Folding@home Consortium scientists have been working hard over the past three months to support experimental collaborators working to develop new therapies for COVID-19.
One of these experimental collaborations is with the COVID Moonshot, an ambitious open science collaboration working to develop low-cost, patent-free small molecule therapeutics that target the main viral protease (MPro) of SARS-CoV-2, the coronavirus that causes COVID-19 disease. This open science collaboration brings scientists from Diamond Light Source, the Structural Genomics Consortium, Oxford, Cambridge, the Weizmann Institute, PostEra, and Enamine, to name just a few.
Illustration of the main viral protease (Mpro) from SARS-CoV-2. Image courtesy Thomas Splettstößer of SCIstyle.
But what exactly is the COVID Moonshot doing, and how did it come about?
This video, from Coffee Break, gives a great overview of the history and goals of the COVID Moonshot:
Birth of a Moonshot
Scientists around the world have been working with unprecedented speed to understand SARS-CoV-2, the coronavirus that causes COVID-19 disease. On 24 Jan 2020, scientists from the China Novel Coronavirus Investigating and Research Team reported the first complete genome sequence of SARS-CoV-2 in the New England Journal of Medicine, barely two months after the first human cases of COVID-19 were reported in Dec 2019. The genome revealed a striking similarity—but key differences—between SARS-CoV-2 and SARS-CoV-1, the virus that caused the 2002-2004 SARS outbreak. While no effective therapies were ever developed for SARS, this similarity allowed scientists to translate their understanding of corresponding components of the virus from SARS-CoV-1 to SARS-CoV-2 and focus on how the differences modified the biology of the virus and—crucially—the consequences these differences had for developing new therapies to combat SARS-CoV-2.
Just a few days after the SARS-CoV-2 genome was published, a team from ShanghaiTech University released the first crystal structure of the main viral protease (Mpro), a SARS-CoV-2 enzyme that cuts a long viral polyprotein into functional pieces necessary for viral maturation. Immediately, there was widespread excitement about the potential for developing drugs that could block its active site, effectively shutting down viral replication, a strategy successfully used in therapies for other viruses like HIV and Hepatitis C.
Scientists at the Diamond Light Source immediately saw the potential, dropping everything to focus single-mindedly on identifying new ways that Mpro could be drugged with small molecules that could block the active site.
Diamond Light Source is an absolutely massive science installation, somewhat resembling a space ship, housing a giant synchrotron that generates incredibly powerful X-rays for studying biology at the atomic scale. The beam is so intense that structures can in principle be collected very rapidly, generating a wealth of new structural information that can be used to design new therapies.
In less than a week, the scientists at Diamond and their collaborators—especially the groups of Martin Walsh and the XChem Lab headed by Frank von Delft—had cloned the viral protease, produced milligrams of protein, coaxed the protein to form tiny but orderly crystals, and solved the first structures by bombarding these crystals with intense X-rays. This massive undertaking was just the start: In just three weeks, they had screened screened thousands of compounds soaked into these tiny crystals and solved their structures, identifying dozens of different small molecules that resembled fragments of drugs binding to different regions of the protease active site.
These fragment structures can be incredibly valuable starting points for drug discovery, since they may mimic the interactions that real drugs might make to tightly lock a small molecule into the binding site, blocking the ability of the protease to function and allow the virus to mature. Much to the excitement of the Diamond researchers, the fragments bound in various overlapping ways, pointing the way to quickly design new molecules that resembled “frankenstein” mergers of the individual compounds that were captured in these crystal structures. Researchers could use this data as a roadmap to quickly develop potent inhibitors of the protease, and with it, a hope at stemming the tide of deaths that would come as the virus makes its way to ever more vulnerable populations around the globe.
Fragment-bound structures of SARS-CoV-2 Mpro solved by DiamondMX/XChem from a fragment screen completed on 24 Mar 2020.
To get this data into the hands of researchers, they posted the data to their website, kicking off what would become a massive open science collaborative undertaking to discover new low-cost drugs for COVID-19.
Robots to the rescue
Solving a single structure of a small molecule bound to a protein usually takes months—how were the scientists at Diamond able to solve thousands in days? The answer lies in an extremely clever strategy pioneered by Frank von Delft (Principal Investigator at Diamond Light Source and the Structural Genomics Consortium at Oxford) and the scientists at the macromolecular crystallography (MX) beamline, who created XChem, a rapid fragment screening technology that uses robots and clever computation to accelerate the discovery of new medicines.
First, protein crystals are prepared in arrays in special plates. These crystals are extremely small—less than 1 mm in diameter—but positioned precisely within the plate to allow an ultrasonic Echo dispenser to soak them with nanoliter droplets of small molecules that resemble fragment of drugs. Some of the molecules—such as a library of chloracetamide-containing compounds from the laboratory of Nir London at the Weizmann Institute in Israel, might even be able to form covalent bonds to the catalytic cysteine residue in the active site. These smaller molecules are able to map out the active site and suggest what kinds of interactions a real drug might make to bind tightly and shut down the enzyme, inhibiting viral replication. The crystals are then rapidly picked, flash-frozen in liquid nitrogen, and placed on the end of a loop in an array inside a puck that can be manipulated by a robot.
Next, a robot in the XChem beamline picks up each loop—stored in liquid nitrogen—and mounts it in the beamline.
The crystal is driven through many different orientations as the sample is moved through an intense X-ray beam created by the synchrotron while diffraction images of the scattered X-rays are rapidly captured. The X-ray beam is so intense and the robot so agile that datasets for many crystals can be collected each hour. The first 600 crystal experiment was completed in just 72 hours.
Once all the data is collected, the computational work of processing hundreds of high-resolution diffraction images from each of thousands of crystals begins. From this enormous amount of data, atomic resolution structures of the viral protease—and the critical atomic positions of compounds that bind in its active site—are extracted.
Of all of this data, a total of 74 structures with compounds clearly resolved emerged. The Diamond scientists rushed to get these structures to researchers as soon as they could to aid in finding an effective cure for COVID-19, making them available online even before any scientific papers had been written. This touched off a firestorm of activity to come.
Fragment-bound structures of SARS-CoV-2 Mpro solved by DiamondMX/XChem.
Crowdsourcing a cure
All of this unfolded rapidly—much more rapidly than typical publication timelines in peer-reviewed scientific journals, which often introduce months or even years of delay. Getting the message out was critical to accelerating progress that could pick up and run with this incredibly valuable dataset. It turns out that all it took was a tweet:
This single twitter thread managed to bring together an enormous, multi-site collaboration. Before long, AI startup PostEra—founded by Cambridge Professor and Forbes Under-30 innovator Alpha Lee, along with co-founder Matthew Robinson—had joined the effort. PostEra’s technology enables computers to rapidly identify strategies for synthesizing compounds, quickly allowing chemists to prioritize those that can be made rapidly from reagents already in stock. Enamine, specialists in rapidly synthesizing new molecules from a virtual synthetic library that includes 14 billion compounds that can quickly be made on demand, soon joined, offering to perform the syntheses at cost. Numerous other collaborators, including UCB Pharma—which offered computational and medicinal chemists to help support the design process without any patent considerations, quickly joined on.
The first step in the process of developing a therapeutic is to find a compound that is big enough and sufficiently well-matched to the active site binding pocket that it effectively inhibits the enzyme at concentrations that could realistically be dosed into patients. The initial fragment hits—while numerous—were all very small, and bound only weakly. But which strategies would be best for building on the more than 60 initial compounds that bound Mpro to progress toward a therapy that would be safe and effective and humans?
Nir London suggested a simple solution: Try all of them. By crowdsourcing ways in which the fragment hits could be elaborated, extended, merged, modified, or otherwise made more potent, they could recruit ideas from the best minds and algorithms all over the world. It would be a Moonshot, an attempt to rapidly progress toward compounds with a potency—measured roughly as the minimum concentration of compound that could be used to inhibit half the functioning proteases—below an initial 1 micromolar threshold before medicinal chemists could take over and work to systematically improve the molecule and ensure it will not be toxic to humans.
With some quick work by PostEra to develop a website where medicinal chemists could quickly sketch ideas and computational chemists or machine learning practitioners could quickly upload ideas, the COVID Moonshot was born.
A call for submissions went out, and was widely circulated to chemists. Before long, hundreds, and then thousands of designs poured in. Over 4,000 compound designs had been submitted by May 1, all viewable on the website, backed by discussion forums in which chemists around the world could discuss the benefits and drawbacks of each compound. Humanity was crowdsourcing the cure to COVID-19.
Work quickly began to identify which molecules could be rapidly synthesized by Enamine, and which molecules had the best chance of reaching the 1 micromolar threshold. It was at this time that the Moonshot scientists got in touch with Folding@home, which quickly sprang into action to help prioritize which designs might be most potent thanks to the help of volunteers all around the world donating their computing time.
While we’ll cover the details of what Folding@home is doing to help prioritize compounds for synthesis in the next blog post, the COVID Moonshot—barely weeks old—has already ordered the synthesis of over 900 compounds. Over 450 of those have already been synthesized, and as of today, more than 400 compounds have been assayed. The real-time status of all compounds can be found on the Moonshot Compound Tracker.
As with the original crystallographic fragment screen, Diamond scientists quickly went to work to soak these compounds into crystals, and collect X-ray diffraction data to see if these compounds bound to the active site of Mpro, and if so, how precisely they were positioned. All this data has been immediately posted online to aid chemists around the world in their own efforts, in addition to helping the Moonshot progress the design of crowdsourced compound ideas.
Meanwhile, Nir London‘s lab at the Weizmann was busy setting up an assay to measure how well the synthesized molecules inhibit Mpro. The assay is incredibly tricky, since Mpro can easily be oxidized if care isn’t taken to mimic the buffer conditions in just the right way, leading to irreproducible results. But in just a couple of weeks, the London lab had beautifully reproducible results, and had screened all the molecules synthesized so far.
As with the Diamond X-ray data, all the London lab assay data went straight to the website to allow chemists who had submitted designs—and everyone else, everywhere—to see how well the designs worked.
Excitingly, several of the designs effectively inhibited Mpro at the tested concentrations, and several molecules had IC50s better than 10 micromolar (a unit of measure of how much compound is needed to effectively inhibit the enzyme), very close to that 1 micromolar goal! While this is still far from a drug, both in potency and other properties critical for ensuring the drug gets to the desired target organ and is safe for humans while still effectively shutting down the virus, this is a very promising initial start.
Initial hits from the first round of COVID Moonshot assay data showing potent inhibitors.
Sharing is caring
While many groups are donating their effort to this project free of charge, the synthesis of new compounds still costs money to buy synthetic reagents. To help fund the costs of compound synthesis for this unprecedented open science, patent-free drug discovery project, the COVID Moonshot is raising funds through a GoFundMe Campaign.
Folding@home has partnered with the Molecular Sciences Software Institute (MolSSI) to share all our data through the COVID-19 Molecular Structure and Therapeutics Hub:
Stay tuned to follow-up posts in the next few days about the datasets Folding@home has generated, details on how we are supporting the COVID Moonshot with physical binding calculations at an unprecedented scale, and more results from the COVID Moonshot.
For those wanting to learn more about the massive crystallographic fragment screen that launched the Moonshot in scientific detail, you can read all the details in the bioRxiv preprint just posted hours ago!