effect
AlphaFold predictions are paving the way for new treatments that could impact more than 10 million people around the world.
It has been a source of hard-won satisfaction after what often felt like an uphill battle. David Komander and his colleagues have finally published the long-awaited structure of PINK1. Mutations in the gene that codes for this protein cause early-onset Parkinson’s disease, a neurodegenerative disease that presents with a wide range of progressive symptoms, particularly tremors and difficulty moving. But when another scientific team published their own structure for the same protein, it became clear that something was wrong.
“The other two structures that came out looked very different from the one our group performed,” says WEHI (Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia). Theirs had unique features that did not exist in the others. The stakes were high: Understanding PINK1 could help find new treatments that address the underlying causes of Parkinson’s disease, which affects more than 10 million people worldwide.
Although the Komander team was confident in their findings, the contrasting results raised some big questions. And in a highly competitive field of research, they knew they would not be alone in their search for answers. “Not only are these nuts really hard to crack, but once they are cracked, it suddenly opens up a whole area where everyone is doing very similar things,” Komander says.
Credit: Jacinta Moore
The team eventually solved the mystery, but it took several years of research, one serendipitous discovery, and the help of AlphaFold, DeepMind’s protein structure prediction system.
Symptoms of Parkinson’s disease occur when someone’s brain can no longer produce enough of the chemical dopamine. Most people with Parkinson’s disease do not know the specific cause, but in about 10% of patients, a specific genetic mutation can be identified. In these cases, Parkinson’s disease tends to have an early onset and affect people before they turn 50.
One of these genetic mutations is in the gene that encodes the PINK1 protein. PINK1 plays a key role in the breakdown and elimination of mitochondria, often referred to as the power plants inside our cells. “As we age, our mitochondria can age and become damaged,” says Gan. “PINK1 is part of the body’s mechanism to recycle old mitochondria to create new ones.”
When this mechanism is weakened, damaged mitochondria accumulate, leading to the loss of dopamine-producing nerve cells and ultimately the development of Parkinson’s disease. So one way to find better ways to treat the condition is to better understand PINK1 and its role.
Credit: Jacinta Moore
When researchers discovered in 2004 that PINK1 could cause Parkinson’s disease, finding its structure became a key goal, but it was never realized, in part because human PINK1 was too unstable to produce in the lab. To cast the net wider, the scientists found that the insect version of PINK1, like the one found in human teeth, was stable enough to be produced and studied in the lab.
Which brings us back to the beginning of our story. The Komander team published the PINK1 structure in 2017. But when other researchers published different structures for the same protein in other insects (flour beetles), they knew that was only part of the story. It wasn’t surprising at all. After all, proteins are dynamic molecules. “They are like machines and can take on many different shapes,” Gan says. What if the revealed structure is one of those shapes, a snapshot of PINK1 at a single step in a longer process?
We had these new structures, and at the time, we were the only people on Earth who knew what PINK1 looks like while it’s active.
David Commander (biochemist)
As part of his doctoral project, Gan took on the ambitious task of finding out what PINK1 looks like at every stage of its activation process. It was during this work that he noticed something strange. It was a molecule that seemed too large to be his target. “Normally you would dismiss it as something that just came together, like scrambled egg whites,” says Komander.
But Gan had a hunch that this mass was worth examining in more detail, and with the help of Dr. Alisa Glukhova, he decided to use cryo-electron microscopy (cryo-EM) to examine the molecule at the atomic scale. The irradiation was conducted using an electron beam. “I remember saying to Zhong, ‘Okay, you can try it, but it’s never going to work,’” Komander admits.
Gan’s persistence paid off tremendously. What he discovered was PINK1, the exact molecule the researchers were looking for. But why is it so big? It turns out that PINK1 likes company. Instead of single proteins, they are grouped together into pairs of molecules known as dimers and arranged into larger forms. “Six dimers of PINK1 came together to form a large bagel-shaped structure,” Gan said.
Credit: Jacinta Moore
This serendipitous discovery meant that cryo-EM, which does not work for molecules as small as single PINK1, could be used to resolve the physical structure of the protein. The team got their answer.
The previously published structure of PINK1 was not a mistake. These were different conformations that the protein took at different stages of the activation process. But there was a problem. All of this experimental work was performed using insect-derived PINK1. Understanding the implications of their findings for humans with Parkinson’s disease requires investigating whether their findings extend to human versions of the proteins.
Komander and his team switched to AlphaFold. “We had these new structures, and at the time, we were the only people on Earth who knew what PINK1 looks like while it is active,” Komander said. So they used AlphaFold to retrieve human-provided predictions of the structure of PINK1, and after a while they appeared on the screen. He said he was “absolutely shocked” by how accurate AlphaFold’s predictions were.
Later, when Gan put the two protein sequences into AlphaFold to predict the human PINK1 dimer structure, the results were virtually indistinguishable from experimental work with insect proteins. “That dimer basically showed us exactly how these two proteins can interact and work together to form some of these complexes that we’ve seen,” Komander said.
Rather than simply dealing with the fact that the protein is broken, we can think about ‘what kind of drug should be developed to repair the protein’.
david commander
The close match between the results of multiple experiments and AlphaFold’s predictive structure gave the team confidence that the AI system can deliver meaningful knowledge beyond empirical tasks. They went on to use AlphaFold to model how specific mutations would affect dimer formation and explored how these mutations could lead to Parkinson’s disease, and their suspicions were confirmed.
“We were able to immediately generate real insights for people with these specific mutations,” Komander said. These insights could ultimately lead to new treatments. “We can start thinking about what kind of drugs we should develop to fix the protein, rather than just dealing with the fact that the protein is damaged,” says Komander.
They submitted their findings on the activation mechanism of PINK1 to the journal Nature in August 2021, and the paper was accepted in early December 2021. Researchers at the Trempe Lab in Montreal, Canada, appear to have reached similar conclusions. Because the team’s paper was published in December 2021, the WEHI authors had to quickly make final revisions. “I was told to complete the paper three days before Christmas so it could be published in 2021,” says Komander. “It was a brutal timeline.”
Credit: Jacinta Moore
Ultimately, these remarkable papers were published within weeks of each other, and both provided important insights into the molecular basis of Parkinson’s disease.
Of course, many questions remain for researchers in the field, and AlphaFold is provided free of charge to help them reach some answers. For example, Sylvie Callegari, a postdoctoral researcher in Komander’s lab, used AlphaFold to piece together small protein fragments to find the structure of a large protein called VPS13C, which is known to cause Parkinson’s disease.
“Now we can ask different questions,” she says. “’What does it look like?’ Instead, we can start asking questions like, ‘How does it work?’ and ‘How do mutations in this protein cause disease?’”
One of AlphaFold’s many goals is to accelerate medical research, and at WEHI it is applying it to the genetic sequences of patients with early-stage Alzheimer’s disease, allowing researchers to investigate the cause in individual cases. “AlphaFold allows us to do this based on a fantastically accurate human model,” says Komander. “It’s really powerful.”