Leica Microsystems
Duration: 14:51 Min
Exploring the Tumor Immune Landscape with Multiplexed Imaging & AI-Powered Spatial Analysis
Transcript
0:01 | Hi everyone, I'm Vasundhara. I'm representing Leica Microsystems and today we'll be talking |
0:07 | about exploring the tumor immune landscape using multiplexed imaging and AI-powered spatial |
0:14 | analysis. So spatial biology allows us to study a variety of cellular landscapes |
0:21 | across different dimensions and it has applications in various fields in molecular biology. |
0:26 | So for example, within neuroscience that you see here, we can look at different cell types, |
0:32 | how they're organized within the brain for example, and then within the space of cancer, |
0:38 | we can look at how different immune cells interact with the tumor to create, |
0:48 | to really go beyond what a typical digital pathology image looks like and create more |
0:55 | of these functional atlases to finally generate data across various different landscapes from |
1:05 | subcellular to micro-anatomical structures. And each layer can be used to inform the next, |
1:13 | either through this bottom-up or top-down analysis method. And for the purpose of this talk, |
1:18 | we'll be focusing more on the tumor immune landscape side of the things. |
1:25 | So we know that it is important to characterize the tumor immune microenvironment, but it contains |
1:33 | distinct immune cell types. So for example, there's immunosuppressive cell types within the TME |
1:41 | and at the same time, these exist in combination with the anti-tumor immune cells. So they kind of |
1:48 | protect us from the, against the tumor cells. So naturally, since there's such a large variety |
1:55 | of cell types, it requires a multiplexed approach to really study such diverse cell types in order to |
2:06 | develop effective cancer treatments. And that's why we introduced CellDive multiplex imaging |
2:13 | solution to interrogate such intricacies of the tumor immune landscape. So it allows us to |
2:20 | perform hyplex imaging within whole tissue sections at a single-cell level, using an iterative |
2:27 | staining and imaging approach. And the key features are that you can look at more than |
2:34 | 60 biomarkers within the same tissue. It's tissue-preserving, so you can always go back |
2:41 | and add as many biomarkers as you want. And at the same time, you get really high signal-to-noise |
2:48 | ratio because of the calibrations that we do, as well as there's like lesser batch-to-batch effects |
2:58 | because of the increased intensity. So what is CellDive? This is like a brief overview. |
3:06 | You start with your FFPE tissue section. You identify which area that you want to image, |
3:15 | either it's the whole tissue or a specific region of the tissue. And that's where the CellDive |
3:20 | process starts, where we start with biomarker staining. So we add four different antibodies, |
3:27 | we image them, and we inactivate those antibodies and then capture the autofluorescence that can be |
3:36 | removed. And then with each iterative round, you keep adding four biomarkers in each cycle and keep |
3:44 | repeating the cycle until you have the final data set that can be analyzed using our analysis |
3:49 | software, which I'll go over in the next step. So what does a typical data set look like? |
3:56 | You start off with DAPI, look at the nucleus. We add, let's say this is a T-cell marker. We look at |
4:03 | immune checkpoint markers. So these four would be within round one. And then you can add |
4:13 | more antibodies, keep layering it. You can look at endothelial cells, macrophages, |
4:19 | proliferation markers, that's round two. And then more tumor evasion markers, |
4:25 | Vimentin, CD11C, GLUT1, which is a hypoxia marker. So you can keep, essentially keep adding as many |
4:32 | markers as possible, which are important to your study. And once we have the entire, |
4:43 | this is like the entire tissue, this is what the image would look like. It's basically like a map |
4:48 | of your tumor tissue here. We're showing a colon cancer tissue from the CD26 wild type |
4:58 | syngeneic mouse tumor. And we're looking at these many biomarkers that are segregated by the kind of |
5:08 | phenotype that they are annotated to. And what's really interesting is that if we zoom in, |
5:15 | you can really appreciate the single-cell resolution that you can achieve with the |
5:20 | CellDive hyplex imaging. And another thing is that the same, pretty much the same set of |
5:28 | antibodies and the imaging calibration settings can be applied across different tumor types. |
5:34 | So it's showing colon, lymphoma, breast. And now we come to like a more important question. So |
5:43 | here you're looking at cytotoxic and helper T cells in this field of view. And you can also |
5:50 | look at other cell types, like B cells within the same field of view. But it gets really complicated |
5:57 | when we're, let's say, looking at more than 25 biomarkers and how they interact with each other, |
6:04 | especially within the whole tissue level. So that's where we ask some of these questions, |
6:10 | like what are the molecular signatures that are associated with the TME? What are the |
6:16 | relationships between each of these biomarkers that we just saw? And what are the spatial |
6:22 | interactions? So what is the geographical information that's carried by these markers? |
6:28 | And that's where our AI analysis platform comes into play to really answer some of these key |
6:36 | questions. So our AI analysis software, it's called AVIA, which stands for AI Visualization |
6:45 | and Image Analysis. So you start off with your original image. And the first step with any kind |
6:51 | of multiplexed analysis is detecting individual cells, so really assigning them into individual |
6:57 | objects. And what the result looks like using our cell detection recipe is you get those individual |
7:08 | outlines of the nuclear and the cell membrane by utilizing all the channels that are available |
7:15 | within our data set. And then the second step is looking at what are the different phenotypes that |
7:26 | are assigned to each of these cell types. So that's called phenotyping. And that's where, |
7:33 | like here, what you can see is the yellow cells are immune cells and the red ones are the blood |
7:39 | vessels, endothelial cells. And with our AI-assisted technique, we can immediately get like a zero-shot |
7:49 | solution where it assigns the marker to each of these cell types. And then the question becomes, |
7:56 | okay, we can do this for, let's say, one marker. Sure, that's easy. But when we're looking |
8:02 | at all the different markers, we want to know what are their interactions with each other. |
8:08 | And that's where the unsupervised clustering that AVIA can do comes into play. So in here, |
8:17 | you see two of the tumor types. And we can look at what are the different cell populations based on |
8:27 | what all markers they co-express. And what's really interesting here is that we know these |
8:33 | T cell clusters and these two cancer types that not only express specific immune checkpoint markers, |
8:41 | but also express these immunosuppressive mechanisms that might be impacting T cell function based on |
8:48 | the regulatory markers that they express, like FOXP3 and IL-2RA. And it really indicates, like by |
9:04 | not just looking at like one cell type, we are also looking at how these cell types interact with |
9:09 | other immune clusters, like in this case, they're interacting with dendritic cells in cluster 5 and 9 |
9:17 | on the right figure. And really, I think the changes between these two tissues for the same cell |
9:24 | population really indicate the heterogeneity that exists within distinct tissue types for tumor. |
9:31 | At the same time, within the same samples, you see these specific myeloid clusters that |
9:38 | express M1 and M2-like functional markers, along with immunosuppressive mechanisms |
9:45 | with PD-L1 expression. And what's interesting is that both these myeloid clusters and these two |
9:52 | different cancer types express GLUT1, which is a tumor hypoxia marker. So that really |
10:00 | got us interested in looking at what are the spatial expression patterns of GLUT1. |
10:07 | So GLUT1, wherever it's enriched, those are the tumor cells that are deprived of oxygen and |
10:14 | nutrients, yet they survive. They provide some resilience to the tissue. And we are interested |
10:23 | in seeing, are there specific cell types that might be enriched in regions that have high |
10:31 | GLUT1 expression versus the ones that don't? So that was really the question. And that's, |
10:37 | I think, kind of the beauty of more of the spatial analysis to really know if there's |
10:42 | any location-based preferences for some of these markers. So in order to answer that, |
10:51 | we can look at, we can cluster cells by the level of expression of markers. So here we're |
10:59 | looking at three different clusters, a red indicating high expression, green indicating |
11:06 | medium, and then blue is little or no expression. And then we can look at what are the immune cell |
11:14 | types that are immune or like functional cell types that are enriched in each of these clusters. |
11:20 | And what's interesting here is that the high GLUT1 regions tend to express |
11:28 | a lot of these myeloid markers, especially |
11:34 | tumor-associated macrophage markers like CD206. And we took a deeper look at some of these |
11:45 | specific markers. Here we're looking at the spatial relational analysis. So it's a heat |
11:52 | map of distances on the right where the red indicates cells that are very close to the |
11:58 | GLUT1 enriched region. And it kind of shows that these tumor-associated macrophages that are noted |
12:06 | by CD206 and CD11b expression, they kind of create these hubs at these tumor hypoxia regions, |
12:17 | which is extremely interesting. And not only can we do this, we can also go beyond that, |
12:23 | really look at what are the percent expression levels of some of these interesting cell types |
12:30 | like T cells that express regulatory phenotypes as well as dendritic cells, M2-like macrophages, |
12:40 | and B cells. And we clearly see that there's an increase of specific cell types within the GLUT1 |
12:46 | rich regions versus the poor ones, which we can also confirm with some of these images that we see |
12:55 | based on our CellDive imaging results. And in conclusion, if I have to summarize, |
13:04 | the take-home message would be that we're really able to capture the tumor immune landscape using |
13:11 | this multiplexed approach. And we can use metabolic markers like GLUT1 to look at how |
13:18 | they regulate immune cell activation within the TME, like I showed with the clustering analysis. |
13:26 | And not only that, we can also look at the complex tumor immune interactions |
13:31 | within the whole tissue using an unsupervised clustering on AVIA. |
13:38 | And finally, we can note the enrichment of specific immunosuppressive cell types like |
13:45 | tumor-associated macrophages or TAMs within the GLUT1 positive |
13:51 | regions, and really that such spatial expression patterns may contribute to therapeutic resistance. |
13:59 | It's important to know how the cells are localized within the tissue. |
14:07 | And I would like to thank our Leica team at the Boston Innovation Hub at Walton, |
14:15 | as well as CST for some of the tissues that they provided for this study. |
14:20 | And if you have any kind of questions and you want to learn more about CellDive or any |
14:26 | of the spatial analysis studies, Jade and Natasha here would be really the best people to reach out |
14:36 | to. And so far, we really looked at the imaging-based methods for molecular profiling. |
14:44 | And in the next talk, we will learn a little bit more about cell and gene therapy. So I'm |
14:51 | just going to take the questions after the next talk and move on to the next one. |