The Dynamic Lives of Intrinsically Disordered Proteins

ABOVE: Intrinsically disordered proteins lack a stable conformation. Instead, they exist as an ensemble of conformations. Anna Ida Trolle, Giulio Tesei; CC-BY 4.0

In Greek mythology, Proteus, son of Poseidon and prophetic shepherd of sea-beasts, could foretell the future. The elusive sea god was difficult to capture as he assumed many forms—a lion, a serpent, water, or a tree—to avoid his pursuers. However, if someone could sneak up on Proteus during his midday nap and wrangle the shapeshifter, he would reveal the truth his pursuer was after. 

Today, scientists are hunting down a different kind of shape-shifting entity in search of answers about the inner workings of cellular life: intrinsically disordered proteins. Unlike their well-known, folded counterparts, intrinsically disordered proteins lack a single, stable three-dimensional structure. Instead, like Proteus, they take on many different conformations.¹ 

These dynamic, ever-changing proteins have long fallen through the cracks of conventional structural biology methods and have been excluded or ignored for their staunch defiance of a central tenet in protein science: structure defines function. However, a growing body of evidence found that these are not rare proteins performing odd jobs in the underbelly of our cells nor are they evolutionary junk hoarded in the proteome. They are well-known entities that are deeply entrenched in regulatory biology.² Yet, scientists still know very little about the dynamic and disordered lives of these proteins that help keep the lights on in our cells. 

As soon as I heard about them, I fell in love. They were these weird little molecules, and they were just so funky, and they challenged a lot of my ideas about biochemistry.

 —Gabriella Heller, University College London

“As soon as I heard about them, I fell in love,” said Gabriella Heller, a biochemist at University College London. “They were these weird little molecules, and they were just so funky, and they challenged a lot of my ideas about biochemistry.” Over the last two decades, scientists like Heller have been marrying computational biology and experimental biophysics approaches to capture these proteins. Along the way, scientists have had to think outside of the (structured) box to study disorder. “I fell into this rabbit hole and haven’t left,” said Heller.

Multifunctional Intrinsically Disordered Proteins Give Cells an Advantage

Proteins drive essential biological processes in the body. From the enzymes that fuel chemical reactions to the antibodies enlisted by the immune system, these tiny molecular machines receive, integrate, and transmit cellular information. In the postgenomic era, scientists are working tirelessly to decipher the functions of the protein sequences encoded in the genome. Fueling these efforts is a central philosophy in biology that an amino acid sequence begets structure begets function; like a hammer’s head is designed to hit nails while its claw perfectly clasps onto nails to remove them, a protein’s three-dimensional structure is designed to interact with specific components in the cell to perform a specialized function. But what if a protein lacks a fixed structure?

Flipping through a textbook, one might think that all proteins are perfectly folded, rigid structures that click together like Lego blocks. But beyond the two-dimensional page, proteins come to life, moving around cells like wet spaghetti monsters with no fixed structure. 

Rather than pigeonholing proteins into either ordered and structured or disordered and unstructured, disorder is better viewed as a continuum. Some proteins have well defined structures with tiny, flexible disordered tails while others are entirely disordered, wiggly strands of amino acids. In a sequence, intrinsically disordered regions (IDR) range anywhere from short (five to 10 residue) snippets to long (1,000 or more) stretches of amino acids.² A single protein can display several distinct IDR. It is estimated that around 30 to 40 percent of eukaryotic proteins harbor some degree of disorder.^3,4  

Proteins with disorder aren’t relegated to the sidelines of cellular activity. On the contrary, disordered proteins are stalwarts of cellular communication. “They have so many different functions. It’s incredible,” said Heller. Their conformational freedom facilitates a kind of functional promiscuity that provides cells with multiplexed and flexible recognition and response systems.^5,6 In line with this, these malleable machines are often hubs for essential cellular processes, including gene regulation, cell division, molecular recognition, and cell signaling.^2,7 “In all of those cases, you need something sensitive to its environment [that] needs to know when to switch on [and] when to switch off,” said Heller. 

         

Protein disorder is everywhere. On a cell’s borders, many cell signaling receptors have structured domains that are embedded into the outer membrane and linked to extracellular and intracellular disordered linkers and tails. For example, many G-protein coupled receptors have disordered regions that link the seven structured transmembrane domains.⁸ These disordered stretches contain sites for post-translational modifications that could tune downstream signaling.  Elsewhere in the cell, RNA binding proteins, histone tails, transcription factors, and nuclear transport receptors are all enriched with disorder.^9-12  

Disordered proteins are also social butterflies of the protein world. “Disordered proteins will typically have hundreds of partners,” said Heller. For example, the hepatitis C nonstructural protein 5A (NS5A) has dozens of binding partners.¹³ It is not that intrinsically disordered proteins never take on structure but rather that they interconvert between many different conformations to adapt to the changing tides.² “The same chemistry that drives proteins to fold is still present in disordered regions or disordered proteins, it’s just there in different amounts and in different ways,” said Alex Holehouse, a computational biophysicist at Washington University in St. Louis. Folded proteins are not exactly frozen statues either; they are also wiggling and moving around the cell. “For disordered regions, that wiggling is just much more pronounced and there is no single reference state that is convenient to talk about or think about,” said Holehouse. He added that a more accurate framework for ordered and disordered proteins alike is a sequence-ensemble-function paradigm where ensemble denotes the collection of states that a protein exists in.¹⁴

Advances in biophysics techniques over the last two decades have helped scientists identify disordered regions within a protein’s amino acid sequence; however, predicting every possible conformation of an intrinsically disordered protein ensemble and their resulting functions has been tricky to pin down. “A lot of the methods for helping us understand this have been really optimized for structured proteins, but it’s very different when you’re working with a disordered protein that’s so dynamic,” said Heller. The usual experimental and computational tools, including X-ray crystallography, cryogenic electron microscopy, and, more recently, AlphaFold, fail to capture these elusive and rebellious regions, leading scientists to develop new approaches to predict and measure disorder. “A lot of the things [that] we’re learning about disordered proteins are making us revisit and rethink assumptions we had about folded things,” said Holehouse.

Intrinsically Disordered Proteins Defy Dogma

In the late 1950s, John Kendrew, a biochemist at the Medical Research Council Laboratory of Molecular Biology, obtained the first ever crystal structure of a protein. Kendrew’s revelation of the globular sperm whale myoglobin structure prompted a structural revolution.¹⁵ Thanks to X-ray crystallography, scientists could examine the folded states of proteins, effectively providing snapshots of their 3D shapes, leading to an explosion in the number of published atomic protein structures. More recently, cryogenic electron microscopy has become a popular technique for determining the 3D structure of molecules. 

However, these go-to methods for determining structure require stability. In contrast to their folded counterparts, intrinsically disordered proteins and protein regions are extremely dynamic, rapidly interconverting between many shapes. The conformations adopted by disordered proteins are highly sensitive to subtle shifts in their environment, including temperature, pH, and the presence of binding partners.¹⁶ This conformational flexibility facilitates multifunctionality; however, it means that they elude several structural determination methods. For example, these unruly regions never crystallize, so biochemists chop them out when possible to get X-ray crystallography crystals that better characterize the folded regions. 

The bioinformatics boom around the turn of the century finally brought recognition for disordered regions amongst some protein biologists. Early in this period, Keith Dunker, now a biophysicist at Indiana University, looked into the literature from the previous 20 years and found examples of disordered regions in proteins that acquired structure, or order, after binding with another substrate, suggesting that a stable structure was not a prerequisite for initiating biological activity.¹⁷ Dunker and his colleagues noted in their paper, “Evidently the many particular examples of important disorder-to-order transitions have failed to register within the molecular biology community as an important generality.”

To survey disorder at scale, Dunker and his colleagues developed neural network predictors to explore the relationship between protein amino acid sequence and regions flagged as disordered. Using this tool, they predicted that 15,000 proteins had disordered regions of at least 40 amino acids in length, evidence they used to bolster their argument that disordered regions should be recognized as a category of protein structure. Dunker was not alone in his efforts. Similar calls to reassess the structure-function paradigm started cropping up.¹⁸  

Around this time, Kresten Lindorff-Larsen was a graduate student at the University of Cambridge studying protein folding dynamics. He had been working in the realm of wet lab experimental biochemistry when he decided to traverse the dryer lands of computational research. Lindorff-Larsen describes himself as, “a protein chemist who happens to use computers.”

Specifically, he was interested in understanding the rules that proteins use to fold as they exit a cell’s translational machinery as long, wiggly strings of amino acids. Lindorff-Larsen ran computer simulations on experimental data collected on this protein folding process.¹⁹ From these data he generated ensembles of conformations taken by the protein in solution and identified interactions between different regions of the protein as it folded. He realized that he could use a similar approach to explore the dynamic properties of proteins, including those that appeared to defy the rules of protein folding by remaining unfolded. When he applied similar molecular simulations to experimental data collected on the intrinsically disordered protein α-synuclein, which is implicated in the pathogenesis of Parkinson’s disease, he demonstrated that, in its native state, the protein has a broad distribution of conformations.²⁰ 

Lindorff-Larsen wanted to develop an efficient approach to characterize all the conformational ensembles of disordered proteins without having to run individual experiments to study them one-by-one. “That turns out to be very difficult to do,” said Lindorff-Larsen. “We put it to the side because we just didn’t have the right tools to tackle the problem.” Although he developed an algorithm to learn the conformational preferences of disordered proteins, he soon realized that he needed more data to make his mathematical models broadly applicable.²¹ However, this data was neither quick nor easy to collect. 

Twisting Proteins to Work Out Disorder

For decades, an air of skepticism has lingered around intrinsically disordered proteins. “I have collaborators that still doubt whether these things exist,” said Heller. Bioinformatic approaches helped identify the pervasiveness of disorder, but many remained unconvinced that these proteins lacked a stable structure or whether they had any functional relevance. Afterall, their existence defied a century of dogma, so scientists wanted more convincing evidence. 

To capture the ensembles of conformations adopted by disordered proteins, researchers needed a method that was better suited to their twists and turns. Nuclear magnetic resonance (NMR) spectroscopy allows scientists to probe an intrinsically disordered protein’s structure and dynamics as it wiggles around in solution. NMR spectrometers use strong magnetic fields—far stronger than Earth’s—to probe properties of hydrogen, carbon, and nitrogen nuclei in a protein. By applying radiofrequency pulses to manipulate nuclear spins and watching how spins return to their equilibrium state, researchers can learn about the chemical environments and dynamics of proteins. “We can play games with our proteins to better understand them,” said Heller. A special probe inside the NMR spectrometer detects radiofrequency signals emitted by the nuclei as they relax back to their equilibrium states. By doing this, researchers like Heller can collect information about the molecular structures and dynamics of intrinsically disordered proteins. Heller combines atomistic information provided by NMR experiments with computer simulations to create a “movie” of the disordered protein’s movements. “For nearly every single atom within our protein, we can learn something,” said Heller.

A lot of the things [that] we’re learning about disordered proteins are making us revisit and rethink assumptions we had about folded things.

 —Alex Holehouse, Washington University in St. Louis

As early as the 2000s, researchers began applying NMR spectroscopy to meticulously study individual disordered proteins as they fold upon interacting with binding partners. In one early study, researchers used NMR to reveal how the transcription factor cAMP response element-binding protein (CREB) molded itself into the binding pocket of one of its protein partners.²² 

NMR has also been useful for studying the unruliest of proteins: those that never fold. In budding yeast, the intrinsically disordered signaling protein substrate inhibitor of cyclin-dependent kinase 1 (Sic1) blocks premature cell division.²³ Only after Sic1 gets phosphorylated at six different sites along its amino acid sequence can it bind to its protein partner cell division control protein 4 (Cdc4), which then sweeps Sic1 to the cell’s “trash bin,” ungating cell division. How Cdc4, with its single phosphate binding pocket, bound the six Sic1 phosphate groups eluded scientists. By combining experimental NMR data with computer modeling, a team of scientists concluded that Sic1 and Cdc4 perform a kind of molecular square dance whereby each phosphorylated site rapidly interacts with Cdc4.²⁴ This dance results in Sic1 picking up a chemical signature that tags it for degradation. These findings further hacked away at the structure-function paradigm by demonstrating how a protein could remain disordered in both its bound and unbound states. As experimental data on disordered proteins accumulated, so did their recognition and acceptance. 

New Tools for Exploring Protein Disorder Take Shape

For the last decade, Lindorff-Larsen has been biding his time, waiting to transform his idea for a large-scale, ensemble prediction tool into a reality. Alongside others, he built simulations and collected experimental data on intrinsically disordered proteins. With more data available to feed his models—and even faster computers to facilitate this extensive training program—he picked the project back up. 

The piecemeal approach taken by Lindorff-Larsen and others over the years to study the conformational ensembles of intrinsically disordered proteins was grueling, but necessary. To build a tool that could analyze disordered regions at scale, Lindorff-Larsen and his team dove into the literature, emailing researchers for raw experimental data on more than 50 disordered proteins. They used this information to teach their simulation model the parameters that dictate ensemble states and then tested the model’s ability to predict conformational ensembles of an intrinsically disordered protein region in the absence of experimental data.²⁵

Around the same time, others in the protein folding field were busy building the revolutionary artificial intelligence (AI) system, AlphaFold.²⁶ Now, scientists can predict a protein’s structure from its sequence with speed and accuracy while bypassing the need to run time-consuming experiments. AlphaFold has been nothing less than transformative in structural biology—for folded regions. “The AlphaFold approach is simply not feasible at the moment for disordered proteins,” said Lindorff-Larsen. 

AlphaFold’s prediction system is fueled by protein sequences and multiple sequence alignment data across species, which identifies evolutionary relationships and shared features between genes. “Essentially everything we as a scientific community have done to understand protein function from sequence has, at some level, rested on evolutionary principles,” said Holehouse.  Across closely related eukaryotes, the folded bits of the proteome look very similar. In contrast, the amino acid residues in disordered regions in a protein move around a lot, changing their address across species. This means that although scientists have access to disordered protein sequences, they cannot use similar tools for aligning them across species.²⁷ 

With an improved version of their model, which Lindorff-Larsen and his team called CALVADOS, the researchers set their eyes on a bigger prize.²⁸ “After we had the tool in hand and validated that it worked pretty accurately, we said, ‘Why don’t we just scale it up to look at all human disordered proteins?’” said Lindorff-Larsen. To select their disordered regions they turned to AlphaFold2—as it turns out, the system’s low confidence score is a good indicator that a sequence is disordered.²⁹ After identifying more than 28,000 disordered regions (corresponding to around 35 percent of the residues in the human proteome) the researchers ran them through CALVADOS to perform molecular simulations and predict conformational ensembles.³⁰ 

Holehouse has also been developing computational tools for predicting the average properties of the conformational ensembles that might take shape from the disordered protein sequence. Around the same time that Lindorff-Larsen was working on CALVADOS, Holehouse and his team were developing a similar tool for large-scale exploration of the sequence to ensemble relationship, which they called ALBATROSS.³¹ This deep-learning model predicts conformational properties of disordered proteins directly from sequence. “It really gives people a way to start thinking biophysically about their disordered regions, whereas if you’re not a biophysicist that has historically been very difficult to do,” said Holehouse. 

Lindorff-Larsen and Holehouse hope that their prediction tools will facilitate efforts to link conformational ensembles with protein sequence, function, evolutionary conservation, and disease variants and drive hypothesis generation. However, additional experimental data is still needed to improve the models.  

Shapeshifters in the Proteome Textbooks often depict proteins as nicely folded three-dimensional structures, but many proteins are far from it.

(1) The lock-and-key model of protein interactions has dominated biology for more than a century, shaping how scientists study molecular communication and approach the development  of novel therapeutics. (2) However, over the last two decades, scientists have come to appreciate that protein structure is a spectrum.  (3) While some proteins exist in a structured or ordered state, others have ordered domains with disordered regions, and some are even fully disordered. (4) Unlike folded proteins, intrinsically disordered proteins lack a stable, three-dimensional structure. Instead, they exhibit conformational heterogeneity and interconvert between different states. (5) Because of their shapeshifting capabilities, scientists struggled to capture intrinsically disordered proteins using conventional approaches for determining protein structure. Over the last two decades, nuclear magnetic resonance spectroscopy and new bioinformatics tools have allowed researchers to explore the conformational properties of intrinsically disordered proteins. (6) The field is now witnessing a paradigm shift in how scientists think about protein interactions whereby an intrinsically disordered protein has an ensemble of possible conformations that allows the protein to respond to the environment and drive different functions accordingly. By dissecting and decoding the biophysical principles that dictate the sequence-ensemble-function relationship, scientists hope to shed light on the role of intrinsically disordered proteins in human health and disease.

Intrinsically Disordered Proteins: The Undruggables

Years ago, while Heller was attending a conference on disordered proteins, a colleague’s remark stuck with her. “Someone said that [disordered proteins] would never be druggable and I was like, ‘Well, why not?’” said Heller, who took on the challenge. For the last decade, she has explored ways of targeting disorder with the hopes that her efforts will unlock an enormous therapeutic opportunity.

The AlphaFold approach is simply not feasible at the moment for disordered proteins.

 —Kresten Lindorff-Larsen, University of Copenhagen

There are few small molecules known to interact with intrinsically disordered proteins and even fewer tools to study these interactions. Therefore, Heller’s team uses an interdisciplinary approach that combines molecular simulations with experimental approaches to explore what happens to an intrinsically disordered protein when it binds a small molecule. Part of Heller’s research portfolio has focused on the amyloid-β peptide, well-known for its association with the onset of Alzheimer’s disease. In its monomeric form, amyloid-β begins as an intrinsically disordered protein; however, it can assemble into long insoluble fibrils, which can clump together to create problematic plaques characteristic of the disease. Heller and her team found a small molecule that interacts with the disordered protein and affects its aggregation behavior.³² When they tracked amyloid plaque formation using a dye, the researchers found that their small molecule could reduce fibril formation. To study these interactions in more detail, Heller turned to NMR.

NMR is a gold-standard method for determining how and where a drug binds to a protein.³³ However, Heller ran into a few problems when she used the technique to measure small molecule binding to a disordered protein. NMR is highly sensitive to picking up changes in folded proteins upon their interactions with drugs. Certain regions of folded proteins are buried in hydrophobic cores, while other regions of folded proteins, such as those on the surface of biomolecules, are more solvent-exposed. “This results in distinct signals in our spectra, like a chemical fingerprint,” said Heller. When the experiment is repeated in the presence of a small molecule, Heller said that the fingerprint clearly changes. She continued, “We can map those changes in the fingerprint back onto the protein to see how the protein changes when it binds.” However, with disordered proteins, most regions are solvent-exposed, resulting in an overlap of signals. “It can get difficult to see what’s happening,” she added. Also, as disordered proteins often remain extremely dynamic in their bound form, Heller struggled to detect changes in the NMR chemical fingerprint, which reports on average conformations for all proteins within a sample.

Heller knew from her previous experiments that this small molecule was interacting with the monomeric, disordered form of amyloid-β, so she and her colleagues turned to computer simulations.³⁴ By modeling the protein, the ligand, and the solvent at the atomistic level, they observed that in the presence of the small molecule, disordered amyloid-β adopted less hydrophobic conformations, which could also explain their previous results showing less peptide aggregation following treatment with the ligand. 

Heller and her colleagues’ simulations agreed with the NMR data to show that the disordered protein was not folding upon binding the small molecule. In fact, the ligand allowed the protein to explore new states. “Our data and those of other labs suggest that this is not traditional “lock-and-key” binding,” said Heller. “It’s almost as if the disordered protein is dancing with the small molecule. It’s totally different from what we’ve been taught in Biochemistry 101,” said Heller.

In a recent paper, Heller and her colleagues demonstrated that dynamic NMR methods, instead of chemical fingerprints, are more sensitive to detecting binding between small molecules and disordered proteins.³⁵ Their approach combined dynamic NMR methods with fluorine detection. While many NMR approaches measure binding changes on small molecules using hydrogen, fluorine offers some advantages: It is more sensitive to its chemical environment and does not suffer from overlapping background signals coming from the solvent or proteins. “Suddenly, we could measure beautiful binding curves, even though the chemical fingerprint approach suggested that there was no binding at all,” said Heller. 

“For a long time, and even still today, people are using this gold-standard chemical fingerprint experiment to look for binders, and they’re not seeing big changes. They’ve been taught that that means there’s no binding,” Heller added. “We really need to go and revisit some conclusions that we’ve made about small molecule binding to disordered proteins because we are starting to have better tools.”

Drugs targeting disordered proteins or regions could be a game changer for treating disease, but they would also provide researchers with new tools to probe protein function, much like what is already available to study folded proteins. However, their multifunctional persona complicates things. “For these disordered regions, specificity is a trickier beast,” said Holehouse. If a disordered protein has 10 binding regions, Holehouse said it could be difficult to design a small molecule that tightly binds to just one of the regions but not the others. 

A Bright, Disordered Future

Reflecting on the last decade of research into disordered proteins, Holehouse said that just 10 years ago, scientists working on disordered proteins were sometimes the only people at their institution who had heard of these molecules. “We’d come together at meetings, and it’d be cathartic because you don’t have to justify your existence to everyone,” Holehouse joked. “There was a tendency of some, perhaps old school structural biologists, to think that this idea of disorder is in competition with structure. That’s simply not the case.”

It’s almost as if the disordered protein is dancing with the small molecule. It’s totally different from what we’ve been taught in Biochemistry 101. 

 —Gabriella Heller, University College London

Scientists are far from understanding disorder-based biology, but they hope that a growing awareness of their prevalence and importance alongside new tools will help uncover how disordered regions mediate cellular function and contribute to disease. These efforts have catapulted disordered proteins from a niche curiosity of biophysicists to entities that are increasingly accepted and appreciated for their regulatory roles in cellular function.

Although some might feel overwhelmed by all the moving parts (literally) with disordered proteins, Holehouse embraces the chaos. “That makes it exciting in that there are so many quite fundamental questions that we don’t really have answers to—or at least convincing answers to—yet,” he said.