Invent: Life Sciences Episode 8
Speakers: Stuart Lowe, Davide Danovi, & James Kusena
[Music Playing]
Stuart: So far in this series, we've spoken to life scientists who are influencing what diseases we target, what sort of cells are under development as therapies, and how the therapies themselves are manufactured.
With all this innovation, we shouldn't be surprised by the increased focus on the means by which processes and products themselves are characterised. Cellular products are distinct from more traditional biologics or small molecule drugs, in that the product itself is alive.
Processing may induce biological changes in the cells that could impact on the effectiveness of the treatment. In extreme cases, the processing itself could impact on cell health or even kill the cells outright, which manufacturers are understandably keen to avoid.
Fundamentally though, cell and gene therapies are engineered products that are designed to have a specific therapeutic effect. And so, you might expect that the product quality would be monitored at every stage of the process.
However, the complexities of cellular function and behaviour are still being uncovered by biologists, and the tools for characterization can be difficult to access. So, process developers can't always measure exactly what they want to measure and are often reliant on other indicators.
Do current process analytical tools provide the insights therapy developers need and how can new technologies be implemented in this fast-moving field? Join me, Stuart Lowe, as we plug into Invent: Life Sciences, a podcast brought to you by technology and product development company, TTP.
Today, we ask could advances in process analytics accelerate the development and manufacture of cell and gene therapies. There are many different cell types in the body, each with a distinct function.
While cells from a single individual might contain identical genetic codes, skin cells are quite different from pancreatic cells. Researchers have built cell atlases to classify the differences between cells to a high degree of granularity. And many of the tools used in this research are being adopted in cell and gene therapy manufacturing.
Someone who deploys these tools extensively in his work is Davide Danovi, who leads the cellular phenotyping department at bit.bio. I asked him for more details.
It's really good to see you and thanks for coming on the podcast. Would you mind introducing yourself and giving us some background to your work in cell biology?
Davide: Thank you, Stuart. Sure, my name is Davide Danovi. I lead the Cellular Phenotyping Department within the research and development in bit.bio. I'm also an academic by training and I have expertise in characterising stem cells from the point of view of imaging, and more recently also being involved in flow cytometry and other functional modalities.
Stuart: In bit.bio, can you tell me a little bit about the sorts of cells that you tend to characterise on a day-to-day basis? What are you doing and how does that impact cell therapy?
Davide: So, in bit.bio we start from a human induced pluripotent stem cell. So, we also provide the methods to make sure that these are pluripotent cells. This means that they can actually being differentiated in different lineages in the human body.
And the kind of work we do sits in three different areas. On one hand, we enable the maintenance of instrument, the training of a people, the kind of infrastructure in terms of data analysis for imaging and flow.
On the other hand, we respond to a specific question of people coming to us asking whether their cell is what they think it is. So, we call this characterization.
And then a separate type of activity we define as capture, is devoted to having modules that allow us to identify one specific cell type from the combination of transcription factor we screen cells on.
And so, the phenotyping team kind of uses these three modalities interchangeably. I have to really give credit to Stefan Milde running the imaging facility and Lori Turner running the flow facility.
And the two kind of main modalities speak constantly, as I mentioned to each other. It's a way to sort of also provide our service to the colleagues without kind of imposing necessarily our preferred modality or our innovation. It's very easy to be captured by your toys.
Stuart: You’re talking about building quite a lot of infrastructure and almost knowing in advance what sort of things that you want to measure. That's quite a fundamental question, really, about how do you decide what you want to measure before you've even potentially completed the process.
Davide: So, my expertise lies in research and development. And so, I have a scientific sense of how kind of key opinion leaders would characterise a certain cell type. This has been recently very enriched by a lot of talents that the company has brought in, that clearly has a much better understanding of the regulatory pathways and the type of information you need to quality control a product.
Which in a sense, it's always seen a little bit as a technicality from the scientists, but at the same time, it's massively important and it is a technicality that can change the destiny of a company. Whether you can produce a cell type that quickly fulfils the requirement of regulatory body or not.
Often, it's kind of a backwards thinking from the ideal situation. And it's also interesting that these dialogues are open. So, both from the point of view of the scientific community and the point of view of regulatory bodies, there is an element of doing as good as you can in terms of answering questions. And it's certainly a learning curve on both sides.
Stuart: Can you give our listeners a few examples of what sort of parameters or CPPs, CQAs the regulators are asking for? What might you be expected to measure the company producing cell therapies?
Davide: There's a lot of expression markers with flow cytometry assays. There are some interesting things we have built around pluripotency in terms of quality control of our starting material. And this is done mostly with flow cytometry, but also now we are building imaging-based analysis of these readouts.
And then you have functional assays like for example, multi lecture arrays. We have publicly shown our work with Charles River Laboratories, and this is great work done in the team.
Often there are interesting behaviours that really distinguish one cell from another. And I guess the closer you get to a readout that really taps into the function of that cell, the less discussion there is about whether this is the right cell type.
Stuart: So, you are almost looking for like a hallmark of a particular phenotype that can manifest in many different ways.
Davide: Correct. Often these approaches are best when combined the supervised and unsupervised type of learning. So, in some instance, we look for one specific indication that is demonstrating function or a particular cell type.
And then in some other instances, we work on building a series of parameter that can help us clustering our products and clustering our research type cells that we obtain.
Stuart: So, paying close attention to the indicators that regulators are interested in, gives scientists a better chance of ensuring their products reach the market. These indicators are subtle and building tools to measure them is expensive.
So, scientists need to be careful when deciding what tools to incorporate into their processes. What can we learn from the world of biologics manufacture and how could this be adapted for cell therapy? To find out, I spoke to James Kusena, VP Operations at MicrofluidX.
Thanks a lot for coming on. It's nice to see you. Would you mind introducing yourself and just telling us a little bit about your background in bioprocessing?
James: Yeah, so I'm James Kusena, I am at MicrofluidX. So, I am vice president of operations there, but I have a strong technical background in the company. So, when I first joined, I was essentially in charge of the bioprocessing and applications within the company.
So, that's really a lot of trying to understand how processes should look. Because we're a equipment provider company, so we're trying to understand essentially what are the processes that are people working on like, and what does the technology need to look like in order to aid that. So, that's been my background.
And before that I was more on the developer side, so more on the therapy side. So, I was working on a stem cell therapy for Parkinson's disease. So, that was using embryonic stem cells and differentiating those into dopaminergic neuro progenitors for then reimplantation into the striatum for Parkinson's disease patients.
Stuart: And so, what would you say the kind of fundamental differences between biologics manufacturer and cell therapy manufacturer would be?
James: I think there's kind of five kind of key differences. So, that would be the nature of the products themselves, the complexity of the manufacturing, the process control challenges as well, the analytical techniques that you can apply. And also, on the kind of more further field side of things, the regulations as well.
Stuart: That was very nice. Tell me a little bit about the nature of the product. So, a biologic, if I think about that, I'm thinking about antibodies, cell therapy is actually the cells itself then, right? It is the product.
James: Yeah, so I think, when we are talking about cell therapy, we are looking at manipulating a living cell and whether we're inputting genetic material, we're using small molecules, whatever it is that we are doing, we are doing it to the living cell. And at the end of whatever that manipulation is, it's that living cell that needs to come out and be used.
Whereas when we're talking about biologics, we're talking about very complex molecules of course. But these are typically monoclonal antibodies, vaccines, recombinant proteins, and things like that.
They are coming from a living organism or from living cells. So, they're the byproduct, which makes it a very different thing because now when you think of things like filtration and purity, et cetera, these are very different challenges because when your cell is the actual product, how you purify that, how you monitor that is very, very different.
Stuart: You're having to handle and process the cells for much longer. Because in the biologics you basically throw them away at some point.
James: Exactly.
Stuart: I suppose why might you want to monitor the performance of a particular unit operation in a bioprocess, what benefits does that give you?
James: So, I think for me, process analytical technologies are really there to ensure product quality. At the end of the day, that is the most important thing when we're looking at both the process development and the manufacturing of these therapies or this area of research, if that's what people are doing.
So, you need to have information that is as real time as possible so that you can have the ability to have timely information that is accurate. And that's important because realistically and fundamentally what you're doing is you're making decisions and adjustments to either a manufacturing run or process development work based on these technologies.
So, I think that's why it's really important to have these process analytical technologies there and have them as robust as possible.
Stuart: If quality is the goal, can you measure product quality directly? Is that possible to do or do we have to use surrogates or how is it done currently?
James: It depends on your processes, but I think a lot of things at the moment are kind of done as an endpoint, which I think is detrimental to the scaling and the progression of the industry. I think what we need to understand is things like surrogate markers and how they come in.
There was a really interesting EU guidance on parametric release quite a few years ago, which I don't really know if it got any legs, but it was really interesting just looking at how that's used in other industries.
And actually, if we can apply something like that, that would be quite useful because having it being endpoint means that things are fairly laborious, they take time. And if we want these therapies to move to being something closer to mainstream, we need to be expediting that QC but not forgoing the quality of course.
And in order to do that, we need things that are in line at line and a lot more dynamic, quicker, easier, small footprints, easy to implement. Those are the kind of key things.
Stuart: What was the EU paper on? Remind me.
James: Parametric release. So, basically, how do you use information gained from historical data? Essentially can you create historical data or evaluate robust surrogate markers that can tell you about your product quality.
So, as an example, some of the work that I did when I was working on the Parkinson's side of things was looking at when you are at the latter stages of your differentiation, if you could have something like a dopamine release assay. So, something that's going to measure if your cells are able to release dopamine.
If you had that, that would be like, “Okay, well I have some flow cytometry data, but I also have some level of functionality data because ultimately that's what these cells need to be able to do.”
So, again, it's going to be very specific for individual therapies and individual cell types, but that's the sort of thing where we can be like, “Okay, as long as my cells hypothetically are releasing X milligrams of dopamine or nanograms or whatever it may be, by day six, I'm getting close to where I need to be.” So, it gives you the ability to be like, “Okay, is my process on track or not?”
Stuart: I've seen similar kind of things where you're trying to do it in 3D models or non-animal testing where you look at liver and you say, well is it producing cholesterol as a proxy for liver function? If so, then yes, we can proceed with the model.
James: Exactly.
Stuart: Is it possible to do that in a real time and robust way? So, do this dopamine release assay.
James: When it comes to process, it's harder because these are not the most routine types of metabolites or analytes that are being measured. So, it's very different to something like glucose or lactate, glutamate, et cetera, which are quite standard now.
And that's I guess sometimes the challenge for developers where you have fairly nuanced and specific things that are your critical quality attribute marker.
Stuart: Sounds like you've been through that loop where you've gone around some of the conferences and seen what's available. What are you seeing that you think would be particularly impactful for the ability to characterise cells therapies?
James: There's a lot of exciting stuff. So, I think there's people looking at what does a potency assay need to look like, I think that's really exciting.
And for a therapy developer, if you can work with a company that is able to make a fairly bespoke but flexible potency assay for you and the pipeline that you're working on, I think that's exciting because then you're not having to spend ages trying to validate processes and understand it.
You've got someone who you're working with that can actually just help tune things and be able to kind of allow that to slot into part of your QC and your general product understanding and your process development.
I think the other exciting things are people who are tagging things on. So, I know for a French company that is essentially looking at having a two-phase approach, which is what can they do online, in line or at line depending on how you define those definitions. And having that information, but then taking the same sample cells further down to actually understanding what's happening at a single cell state.
So, you get the information you need to make that decision within that moment in time. But then when you're doing your process development, you also then get that really deep-down information in terms of what's happening in terms of the sequences, in terms of the expression and all that more nuanced and highly detailed information as well.
Stuart: So, you're sort of getting some development friendly information while you're actually running the process as well.
James: Exactly.
Stuart: Surrogates can be useful as long as you link them back to the underlying biology. As James explained in cell therapy, you should ideally try to incorporate some kind of readout of cellular activity since this is the ultimate measure of product quality.
It's easier said than done, but it's an approach with massive potential impact. I asked James about where he saw the measurement space evolving.
James: If I think of companies, there's a few that are looking at almost basically like circulatory systems. So, it will typically integrate this system to your bioreactor, it will sample, take the measurement, and then put it back. At the moment that's been one of the approaches.
Holographic is also something that has been growing. So, essentially being able to use the diffraction that happens to be able to give you an understanding of that cell density. Slightly more mature, I'd probably say is confluency based, which isn't exactly your density, but correlates to your density.
Stuart: And also, is a useful measure in itself.
James: It is a useful measure in itself as well. And also, if you can automate it, it removes that variability because my 80% confluence and your 80% confluence and the next person's are completely different.
And also, it depends which part of the field of view that you have. So, you might have seen the most confluent part, and I might see the least confluent part.
I think what we're seeing is at the moment the way that some of these technologies are being developed is it works very well, stood next to one machine and it's tied up there and that's fine. When we are working on an autologous scale, when we're working with a few batches being made, then that's fine.
But as we move forward, I think we want things that are essentially able to have mutuality in terms of the processing that they can do so that they can actually be deployed across a range of different batches.
And if we think of a dream future cell therapy where we can have our own versions of factories, you want that very easy to move analytic system that is going to go to each of those manufacturing batches and give you that information when you need it instead of having to buy that a hundred times to be able to monitor and process a hundred batches.
Stuart: Sort of want to have a high measurement density per piece of equipment.
James: Exactly. And that's really important and something that we've tried to work on within MicrofluidX where one of our platforms can do up to 30 different conditions at process development.
Now what you wouldn't want is a situation where you have to have 30 different bits of analytical equipment to be able to monitor that. So, mutualizing that really helps to be able to control in terms of the Capex that you're spending, but also ensure that you've got information rich systems.
And when I think of something like cellular origins, the way in which that is robotized, that's a really cool way of being able to look at whether that's moving a microscope or a sensor, whatever it is to different areas and taking that measurement when required. That would be something that's quite cool as well.
Stuart: Imagine we are gaining useful insight as we collect these masses of data. Ultimately, what's the benefit of being more informed about the process?
James: Being more informed about the process means you understand your process, then you can understand how you control it and how you monitor it. And those are very important principles when you're looking at quality by design. And quality by design is important when you're trying to create a quality product. So, I think that's where it stems from for me.
Stuart: Are we going to kind of end up with closed feedback loops or is that a bit too far-fetched?
James: I don't think it's too far-fetched at all. And I think we are seeing some levels of that trying to happen. So, we're having a lot of companies start to make the platforms that they're developing be integratable so you can actually have let's say a bioreactor company interface with a analytic technology to be able to do that feedback loop.
So, for instance, “Hey my glucose level is X percent,” and then that goes back to the bioreactor company and says, “Oh actually pump more of the media and perfuse so that we get back to that level.” Same thing with your oxygen or your pH pump more CO2 in. So, I don't think it's farfetched at all. I think that is where most people are heading.
And I think it's also super important because what we're also trying to do as an industry is really increase that automation piece. So, if we can understand this whilst humans are still heavily involved and we can build trust in understanding that these feedback loops come in, we can now add on that AI layer and be like, “Okay, this is my process.
These are the tolerances that I want to be in. Go do your thing and if there's deviations, let me know. If everything is staying within the process, then run it as you see fit.”
Stuart: So, having access to real time information about your bioprocess could help in the development of feedback control to ensure dynamic process improvement. And as James mentions, technologies such as AI could prove instrumental in interpreting and delivering the data within these novel control strategies.
I wondered how advances in cellular phenotyping might be fed into process control by making these complex measurements more amenable to integration. I asked Davide for his thoughts.
Do you have any examples of technologies that are just emerging that you think are going to have a real impact on the field of phenotyping? I think you've touched on some of the imaging, but perhaps there's something that you want to go into.
Davide: So, I think the hypes of getting the tissue culture closer to the relevant cells, they go in cycles. And I always think that of course the cells proliferate in 3D, but they also proliferate in low oxygen, and they also proliferate in live conditions.
So, when you push one of these three forward as your sort of holy grail of having the perfect cell type, you implicitly dismiss the two others in many cases. So, I think these things need to be put in context of the market again and again, products can be very detrimental in this.
I think for imaging, it's super interesting how quantitative phase imaging and like typography and holography have allowed the cells to be observed in a less invasive way, less light is used. It was for a long time a matter of just imaging a single cell with very little throughput, now that is filling up as well.
I think there is still a scope to have, for example, parallel imaging. There are some attempts now, you could imagine 96, well, at the very same time. So, not having a camera that moves these days it's actually cheaper to have L96 cameras all synced than to have like a moving piece from L to L. So, this is an interesting type of development.
More in general, spatial transcriptomics have really got the readouts of RNA and morphology closer and for flow cytometry there are interesting ways of either sort of going cell by cell even with a confocal. Now there is some attempting recently or like using an array of pinholes and you can get major information from any cell moving.
Stuart: And this is what you mean more like imaging.
Davide: Exactly. I'm also very fascinated by the type of microfluidics that one can use. For example, micro carriers I think have big advantages in different types of approaches. And you can think of using the kind of growing culture almost as a readout.
There's interesting new companies that flip this concept and they allow you to have like on the culture elements that offer the readout. So, in a way you don't sort the cells, you actually have wells that light up saying, this is the condition I require, et cetera.
So, I think it's a complex type of environment and I think to some extent it's also like change management all the time. So, before you jump full speed into one new type of field, you need to have a sense of where you're going.
But I like the fact of observing it under the lens of what is needed and where are the blockages. I think it sort of makes the king a bit naked.
So, for example, I really cannot stand this kind of, “Oh, it's 3D the way forward. Oh, it's low oxygen the way forward. Oh, it's live.” Actually, live very few people mentioned generally, it's always this 3D and low oxygen that stand out.
But I think one needs to be just a little bit more stepping back saying, “Okay, let's accept the model we have is far away from the cell in vivo and how close can we get?”
And for example, bit.bio has this way forward very clearly by choosing direct reprogramming by a transcription factor, which is a different thing from kind of directing the differentiation with peptides.
And it has incredible advantages from the point of view of cell therapy, but at the same time it's almost embracing what you're doing and not kind of trying to have solutions that fit all necessarily.
Stuart: So, you're saying that it's also easier to verify the phenotype if you've done the reprogramming in that way?
Davide: It's much easier because you have homogeneity of the cell types derived. So, you can see our movies, you have a hundred iPS cells that become a hundred neurons or a hundred iPS cells that become a hundred myocytes and it's a one-to-one stoichiometry, so to say.
Stuart: More deterministic.
Davide: Exactly. Exactly.
Stuart: Davide's example is a good illustration of how new technologies can be embraced by manufacturers to improve process understanding. Could we be starting to see measures of cellular phenotype being incorporated into process control? Davide elaborates.
Is a general consensus that I suppose cell phenotyping is going to become more and more a pillar of the characterization in this field.
Davide: I think imaging is still underrated. Like on my academic hat, we have published some recent work in collaboration with UCL kind of proposing the use of imaging to guide the quality controller for cell therapy.
There is a lot of this thing happening, which is very exciting, but I think the whole cell therapy has sort of built on top of existing pharma type model. So, it's still quite siloed from the stem cell research sometimes. And I think this is happening, but maybe not as fast as it could happen.
Stuart: And the flow cytometry seems to be basically the bedrock technology and imaging is coming on top. Do you think flow cytometry itself will also evolve over the coming years?
Davide: Oh, for sure. Imaging is kind of becoming more specialised and it's getting closer to transcriptomics, flow cytometry is also getting close to imaging in many ways. There are interesting examples with different type of approaches.
I think the possibility also, not just from the point of view of acquiring information about a cell, but also from the point of view of cross analysing the different modalities, this is still a big bottleneck.
And there's still not a clear understanding on how to do multi-omics effectively. And most of the times it's kind of a case of applying it to a specific place or a specific set of data.
Stuart: I was going to ask what you would like to measure that you cannot currently measure. What do you think would be really useful outputs that you just can't access right now or it's hard to access?
Davide: Well, for example, some of the work we have done academically has aimed at predicting the graft ability of the product, so to say. And I think in cell therapy, this particularly on autologous is still quite unclear.
Stuart: So, when you say graft ability, what do you mean by that?
Davide: I mean the successful implantation of cell therapy product in a patient with our rejection and the cell doing the job. I think there's quite a lot of interesting stuff happening in the sort of precision medicine space. There has been example of pharmacoscopy, for example.
So, you can sort of think of taking cells from people and test in vitro drugs and then kind of understand the reaction. And something like this in cell therapy is attracting because it offers a quality control that's very, very close to the final product.
And I think it can be built in manufacturing. And I think if you really combine these approaches with our mass production of potentially allogeneic therapy using proprietary transactional with … x, you can think of getting indication by indication the right type of products. And also, for example, stratify clinical trials, there's a lot of advantages in having this awareness.
Stuart: It wouldn't be a massive burden, you wouldn't think to take a patient sample, just a small sample to then be used in the manufacturing process.
Davide: Well, it depends on what your starting product is. I think choosing human iPS, so induced pluripotent stem cell as a starting product is a great thing because it allows the scale, but yeah, there are other type of strategies that involve starting from patient derived material.
Stuart: So, if you had a vision for how the future of phenotyping, how the future of analytics might evolve, what would you like to see happen in the next 10 years say?
Davide: Well, I think the cell therapy is now extending over the borders of CAR Ts and I think some of those indications will get there. I think the possibility to really tap into the quality control of the final product in terms of efficacy and characteristics is probably the frontier.
And I think it kind of goes hand in hand with a more precision medicine approach that also happens outside cell therapy, and these are very exciting.
On the other hand, I think the lab as such is kind of having a big revolution, which is still a bit too slow, but it will be now accelerated a lot.
Stuart: Do you mean the QC lab is going to be looking different in the future?
Davide: I think so. Even the shift from office space and lab space, footprints on research institution and on biotech I think is shifting a lot. And for example, there are kind of low hanging fruit like use of the height in a space. And there's quite a lot of interest in feeling that.
So, automation is a no brainer in many cases, especially the need for close environment. All of these things will happen and are happening.
In terms of characterization, there's still quite a bit of to do on the analysis with the type of acquisition data, we have already. And once that is done, it'll really make us more certain that one particular indication can be predicted in efficacy right at the start of the production.
Stuart: When you talk about data challenges, AI kind of springs to mind is there a move to bring AI to bear onto these challenges as well?
Davide: Well, so things like segmentation of objects have been almost sold by AI in a sense, like it's really near human. We were doing things around 10 years ago where we're comparing the cell profiler analyst readouts with a PhD or a postdoc readout. And I was a postdoc at the time. And so, you could actually cross compare the three. Now it's really human-like, so the segmentation is kind of done.
The further analysis is more complex because often AI can still be quite misled by, I'm thinking about for example, those image recognition that — and can often recognize a completely different type of path and these kind of things can happen.
But it's super interesting and also, I think it's a wonderful space where information experts interact with biologists and with people who are very aware of the market and the kind of workflows and the pipeline.
So, for example, for me being surrounded by talents in manufacturing commercial and my colleagues in science in bit.bio is fantastic.
Stuart: I think there's a lot of interesting data generated from biology, which is great raw material for artificial intelligence techniques. And we're probably going to see as more data's analysed, those models will start to improve, as you train it on more data, we'll probably see improvements come soon.
Davide: And I think some modalities, like the cells not being stained, for example. And live imaging and also flow is also moving to similar territories. Things like holographic imaging or typography, they nicely compliment the approaches that look for a specific market. And you can also cross train one modality with the other in a way that is kind of supervised but not human supervised.
Stuart: So, you're sort of gaining the insights from gold standard and porting it over to this developing technology. As Davide points out, one of the barriers to integration for cellular phenotyping is the need for detailed scientific interpretation of the data.
AI tools are being developed to speed up this part of the process, freeing up scientists to focus on other aspects. And with scientists working on even better means of characterization, what other advances can we look forward to? I asked James what he thought about the future of this industry.
Do you think the way that we do process analytics is going to be different in the future from how we're doing it now?
James: Yeah, I think it will be because I think at the moment it's very endpoint based. When I look at kind of what is the state of the art at the moment it is really endpoint based.
We look at endpoint of viability. We look at metabolites slowly moving to being in line. But again, that's normally, “Oh, okay, I'll take a sample and then I'll analyse it.” Same with things like transduction efficiency and a lot of potency assays. So, a lot of it is very endpoint based.
Especially what I would say is the more meatier bits of data or analytics are definitely more endpoint at the moment. You definitely have ease of integration when it comes to pH and DO viability to some extent.
Some vice cell will do inline cell counting if you've got the right equipment, et cetera. And that's the vanilla stuff. So, when you are talking about, we've got the glucose sate, et cetera, and that's the baseline, that's the vanilla stuff, but the actual things that help you make your specific process are still definitely endpoint based.
Stuart: And is there a sort of a cell therapy sensors company looking at all of this or is it still quite fragmented?
James: In my opinion, still fairly fragmented. So, when I look at novel approaches out there, so I know of companies that are specifically looking at cell content viability. I know companies that are looking specifically at amino acids, some are looking at high through potency assays.
Some are looking at high resolution fluorescence, some are looking at phenotyping and sequencing others at metabolites, transduction efficiency and cell fingerprinting. But can anyone do all of that? I don’t know. But no one is to my knowledge.
Stuart: It would be a very bold push to do all of that and say I'm not going to do anything else. Because there’s are quite disparate domains, from sequencing to amino acid content is you're talking about quite a vastly different technologies.
James: Exactly.
Stuart: So, is that a challenge then for you guys when you're trying to, I suppose, build a platform that integrates lots of sensors? You end up having to talk to a lot of people?
James: Yes, it is a challenge. And for us at MicrofluidX, it's really about how do we grow with people in terms of the developers and even the people providing these solutions. I would say we're fairly ecosystem based.
As a company, we want to work with people, and we want to grow the sector together so we don't think there's a one answer that will fit everyone. And I think it's really about having that ability to work with as many people as you can.
Obviously, you won't be able to work with everyone, but how many people can you work with? And that really has been thought about when we've been developing our bioreactor, but also the hardware and software elements that complemented as well.
So, we're trying to be going back to one of the earlier points about that legacy information, about those legacy technologies. Part of it is they just didn't have to integrate because that just wasn't a thing. Whereas from the forefront for us it's always been how do we integrate and how can we keep it as flexible as possible?
Stuart: What have you done differently from how you might imagine the bioreactor designers of the past?
James: I mean, personally for me, obviously in the context of the company, I've spoken to a lot of people in terms of the different technologies that they look at. So, I will look at, as an example, who are the people working on metabolites? Can we do some sort of mini trial? Can we just have a conversation to understand how it works and what are the limitations?
And if you know a potential future user was to use it, would it work? What would be the limitations? What would be the challenges? And it's just thinking about that and the more people you talk about in the space of analytics and what they need from the bioreactor side, you start to understand, “Oh, actually 8 out of 10 of them need the same sort of type of connection or the same sort of sample volume or you start to see those commonalities creep up.”
Stuart: Buildups and commonalities.
James: Exactly.
[Music Playing]
Stuart: What's interesting about the approaches that Davide and James are pursuing is the desire not only for improved analytical tools, but also better process integration and better links to actionable information. And if we solve these challenges, the industry will be in a much better place.
As Davide mentions, a good place to start a process is knowing your starting material. And if that material evolves throughout the process, we ought to have tools to measure that.
We're going to see some of the familiar cellular phenotyping tools, Davide described being adapted for process characterization purposes, which should improve product quality. And James's thoughts on how we get there touched on a topic we've seen across this series, which is the power of collaboration and networks.
We've still got a lot to learn as an industry and his work helps to ensure that everybody has access to the highest quality information, which will ultimately improve bioprocess efficiency and therapy efficacy.
That's all for today and for the series. Thank you for joining us, and many thanks to both Davide and James for their insights on this episode.
Speakers: Stuart Lowe, Davide Danovi, & James Kusena
[Music Playing]
Stuart: So far in this series, we've spoken to life scientists who are influencing what diseases we target, what sort of cells are under development as therapies, and how the therapies themselves are manufactured.
With all this innovation, we shouldn't be surprised by the increased focus on the means by which processes and products themselves are characterised. Cellular products are distinct from more traditional biologics or small molecule drugs, in that the product itself is alive.
Processing may induce biological changes in the cells that could impact on the effectiveness of the treatment. In extreme cases, the processing itself could impact on cell health or even kill the cells outright, which manufacturers are understandably keen to avoid.
Fundamentally though, cell and gene therapies are engineered products that are designed to have a specific therapeutic effect. And so, you might expect that the product quality would be monitored at every stage of the process.
However, the complexities of cellular function and behaviour are still being uncovered by biologists, and the tools for characterization can be difficult to access. So, process developers can't always measure exactly what they want to measure and are often reliant on other indicators.
Do current process analytical tools provide the insights therapy developers need and how can new technologies be implemented in this fast-moving field? Join me, Stuart Lowe, as we plug into Invent: Life Sciences, a podcast brought to you by technology and product development company, TTP.
Today, we ask could advances in process analytics accelerate the development and manufacture of cell and gene therapies. There are many different cell types in the body, each with a distinct function.
While cells from a single individual might contain identical genetic codes, skin cells are quite different from pancreatic cells. Researchers have built cell atlases to classify the differences between cells to a high degree of granularity. And many of the tools used in this research are being adopted in cell and gene therapy manufacturing.
Someone who deploys these tools extensively in his work is Davide Danovi, who leads the cellular phenotyping department at bit.bio. I asked him for more details.
It's really good to see you and thanks for coming on the podcast. Would you mind introducing yourself and giving us some background to your work in cell biology?
Davide: Thank you, Stuart. Sure, my name is Davide Danovi. I lead the Cellular Phenotyping Department within the research and development in bit.bio. I'm also an academic by training and I have expertise in characterising stem cells from the point of view of imaging, and more recently also being involved in flow cytometry and other functional modalities.
Stuart: In bit.bio, can you tell me a little bit about the sorts of cells that you tend to characterise on a day-to-day basis? What are you doing and how does that impact cell therapy?
Davide: So, in bit.bio we start from a human induced pluripotent stem cell. So, we also provide the methods to make sure that these are pluripotent cells. This means that they can actually being differentiated in different lineages in the human body.
And the kind of work we do sits in three different areas. On one hand, we enable the maintenance of instrument, the training of a people, the kind of infrastructure in terms of data analysis for imaging and flow.
On the other hand, we respond to a specific question of people coming to us asking whether their cell is what they think it is. So, we call this characterization.
And then a separate type of activity we define as capture, is devoted to having modules that allow us to identify one specific cell type from the combination of transcription factor we screen cells on.
And so, the phenotyping team kind of uses these three modalities interchangeably. I have to really give credit to Stefan Milde running the imaging facility and Lori Turner running the flow facility.
And the two kind of main modalities speak constantly, as I mentioned to each other. It's a way to sort of also provide our service to the colleagues without kind of imposing necessarily our preferred modality or our innovation. It's very easy to be captured by your toys.
Stuart: You’re talking about building quite a lot of infrastructure and almost knowing in advance what sort of things that you want to measure. That's quite a fundamental question, really, about how do you decide what you want to measure before you've even potentially completed the process.
Davide: So, my expertise lies in research and development. And so, I have a scientific sense of how kind of key opinion leaders would characterise a certain cell type. This has been recently very enriched by a lot of talents that the company has brought in, that clearly has a much better understanding of the regulatory pathways and the type of information you need to quality control a product.
Which in a sense, it's always seen a little bit as a technicality from the scientists, but at the same time, it's massively important and it is a technicality that can change the destiny of a company. Whether you can produce a cell type that quickly fulfils the requirement of regulatory body or not.
Often, it's kind of a backwards thinking from the ideal situation. And it's also interesting that these dialogues are open. So, both from the point of view of the scientific community and the point of view of regulatory bodies, there is an element of doing as good as you can in terms of answering questions. And it's certainly a learning curve on both sides.
Stuart: Can you give our listeners a few examples of what sort of parameters or CPPs, CQAs the regulators are asking for? What might you be expected to measure the company producing cell therapies?
Davide: There's a lot of expression markers with flow cytometry assays. There are some interesting things we have built around pluripotency in terms of quality control of our starting material. And this is done mostly with flow cytometry, but also now we are building imaging-based analysis of these readouts.
And then you have functional assays like for example, multi lecture arrays. We have publicly shown our work with Charles River Laboratories, and this is great work done in the team.
Often there are interesting behaviours that really distinguish one cell from another. And I guess the closer you get to a readout that really taps into the function of that cell, the less discussion there is about whether this is the right cell type.
Stuart: So, you are almost looking for like a hallmark of a particular phenotype that can manifest in many different ways.
Davide: Correct. Often these approaches are best when combined the supervised and unsupervised type of learning. So, in some instance, we look for one specific indication that is demonstrating function or a particular cell type.
And then in some other instances, we work on building a series of parameter that can help us clustering our products and clustering our research type cells that we obtain.
Stuart: So, paying close attention to the indicators that regulators are interested in, gives scientists a better chance of ensuring their products reach the market. These indicators are subtle and building tools to measure them is expensive.
So, scientists need to be careful when deciding what tools to incorporate into their processes. What can we learn from the world of biologics manufacture and how could this be adapted for cell therapy? To find out, I spoke to James Kusena, VP Operations at MicrofluidX.
Thanks a lot for coming on. It's nice to see you. Would you mind introducing yourself and just telling us a little bit about your background in bioprocessing?
James: Yeah, so I'm James Kusena, I am at MicrofluidX. So, I am vice president of operations there, but I have a strong technical background in the company. So, when I first joined, I was essentially in charge of the bioprocessing and applications within the company.
So, that's really a lot of trying to understand how processes should look. Because we're a equipment provider company, so we're trying to understand essentially what are the processes that are people working on like, and what does the technology need to look like in order to aid that. So, that's been my background.
And before that I was more on the developer side, so more on the therapy side. So, I was working on a stem cell therapy for Parkinson's disease. So, that was using embryonic stem cells and differentiating those into dopaminergic neuro progenitors for then reimplantation into the striatum for Parkinson's disease patients.
Stuart: And so, what would you say the kind of fundamental differences between biologics manufacturer and cell therapy manufacturer would be?
James: I think there's kind of five kind of key differences. So, that would be the nature of the products themselves, the complexity of the manufacturing, the process control challenges as well, the analytical techniques that you can apply. And also, on the kind of more further field side of things, the regulations as well.
Stuart: That was very nice. Tell me a little bit about the nature of the product. So, a biologic, if I think about that, I'm thinking about antibodies, cell therapy is actually the cells itself then, right? It is the product.
James: Yeah, so I think, when we are talking about cell therapy, we are looking at manipulating a living cell and whether we're inputting genetic material, we're using small molecules, whatever it is that we are doing, we are doing it to the living cell. And at the end of whatever that manipulation is, it's that living cell that needs to come out and be used.
Whereas when we're talking about biologics, we're talking about very complex molecules of course. But these are typically monoclonal antibodies, vaccines, recombinant proteins, and things like that.
They are coming from a living organism or from living cells. So, they're the byproduct, which makes it a very different thing because now when you think of things like filtration and purity, et cetera, these are very different challenges because when your cell is the actual product, how you purify that, how you monitor that is very, very different.
Stuart: You're having to handle and process the cells for much longer. Because in the biologics you basically throw them away at some point.
James: Exactly.
Stuart: I suppose why might you want to monitor the performance of a particular unit operation in a bioprocess, what benefits does that give you?
James: So, I think for me, process analytical technologies are really there to ensure product quality. At the end of the day, that is the most important thing when we're looking at both the process development and the manufacturing of these therapies or this area of research, if that's what people are doing.
So, you need to have information that is as real time as possible so that you can have the ability to have timely information that is accurate. And that's important because realistically and fundamentally what you're doing is you're making decisions and adjustments to either a manufacturing run or process development work based on these technologies.
So, I think that's why it's really important to have these process analytical technologies there and have them as robust as possible.
Stuart: If quality is the goal, can you measure product quality directly? Is that possible to do or do we have to use surrogates or how is it done currently?
James: It depends on your processes, but I think a lot of things at the moment are kind of done as an endpoint, which I think is detrimental to the scaling and the progression of the industry. I think what we need to understand is things like surrogate markers and how they come in.
There was a really interesting EU guidance on parametric release quite a few years ago, which I don't really know if it got any legs, but it was really interesting just looking at how that's used in other industries.
And actually, if we can apply something like that, that would be quite useful because having it being endpoint means that things are fairly laborious, they take time. And if we want these therapies to move to being something closer to mainstream, we need to be expediting that QC but not forgoing the quality of course.
And in order to do that, we need things that are in line at line and a lot more dynamic, quicker, easier, small footprints, easy to implement. Those are the kind of key things.
Stuart: What was the EU paper on? Remind me.
James: Parametric release. So, basically, how do you use information gained from historical data? Essentially can you create historical data or evaluate robust surrogate markers that can tell you about your product quality.
So, as an example, some of the work that I did when I was working on the Parkinson's side of things was looking at when you are at the latter stages of your differentiation, if you could have something like a dopamine release assay. So, something that's going to measure if your cells are able to release dopamine.
If you had that, that would be like, “Okay, well I have some flow cytometry data, but I also have some level of functionality data because ultimately that's what these cells need to be able to do.”
So, again, it's going to be very specific for individual therapies and individual cell types, but that's the sort of thing where we can be like, “Okay, as long as my cells hypothetically are releasing X milligrams of dopamine or nanograms or whatever it may be, by day six, I'm getting close to where I need to be.” So, it gives you the ability to be like, “Okay, is my process on track or not?”
Stuart: I've seen similar kind of things where you're trying to do it in 3D models or non-animal testing where you look at liver and you say, well is it producing cholesterol as a proxy for liver function? If so, then yes, we can proceed with the model.
James: Exactly.
Stuart: Is it possible to do that in a real time and robust way? So, do this dopamine release assay.
James: When it comes to process, it's harder because these are not the most routine types of metabolites or analytes that are being measured. So, it's very different to something like glucose or lactate, glutamate, et cetera, which are quite standard now.
And that's I guess sometimes the challenge for developers where you have fairly nuanced and specific things that are your critical quality attribute marker.
Stuart: Sounds like you've been through that loop where you've gone around some of the conferences and seen what's available. What are you seeing that you think would be particularly impactful for the ability to characterise cells therapies?
James: There's a lot of exciting stuff. So, I think there's people looking at what does a potency assay need to look like, I think that's really exciting.
And for a therapy developer, if you can work with a company that is able to make a fairly bespoke but flexible potency assay for you and the pipeline that you're working on, I think that's exciting because then you're not having to spend ages trying to validate processes and understand it.
You've got someone who you're working with that can actually just help tune things and be able to kind of allow that to slot into part of your QC and your general product understanding and your process development.
I think the other exciting things are people who are tagging things on. So, I know for a French company that is essentially looking at having a two-phase approach, which is what can they do online, in line or at line depending on how you define those definitions. And having that information, but then taking the same sample cells further down to actually understanding what's happening at a single cell state.
So, you get the information you need to make that decision within that moment in time. But then when you're doing your process development, you also then get that really deep-down information in terms of what's happening in terms of the sequences, in terms of the expression and all that more nuanced and highly detailed information as well.
Stuart: So, you're sort of getting some development friendly information while you're actually running the process as well.
James: Exactly.
Stuart: Surrogates can be useful as long as you link them back to the underlying biology. As James explained in cell therapy, you should ideally try to incorporate some kind of readout of cellular activity since this is the ultimate measure of product quality.
It's easier said than done, but it's an approach with massive potential impact. I asked James about where he saw the measurement space evolving.
James: If I think of companies, there's a few that are looking at almost basically like circulatory systems. So, it will typically integrate this system to your bioreactor, it will sample, take the measurement, and then put it back. At the moment that's been one of the approaches.
Holographic is also something that has been growing. So, essentially being able to use the diffraction that happens to be able to give you an understanding of that cell density. Slightly more mature, I'd probably say is confluency based, which isn't exactly your density, but correlates to your density.
Stuart: And also, is a useful measure in itself.
James: It is a useful measure in itself as well. And also, if you can automate it, it removes that variability because my 80% confluence and your 80% confluence and the next person's are completely different.
And also, it depends which part of the field of view that you have. So, you might have seen the most confluent part, and I might see the least confluent part.
I think what we're seeing is at the moment the way that some of these technologies are being developed is it works very well, stood next to one machine and it's tied up there and that's fine. When we are working on an autologous scale, when we're working with a few batches being made, then that's fine.
But as we move forward, I think we want things that are essentially able to have mutuality in terms of the processing that they can do so that they can actually be deployed across a range of different batches.
And if we think of a dream future cell therapy where we can have our own versions of factories, you want that very easy to move analytic system that is going to go to each of those manufacturing batches and give you that information when you need it instead of having to buy that a hundred times to be able to monitor and process a hundred batches.
Stuart: Sort of want to have a high measurement density per piece of equipment.
James: Exactly. And that's really important and something that we've tried to work on within MicrofluidX where one of our platforms can do up to 30 different conditions at process development.
Now what you wouldn't want is a situation where you have to have 30 different bits of analytical equipment to be able to monitor that. So, mutualizing that really helps to be able to control in terms of the Capex that you're spending, but also ensure that you've got information rich systems.
And when I think of something like cellular origins, the way in which that is robotized, that's a really cool way of being able to look at whether that's moving a microscope or a sensor, whatever it is to different areas and taking that measurement when required. That would be something that's quite cool as well.
Stuart: Imagine we are gaining useful insight as we collect these masses of data. Ultimately, what's the benefit of being more informed about the process?
James: Being more informed about the process means you understand your process, then you can understand how you control it and how you monitor it. And those are very important principles when you're looking at quality by design. And quality by design is important when you're trying to create a quality product. So, I think that's where it stems from for me.
Stuart: Are we going to kind of end up with closed feedback loops or is that a bit too far-fetched?
James: I don't think it's too far-fetched at all. And I think we are seeing some levels of that trying to happen. So, we're having a lot of companies start to make the platforms that they're developing be integratable so you can actually have let's say a bioreactor company interface with a analytic technology to be able to do that feedback loop.
So, for instance, “Hey my glucose level is X percent,” and then that goes back to the bioreactor company and says, “Oh actually pump more of the media and perfuse so that we get back to that level.” Same thing with your oxygen or your pH pump more CO2 in. So, I don't think it's farfetched at all. I think that is where most people are heading.
And I think it's also super important because what we're also trying to do as an industry is really increase that automation piece. So, if we can understand this whilst humans are still heavily involved and we can build trust in understanding that these feedback loops come in, we can now add on that AI layer and be like, “Okay, this is my process.
These are the tolerances that I want to be in. Go do your thing and if there's deviations, let me know. If everything is staying within the process, then run it as you see fit.”
Stuart: So, having access to real time information about your bioprocess could help in the development of feedback control to ensure dynamic process improvement. And as James mentions, technologies such as AI could prove instrumental in interpreting and delivering the data within these novel control strategies.
I wondered how advances in cellular phenotyping might be fed into process control by making these complex measurements more amenable to integration. I asked Davide for his thoughts.
Do you have any examples of technologies that are just emerging that you think are going to have a real impact on the field of phenotyping? I think you've touched on some of the imaging, but perhaps there's something that you want to go into.
Davide: So, I think the hypes of getting the tissue culture closer to the relevant cells, they go in cycles. And I always think that of course the cells proliferate in 3D, but they also proliferate in low oxygen, and they also proliferate in live conditions.
So, when you push one of these three forward as your sort of holy grail of having the perfect cell type, you implicitly dismiss the two others in many cases. So, I think these things need to be put in context of the market again and again, products can be very detrimental in this.
I think for imaging, it's super interesting how quantitative phase imaging and like typography and holography have allowed the cells to be observed in a less invasive way, less light is used. It was for a long time a matter of just imaging a single cell with very little throughput, now that is filling up as well.
I think there is still a scope to have, for example, parallel imaging. There are some attempts now, you could imagine 96, well, at the very same time. So, not having a camera that moves these days it's actually cheaper to have L96 cameras all synced than to have like a moving piece from L to L. So, this is an interesting type of development.
More in general, spatial transcriptomics have really got the readouts of RNA and morphology closer and for flow cytometry there are interesting ways of either sort of going cell by cell even with a confocal. Now there is some attempting recently or like using an array of pinholes and you can get major information from any cell moving.
Stuart: And this is what you mean more like imaging.
Davide: Exactly. I'm also very fascinated by the type of microfluidics that one can use. For example, micro carriers I think have big advantages in different types of approaches. And you can think of using the kind of growing culture almost as a readout.
There's interesting new companies that flip this concept and they allow you to have like on the culture elements that offer the readout. So, in a way you don't sort the cells, you actually have wells that light up saying, this is the condition I require, et cetera.
So, I think it's a complex type of environment and I think to some extent it's also like change management all the time. So, before you jump full speed into one new type of field, you need to have a sense of where you're going.
But I like the fact of observing it under the lens of what is needed and where are the blockages. I think it sort of makes the king a bit naked.
So, for example, I really cannot stand this kind of, “Oh, it's 3D the way forward. Oh, it's low oxygen the way forward. Oh, it's live.” Actually, live very few people mentioned generally, it's always this 3D and low oxygen that stand out.
But I think one needs to be just a little bit more stepping back saying, “Okay, let's accept the model we have is far away from the cell in vivo and how close can we get?”
And for example, bit.bio has this way forward very clearly by choosing direct reprogramming by a transcription factor, which is a different thing from kind of directing the differentiation with peptides.
And it has incredible advantages from the point of view of cell therapy, but at the same time it's almost embracing what you're doing and not kind of trying to have solutions that fit all necessarily.
Stuart: So, you're saying that it's also easier to verify the phenotype if you've done the reprogramming in that way?
Davide: It's much easier because you have homogeneity of the cell types derived. So, you can see our movies, you have a hundred iPS cells that become a hundred neurons or a hundred iPS cells that become a hundred myocytes and it's a one-to-one stoichiometry, so to say.
Stuart: More deterministic.
Davide: Exactly. Exactly.
Stuart: Davide's example is a good illustration of how new technologies can be embraced by manufacturers to improve process understanding. Could we be starting to see measures of cellular phenotype being incorporated into process control? Davide elaborates.
Is a general consensus that I suppose cell phenotyping is going to become more and more a pillar of the characterization in this field.
Davide: I think imaging is still underrated. Like on my academic hat, we have published some recent work in collaboration with UCL kind of proposing the use of imaging to guide the quality controller for cell therapy.
There is a lot of this thing happening, which is very exciting, but I think the whole cell therapy has sort of built on top of existing pharma type model. So, it's still quite siloed from the stem cell research sometimes. And I think this is happening, but maybe not as fast as it could happen.
Stuart: And the flow cytometry seems to be basically the bedrock technology and imaging is coming on top. Do you think flow cytometry itself will also evolve over the coming years?
Davide: Oh, for sure. Imaging is kind of becoming more specialised and it's getting closer to transcriptomics, flow cytometry is also getting close to imaging in many ways. There are interesting examples with different type of approaches.
I think the possibility also, not just from the point of view of acquiring information about a cell, but also from the point of view of cross analysing the different modalities, this is still a big bottleneck.
And there's still not a clear understanding on how to do multi-omics effectively. And most of the times it's kind of a case of applying it to a specific place or a specific set of data.
Stuart: I was going to ask what you would like to measure that you cannot currently measure. What do you think would be really useful outputs that you just can't access right now or it's hard to access?
Davide: Well, for example, some of the work we have done academically has aimed at predicting the graft ability of the product, so to say. And I think in cell therapy, this particularly on autologous is still quite unclear.
Stuart: So, when you say graft ability, what do you mean by that?
Davide: I mean the successful implantation of cell therapy product in a patient with our rejection and the cell doing the job. I think there's quite a lot of interesting stuff happening in the sort of precision medicine space. There has been example of pharmacoscopy, for example.
So, you can sort of think of taking cells from people and test in vitro drugs and then kind of understand the reaction. And something like this in cell therapy is attracting because it offers a quality control that's very, very close to the final product.
And I think it can be built in manufacturing. And I think if you really combine these approaches with our mass production of potentially allogeneic therapy using proprietary transactional with … x, you can think of getting indication by indication the right type of products. And also, for example, stratify clinical trials, there's a lot of advantages in having this awareness.
Stuart: It wouldn't be a massive burden, you wouldn't think to take a patient sample, just a small sample to then be used in the manufacturing process.
Davide: Well, it depends on what your starting product is. I think choosing human iPS, so induced pluripotent stem cell as a starting product is a great thing because it allows the scale, but yeah, there are other type of strategies that involve starting from patient derived material.
Stuart: So, if you had a vision for how the future of phenotyping, how the future of analytics might evolve, what would you like to see happen in the next 10 years say?
Davide: Well, I think the cell therapy is now extending over the borders of CAR Ts and I think some of those indications will get there. I think the possibility to really tap into the quality control of the final product in terms of efficacy and characteristics is probably the frontier.
And I think it kind of goes hand in hand with a more precision medicine approach that also happens outside cell therapy, and these are very exciting.
On the other hand, I think the lab as such is kind of having a big revolution, which is still a bit too slow, but it will be now accelerated a lot.
Stuart: Do you mean the QC lab is going to be looking different in the future?
Davide: I think so. Even the shift from office space and lab space, footprints on research institution and on biotech I think is shifting a lot. And for example, there are kind of low hanging fruit like use of the height in a space. And there's quite a lot of interest in feeling that.
So, automation is a no brainer in many cases, especially the need for close environment. All of these things will happen and are happening.
In terms of characterization, there's still quite a bit of to do on the analysis with the type of acquisition data, we have already. And once that is done, it'll really make us more certain that one particular indication can be predicted in efficacy right at the start of the production.
Stuart: When you talk about data challenges, AI kind of springs to mind is there a move to bring AI to bear onto these challenges as well?
Davide: Well, so things like segmentation of objects have been almost sold by AI in a sense, like it's really near human. We were doing things around 10 years ago where we're comparing the cell profiler analyst readouts with a PhD or a postdoc readout. And I was a postdoc at the time. And so, you could actually cross compare the three. Now it's really human-like, so the segmentation is kind of done.
The further analysis is more complex because often AI can still be quite misled by, I'm thinking about for example, those image recognition that — and can often recognize a completely different type of path and these kind of things can happen.
But it's super interesting and also, I think it's a wonderful space where information experts interact with biologists and with people who are very aware of the market and the kind of workflows and the pipeline.
So, for example, for me being surrounded by talents in manufacturing commercial and my colleagues in science in bit.bio is fantastic.
Stuart: I think there's a lot of interesting data generated from biology, which is great raw material for artificial intelligence techniques. And we're probably going to see as more data's analysed, those models will start to improve, as you train it on more data, we'll probably see improvements come soon.
Davide: And I think some modalities, like the cells not being stained, for example. And live imaging and also flow is also moving to similar territories. Things like holographic imaging or typography, they nicely compliment the approaches that look for a specific market. And you can also cross train one modality with the other in a way that is kind of supervised but not human supervised.
Stuart: So, you're sort of gaining the insights from gold standard and porting it over to this developing technology. As Davide points out, one of the barriers to integration for cellular phenotyping is the need for detailed scientific interpretation of the data.
AI tools are being developed to speed up this part of the process, freeing up scientists to focus on other aspects. And with scientists working on even better means of characterization, what other advances can we look forward to? I asked James what he thought about the future of this industry.
Do you think the way that we do process analytics is going to be different in the future from how we're doing it now?
James: Yeah, I think it will be because I think at the moment it's very endpoint based. When I look at kind of what is the state of the art at the moment it is really endpoint based.
We look at endpoint of viability. We look at metabolites slowly moving to being in line. But again, that's normally, “Oh, okay, I'll take a sample and then I'll analyse it.” Same with things like transduction efficiency and a lot of potency assays. So, a lot of it is very endpoint based.
Especially what I would say is the more meatier bits of data or analytics are definitely more endpoint at the moment. You definitely have ease of integration when it comes to pH and DO viability to some extent.
Some vice cell will do inline cell counting if you've got the right equipment, et cetera. And that's the vanilla stuff. So, when you are talking about, we've got the glucose sate, et cetera, and that's the baseline, that's the vanilla stuff, but the actual things that help you make your specific process are still definitely endpoint based.
Stuart: And is there a sort of a cell therapy sensors company looking at all of this or is it still quite fragmented?
James: In my opinion, still fairly fragmented. So, when I look at novel approaches out there, so I know of companies that are specifically looking at cell content viability. I know companies that are looking specifically at amino acids, some are looking at high through potency assays.
Some are looking at high resolution fluorescence, some are looking at phenotyping and sequencing others at metabolites, transduction efficiency and cell fingerprinting. But can anyone do all of that? I don’t know. But no one is to my knowledge.
Stuart: It would be a very bold push to do all of that and say I'm not going to do anything else. Because there’s are quite disparate domains, from sequencing to amino acid content is you're talking about quite a vastly different technologies.
James: Exactly.
Stuart: So, is that a challenge then for you guys when you're trying to, I suppose, build a platform that integrates lots of sensors? You end up having to talk to a lot of people?
James: Yes, it is a challenge. And for us at MicrofluidX, it's really about how do we grow with people in terms of the developers and even the people providing these solutions. I would say we're fairly ecosystem based.
As a company, we want to work with people, and we want to grow the sector together so we don't think there's a one answer that will fit everyone. And I think it's really about having that ability to work with as many people as you can.
Obviously, you won't be able to work with everyone, but how many people can you work with? And that really has been thought about when we've been developing our bioreactor, but also the hardware and software elements that complemented as well.
So, we're trying to be going back to one of the earlier points about that legacy information, about those legacy technologies. Part of it is they just didn't have to integrate because that just wasn't a thing. Whereas from the forefront for us it's always been how do we integrate and how can we keep it as flexible as possible?
Stuart: What have you done differently from how you might imagine the bioreactor designers of the past?
James: I mean, personally for me, obviously in the context of the company, I've spoken to a lot of people in terms of the different technologies that they look at. So, I will look at, as an example, who are the people working on metabolites? Can we do some sort of mini trial? Can we just have a conversation to understand how it works and what are the limitations?
And if you know a potential future user was to use it, would it work? What would be the limitations? What would be the challenges? And it's just thinking about that and the more people you talk about in the space of analytics and what they need from the bioreactor side, you start to understand, “Oh, actually 8 out of 10 of them need the same sort of type of connection or the same sort of sample volume or you start to see those commonalities creep up.”
Stuart: Buildups and commonalities.
James: Exactly.
[Music Playing]
Stuart: What's interesting about the approaches that Davide and James are pursuing is the desire not only for improved analytical tools, but also better process integration and better links to actionable information. And if we solve these challenges, the industry will be in a much better place.
As Davide mentions, a good place to start a process is knowing your starting material. And if that material evolves throughout the process, we ought to have tools to measure that.
We're going to see some of the familiar cellular phenotyping tools, Davide described being adapted for process characterization purposes, which should improve product quality. And James's thoughts on how we get there touched on a topic we've seen across this series, which is the power of collaboration and networks.
We've still got a lot to learn as an industry and his work helps to ensure that everybody has access to the highest quality information, which will ultimately improve bioprocess efficiency and therapy efficacy.
That's all for today and for the series. Thank you for joining us, and many thanks to both Davide and James for their insights on this episode.
