With lessons learned from computers, a new platform could help boost production of lifesaving biological therapies

CAMBRIDGE, United Kingdom – In recent years, companies have figured out how to engineer bacteria to make cement, helping reduce the pollution involved in traditional manufacturing. Using more advanced techniques, scientists have even programmed patients’ immune cells to recognize and kill leukemia cells, giving children who had virtually no chance of survival years of prolonged life.

That’s all possible because introducing a manufactured sequence of DNA into a living cell can make it behave in new and transformative ways.

Researchers believe that this ability to program biology has enormous potential to transform how we produce everything from medicines and chemicals to food and fuel. But predicting which out of the millions of combinations of genetic and environmental factors will unlock a desired result is still an expensive, arduous and largely artisanal endeavor. It’s certainly many steps beyond blind luck, but each success can take months or even years of failed experimentation before it works.

“Imagine you’re trying to land a rocket on the moon, but you don’t sufficiently understand the laws of physics or have the means to precisely control it. You’d be lucky to even get close,” said Andrew Phillips, who leads the Biological Computation Group at Microsoft’s research lab in Cambridge, United Kingdom. “But if you understand the equations that govern how the rocket will behave and are able control it, you can determine where it is going to go.”

For thousands of years, people have been taking advantage of biological processes to make things we cannot make ourselves, such as using brewer’s yeast to ferment beer. The instructions that tell the yeast what proteins to make and how to perform the complex operations that turn raw ingredients into a frothy beverage are encoded in the yeast’s DNA. But if we insert new DNA into the yeast, we can program it with new instructions to make all sorts of other things.

Microsoft researchers have spent years studying those underlying processes and learning how to influence them. Now they are launching collaborations with Princeton University and UK-based companies Oxford Biomedica and Synthace to develop and test an integrated platform that’s designed to help other companies and research teams perform that work more reliably. The goal: To reduce the amount of trial and error required to make beneficial scientific breakthroughs and enable companies and academic researchers already established in the field to operate more efficiently and cost-effectively.

The project, called Station B, aims to develop an end-to-end platform — including a software stack, a means to automate lab experiments and machine learning methods that run in the cloud on Microsoft Azure — to help scientists more efficiently and predictably channel the power of life’s ultimate information processing machines: living cells.

Andrew Phillips stands in front of a glass wall with his arms crossed
Andrew Phillips, head of Microsoft’s Biological Computation Group that developed the Station B platform. Photo by Jonathan Banks.

The Station B researchers are working to apply the same rigor that provides predictable results in computer programming to biological systems. The platform draws on Microsoft’s expertise in developing programing languages and compilers and applies new principles to generate sequences of As, Ts, Cs, and Gs that represent the building blocks of DNA instead of binary code. It uses computational models to describe interactions within a cell that are too complex for humans to parse using notebooks and spreadsheets alone.

The first academic collaboration to pilot the Station B platform, with microbiologists and physicists from Princeton University, will investigate the formation of biofilms — thin, slimy layers of bacteria that build up on surfaces and contribute to processes ranging from medical infections to industrial fouling. The team plans to rapidly assemble and test genetic constructs to help researchers understand and, ultimately, learn to disrupt these bacterial communities that are believed to kill as many people as cancer and are a leading cause of infection worldwide.

Microsoft is also launching a collaboration with Oxford Biomedica, a company that develops and manufactures gene therapies that enable a patient’s cells to combat debilitating and fatal diseases. By working with Station B researchers to identify which genetic and environmental combinations will help make its manufacturing processes more productive, Oxford Biomedica hopes to “dramatically lower the costs of those life-changing treatments and put them within reach of more patients,” said Jason Slingsby, the company’s chief business officer.

“If we are going to tackle diseases that are more common, we need to go from making hundreds of doses of targeted therapies per batch to thousands of doses per batch with the same effort,” Slingsby said.

The platform also relies on software developed by Synthace, a London-based company that uses Microsoft Azure to automate biological experiments instead of relying on scientists to do them by hand. This allows scientists or drug manufacturers to test more complicated scenarios and, crucially, to reproduce the same experiment in different settings.

Microsoft’s Azure cloud infrastructure and machine learning tools can quickly analyze experimental data and improve models that predict how cells will react when a particular sequence of DNA is introduced. That could help users zero in on the best conditions for engineering a lifesaving drug, or bacteria that fix dyes onto textiles through a non-toxic process.

Ultimately, a shared knowledge base could let people predict how genetic devices — essentially, segments of fabricated DNA introduced into a cell — will work in new situations, cutting down on the amount of trial and error in developing new products or processes.

By modeling those biological processes in silico, much like meteorologists use computer models to predict tomorrow’s weather or where hurricanes may land, the Station B tools could help scientists obtain promising results or speed up production processes without having to laboriously test every scenario in the lab.

Associate scientist Shama Chilakwad uses Synthace’s Antha software and automation platform to perform experiments in the company’s London laboratory. Photo by Jonathan Banks.

Turning cells into superfactories for lifesaving drugs

Oxford Biomedica’s headquarters are configured in the shape of an antibody, a holdover from the pharmaceutical company that designed the building and used those proteins to elicit helpful behaviors in cells.

These days, drug manufacturers are seeking out personalized treatments. That’s driven up the costs to discover and bring new drugs to market. One of the fastest growing areas of drug development is gene therapy, which uses biological delivery tools that are often made by genetically engineered cells in culture.

Oxford Biomedica, established more than two decades ago as a spinout from Oxford University, has many of these cutting-edge gene and cell therapies in development and also helps strategic partners develop and manufacture their own. It specializes in making the delivery system that gets engineered genes into a cell — essentially a virus particle that’s been stripped of its ability to cause disease or grow but uses its clever cell entry mechanisms to deliver a therapeutic payload of helpful genes.

For instance, a single delivery into the brain of Parkinson’s disease patients introduces three genes that induce target cells to produce a missing neurotransmitter called dopamine. Early clinical trials show promise for helping reverse the illness’ debilitating symptoms.

The company also is helping a partner develop a drug that fixes a gene mutation in the stem cells of children with severe combined immune deficiency (SCID) syndrome — sometimes known as the “Bubble Boy Disease.” The treatment has restored immune function to all the children who have been treated to date, transforming their lives and often allowing them to attend school for the first time.

Oxford Biomedica also secured a deal with Novartis to produce the first treatment approved in the U.S. and the E.U. that reprograms a patient’s own immune cells to recognize and kill cancer cells in patients with leukemia and lymphoma. The drug must be specially made for each individual and costs nearly half a million dollars to treat a child with acute lymphoblastic leukemia.

Before the treatment, those children typically had weeks or months to live. After receiving the cell therapy, 81 percent of the children in clinical trials went into remission, and the first patient to receive the treatment is still alive six years later.

“There’s a lot of excitement around what gene therapies can achieve. Being able to save or significantly prolong the life of a child for whom nothing else has worked is incredible,” said James Miskin, Oxford Biomedica’s chief technical officer. “But there’s also been a lot of desire to reduce manufacturing costs to make these transformative therapies more accessible to patients suffering from more common diseases.”

A woman handles a syringe in an Oxford Biomedica lab
Gene therapies, like the ones Patience Brace develops in this Oxford Biomedica lab, enable a patient’s own cells to fight diseases ranging from cancer and Parkinson’s Disease to severe immune deficiencies. Photo by Jonathan Banks.

Oxford Biomedica’s current challenge is ensuring that it can make enough doses from its LentiVector® platform — the system that delivers engineered genes into human cells — for its own products and for those of its partners. The larger goal is to lower the cost of existing therapies and meet ballooning demand for new ones, including therapies that target larger organs or diseases which affect millions of patients.

“To get democratic and widespread adoption of gene therapy products, we need to drive the economics downward so the cost is reasonable for patients and so that healthcare systems will adopt them,” Slingsby said. “This new collaboration with Microsoft should allow us to discover new insights into how best to configure our systems to further ramp up our production and move in that direction.”

Oxford Biomedica is investing in new facilities and has already managed to improve its output by a factor of ten. To make the next big leap in productivity, it’s looking to the Station B collaboration to better understand, and ultimately engineer, optimal producer cells for its LentiVector platform.

The company is able to meticulously test thousands of combinations of cell lines, genetic material and environmental conditions and pick a winner. But its scientists still can’t say why that one performed better than all the others.

“These screening methods commonly used by pharmaceutical companies get you a result, but they don’t get you understanding,” said Paul Grant, a synthetic biologist who runs the experiments at the Station B lab in Cambridge. “You can’t build on those successes to make the next thing because you don’t really know how you did it. We are building the infrastructure to change that.”

Oxford Biomedica collects many thousands of data points from experimentation and each manufacturing run — from movements of liquid handling robots to conditions inside industrial bioreactors and machines that mix production systems like a cocktail bartender shakes a martini.

The Station B team will first apply its modeling and machine learning expertise to that data to better understand and predict what experimental conditions will produce more of the gene therapy vector. Boosting that manufacturing productivity, producing more doses more reliably with less effort, should help reduce the overall cost of the therapies.

In a second phase, the teams will work to systematically engineer genetic devices — the biological equivalent of computer programs — to enhance helpful interactions and tweak behaviors that can turn the company’s cell lines from reliable performers into superfactories.

Paul Grant stands with arms on a table in a Microsoft Research Cambridge (UK) lab
Microsoft Research Cambridge, UK, biologist Paul Grant runs experiments in the Station B lab. Photo by Jonathan Banks.

Building the Station B platform

Microsoft’s Biological Computation Group has been assembling the Station B platform’s building blocks for years.

Nearly a decade ago, Phillips and colleague Michael Pedersen developed a programming language that compiles biological algorithms to DNA instead of binary code, to influence how a cell behaves. Building on this initial work, Microsoft scientist Boyan Yordanov, who specializes in computer science and biochemistry, worked with the team to develop a prototype platform for designing, implementing and analyzing biological algorithms in cells. It also incorporates the ability to predict behaviors based on training simulation models with data, using a machine learning system developed by Neil Dalchau, a mathematician with a history of exploring diverse biological applications. Grant, a biologist who once studied how zebrafish cells organize to make stripes, uses this prototype to engineer and test genetic constructs to learn how cells make decisions.

Sara-Jane Dunn, a Microsoft scientist and mathematician, also develops computational models to understand how stem cells maintain their ability and readiness to transform into other types of cells: brain, pancreatic, skin, nerve, blood. By reverse engineering that process, scientists in this domain hope to create stem cell therapies to treat a host of illnesses, from regenerating healthy organs to replacing missing neurons that contribute to Alzheimer’s disease.

But influencing biological systems isn’t like programming an operating system or video game. It’s less like writing a recipe — which involves following a logical sequence of steps — than trying to unravel the many interactions and parallel processes that occur in a noisy molecular soup.

“When your cells are trying to decide between something that’s healthy or foreign or deciding when to divide or how to grow, this is essentially information processing at a molecular level,” Dunn said. “We’ve had to come up with new methods to understand this biological computation and how to make the system do what we want it to do.”

These screening methods commonly used by pharmaceutical companies get you a result, but they don’t get you understanding. You can’t build on those successes to make the next thing because you don’t really know how you did it. We are building the infrastructure to change that.

The power of the Station B platform lies in pulling all those pieces of the puzzle together in one integrated system, Phillips said. Both initial deployments will occur in labs that are overseen by health, safety, ethical and medical regulators.

“It marries Microsoft’s deep expertise in programming languages, modeling capabilities and machine learning with lab automation and the power of the cloud and intelligent edge — that combination of tools doesn’t exist anywhere in this industry today,” Phillips said.

To solve one key challenge, the platform uses Synthace’s lab automation system to allow users to run experiments from the cloud and precisely replicate each step in complicated scientific protocols.

Synthace’s Antha software allows the user to replace subjective instructions like “shake a test tube vigorously” with digital language that isn’t open to misinterpretation and that lab robots can execute. Building on top of Azure IoT, Antha is a high-level language for describing biological experiments that allows an array of lab machines made by different manufacturers to run them, much like printer drivers allow any make or model of printer to print PDF documents.

That ability to run experiments exactly the same way each time gives users confidence that the results they’re seeing are meaningful, and not just a fluke in the way the experiment happened to be set up that day.

Synthace’s system — which can handle experiments that simultaneously test dozens of different parameters or genetic constructs rather than one or two at a time — speeds up the research process exponentially. Combined with machine learning capabilities, it also gives customers the ability to pose and learn from much more sophisticated lines of inquiry.

“The near infinite power of biology can only be unlocked by bringing software abstraction and automation to biological R&D and manufacturing, and by enabling biologists to build atop their collective work. That is what the Antha platform does successfully,” said Tim Fell, Synthace chief executive officer.

Sarah-Jane Dunn stands in front a mural with her arms crossed
Sarah-Jane Dunn, scientist for Microsoft Research Cambridge, UK. Photo by Jonathan Banks.

’This could have huge reach’

The Station B platform will be tested first in the lab of Bonnie Bassler, chair of Princeton’s Department of Molecular Biology, a Howard Hughes Medical Institute Investigator and recipient of a MacArthur genius grant, who studies how bacteria wield outsized power by acting as collectives. The Princeton team includes Bassler’s longtime collaborator Ned Wingreen, a physicist and professor in Princeton’s Lewis-Sigler Institute for Integrative Genomics.

“Historically we’ve thought of bacteria as only having harmful behaviors, like infecting us and causing disease, but more recently scientists have discovered the microbiome, a rather magical bacterial community that lives in and on us and that keeps us alive,” Bassler said. “What my lab has always wondered about is how do bacteria manage to either kill us or keep us alive? They’re so tiny.”

Bassler discovered the widespread use of a phenomenon called quorum sensing in the bacterial world. It’s a form of molecular communication that bacteria use to determine when their numbers have reached a critical mass. When they reach the “quorum,” together they trigger behaviors that are only successful when bacteria act as a coordinated group — such as unleashing virulent diseases.

In a proof-of-concept pilot, the team will deploy the Station B platform to investigate how cholera bacteria use quorum sensing to form biofilms, thin layers of bacteria that grow on almost all surfaces. Bacteria living in biofilm communities can be 1000 times more resistant to antibiotics than non-biofilm bacteria.

Princeton researchers will use the Station B platform and Synthace’s lab automation tools to construct and test different versions of two proteins that are key to biofilm formation — which are also genetically programmed to light up. The light allows the scientists to see and measure how much of each protein is produced under many different conditions and in different regions of the biofilm.

Bassler compares the working microbiologists in her lab to master craftspeople, creating elegant and complicated genetic constructs to produce a desired result. But that artisanal process yields only a few prospects at a time and doesn’t allow the team to massively attack the problem.

The Station B platform will be able to build and test dozens of engineered proteins at once — in whatever combinations a researcher can dream up and type into the system for a liquid handling robot to produce. The platform will then help the scientists learn which of the protein constructs behave most like the natural proteins and yield an accurate picture of how biofilm cells organize, Bassler said.

The goal is to build on that basic understanding and find an Achilles heel that might weaken virulent biofilms or increase their sensitivity to antibiotics.

“The platform will allow us to ask more questions, get more results and do more experiments than a graduate student or postdoc, no matter how clever, can do today. So, it gets us to the winning genetic constructs faster,” Bassler said.

Equally important, the platform will also collect and help analyze data from every single lab experiment — including ones that fail, Bassler said. By necessity, scientists have to pursue their most fruitful lines of inquiry, but that can leave an untapped trove of information about why something didn’t succeed.

“If this extra information can help us discover the underlying patterns in what works and what doesn’t work and why, that would be a transformative leap for us,” she said.

The value of deploying the Station B platform in Bassler’s lab is that those researchers have already built an extensive inventory of genetic components, chemical mixtures and models in the years that they’ve been studying bacteria like cholera.

If the team can begin to uncover the rules and principles that govern those systems, Wingreen said, they may be able to program them in transferrable ways. That could potentially enable a doctor who studies cancer or an engineer working on low-carbon fuels to imagine a genetic construct that they’d love to test and get an exact blueprint for assembling it — without spending years at a lab bench.

“From my perspective, this could have a huge reach,” Wingreen said. “Just as the tech sector was democratized by software that lets you ask for what you want in a microchip design and have someone make it, we need that same revolution in biology.”

Top image: Breech Odu works in an Oxford Biomedica lab, where the Station B platform will be deployed to accelerate discovery and manufacturing of gene and cell therapies. Photo by Jonathan Banks.

Related:

Jennifer Langston writes about Microsoft research and innovation. Follow her on Twitter.