Textbook illustrations don’t do the human cell any justice. Maps and diagrams can give you a sense of what a cell is and how it works. But to truly understand the basic unit of life, you need to grasp how wildly alive it is.
A cell is hot and never still. Sloshing layers of external fat are anchored to a stiff scaffold of filaments that extend and retract as they renew themselves, a network of twinkling spikes. Inside, vibrating water excites a thick traffic of molecular machines — millions of them, in dizzying variety. Some spin like rotors, some squeeze like fists, and some move around on assemblages that look like running feet. A chaos of raw materials circulates, pelting larger structures in a constant hailstorm of energy. There’s barely any room to move. Everything bumps into everything.
The body is a crowded scene, and within it, no molecular machine makes a greater impression than proteins, those folded chains of amino acids that evolved to bump into things in remarkably useful ways.
The usefulness of proteins far exceeds the muscle building for which they are best known. Cells manufacture proteins in highly specific three-dimensional structures to accomplish the tens of thousands of biological tasks programmed into our DNA. Think: Y-shaped disease hunters, tightly packed spirals of oxygen transporters, horseshoe communication systems, and hollow iron-storing spheres.
Image
These jellylike microscopic tools bounce around the body’s teeming landscapes — inside and outside the cell, depending on type — where their distinctive shapes provoke varied reactions. Crash into another molecule that doesn’t match their shape? No reaction. Knock against a partial match? Small reaction. Encounter a series of complementary pockets? Something happens.
Immunoglobulins claw onto viruses to stop them from invading. Insulin binds to cell receptor slots to regulate blood sugar. Amylase, protease, and lipase break down chemical bonds in food with atomic precision. With ornate formulas and shapes, all these proteins rely on both chemistry and physics to execute our DNA’s meticulous instructions.
For tens of thousands of distinctive functions, we have a mere 20 amino acids to thank. Alone, an amino acid can’t do much. But when these clusters of atoms combine, they form molecules that spontaneously recoil from their watery environment and fold up into tools honed by eons of evolutionary selection.
Evolution hasn’t come close to exhausting every possible amino acid combination, a number so large it likely exceeds the total number of atoms in the universe. But could science explore these untapped potentials? Is it possible to create new proteins to do new things?
In medicine, protein design would mean treatments with capabilities we’ve only ever dreamed of — drugs able to match the chaotic movements of biology and control them with sensitivity, rationality, and precision.
Artificial intelligence has vaulted this once far-fetched idea into near horizons of possibility. At UC San Francisco, a cluster of labs is focused on this opportunity, and together, they’re building the systems needed to enrich newly powerful AI predictions with real-world proof.
Image
DeGrado Lab
The First Protein Designer
In the late 1980s, William DeGrado, PhD, was an industry chemist with time on his hands. His work in drug development had always been productive, and management was supportive of scientists ambitious enough to pursue their own open-ended projects on the side.
DeGrado, now UCSF’s Toby Herfindal Presidential Chair for Entrepreneurship and Innovation, came of age as plastics were becoming ubiquitous in everyday life. His research focused on small-molecule medicines, but his fascination with chemistry was intertwined as much with lifesaving drugs as with the way the quiet science of constructing huge polymer molecules — plastics among them — had changed the world.
It was pure scientific curiosity that led him to design proteins and to coin the Latin prefix — de novo, meaning “from the beginning” — that has become the definitive scientific name for proteins built from scratch.
Image
DeGrado envisioned de novo protein design as a chance to rethink the relationship between protein structure and function. Our bodies’ native proteins achieved their precise forms and roles over billions of years of trial and error. Evolution is unmotivated and impersonal, a glacial process of response to biological and environmental pressures.
As a result, protein structures are florid and overcomplicated things full of hasty fixes, long ways around, and layers of historical baggage. Instead of modifying or imitating these Rube-Goldberg-like designs, DeGrado wanted to create completely different and radically simple amino acid structures. De novo proteins would be tailored to their goals with scientific intention.
Theoretically, these streamlined proteins would be easier to build and control. And, even more significantly, creating proteins from scratch could allow for benefits that evolution hadn’t had time or reason to give us. We could imagine the way we wanted to influence molecular behavior — control immune responses, identify diseases, destroy tumors, preserve brain tissue — and then design a protein to make it happen.
When DeGrado brought an early design to a senior scientist at his pharmaceutical company, the reception was less than enthusiastic.
DeGrado laughs, recalling it. “He said to me, ‘Bill, I don’t know what you’ve designed, but it’s not a protein. Humans can’t design proteins.’”
He said to me, ‘Bill, I don’t know what
you’ve designed, but it’s not a protein.
Humans can’t design proteins.’ It was just
so motivating to think that he was wrong.William DeGrado, PhD
The exchange lit a fire under him. “It was just so motivating to think that he was wrong,” DeGrado says.
Very wrong, it turns out. Today, DeGrado’s notion has developed into one of the most innovative and rapidly growing research fields, supercharged in recent years by the speed at which AI can now explore and test biological variables. Biotechnology and, yes, pharmaceutical companies are hard at work refining de novo protein capabilities into medicines that could soon vastly improve upon our current generation of modified natural ones — eliminating side effects, increasing effectiveness, and expanding the range of diseases they can treat.
While DeGrado’s breakthroughs have contributed to these near-term wins, his sights are set much higher. Enzymes — the holy grail of de novo protein design — are the latest focus of his lab. Unlike other types of proteins, which accomplish their functions by binding to things, enzymes also do chemistry: They catalyze their target molecules, transforming them into different chemicals.
This type of dynamic molecular machine is hellishly difficult to create. It requires a population of atoms to move in just the right way and at just the right speed to recognize and modify only the most precisely defined targets.
DeGrado’s lab is getting there. His team of scientists recently collaborated with James Fraser, PhD, to create one of the most efficient and successful de novo enzymes on record.
Scientists in the DeGrado Lab are paving the way to an entirely new pharmaceutical era. The ability to build de novo enzymes — ones that can do what no natural enzyme can — has the potential to transform not only clinical care but also drug manufacturing, eliminating the need for the expensive and toxic materials required to make every medicine you can imagine.
Those same advances could also help address environmental pollutants and their impacts on health. It’s funny, the way that medicines and the rise of affordable synthetic materials inspired all this. In the end, de novo proteins could be just as lifesaving and world-changing. Maybe even more so.
“They’d be perfect for cleaning up plastics,” DeGrado says.
Image
Fraser Lab
The Cinematographer of Life
James Fraser is famous for shooting things with lasers. A single pulse, a flash of energy, and the molecule on the receiving end responds — often in unanticipated and informative ways.
An imaginative empiricist, Fraser, UCSF’s Ernest L. Prien Professor, is known for this and other creative methods he’s devised to capture the slipperier truths of life. For proteins, this means discovering how and why they like to move.
Natural proteins evolved over billions of years in an environment that rewarded flexibility. A protein that could shape-shift made it more adaptable, and cells put this flexibility to work.
And yet, for most of the time that biologists have been investigating protein structures, they have ignored this essential dynamic nature. Standard practice valued structural clarity: Freeze a protein into a stable crystal form and take an X-ray image.
As the science matured, it became clear that a static image provided as many obstacles as insights. Excited by water and heat, proteins are in a constant state of chaotic motion. They don’t hold a single fixed shape. Rather, proteins exist in several possible shapes at once, some likelier than others.
Scientists represent this strange molecular state of affairs with what they call “energy landscapes” — mathematical representations of every possible form and gesture a protein can express. In these landscapes, “valleys” represent shapes that the protein can take, while “hills” show the amount of energy required to travel between them. The deepest valleys show the shapes a protein is statistically most likely to be in, while the tallest hills index unlikely or rare forms.
Image
In their native environment, proteins are always sampling these different formations, responding to shifts in energy that, from our human-scale point of view, are absurdly small and imperceptibly fast. Over time, their random movements have evolved to become useful responses to cellular cues: Get a signal, change shape, do something new.
This animated reality is exactly what Fraser’s experimental techniques are designed to capture. Where we once had only photographic stills of detailed protein structures, he gave us feature films.
It turns out, the malfunctioning proteins involved in our most difficult-to-treat diseases — cancer, Alzheimer’s, and Parkinson’s, for example — have very intricate energy landscapes. They don’t present obvious chemistries and binding sites for drugs. Instead, these potential targets seem to be hiding in rare dynamic states. But how is it possible to find and understand these rare states? And how can we interact with molecular structures that flicker in and out of presence, there one moment and gone the next?
To answer these questions, Fraser stopped freezing molecules, instead creating ways to perform X-ray crystallography at room temperature so the proteins could remain active. He learned to treat proteins like probabilities, layering data about their many possible forms rather than smoothing out the math with averages. Using lasers, he blasted proteins with tiny jolts of energy to reveal hidden shapes. And, in his collaboration with DeGrado, he invented a way to expose different iterations of their de novo enzyme to a range of drug molecules, learning from their interactions.
Scientists in the Fraser Lab explore natural protein dynamics, providing insights into the fundamental mechanics of disease. Their work is crucial to de novo protein designers, who build new kinds of drugs that can take on these moving targets.
Importantly, the Fraser Lab also provides real-life verification for AI-designed molecules. These experiments test whether de novo proteins behave as predicted and identify how to improve them when they don’t.
Before AI, the failure rate for de novo proteins was “something like 99.99999 percent,” Fraser says. Not heartening to most, but normal and interesting for science. “Failure is exciting. When things fail, we get to figure out why.”
Now, with AI, the failure rate is low enough that researchers are pivoting to learning from the flood of new ways that designs can succeed. “Things have never been more exciting,” Fraser says.
Image
Kortemme Lab
The Logician of Nature
Tanja Kortemme, PhD, is, at heart, two very different kinds of researcher. She’s a biochemist who has spent decades happily immersed in the chaotic molecular life of the cell, delighting in discovery. And she’s also an engineer, someone who looks at nature’s infinite capacity for surprise and sees a design opportunity.
It’s the perfect combination for a scientist who creates de novo proteins that interact with mutable biological environments. Like natural proteins, Kortemme’s molecules change shape. But unlike natural proteins, their energy landscapes are orderly and disciplined, devised for control.
“I’ve always cared about how proteins move,” says Kortemme, a professor of bioengineering. “At first, it was because protein designs kept failing; they were either too flexible or not flexible enough.”
I’ve always cared about how proteins move. At first, it was because protein designs kept failing; they were either too flexible or not flexible enough.
Tanja Kortemme, PhD
Current protein therapeutics, made with modified natural proteins, have shortcomings that arise from their single active state. Like keys drifting around the body and smashing into locks, these proteins achieve their goals bluntly and without much discernment.
These are drugs and therapies that have changed lives and cured diseases, but they can also cause side effects that can be painful, life-altering, or fatal. The single active state means that the proteins can enter a range of similarly shaped locks and wreak havoc where they aren’t needed. They have no way of sparing healthy cells, adjusting potency, sensing location, or shutting down when their work is done — all things a natural protein would do by shifting shape.
Image
“Plus, the only molecular targets these drugs can typically reach are ones that happen to be stable enough to be treated as static,” Kortemme says. “This kind of target is very uncommon. Most druggable targets are dynamic. We need smarter medicines to interact with them.”
Today, the Kortemme Lab is using AI to design these smarter medicines, creating molecules flexible enough to switch between two distinct forms but stable enough not to waver into other shapes. These new machines will speak the language of biology, but their grammar and syntax will be noticeably different.
The de novo proteins created in Kortemme’s lab will be able to toggle into active and inactive states, only deploying where and how they’re needed. They could hinge, sporting an open, lid-like structure that wouldn’t snap shut and assemble the active protein until it was certain it was in the presence of a tumor. Or they might slide along a natural protein’s surface, exposing hidden binding sites or amplifying a distress signal that’s so faint, the immune system is unable to properly respond to it.
To help us graduate from rigid molecular fixes to intelligent therapeutics, Kortemme is working to engineer a suite of modular de novo protein components — defined movements and structures that can be assembled into predictable machines capable of correcting the damaging cell behaviors that make us sick.
For Kortemme, what began as an interest in understanding how proteins move has unfurled into an entire scientific vision: to program logic into life.
Special thanks to Sofia Bali, PhD, Sagar Bhattacharya, PhD, Yuda Chen, PhD, and Stephanie Crilly, PhD.
