the body's blindness to misfolded proteins sits at the heart of the aging process and the development of these diseases.

"Stephen is looking at the whole problem of aging from a completely different angle," says Bindu Paul, an associate professor in the Department of Neuroscience at the Johns Hopkins School of Medicine who is collaborating with Fried on neurodegeneration and aging. "We have been finding a lot of new stuff with his angle."

Rather than serving as a professional death knell, the arrival of AlphaFold would help propel Fried's work.

Fried didn't want to focus on a single protein and how its peculiar creases and blobs congealed out of the cellular soup. Instead, he has spent the last decade working to build a new field of science called structural proteomics. His goal is to investigate three-dimensional protein structures on a massive scale to identify common mistakes they make when folding. Fried also has begun to investigate how aging impacts the cell's ability to detect these errors and chop the protein up for recycling. This work requires him to look at thousands of proteins and millions of conformations, a diversity he says is crucial to understanding the rules of our protein playbook.

Recent improvements in technology mean that Fried can interrogate thousands of proteins at once, giving him a clearer picture about how life works.

The work can be perplexing, even tedious, but Fried's colleagues say that there's no one better suited to the task.

"Stephen wants to understand the basic science," Paul says. "He wants to understand correlations and connect the dots."

No sooner had Rosalind Franklin, James Watson, and Francis Crick solved the structure of DNA than many biochemists began turning their attention to proteins. Unlike DNA, which spiraled neatly into a double helix, proteins had myriad shapes. DNA's consistency makes it ideal for its function as information storage. Countless protein structures have evolved across Earth's lifeforms over billions of years until these collections of carbon, hydrogen, oxygen, and nitrogen took on a unique, three-dimensional shape. The question in the halcyon post–Watson and Crick days was whether a mysterious factor controlled a protein's form.

While still stationed at the National Institutes of Health in Bethesda, Maryland, biochemist Christian Anfinsen began a series of experiments in 1961 that eventually earned him the Nobel Prize in Chemistry in 1972 (Anfinsen would join the Johns Hopkins faculty in 1982). After British scientists determined the structure of myoglobin, a protein that stores oxygen in muscle, Anfinsen wanted to know how such proteins took on their final, complex shape. He selected a small, simple protein from the bacterium Staphylococcus aureus known as a ribonuclease, which chops up RNA into smaller pieces. When he isolated the protein and measured its activity in a test tube, the ribonuclease minced RNA as expected. So far, so good. Then Anfinsen treated the protein with a chemical called a denaturant, a certain type of compound that is good at breaking up some of the bonds that stabilize a protein's shape—while not affecting the stronger bonds between amino acids that link these building blocks into a complete protein. The protein's sequence remains identical; only its shape changes. Instead of its usual kidney shape, ribonuclease becomes an amorphous blob of goo, despite containing the same amino acids in the same order.

When Anfinsen measured how well the denatured ribonuclease could break down RNA, he found nothing happened. The chemical reaction simply didn't happen. But when Anfinsen removed the denaturant and repeated the experiment a third time, he found the ribonuclease had refolded itself into its beanlike shape and was once again able to chop up RNA. From this work, Anfinsen deduced one of the foundations of molecular biology: A protein's amino acid sequence determines its shape, and a protein's shape determines its function. In other words, if you know the precise order of amino acids in a protein, you can guess its shape.

As an undergraduate student, Fried learned these facts in his lectures at MIT, but they didn't grab his attention. The son of a physician in Kansas, Fried loved science but vowed to stay away from anything that veered too close to cells or DNA or anything else that smacked of medicine. He eschewed organic and biochemistry for physical chemistry. After graduation, he moved cross-country to study fuel cells and alternative energy at Stanford. But as Fried dived into the topic, he found he didn't have the passion for the work that he anticipated. So, he spoke with all the PIs in the Chemistry Department to try to find a new doctoral home. One researcher asked Fried if he had considered biophysics.

"I always thought of biology and biochemistry as the squishier side of science. And there's this whole field of biophysics that is all about trying to understand how biology works on a molecular, fundamental level. It's not giving life a pass and saying evolution can do anything it pleases," Fried says.

For a self-described big picture thinker and idea man, biophysics was the perfect fit. Fried focused on the precise atomic mechanics by which enzyme catalysts could speed up chemical reactions. If the molecules involved in these chemical reactions were left to their own devices, it could take longer than the history of the universe to complete a single chemical reaction. Add in a catalyst, however, and the reaction could be completed, literally, in the blink of an eye. What no one had yet done was pull back the metaphoric curtain on this chemical magic trick. Fried's lack of background in biophysics allowed him to be more unprejudiced by accepted principles, he says, and his research quickly began challenging some of the field's top scientists.

As Fried began teaching biochemistry to undergrads during his PhD studies and as a new PI at Johns Hopkins in 2018, he turned this same skepticism to Anfinsen's experiments. More than a half century after Anfinsen initially studied ribonuclease in a test tube, researchers uncovered the important role played by chaperones, proteins whose sole function was to ensure that other cellular proteins folded properly. If proteins could readily refold after being denatured, as Anfinsen found, there wouldn't be a need for chaperones. Surprisingly, no one had repeated Anfinsen's experiments on a large scale to see whether ribonuclease's ability to refold on its own was relatively standard or whether Anfinsen had happened upon one of life's many exceptions. Fried's curiosity was piqued, and he launched in on the question.

"He thinks very deeply about putting everything together, and he also does really, really good experiments," says Vincent Hilser, the Ralph S. O'Connor Chair of the Biology Department.

"He has his own ideas about things, but he's not afraid to challenge them. It takes a rare sort of courage to destroy your own idea."

Repeating someone else's experiments might be an unusual way to start a lab, but Fried knew that asking these sorts of basic questions would also provide clues into life's fundamentals.

In the 1960s, sans AI, Anfinsen only had the capacity to study the activity of a single protein. By the late 2010s, Fried had a range of technologies at his fingertips that would let him replicate these groundbreaking experiments tens of thousands of times. Fried didn't need to determine each protein's precise three-dimensional structure. To answer this question, all Fried needed to do was compare a protein's original structure with its refolded one. This was still a huge task, but Fried simplified it by using a protein-cutting enzyme called a protease that he knew would make cuts only at specific points. Proteins that refolded correctly would return the same selection of bits and pieces after protease treatment. Those that didn't refold would yield a different array of building blocks. By making these comparisons across the entire proteome, Fried could test Anfinsen's hypothesis that the simple laws of physics were adequate to explain protein folding, and do so without inserting any of his own preconceived notions or hypotheses by studying as many examples as he could find.

"It's just really useful to sort of look at as many proteins in the cell as possible without prejudice because the truth is that every protein kind of has its own story to tell," he says.

Fried began with the 4,300 proteins synthesized by the bacterium Escherichia coli, a microbe commonly found in the human intestinal tract that is easy to grow in large numbers. His team split the cells open, causing the dead E. coli to spill its guts into the test tube. He collected all the proteins in the cell and then analyzed them using mass spectrometry, which separates molecules based on their size and charge. Like Anfinsen, Fried then unfolded the proteins with a chemical, which transformed the complex protein origami into amorphous, floppy strings of amino acids. Then he removed the denaturant and waited to see what the proteins would do, without any additional inputs. Would the proteins be like Humpty Dumpty after his fall and not be able to put themselves back together again? Or could the proteins refold? Fried's results were telling: 60% of the proteins in E. coli could refold without assistance, but 40% couldn't.

"It's the classic question of, Is the glass half empty or is the glass half full?" Fried says.

Anfinsen's deduction that proteins can readily refold was owing, in part, to his chance selection of a protein that did, in fact, return to its initial shape without assistance. Other proteins from bacteria would have given the exact opposite answer. Depending on which proteins you were studying, you could convince yourself that either option was correct.

"If you're the type of biophysicist who really likes small, simple proteins, you really could have gone about your whole life thinking that every protein can refold," he says. "The moment that you think you know everything, that's kind of when you dupe yourself, and you don't make as many interesting findings."

His results, presented at the [February 2020 meeting of the American Biophysical Society]) https://www.biophysics.org/2020meeting#/), marked the completion of the first stage of what Fried nicknamed the Anfinsen Challenge. Fried's excitement wasn't focused on one-upping a Nobel Prize winner but rather on showing that Anfinsen's work was incomplete. Proteins didn't all magically refold themselves. That wasn't how life worked. The project's success, however, left the freshly minted PI with more questions than answers. His original experiment had shown that about half of proteins couldn't refold without chaperones, but it didn't answer anything about how important those chaperones were and which ones were most important. After all, in real life, an unfolded protein would encounter a chaperone to help. So, he repeated his Humpty Dumpty experiment, only this time he added molecular chaperones from one of three different E. coli chaperone systems. Half the proteins that didn't naturally refold were able to assume the correct conformation by adding chaperones. That still meant that one in five unfolded proteins couldn't refold even with the appropriate chaperones. Those were the true cracked eggshell proteins that couldn't be put back together again. It was a major blow to one of biochemistry's oldest assumptions and given by an as-yet untenured professor.

Fried credits his discovery to the support of a special National Institutes of Health Director's New Innovator Award that provides no-strings attached funding to new professors. Rather than requiring large amounts of preliminary data and forcing scientists to take an incremental approach, New Innovator Awards fund high-risk, high-reward platforms such as Fried's. It was a gamble that paid off with dividends to spare.

Getting others in his field to recognize the significance of the findings was far more challenging. When Fried and his team submitted his results to the Proceedings of the National Academy of Sciences, one reviewer was especially skeptical, arguing that the research team must have done something wrong. The anonymous reviewer simply couldn't believe the validity of Fried's findings, that adding in molecular chaperones didn't correctly refold the entire range of proteins, as posited by a famous German biochemist in the 1990s. Only after months of back-and-forth was his article finally accepted and published in November 2022.

As Fried went over his results with a fine-toothed comb, he found that a small handful of E. coli proteins were able to fold correctly on their own but ended up misfolding in the presence of chaperones. The "helper" had inadvertently made the problem worse, something Fried finds analogous to when his own young children attempt to help him with household chores. They mean well, but even the most helpful toddler often makes more of a mess than they will ever be able to clean up.

Since a protein's structure controls its function, evolution should quickly weed out chaperone-enabled misfolding. But that's not what Fried and current JHU PhD student Divya Yadav found when they began investigating this phenomenon in 2018. Yadav and Fried have continued this line of inquiry to understand how frequently chaperones don't help as expected and how this impacts bacterial biology.

"It's new and novel and interesting, but at the same time, it's outside the traditional perspective scientists have for chaperones," Yadav says.

But no sooner had Fried begun to pry into the secrets of protein folding than the headlines announcing the success of AlphaFold began to appear. Overnight, the future of the Fried lab seemed bleak.

Tolstoy, meet proteins

As many protein scientists began uncorking bottles of Champagne now that researchers had created the first system that could accurately predict a protein's three-dimensional structure from the string of amino acids that built it, Fried began thinking about what AlphaFold couldn't tell scientists. He quickly realized that the AI system used to build AlphaFold was trained on the structures of correctly folded proteins archived in the Protein DataBank at the NIH. AlphaFold couldn't provide information on the structures that had improperly folded.

It was the Anna Karenina principle at work. (Stick with me.) Tolstoy opened his novel by positing: "Happy families are all alike; every unhappy family is unhappy in its own way." In other words, properly folded proteins all follow certain rules, but there are an infinite number of ways by which a protein can misfold. It's why neither Fried nor anyone else is liable to create an AlphaFold for Proteins Gone Wild. But that doesn't mean that Fried can't glean valuable information on the topic.

When PhD student Haley Tarbox joined the Fried lab in 2019, she expressed interest in understanding protein misfolding, especially in neurodegenerative disease.

Scientists have created animal models of these diseases in mice, fruit flies, yeast, and worms. But these models have limitations, including the fact that researchers have to genetically engineer most of these animals so that they develop the disease, often at a young age. Humans, however, spontaneously develop many of these neurodegenerative conditions as they age. Tarbox wanted to study a more humanlike model of disease and identify whether proteins developed folding problems during aging, or whether misfolding was confined to the proteins that had become hallmarks of neurodegenerative disease, such as amyloid, tau, and α-synuclein.

"What if amyloid beta is just the tip of the iceberg? What if there is a whole bunch of proteins that are misfolding, and we just don't know about them?" Fried says. "There's got to be some molecular driver for 'old.' What is that, and can it be slowed down?"

In recruiting other faculty members to be a part of her thesis committee, Tarbox came across work on aging and biochemistry in neurons done by Michela Gallagher, Krieger-Eisenhower Professor of Psychology and Neuroscience in the Department of Psychological and Brain Sciences. Not only did Gallagher agree to serve on the committee, she also mentioned that her lab at the School of Medicine had created a colony of genetically diverse, aged rats. Other than the beady eyes and long tails, this colony was about as close as you could get to a randomly selected group of older adults. Even more exciting to Tarbox was data from the Gallagher lab that showed some of these mice remained cognitively intact until they died, whereas others showed memory problems and other impairments. For Tarbox and Fried, this was the perfect setup to measure how protein folding impacted cognition during aging.

The pair focused on proteins in the hippocampus, a seahorse-shaped region at the center of the brain controlling memory, navigation, and emotion. Using the same mass spectrometry– based technique that Fried pioneered in E. coli, Tarbox looked at how protein folding changed in the brains of aged rats. The team identified several hundred candidates in cognitively impaired aged rats but not the healthy aged animals. When Gallagher, Fried, and Tarbox analyzed these proteins, they found a disproportionate number of them were those that couldn't refold after being denatured.

"The network of processes that keep our proteins in their correct shape is known to be disrupted in aging. Our data supports the idea that maybe, if some proteins misfold, they can't get back into their native state on their own," Tarbox says.

A follow-up study that performed a similar analysis in the yeast Saccharomyces cerevisiae also found a higher proportion of misfolded proteins in cells during their second half of life.

Fried emphasized that these proteins did not have major folding mistakes. The cell hadn't folded into an origami butterfly instead of a crane. Instead, the differences were far more subtle—perhaps the crane's beak was a hair too long or the wings a smidge too small. The cranes could also technically fly but perhaps not as quickly, or they might tire more rapidly. Any cell can make these errors, Fried says, but healthy cells have robust mechanisms to break down and recycle any faulty proteins, a process that declines during aging. It's a molecular problem that Fried thinks may have a solution. A class of drugs called molecular glues or PROTACs (proteolysis-targeting chimeras) can target specific proteins for recycling. Already in clinical trials for some types of cancer and neurological disorders, Fried believes that this class of drug may also hold promise for treating some of the molecular culprits of aging.

"This type of science happens because people are working together and putting different skills together," Fried says. "You have to be open-minded."

A biology-avoiding physicist who now makes his home in the Chemistry Department seems like an odd source for the Grand Unified Theory of Aging, but Fried's mentees aren't surprised that he's making such an impact.

"He's the smartest person I know," Yadav says. But what makes Fried so successful isn't just his raw intelligence; it's his willingness to collaborate and look at life from different viewpoints.

"Collaborations happen when you have the ability to talk the other person's language. Stephen has the initiative and the ability to do that," Gallagher says.

These collaborators mean that Fried's opinions don't stagnate and that he can perceive breakthroughs others may have missed. His commitment to diversity isn't just lip service, Hilser says.

"Just like people, proteins aren't all alike," Fried says. "There's really new stuff out there that is still going to be found."