A Next-Generation Baker’s Yeast Biofoundry Presents Transformative Potential for Biomanufacturing

Keith Williams, Founder and Managing Director of Phenotypeca, discusses how the company is combining breeding and rational engineering technologies to rapidly generate next-generation strains of Saccharomyces cerevisiae that optimize production of even complex proteins and related biomolecules, reducing cost and time to market for drug developers, in conversation with Pharma’s Almanac Editor in Chief David Alvaro, Ph.D.

David Alvaro (DA): To begin, what do you view as the key attributes of Saccharomyces cerevisiae that make it a robust biomanufacturing system?

Keith Williams (KW): Humanity has had a close relationship with baker’s yeast (Saccharomyces cerevisiae, also referred to as brewer’s yeast) for over 10,000 years, so it’s not like we’re suddenly trying to find some new platform that never has been in contact with humankind. There’s an incredibly long history regarding safety with yeast. Only a single branch of the yeast family tree has really been used for industrial applications — a wine-related strain family used by all U.S. and European labs. As a result, there’s never really been an open manufacturing platform for use of baker’s yeast, which has been both a strength and a weakness.

It was a strength owing to many advances made by Big Pharma companies using baker’s yeast platforms, including extensive research on genetic knockouts and other modifications. However, baker’s yeast wasn’t universally adopted as a starting point, which is a weakness. On the other hand, another species of yeast, Pichia pastoris, became an open platform, accessed by many academic groups, and their work is starting to filter through to the next generation of possible products.

Baker’s yeast has many positive attributes that make it an ideal system for industrial manufacturing. It grows readily on sugar, water, and salts and is generally easy to work with. There is also no limit to process scale-up — S. cerevisiae could be grown in 200,000-L fermenters if someone needed that scale. For products that are secreted outside of cells, there are typically few impurities co-produced. Products produced inside the cell present some complications, but approaches to manage those have also been developed, especially for vaccine manufacture. Importantly, baker’s yeast, being a single-cell eukaryote, offers the ease of use and manipulation of microbial systems with intracellular processes that are far closer to human cells, enabling production of products bacteria can’t handle and would otherwise require expensive mammalian cell culture.

Phenotypeca has made baker’s yeast even more flexible by adopting our own stable 2-µm plasmid. In addition to establishing key strains that do certain things, we have plug-and-play genetic cassettes, which simplifies engineering because they avoid the need for integration. Most of the Pichia platforms require integration into the genome, and you don’t always know where they will insert and what the consequences will be. Another strength is that there is a real wealth of data on the baker’s yeast genome, so we have a lot of information to help us understand what is going on.

The biggest weakness for S. cerevisiae to date, however, is that it was kept as a closed platform that wasn’t available for use by everyone. Phenotypeca is here to change that.

DA: What can you tell me about the genesis of Phenotypeca?

KW: Our three founders: Ed Louis, Chris Finnis, and I — all came from very different backgrounds. My first job out of university was working at Delta Biotechnology, where I ran the recombinant albumin manufacturing plant, effectively building and automating the GMP facility. Chris engineered the strains that we used at the time. There was an excellent feedback loop between the actual manufacturing environment and the yeast strain development people. That was 30 years ago.

I then moved into consulting around automation and established software and other businesses, all within the biotechnology space. Chris remained at Delta Biotechnology, which was acquired by Novozymes and then spun off as Albumedix, strengthening the baker’s yeast platform and generating many patents, so he's really an expert in that area. He met Ed Louis, who is a giant in yeast genetics, about eight years ago at a conference, where they discussed the potential of combining site-directed mutagenesis with breeding to produce better yeast strains.

Those two stayed in touch, and when the Albumedix research group closed down, Chris developed an idea for a company that could implement the concepts they had discussed. Ed was excited by the idea, and Chris brought me in, because he knew I had just sold a company and was looking for a new venture. My contribution was to keep it as a technology business rather than a manufacturing organization. I knew that vast quantities of data would be generated from the billions of strains that we would be creating every breeding cycle and thought that licensing would be the best route.

We each invested some of our own money to get the company off the ground, and we supplemented that with grants from Innovate UK. We then attracted our first customers and soon after were approached by the Bill & Melinda Gates Foundation, which funded the development of a vaccine product for low-income countries.

DA: What can you share about how your platform has created industrially relevant improvements?

KW: Most of the improvements are subject to patent applications at the moment, so I can’t discuss them in detail, but a representative example of a phenotypic trait we are building into all of our baker’s yeast strains is the ability to be high-producing at higher temperatures, such as 34 °C, which is closer to body temperature. Most strains function optimally at 30 °C. Those four additional degrees mean that proteins are being produced at a temperature much closer to the temperature they are made in the human body. It also means that growth can occur at a higher maximum so that there is a reduced need for cooling at scale, both of which have cost, time, and environmental implications.

On the software side, we collaborated with Tobias von der Haar at the University of Kent, who is an expert on codon optimization. The collaboration resulted in a very robust algorithm for codon optimization in baker’s yeast. It has been extremely advantageous at the molecular level when developing custom strains and expression cassettes for customers. The algorithm is used to optimize the genetic coding sequence for our customers’ proteins, out of which we pick the top performers.

DA: Can you elaborate on the Biofoundry approach and its advantages over simply engineering a small set of optimized strains?

KW: The genetic permutations that are possible if you start mixing up genomes, even at the scale of yeast, are probably greater than the number of atoms in the universe. You are never going to get every single one, but you can generate around a billion, which is enough for our purposes. Think of the game Battleship, with the ships representing the five optimal strains that you want to find. You can guess points on the opponent’s grid, hoping to locate those ships, but most times you have a miss. With our approach, the breeding cycles take four months, but during that time we have effectively carpet-bombed the entire Battleship grid and know where all five of the opponent’s ships are located. That is the power of what we are doing here. Then that is followed by rational engineering to tweak the strongest candidate strains to bring them up to their optimum potential. If there is the opportunity for a 10-fold cost-of-goods (COGs) reduction, we’ll find it.

DA: How does the screening process work?

KW: Using flow cytometry, we can screen a billion strains in four hours, depending on the specific trait we are looking at. We generally pick from the top million strains and analyze these individuals further. Keep in mind that the best parents are far exceeded by the best offspring. Through breeding, we often observe new strains that exhibit quantum shifts in performance.
That process can take up to six months in total. Every time we’ve done it, we’ve observed some level of improvement from the breeding process alone. Then we screen the generated strains as quickly as possible in fermentation conditions to rapidly identify those that will be suitable. A million individual cells or colonies can be screened under simulated fermentation conditions in a few weeks. We are also working with a company that is building a platform comprising 1,000 1.5-mL mini-bioreactors, each with extensive sensors for monitoring fermentation processes, for high-throughput screening of individual strains.

Once we have identified 100–300 strains that perform well under typical fermentation conditions, we examine the quality of the proteins that they produce using high-performance liquid chromatography–mass spectrometry. We also sequence the genomes of those strains. With this data, it is possible to identify the areas that are common in those different strains — typically five or so that are more expected and another 25 or so that are totally unexpected.

This process currently takes about two weeks with two experts using our current platform. We plan to do many of these tasks in parallel so that all could be tested within the two weeks. For this approach, Lab 4.0 could reduce the identification time from two weeks to a day, but we will still rely on external sequencing. I want to stress that this work is being done without using CRISPR or other gene-editing tools –– it is accomplished purely through breeding followed by some rational engineering.

DA: Do you share any of the data you generate that is outside your interests with academic labs?

KW: Yeast isn’t the answer to everything; monoclonal antibodies are currently best produced in Chinese hamster ovary (CHO) cells, but this might change in the future. But you can efficiently produce other complex proteins using yeast strains. Albumin, which is the product developed by Delta Biotechnology 35 years ago, has a complex multi-domain structure with many internal structural bonds. Baker’s yeast in particular is best suited for applications that require large quantities of proteins.

Phenotypeca is currently focused on vaccines and recombinant therapeutic proteins, including those based on single-domain antibodies, such as scFv, VHH, and VNAR scaffolds, that can be used as therapeutics or affinity ligands and in other applications.

We also see potential with our strains for the production of proteins and enzymes used in the manufacture of artificial meats. bovine albumin will need to be animal-free and available in very large quantities. Production via fermentation in yeast is a viable approach. There is also a possibility of using yeast to produce small molecules; nothing is ultimately off limits.

For the present, we are focusing on areas where we have expertise, which is in the therapeutic space. As we bring new people and projects on board and our skill set expands, we aim to make the plug-and-play platform as flexible as possible.

DA: What can you tell me about Phenotypeca’s evolving business model?

KW: In the therapeutic space alone, there are around 40–50 FDA-approved products manufactured using baker’s yeast, with over a $40 billion market, so yeast is proven even without our advanced strains. We offer a proprietary next-generation baker’s yeast strain development platform developed by domain experts and augmented by artificial intelligence and machine learning technologies. Our mission is to enable treatments to be brought to market more quickly and more cost effectively for all who need them, not just those that live in the wealthiest nations.

Our business model is to onboard as many customers as possible at the feasibility stage using our PhenoStart™ product, which is one of the three products within our Phenotypeca® Biofoundry. At this stage, we are looking to determine whether we can express the protein. We aim to keep that at cost, because we are essentially verifying that our platform will work.

Customers can then carry on the project internally with an R&D cell bank or continue working with Phenotypeca using our PhenoDev™ process to create and industrially ready set of strains. From this, we can provide 6–10 strains that work and establish a cell bank and the documentation that goes with that; they can take those strains to a CDMO partner. They may also in parallel choose to do a PhenoFull™ project with us, which involves development of the optimal industrial production strain with a scalable fermentation, separation, and downstream process. The IP for the strains for their proteins and their process is then transferred to the customer for implementation with a CDMO or internally if they have their own capacity. Through the PhenoDev™ and PhenoFull™ products, Phenotypeca will hope that these strains and processes are successful for their customers to generate future licensing income.

Customers usually come to us with a process in mind, unless they are new biotechs that have one or two target proteins and no process. Because we have previously developed many processes in similar areas, we can really help these companies, especially with regard to industrialization and regulatory compliance.

With our next-generation baker’s yeast platform, customers benefit from significantly reduced COGs (we are targeting 10 times lower than mammalian cell culture and three times lower than bacterial processes), reduced risk in achieving regulatory approval, and eventually reduced time to market for their products.

Phenotypeca is currently in the process of filing four strong patents and has identified an additional 11 patent opportunities, with more expected. By 2027, we conservatively estimate £10M in annually recurring revenues from customers who have licensed our strains for commercial production. Phenotypeca already has two such customers working on PhenoDev™ projects.

DA: Is the approach the same if you are optimizing an established process versus identifying the best strains in a process-agnostic manner?

KW: Companies with licensed processes for approved products may not come to Phenotypeca, because the cost to change processes is too great, even if five-fold improvements can be achieved. However, there are many large pharmaceutical companies with projects that were halted because the COGs didn’t work out. Some of those might be appropriate to bring to us, because a five-fold improvement might make a project viable in such cases.

Smaller biotechs with one or two proteins want to get to the next round of funding quickly so they can enter the clinic. They need a production strain in hand in as little time as possible so they can produce material for clinical testing. For some proteins, developing a CHO process may be faster, but mammalian cell culture isn’t applicable for others. In those cases, the choices are typically Escherichia coli or yeast, which includes both P. pastoris and S. cerevisiae.

DA: Do you have a framework of how you suggest the platform can benefit people in terms of product development time, risk, cost of goods, and so on?

KW: We have some very good cost modeling data for industrial processes. Of course, the actual benefits will always be dependent on the specific process — how many chromatography columns are required for purification, and so on. Given some basic information, however, we can apply the modeling approach and quickly estimate the COGs for a specific project. If someone has a specific cost target, we can determine whether it is possible to achieve that goal given a certain titer level and conservative yields for each downstream processing step. What we would hope is that potential customers come to us early in their development projects, rather than wait until they have already reached phase I or phase II.

In the end, it is all about risk. Companies developing biosimilars probably already know the sizes of the markets they are going to target, and generally would want to pay for the cost of a commercial license to get to approval and then not come back again. Companies developing novel drugs that don’t have an established market, however, might be inclined to sign an agreement based on a percentage of revenue, because they don’t know how big that will be.

The vaccine for human papillomavirus (HPV), which is manufactured via fermentation using baker’s yeast, is a good example. The market was initially estimated to be on the smaller side, but it has turned out to be quite large — over $4 billion. The process, however, still is not sufficiently cost-effective to enable distribution to everyone in the world who should have it. We aren’t a non-profit organization, but we want to help make medicines cheap enough that access is increased while still making a return for our investors.

DA: Can you tell me about the partnerships Phenotypeca has established to date?

KW: We started with really strong support from local universities, who gave us access to key technologies, such as flow cytometry, that enabled Phenotypeca to build very efficient workflows. Our government grants have also supported the development of our production platform. All of the partners in the Innovate UK grant program have been really helpful, encouraging, and collaborative. It fills me with great pride that the scientific community is still open to that level of collaboration.

We have also had some great investors from the beginning who have supported us through the seed rounds, and the initial customers we have worked with have been tremendous. They appreciate our motto about problem solving: “If data is unexpected, share it quickly.” That enables twice as many brains to work on a solution. That approach has been very effective and has led to the discovery of new concepts and improvements that we would not have ever known about otherwise.
We are currently looking to raise a significant amount for the next round of development. In addition, we will be announcing some significant customers coming on board soon.

DA: Are there any areas in which you’re looking for a partner in order to leverage complementary technology?

KW: We don’t perform metabolic engineering for metabolites, so there could be a situation where combining our strains with new metabolic engineering could lead to novel developments. Companies developing complex molecules like cannabinoids should consider yeast as a starting point rather than the approaches they are currently using. How many fields of hemp do you need to match the output of a 100,000-L fermenter? That situation seems to present opportunities for strategic partnering. I think our Big Data activities also complement what they’re doing.

We are also working with partners on our Lab 4.0 initiative to automate appropriate systems. We want to automate the parts that are worthy of automating and gather data that will support decision-making so that we can realize our goal of onboarding up to 20 customers per year while maintaining a scientific staff of 20. We’re using off-the-shelf commercial software where possible and equipment with which it has already been integrated. Where neither of those things are available, we’ll use specialized technology.

To design the lab, we are partnering with Bryden Wood, an architectural firm that specializes in high technology, and they are working without a preconception of how the lab should look and with very few constraints, which will enable us to develop an optimal solution. Synthace, Google, and Microsoft are helping with the development of input–output DoE, basic operating, and AI software, respectively; each is the best at the respective technologies.

Working with this consortium is really exciting. The only thing constraining us at the moment is a few million dollars to get the right bits and stick it all together. We are looking for an investment of around $10 million to accelerate business development activities, grow the operational and scientific team, and further develop AI and automation solutions for our existing workflows, IP filings, and R&D kits to seed the next customers.

DA: When do you anticipate hitting those coming milestones?

KW: We are looking for an investment over the next six months. We want to build the prototype of Phenotypeca 1.0 within 18 months and to have the production unit in place in 24–30 months. In five years, we are targeting a valuation of $180 million. The platforms for PhenoStart™ and PhenoDev™ have been built, and we are presently ramping up business engagements to fill the platforms. We can currently take on about five to seven customers in a year –– we have got two already and a third hopefully that will come onboard in late April.

We are also building the Phenotypeca team. We are looking for an operations director that can convert workflows in the lab state into an efficient operational state that can be transferred to customers. Once the Lab 4.0 project is in full swing, we will need a CIO as well; we currently have someone working one day a month until we can support a full-time position. We will also need technical folks to be responsible for data enrichment and analysis, which are now being performed at the University of Nottingham through a joint venture program — we keep the IP related to the yeast and they keep the AI-focused IP. Altogether, we plan to recruit another six or seven people to add to the 13 we have now.

DA: Over the longer term, do you see Phenotypeca staying privately owned or might you look for a buyer or go public?

KW: In five years’ time, if Phenotypeca is a $150 million valuation business and someone can take it to half-a-billion — and it will be able to go to half-a-billion — my work will be done at that point. Reaching $25 million a year in sales will be easily achievable as the licenses start to bear fruit in the biopharmaceutical markets. The IP we are generating now — some of it is defensive, but some of it is truly disruptive will really add to the value of the company. In fact, much of what we are discovering in yeast equally applies to CHO and other systems, so the potential for the company is truly limitless.

The current plan is to exit by late 2027 and have someone else growing the business on from there, probably in other areas, possibly industrial enzymes and nutraceuticals. That could be a specialist CDMO who would want a baker’s yeast platform, a synthetic biology company or larger biofoundry needing a baker’s yeast platform and wanting the IP we have generated, or possibly a large biopharmaceutical company wanting to own the best second-generation platform. There is also the option for an IPO at this stage. We are really excited about all of those prospects, and the journey up until now has been both fascinating and rewarding.

Keith Williams

Keith is an experienced entrepreneur, innovator, and business manager with extensive life science and technology experience in the UK, European, and U.S. regulatory frameworks. He receive an MSc in biochemical engineering and a BSc in biological sciences from Birmingham University. Additionally, Keith is a certified cricket nut.