On this episode of Translating Proteomics, host Parag Mallick discusses spatial proteomics with special guest Fiona Ginty Ph.D. Fiona is a Senior Principal Scientist in Precision Diagnostics at the GE Healthcare Technology & Innovation Center. She is a leader in the development of spatial proteomics technologies and their application in precision diagnostics and medicine.
Their discussion covers:
· How Fiona began working in spatial proteomics
· Why spatial biology is important
· What the future holds for spatial biology
00:00 – Introduction
01:54 – Fiona’s journey to biology
05:26 – Fiona’s transition to tool development
07:20 – Working at GE Research
11:26 – Identifying the importance of spatial biology
14:43 – How Cell DIVETM works
18:25 – The importance of single cell
23:01 - When Fiona realized the technology worked
28:04 – Spatial biology projects Fiona is excited about
30:08 – Fiona’s role in HuBMAP
36:50 – Learnings from HuBMAP so far
38:35 – The future of spatial proteomics in the clinic
46:56 – Current limits on spatial proteomics
49:56 – Current and future uses of AI in spatial proteomics
53:30 – The most exciting thing Fiona learned in her spatial proteomics journey
56:08 – Outro
Method of the Year 2024: Spatial Proteomics
Paper covering the spatial proteomics technology Fiona worked on at GE Healthcare
HubMAP – Human BioMolecular Atlas Program
Cell DIVE Multiplex Imaging Solution
Papers discussing what makes colorectal cancer cells undergo apoptosis in response to chemotherapy
Paper discussing how the distance between tumor cells and cytotoxic t-cells correlates to caspase level
Paper showing it takes 3 hits from cytotoxic t-cells to kill cancer cells
Paper focused on generating new pathology image patches using AI
Dr. Ginty wanted to acknowledge the many people she’s worked with on Cell DIVE and as part of HuBMAP. She stresses that these are highly collaborative projects, and just some of the many great team members she's worked with are listed below.
Cell DIVE:
Anup Sood, Sean Dinn, Yunxia Sui, Sanghee Cho, John Graf , Kathleen Bove, Musodiq Bello, Sudeshna Adak, Michael Lazare, Anirban Bhaduri, Thomas Treynor, Dileep Vangasseri, Alberto Santamaria-Pang, Maureen Bresnahan, Micheal Gerdes, Megan Rothman, Colin McCullough, Chris Sevinsky, Liz McDonough, Christine Surrette, Max Seel, Alex Corwin, Bob Filkins, Kevin Kenny, Brion Sarachan, Mike Montalto, Christoph Hergersberg, John Burczak, Jens Rittscher, Steve Zingelewicz, Dr. Sunil Badve (Emory U), Dr. Yesim Gokmen Polar (Emory U), Eli Lilly collaborators
HuBMAP:
Lou Falo (U. Pitt), Arivarasan Karunamurthy (U. Pitt), John Hickey (Duke), Chenchen Zhu (Stanford), Ioannis Vlachos (BIDMC), Liz McDonough (GE HealthCare Technology and Innovation Center [HTIC]), Sammi Abate (GE HealthCare Technology and Innovation Center [HTIC]), Soumya Ghose (GE HealthCare Technology and Innovation Center [HTIC]), Christine Surrette (GE HealthCare Technology and Innovation Center [HTIC]), Dr. Katy Borner (Indiana U)
Foreign.
Speaker B:Welcome back to Translating Proteomics.
Today I'm delighted to introduce you to Fiona Ginti, who is a senior principal scientist in precision diagnostics at the GE Healthcare Technology and Innovation Center. She is a leader in the development of spatial proteomics technologies and their application in precision diagnostics and medicine.
ics the method of the year in:Additionally, Fiona is a leader on projects within NIH's Human Biomolecular Atlas Program or HUB MAP, which aims to create an open global atlas of the human body at scale cellular molecular resolution. Fiona, I'm really honored to have you here today and welcome to Translating Proteomics.
Speaker A:Thank you, Parag. And it's great to see you again. We work a lot together.
Speaker B: ealized it was in February of: Speaker A:Yes.
Speaker B:So it was a while ago.
And so you had come to visit us at Stanford at the Canary center and shared for the first time I'd ever seen this technology for studying multiplexed single cells. And I remember thinking, and I may have even said out loud, wow, that is the coolest thing I've ever seen.
And then that led to an amazing partnership and multiple grants together and so much fun.
I'd love to go back a little bit in time just to understand how you got to where you are and how you first got into biology and your early years studying bone biology at University College Cork. Maybe just share a little bit about your, the earliest part of your journey.
Speaker A:Sure, yeah. Thank you for, yeah, going back down memory lane. It's always fun. So, yeah, I mean, you know, biology has always been of great interest.
I think when, you know, at school level in high school, it's very much focused on, you know, zoology and bear form and, you know, you're not really thinking too far into the future, but it's, it's cool and it's interesting. And so I think, yeah, going to university College Galway, University of Galway we call now.
I studied microbiology and just became, you know, passionate about bacteria and what bacteria could do. I just thought that was incredibly amazing that you could use bacteria for so many things and.
But where I sort of got my interest more in the applied side of biology within. I did a course on nutrition for one year and just became fascinated by the idea of biomarkers.
And at the time, I remember my professor, at the time, he, Professor Botchwell, he talked about nutrition biomarkers and I thought, what are they? That just sounds cool. So then I ended up going to Cork and working in bone, bone biomarkers.
And that sort of led me to India and did a PhD in nutrition science and, you know, became more appreciative of what biomarkers can do in science, in biology and in clinical translation. So I just, Yeah, I think biomarkers has sort of been my north star and along the way. And then for.
From there went and worked for Nestle for five years or two years and sold more of the applied side of, of nutrition, you know, how you food products and modify health. And you know.
Speaker B:Well, that, that's already pretty interesting because I, I think most people, when they think about Nestle, they think candy bars. But of course Nestle has just an incredibly deep amount of research going on.
Speaker A:Yeah, it was a. Yeah, it was an incredible research center. It is an incredible research center.
And one thing I do remember when I went there for my job interview, there was chocolates there. You know, you could have like, you know, I don't know if they still do that, but that was a very impression, left a big impression on me.
Speaker B:But when you're a young scientist, that is a huge draw.
Speaker A:Yeah. So.
But yeah, I think what was fascinating about the research center which stuck with me was it's multidisciplinary, it encompassed nutrition, whether it was food science, food technology, as well as the whole sort of field of biomarkers and how you could measure the effects on health. So I think that is something that stuck with me too, is just the team science aspects of how important it is to work with other disciplines.
Speaker B:I'm so curious about how you went from first studying and learning about fundamental aspects of how we detect what's happening in the body and how we alter that with nutritional programs to then building entirely new technologies.
And so I'm curious about, was that an explicit mind shift or that you like, wow, I really wish we had better tools or how did that transition come about?
Speaker A:Yeah, so I think it was, it was gradual. So I.
One thing I came to appreciate through nutrition research and still do is how, you know, it's not one measurement that determines your health, it's not one nutrient that determines your health. You know, sort of the balance of nutrition, different, you know, different vitamins, you know, so forth fat, protein.
So it's a sort of a system of systems. And then the Measurement tools are so important.
So sort of like my mindset was already kind of like, okay, all these things have to work together to produce a perfect outcome or near whatever. Near a perfect outcome.
But I think the thing that's hard when as a biologist you're, you know, for me this is what kind of drove me, I think into the new technology space was that you are, you know, you're, you're, you're, you're limited to the tools that you can buy off the shelf. And so, and you always just have more questions, you know, how, you know, if I could, one more measurement, I could do this.
But you're still limited by, you know, what's available. But an opportunity came to move to the US and moved to ge.
And yeah, at the time it was like when I first started considering ge, I didn't realize what.
Speaker B:Right.
Again, when people think about ge, they think light bulbs, but they forget that GE has this whole other aspect to it with MRI scanners and so much wider. And also where you went in GEG research. Most people are probably not familiar with GE research.
So tell us a little bit about what GE research was and its mission and how it's sort of, sort of academic, but also sort of industry and living in the middle.
Speaker A:Yeah, I think it's a very unique model. I mean often it's really compared to Bell Labs, you know, was one of the last industrial labs in the country.
You know, it's thousand scientists, engineers on site joined with the, I joined the biology biosciences group which GE had just bought Amersham and you know, known for pharma tools, molecular imaging.
And so we were building a new department and but the fact that there were so many skills that came from other backgrounds that were non biology actually started to bring together a lot of new ideas. We were tasked with, our mission was to, to do for pathology what had been done for radiology in terms of going from film to digital radiology.
And so our, our mission was to go from slides to digital pathology.
Because the problem with pathology as you know, is, you know, you've got one piece of tissue on a glass slide and that was used by pathologists to make the diagnosis. And then if there needed to be a second diagnosis or referral, then that glass slide had to go be undelivered. Essentially.
Speaker B:People were physically mailing around slides.
Speaker A:Exactly. You know, when you think about it, I mean, I know it still happens and you know, there's different.
But at the time that was the only, pretty much the only way it was done. And, yeah, I mean, and there's so much data in, in, in an image. You know, that was the thing, I think, that was compelling for us.
On the one hand, we sort of take the engineering view. Okay, we're going to scan this slide and we're going to do it really quickly. I mean, you know, we knew what we needed to do for it to be successful.
It had to be scanned in a very short amount of time. And then all the QC around that and, you know, there was a big.
Speaker B: at. So timeline wise, this is: Speaker A: It's: Speaker B:Wow. And so, so just even the, the technology to take a picture and digitize it was not what we're accustomed to today.
Speaker A:No. And, you know, the team was just like, I mean, it blew my mind, you know, that the. None of us were pathologists. We were not trained in pathology.
There were certainly people on the team who had expertise and so they were able to, you know, sort of help guide at least the initial questions. But, you know, we worked with pathologists who helped guide us in terms of what they needed.
And we watched their workflows and went on site to look how they, how they did things, what the, you know, all the barriers along the whole workflow, how long it took to read a slide. And they're just very simple things that, you know, just get. Gave the team the ideas that needed to be implemented.
So, you know, and I think that sort of quickly evolved from just digitizing the slide to the imaging of the slide of the proteins and the information that were sort of locked within the image. And that sort of was the eureka for me at first, you know, I was just fascinated. Okay, this is cool.
I had never worked in pathology, but the fact that this was, you know, these are, you know, they're unvisual persons, so you can see there's information there, but, you know, what does it mean?
Speaker B:Can you maybe walk me through a little bit that, that transition from I'm just trying to, you know, make pathology digital to I, I actually see that there's this space of spatial biology and molecular spatial biology that happened in your.
Speaker A: of that work is going back to:And, you know, we were certainly, you know, just didn't know what to expect in terms of how it would look. The early images were, you Know, not great quality. And so once we kind of got.
Speaker B: So it was already: Speaker A: produced the first images in:I mean, we were doing several things at the same time and you know, some of the team were focused on the chemistry, you know, how do we exploit the fact that these dyes are not stable. And so we had an amazing team. Anuxud, you know, he came up with the original concept of how to, you know, inactivate the dyes.
And so we, you know, we tried other methods too. We, you know, we tried light, we tried, you know, different compositions.
But the, the dye, the chemical inactivation was the most conducive to doing a whole slide, you know, versus light, you know, it's more focal. And so, you know, those first experiments, you know, we, we had to prove out how many times could we iterate over, you know, 100 rounds of staining.
And I remember another team member, Max Seal, he, he, he said, let's just keep going. I'm just going to keep staining and staining and staining. And we were like, okay, are you done yet? Are you done yet?
ll. This was, you know, maybe: Speaker B:Oh my gosh.
rst multi omics paper was the: Speaker A:That's right. So we actually. Yeah, so Michael Gurde, he's another critical team member, he had the vision around combining it with fish and amongst other things.
So we had a lot of really great brainstorming during that time and then we were trying to couple it with the imaging workflow. So there was always this what if, but we need to do this, but we need to register the images.
And so we had, you know, another engineer in the team, Kevin Kenney, who was helping us register all the images. And then there was the auto fluorescence. So we knew that that was going to be a co founder.
et I think probably was about: Speaker B:On just to just explain for a second, sort of come forward in time just what the general workflow and approach of cell dive multiomics is and then we'll go back and dive Into a little bit more how we got there, Because I think this concept of iterative staining and dye inactivation, how does cell dive actually make a multiplex single cell measurement?
Speaker A:So in conventional immunohistochemistry, staining standard, what you do is use an antibody that's labeled, has got a tag on it, some kind of signal that can be detected by light.
And the basic principle is you use an antibody, but attached to a fluorescent dye, and then that antibody binds to the protein on the tissue, so it lights up wherever that protein is on the tissue, somewhere in the cell or multiple cells. And so that gives you one photograph or one picture.
But then you can get rid of the fluorescence dye that, you know, the tag that's lighting up that cell. And then you can add another antibody with maybe the same dye or a different dye.
So you're basically taking multiple pictures of the same sample in an interest of workflow, and you're inactivating the dye every time it dies. And that way you're building up a montage of pictures of the same sample.
Speaker B:And then because you register them, you're then able to say, I had this marker, There was this much of it in this particular cell at this coordinate, and then marker 2 and marker 3 and marker 4. So you build up this data cube where you've got, you know, some large number, 40, 50, 200 different proteins measured on a per cell basis.
Speaker A:Exactly.
And I think that was sort of one of our key things that we got into early, because there have been other multiplexing methods before, is that did things a little differently, either with light or had a chemical removal step. But we, one, we needed to know we needed a dental process because we needed.
Sample needed to survive, you know, 30, 40 times where it was going to be stained and restained. So it had to be a gentle process. And then secondly, we didn't want to.
We wanted to work at a single cell level rather than just cover kind of a lot of data, pixel level. You know, that would be hard.
Speaker B:That was an explicit choice to say, we actually want to view this from a cell perspective, not just a pixel perspective.
Speaker A:Exactly. And I think the first thing that drove us as well was we were really thinking from a clinical point of view at the very beginning.
Meaning, you know, if you take breast cancer, for example, you know, that takes, you know, her to estrogen receptor. There's three or four proteins that are important from a diagnostic perspective.
So our view, at least in the beginning, was, well, what about, wouldn't it be cool if we could do all that on the same slide. That's not actually where we ended up, but that was our mindset that we wanted to measure at a cell level.
It wasn't even at the very beginning, a single cell level. That came a little later. We were just looking at sort of overall signal within a sample. You know, was it above a threshold or below a threshold?
And then. But then we realized, oh, no, we need to actually go single cell.
Speaker B:So that's really interesting because there was a. There's a moment in time, and I think my lab was.
Was converging around this moment in time where it felt like as a community, we came to appreciate tumor heterogeneity. And so I'm wondering if this. If it was that moment where you're like, oh, actually, it's not just the total signal.
We've got all these different variations from cell to cell that we want to capture.
Speaker A:Yeah, I think that was. We once. We sort of.
ively with Eli lilly in about:We had sort of rudimentary tools developed, but we. We needed to show them off.
We needed people to sort of see, try to understand from a pharma perspective what was important from, you know, drug response perspective. And heterogeneity sort of slowly evolved out of that.
But once we knew how to do single cell analysis, we were able to start asking those questions because you could see it in the images. It's not like, you know, sort of start to look at these stains or relate with each other. You're like, oh, my. There's a cluster of cells there.
And there's. There's sort of similar cluster cells there. And, oh, there's immune cells there too.
And so you start to think about things in a little different way that we'd ever kind of thought of before this. And. And then how do you quantify this? It's like that was our next kind of conundrum. How do you convert what you see into a number?
Speaker B:That's right. Because the difference between, hey, percent positive, there's no quantitation.
There's just like, is it above threshold or below threshold or percent positivity. It's a really straightforward quantification.
Speaker A:Yeah, exactly.
And I think that's where we were still kind of thinking, I suppose, bound by thresholds, you know, that being the only way to understand like the biology, you know, above this, then it means this. If it's below this, it means that. And so we're able to segment every.
Speaker B:Cell in the image, which is a non trivial problem. Like that is an incredibly hard problem. And for some technologies today, that is still a mostly unsolved problem.
Speaker A:Yeah, that was really challenging. We tried to do something fairly straightforward, which was to use a membrane marker.
So we had DAPI for the nucleus, we had cytokeratin and we had the membrane marker. And so we had sort of the key elements of a cell, if you like. I mean it wasn't perfect, but we were able to at least segment the cells.
And what we realized in doing so was that you could conserve the spatial coordinate and generate an ID for each cell. But we still didn't know what to do next. We had a scientist on the team who was working with another scientist.
This is often how things work in that environment. Who had worked in geospatial map. Colin McCall, I was his name.
He, he, he said, okay, let's talk to Tom, because Tom has done this geospatial analysis, I think that might help us.
So we explained to Tom what we wanted to do, that we wanted to map the cells and look at their, you know, the neighborhoods or see what they were co localized with. And Tom said, oh, that's easy. You know, that's, we do that all the time.
Speaker B:We frequently register thousands of images together. No big deal.
Speaker A:Yeah. And so we were like, okay, here's our data set.
And then he came back like the next day and showed us, you know, again, very rudimentary co localization of the different cell types. And we were like, oh wow, that's, that's what we wanted. And so that sort of was our, that was our step one. But they were all really long steps. Yes.
Speaker B:And was there, was there a moment there and, and when was it that you sort of got that data set that you're like, wow, this, this works?
Speaker A:I think one of the Eureka moments was we were working with Eli Lilly at the time. They had given us samples that, you know, were drug treated in some way. So you know, in theory there was going to be a heterogeneous response.
It was, you know, we kind of knew that going in. And the question was whether we could quantify that or not.
Taking all the learnings that we had around single cell segmentation, mapping them Back to the image. And then I remember another one of our team members, Megan Rothman, she had applied a clustering K means clustering method to identify locations.
So to find the common groupings of cells or cells of common characteristics. And then because we have the spatial location, we could see where they map back to. On the. On the sample. And sort of was amazing.
Like, there was sort of. You could see this gradation of, you know, where there was, you know, pockets of, you know, either response or cell types.
And so we thought, wow, that. That looks. That looks really cool. And then I remember we showed this to Lily and they were, wow, this is the Holy Grail.
We were like, oh, if it's the Holy Grail, that's amazing. If that's the Holy Grail. And I remember Chris Vinsky coming in one day. He says, hey, we really need to start adding immune cells into our workflow.
Speaker B:Because you were focused so much on the epithelial cells, probably not even the endothelial cells. You were focused on the cancer cells.
Speaker A:Exactly. And, you know, okay, let's put them in as well. And then we started getting these beautiful images.
PNAS paper that published in:And so we're like, oh, this is actually really big. This, you know, we need to understand. We need to have immune themselves as part of every study that we do going forward. And so, yeah, we were.
We were. We were having fun with this, but we also knew we were. We were working towards. We needed to develop something, you know, for GE as well. You know, we.
Building a workflow, and it needed to.
Speaker B:Evolve from being a, hey, look, we can do it into something that could be more mature and productized and.
Speaker A: arion, which GE had bought in:And that was sort of our technology transfer moment that we. We started to sort of think more practically in terms of how this could work in the clinical context.
And that was also very important from a QC perspective, because now we really have to start thinking about workflow.
And the team members, Alex and Bob, and they really guided the team in terms of how do we make something that can be used in a very robust, reproducible context. So that was another big transition.
Speaker B:It sounds like there was a first vision which was just how do we make pathology digital? And then that evolved into how do we get more information density into, into pathology?
And then you accomplish that and then had sort of two parallel paths. One was let's use this and learn things and do science. And then the other was let's continue to advance and mature the technology as well.
Speaker A:Yeah, exactly. I think. And that was, you know, that was sort of, you know, moments where we were able to do both or.
And then, you know, we started to work with Michael Bernie started working Vanderbilt and we got one of our first NIH grants with, with the team there and Bob Coffey. And, you know, it was a matter then of being able to prove out the value for the research community as well.
That was one of the first single cell analysis programs that the NIH was running.
I think that was maybe: Speaker B: rom the start of the journey,: Speaker A:Yeah, exactly.
Speaker B:If you think just in the last sort of three years, what, what studies, what projects have you been most excited about?
Speaker A:I think that one of the biggest, most exciting areas that we've worked on in the last, actually it's been seven years or six to seven years, is the HUBMAP program, which is the human Biomolecular Mapping project. This is a massive effort that sponsors Bangs. I mean, we have hundreds of folks on this project.
We're mapping the human reference map for healthy organs.
Myself, together with upit, we're leading the skin atlas and creating spatial map of cell types using spatial protein, spatial proteomics, cell drive, codex, other platforms, similar thing, map out all the different cell types in our healthy organs, healthy reference organs, and then also combine that with spatial proteomics, spatial transcriptomics, as well as single cell signal nuclear RNA seq. So really multimodal, which I think is where, you know, if you'd asked us 10 years ago or plus, you know, where, where were we going?
I mean, in terms of building these types of spatial maps. And you know, that would, we wouldn't have predicted that. So I think that in and of itself has been very. We can chat more about that.
But that's been super exciting because it's really starting to show the value of spatial mapping versus dissociated analysis, which is still extremely important. From, for cataloging and analyzing all single cells in any given organ.
But with the spatial methods, you're really seeing the different layers of cell types, the cell interactions, and then combining that with spatial transcriptomics, then you're really exploding the data.
Speaker B:This sounds like a gargantuan program. Tell me a little bit, how did you become involved in this program?
Speaker A: e, was first approved back in:So we. One way is about setting up the data infrastructure.
All the data is being processed centrally, which is also a huge asset from a community perspective because it's allowing the data that will be used by the end users. It will be, you know, we'll have all the provenance and quality control, and it'll be, you know, have a stamp of a QC stamp on it.
And we'll also have data that's been analyzed by each of the groups themselves. But this was a massive effort to create that infrastructure, that machine, if you like.
We're now in phase two, which is the production phase, where we're gearing up, ramping up all the kind of. With the standardized workflow. So ironed out a lot of the kinks in phase one also.
The other interesting thing about this program is that not only are we working in 2D, but we're also working in 3D. And so that's another cool aspect to this where.
Speaker B:So you have multiple sections of the same area so that you're able to dive. Dive through different planes.
Speaker A:Exactly. So in our hands. So we, yeah, we published our skin atlas a couple of years ago. It was in 3D. So we did serial sections, 26 serial sections that were.
Each section was multiplex. So we have 14 or so cell types per section, serial section. Those are registered, and then a 3D volume is created.
So that work was done by Sonia Bose and our team, and he. Then we worked with Indiana to create the tools to interrogate in 3D the distances between, say, endothelial cells and nearest.
Whatever, you know, your hub cell is. So in our case, we're looking at endothelial cells versus immune cells.
That was really at the, you know, the sort of the proof of concept that you could take this data, register it in 3D and now start to do some analysis.
So that was the basis then for a bunch of tools, including Cell Distance Explorer, which was part of PubMap, which is now being used across all other organs in, you know, with different questions about different cell proximity. And so I think that's the other important thing about that. I'm happy to jump into any aspect of it.
But the 3D was important because whilst we're used to 2D and we're familiar with looking at 2D images, organs are 3D, of course. And so having a 3D reference map is, you know, it's an important step. It's not and it's not the end step.
There's going to be much more to do in the future. But that was the goal of the program to try and all to do some degree of 3D.
Speaker B:That sounds incredible. I'm curious just from. With methods like cell dive, you have to define upfront what markers to look at.
And so how did you go about the process, I guess first of deciding that you wanted to look at skin versus liver or lung or you know, pick your organ and. And then two, how did you pick what markers to look at even within. Within skin?
Speaker A:Yeah, it's a great question. So, so for skin, I mean, skin is the largest organ in the body. So we are collaborators at yupic.
We've worked with UPIC for a long time and very fortunate to have great collaborators there and you know, discussing with them, you know, there was any number of organs we of course could have gone after, but we decided skin because, you know, skin disease, inflammation, skin cancer, it's a huge issue that affects everyone at some point in their life. And it's also very accessible.
I mean, we use samples that in the first few studies that were archived, but then later we were able to get biopsies, which was really nice, from donors who provide us with samples from sun exposed and chronically sun exposed, less unexposed regions younger and older. So we're able to, we're not only doing an atlas, but we're also, at least at a small scale, able to compare anatomically within donors.
And so it's sort of, we wanted to bring science into it as well. That was really came about that whole thinking. You know, we work with Dr. Luffelo and Dr.
Arivarson and John and Ho at UPIT and they just helped sort of form these ideas around how we could approach, you know, skin as an atlas with some interesting questions as well. And then how do we choose the antibody?
Well, you know, the good thing is, I mean when you work with derma pathologists, they know exactly what biomarkers will mark, which cell types they'll you know, so we wanted, you know, all the epidermis, the dermis, the immune cells, the Langerhan cells. So, so we, so we documented with them what all the likely markers needed to be. I'm sure we should include from an atlas perspective.
And then separately to this as a sort of parallel strand to HubMap, there's a. The Human Reference Atlas effort, which is led by Katie Boyner in Indiana.
And they're, you know, creating this monumental map of all the known cell types and associated biomarkers into one sort of one master database, if you like. So that again, from an external.
Both from our perspective, could we standardize which markers we were going to select for the most common cell types, map that to what's called the anatomical structures and cell types. Cell types and biomarkers table. That they were documenting with all the other organs.
Speaker B:Yeah, that's really interesting. And in terms of learning so far, I know you're still mid straight dream, but what was the most surprising result for you?
And I'm particularly curious in the homogeneity or heterogeneity aspects, particularly in three dimensions. It seems like there's potentially some very surprising results that could result from that.
Speaker A:I think one of the really interesting things which we maybe, you know, we're used to seeing sort of the classical pictures of skin in our textbooks and it's very fixed and it always, always looks the same.
But of course, when you're working with samples from younger, older, sun exposed, less exposed, then you start to see things like, you know, the skin cell damage and immune cell infiltration. And you know, in a couple of cases there were. There was like a minor cyst in the sample and just so, you know, blood of immune cell.
When you come to appreciate how important, of course, skin is and how well designed it is from the vascular perspective and fatigue and just, you know, the hair follicles, you know, the different cell types, stem cells, immune cells. So I think it's sort of, you know, we probably oversimplify skin, but yeah, there's 20 billion T cells in skin.
I mean, we, we caught, you know, maybe one million.
Speaker B:Wow, that's. That's incredible. That's utterly incredible. They.
So I, I guess I'd like to maybe change gears a little bit and just talk about with all of these learnings, what you were doing in hubmap was very much basic science, learning the foundations, but with these areas of looking at damage and you're transitioning to thinking about the more clinical side of spatial proteomics. And so I'd love to hear your thoughts on where spatial proteomics is going from a clinical perspective.
What are the questions that we're going to be asking with it? What are the places where this might be part of clinical workups in the future?
Speaker A:I think, you know, what's really stayed the course in terms of interrogation, maybe clinical value, translational value is immuno oncology.
Speaker B:And.
Speaker A:Being able to classify or to image different immune cells and you know, their exhaustion, their functional status. I think that's still a non, it's still, it's not in the clinic yet. I mean it will take some time, but it will.
To me that's still probably the, one of the most promising areas as we try to understand why patients don't respond to immunotherapies. I think it still takes more time.
I mean there's work on going on in simulation and modeling to understand, you know, you know, why some tumors are more immunoresponsive than others. You know, what changes that status over time? Why is, you know, our older versus younger patients maybe have a different type of so on.
So it's still, I think there's still a lot of research to do.
On the other hand, you know, one of the other, one of the earliest projects, clinical projects that we were peripherally involved with was with clariant back in 10 years ago where, 12 years ago actually where pathologists we work with there felt lymphoma was really a perfect use case for multiplexing because it can take 20 to 30 different serious actions to determine what subtype of lymphoma. There's over 30 cells, subtypes of alevo of lymphoma.
So, you know, you need to do a lot of sectioning in order to find out, you know, what type of lymphoma that is, meaning flow cytometry is possible. But you still, you know, you might not have enough sample to do that.
So I think, you know, I, and we, you know, at the time there was a test that was launched for lymphoma, but I feel you might have been a little early and I think that there's still a need. We, you know, in the, in the, in the later years, we, it does come back up again that lymphoma is one that still could be, you know, very valuable.
And the good thing is that there's a lot more people working in this space now.
Speaker B:So is the thought then that, that in the future it may be that we instead of doing sort of normal single marker pathology, we instead might move to a world where we're, we're doing, you know, full on spatial proteomics and learning much more sophisticated patterns than just how many cells are expressing X marker.
Speaker A:Yeah, I mean, I think there'll be a few steps still to get there because it's the relationship between cytotoxic T cell infiltration and outcomes is still a very powerful predictor. We see it in all of our studies, you know, even depending on how many patients you have.
But the trend is always there that the higher the cytotoxic T cells, the better your outcome.
But then when you start to split it down a little bit more and you start to look at, let's say, the phenotype of those immune cells, you start to see a little bit more gradation there in terms of, you know, okay, the patient. There will still be the patients who always do worse, who have, let's say, cold to tumors.
But then I think you can split it out further with additional information about those biomarkers, about the functional status of those immune cells.
And also, you know, as we get into, you know, the different types of immuno oncology therapies, knowing, you know, what type, you know, the exhaustion status of those, are they functional, are they able to actually so you could imagine, you know, I don't know whether it'll be a 30 marker assay, but it probably would be something closer to maybe 10 at least where you're really, you need to get sort of, you know, very automated counts. Yeah. And for sure there's also, and we showed this actually in one of our papers we published last year on colorectal cancer.
The tumor cells themselves, you know, they have to be responsive to the effects of the cytotoxic C cells.
And if they're what we showed in a couple of papers, one was both with the World Cultural Surgeons in Ireland and Queen's University Belfast, we looked at heterogeneity of apoptosis sensitivity and response to chemotherapy. So there's sort of this persistent apoptosis resistant cell cancer cell type that may be not responding to chemotherapy and may also.
And in another paper we show that the proximity of cytotoxic T cells to their neighboring cancer cells, the neighboring cancer cells, you can directly quantify the level of Caspase 3 level.
Speaker B:And so, so there's, there's a neighborhood effect. So. So me killing this cell actually has an effect on its friends and neighbors as well.
Speaker A:Yeah, and I think that whole space is very fascinating because, you know, not work we've done, but others have shown you know, the sort of the, the killing time of a cytotoxic T cell. You know, this sort of takes three shots. And you know, not just the proximity even you have to induce.
Speaker B:Well, and I think what you're talking about goes even beyond immuno oncology, because what you're referring to there's sort of the historical dogma was hey, I have a kinase inhibitor or I have a cytotoxic agent, and then it gets to the cancer cell. And the cancer cell is just like, I'm done, I'm out, goodbye.
And it dies versus there being a necessary interaction with the immune system, even for drugs that are acting on the cancer cell.
Speaker A:Right, Exactly. Yeah. The other very interesting thing, I think is that a lot of these studies that we and others have done, these are adjuvantly treated patients.
So the cancer is actually removed, or at least the colorectal cancer. It's the residual cells that we're making a prediction or that we're making predictions.
Speaker B:On a tumor that was, that's not there anymore.
Speaker A:So, you know, the sort of very interesting thought that invasion has already occurred, but still we can make some level of prediction. And maybe that's because in the end, maybe the tumors are more uniform from the perspective of, you know, what's left behind.
Speaker B:That's an interesting deep question because you're, you're wrestling with the reality that the microenvironment and the metastatic niche is very different than the microenvironment in the primary tumor.
Speaker A:Right. Yeah.
Speaker B:And yet you still have this cell endogenous characteristic and how much is it influenced by its friends and neighbors and when.
Speaker A:Yeah. And how much training has the immune system already received? There has to have been training that is long lasting. It's a fascinating.
I don't know how to solve it. So it's like.
Speaker B:Well, that, that's actually, that's a great segue. So my, my next question for you was going to be about what can't we do yet?
So the technology has advanced incredibly over the last 20 years, but there's still things that we might like to be able to do that we can't do today. So where are the limits of today's technology that you like? Okay, 20 years from now we should be able to do X. And today it's a problem, I think.
Speaker A:Speed, throughput, you know, those are, and cost. I mean, they're, they're, they're not the most exciting things to solve, but those are the ones that, those are, those.
Speaker B:Are the particular pain Points that, that when you think about the technology achieving ubiquity, those are the, those are the big barriers today.
Speaker A:I think, you know, you're competing today against single marker and you have to, you know, you have to prove out the values and we've seen scenarios where you know, you're willing to wait as a, you know, if you're getting good information and it's getting, you're getting actionable information, sequencing is, you know, still takes time to do a gene panel. So you know, it would be down to the value of the data that you're getting as well.
So 20 years from now, I mean it would be amazing if it was as common to do multiplex imaging as it is to do flow cytometry or sequencing that it's just built into the workflow. I think we would need to have, you know, you know, a sort of, at least a sort of a 24 hour turnaround.
I think we, things like HubMap, HTAN, Human Tumor Atlas network and these big efforts will to some extent normalize how we think about these new technologies. These are going to be massive amounts of data that are going to be available for further research.
And I think that's a tipping point too because if you look at cancer genome atlas.
Speaker B:That's right, the value of that has been exponentiated because it's been analyzed and reanalyzed and reanalyzed a billion times by so many different researchers and brought in to supplement so many other parallel studies.
Speaker A:Exactly. Yeah. And I think you know, that's in a sense what it's going to take as well to get us to the aha's.
Yeah, I don't think we've even our brains have even got there yet.
But I can see, I mean there's you know, heterogeneity types of tumor cells themselves, the different types of immune cells are present or infiltration location, you know, those are being studied a great depth I think across you know, many publications starting to come out now as well as sort of like that more basic level of research of understanding, you know, what the transitions from early stage cancer to aggression, you know, they're having these types of spatial insights will give us new idea, other new ideas that we haven't been thought of.
Speaker B:So on that topic. So my last question for you today, you mentioned soumya earlier and this of course the topic du jour is AI and how this interacts.
And so I'm curious to hear your perspectives as we start building these libraries, as we start building these resources as we start looking not just in 2D but in 3D. What are you already seeing in terms of AI tools coming to bear and what do you envision for, for AI in the future?
Speaker A:Yeah, I mean, somebody goes, he's the right person to talk about this. He's one of our team members and has always been the head of the curve on how we can analyze this data with the latest AI methods.
I mean, some of the stuff that we've done in the last few years have include the GAN approaches, Generative Adversarial Network where using that to simulate more data beyond what we even have.
But I think it's the general idea, you know, and we published that with Emory a couple years ago, looking at H and E images and creating additional patches derived. So you, you start with N number of samples and you know, let's say 20 or 100 samples.
But then you can start to identify which regions within the sample are associated with aggressive disease and which are, let's say more innocuous regions. So you're trying to essentially balance your data set by creating more of these aggressive patches.
And you know, one of the problems with these types of analysis, predictive analysis, is imbalanced a very small amount of aggressive regions and then you have a lot of normal. So I think it's sort of fascinating to think that we can, you know, create more data. You know, you sort of wonder, okay, is it, it's real?
Because it's based off of something that was real. But you know, how far can you go? And I, you know, that's, I guess a general question for the field, but.
Speaker B:It'S a deep question because on it, it was actually quite a surprising result that you could, could add noise, rotate, do these very simple transforms and the AI methods would actually learn better. And as a human we're like, well, it's the same image, it's just rotated 90 degrees. Why should the AI do better? And yet they did. It was astonishing.
Speaker A:Yeah, no, it is, it is astonishing and as well, because these technologies are expensive. So can you create more data, you know, without having to spend a lot of money? But you know, the question is how diverse can you make your data sets?
And you know, are rare cell types, they may not be captured. So I think it's, it's sort of got its place depending on what your question is.
And then I think also, you know, the deep learning methods for cell analysis and you know, features within tumors using the H and E, that's another in a whole other area. But you know, there's A lot within the H and E image that we, you know, is untapped as well. So that could also be, you know, very nice hybrid.
But yeah, I think AI can help on all those ones. Also the large language models in trying to just distill down, you know, what information, actions, you know, what does it all mean?
Speaker B:That makes a lot of sense. So I'm going to put you on the spot with my last, last question which is really again a little bit of a forward looking.
If you think about yourself 10 years from now, 15 years from now and you think about, you're looking back, what have you learned? So you've had all of this time, you've had all this infrastructure, you've done so many of these big projects.
What is the, what is the, that exciting thing that you've learned over the course of this journey?
Speaker A:Probably the immune response is really something very remarkable, not surprising, but how much you know, I think there's so I hope in the next 10 years that that's even more solved than it is today.
But just when I look back, you know, over the last 10 years when it was even maybe more than last 10 years, maybe that's 15 years ago, it, we did not even analyze the insights.
Speaker B:You just ignored them. They were just, they were there, they were annoying, they were like no, no.
Speaker A:You don't, they don't need to look at those. And then.
But I think just, it's just profound how that's changed and just the, you know, the variability amongst the different, the functions of the different types of immune cells, their interaction, their cross talks, their metabolism, how that plays into vaccines for, for cancer vaccines. I think, you know, that's still to pan out, but I hope that's another holy ground.
Speaker B:Anything else that you're, you're excited about as this field develops and evolves?
Speaker A:Yeah, I think the really wonderful, phenomenal thing is just all the next generation of scientists, junior investigators who are now working on this data in totally different ways. You know, we see this on hopmap with the junior Investigators and their Jumpstart program and just phenomenal.
I mean they're coming up with amazing ideas around analyzing cells, neighborhoods.
You know, you've got people like John Hickey who's now in Duke and Chen Shenzhou in Stanford and others and that there's like a list of folks who just, they get it and they're just thinking in a totally different way. And I think that's going to change the field as well. So that's, that's very exciting. I can't wait to see what they do in the next phase.
Speaker B:That's awesome. Well, thank you. Thank you again.
So for our listeners, you've been sitting here chatting with Fiona Ginti, had an incredibly fun conversation, learning about the Hubmap Consortium and the role of spatial proteomics in the clinic and where it's going, as well as the future involving AI and the the immune system in cancer as well. So thank you so much, Fiona, for joining us today.
Speaker A:Absolutely. Thanks, Tarek.
Speaker B:All right, thanks, everyone, for joining us. And I hope to see you again soon on another episode of Translating Proteomics.
Speaker A:We hope you enjoyed the Translating Proteomics podcast brought to you by Nautilus Biotechnology. To contact us or for further information, Please email translating ProteomicsAutilus bio.