Welcome, everybody. My name is Fabian Schneider. I’m the Product Manager for the research product at Visiopharm, and welcome to my talk, Multiplexing Phenotyping, Enabling Tissue Microenvironment Insights.

My talk is split into three main sections.

I will start with showing you a platform capabilities overview, then some examples for multiplex image analysis and the possibilities within our software, and then demonstrate quickly our interactive data insights tool, which is a new capability feature that will be released soon.

Let me start with the platform capabilities overview.

Let me briefly introduce you the Visiopharm platform. The Visiopharm platform is an easy- to-use toolbox for image analysis scientists and you can use it across all modalities of images and even file formats.

All outputs that you would like to generate are fully customizable in the module and no coding is required, which is great for biologists like I’m You can extend your platform with modules like tissue array, where you can de array tissue microarrays and give each core an individual analysis.

You, can extend it with multiplex phenotyping, what we will be talking today about.

You can use, the Tissue Align module to align serially cut sections and to, virtual multiplex stains if you have several of your stains in order, and you can use, the deep learning platform, that we are having by extending your core platform with AI engine and AI author. If you’re not using the multiplexing assays on your tissue samples today, you know the pain of how much tissue you need to consume to, get to all the biomarkers that you’re interested in in your, studies. As you will have to serial co register all your biomarker sections, This will make it impossible to understand the tissue microenvironment in detail.

It’s then really difficult to understand the spatial distribution of your cell populations and their cell cell interactions, and that means that you’re lacking the granularity of cellular phenotypes, like which lineage markers express which functional markers, so where are my pairs of receptor and ligands. And answering in-depth hypothesis with low plex biomarker data is really difficult.

And it’s then really tricky also with statistical, methods to obtain novel biomarker signatures. When you want to analyze your multimarketer images, you need to undergo several steps that you can either do in several softwares or in our platform in one workflow.

So you need to do tissue segmentation.

You need to do a robust cell segmentation that accounts for the different sizes in the tumor or tissue microenvironment of your different cells.

You need to do the phenotyping, so you need to have the mask of your cells and use the underlying channel information to get to all the phenotypes.

You need to be able to give spatial readouts and intensity readouts and x y positions.

And you need to be able from the platform to export this data for downstream on statistical analysis.

And our workflow is that you can use this in an app sequence, where in the first app you do the tissue segmentation, in a second app you do a robust cell segmentation, and then a third app you would do the phenotyping individually for the assay that you, and the marker combination that you’re using. And then you can do the data export.

So taken together, Visiopharm Phenotyping offers you a combination of machine learning and AI deep learning for tissue and cell segmentation, automated phenotyping, setup of output classes to your needs, and also the export of all raw data. After showing you the capabilities, I will now move into some example of multiplex imaging analysis.

The example image analysis I want to show is coming from an Ultivue io eight, image that was provided by Ultivue.

It has eight biomarkers as a ready to use kit. And for more information, please visit the website from Ultivue or their boost.

And what we did is we we looked at all of these biomarkers, but we want to show you one question that you might have when you are looking with into these biomarkers. And the question is, where do CD8 and PD-L1 positive macrophages and cancer cells interact with each other. So where are the cytotoxic T cells and where are the prime cytotoxic T cells, as shown here in green with which express PD-1, and where are my macrophages and tumor cells and which of them express PD-L1 in my case, and where are they at distance wise.

The first thing in the app sequence is that you need to, detect tissue and do your, tissue compartmentalization.

So for this we have used an app for the tissue detection in green, and then did a manual drawing of the region of interest for the delineation of the tumor versus host tissue.

And that can also be done by an expert pathologist, either directly on the MIF images, or on an agent E image that, Ultivue, kits allow to do an agent E in the end. And then you can transfer the annotations from the H and E onto the masks of the IF channels.

Next would be that within your region of interest, so your tumor annotation, you would do your tissue segmentation by segmenting your cytokeratin positive tumor cells here using the keratin layer, from the IO eight plex, but you can also train a deep learning classifier to detect the tumor cells versus the stromal cells.

When visualizing eight markers at once, it’s almost impossible for me to understand what is going on in this cancer case. So when we are zooming in and look at only the lineage markers, we can see that the tumor is delineated and almost shielded from other cells with, a line of CD68 positive macrophages, CD4 expressing T helper cells, and intermingled in between you have CD8s, and the CD8s are also recognizing, as it seems, the tumor, and are invading into the tumor epithelium, and maybe doing their job by, killing the cancer cells. When, at this, spot, we are doing now the cell segmentation and therefore doing AP2.

So here we have already done the nuclear segmentation and now did already the dilation of the cells to show the whole body of a cell.

And when we are then adding AP3 on top, then we can do the, phenotyping.

And here we have already started phenotyping the lineage markers, and this is a similar image to what we have shown in the beginning, but now we have already implemented one functional marker, which is FOXP3, to show us the t regular cells here in blue. And when we add the other two membrane, markers, PD-L1 and PD-1 for the functional system, then we can, mark them, for example, by only showing them within the cytoplasm of the image object that we have identified. So in red you see PD-L1, so you see a lot of PD-L1 expressing macrophages.

You see, in this light green PD-1 expression, and there you see that a lot of your CD8s are expressing PD-1, and that many of these CD8s that are on the tumor epithelium or within, that they are also expressing PD-1.

To reduce the complexity of this cell segmentation, which is also overwhelming for my eyes, you can only show, for example, these dots. And if you then remove all your image channels or, do it with only one, then you can look at this distribution of all these beautiful colored, boxes.

And then it’s much more easy for me as a human to understand where are my CD8s, and that they are really and clearly trapped on the one hand outside of this line of PD-L1 expressing macrophages, and then you clearly also have them inside of the tumor epithelium.

After we were done with the basic image analysis and had all the phenotypes, and we wanted to know where are these phenotypes of the macrophages, so the CD68s and my cytotoxic T cells, and where are they within the whole tissue? Are they homogeneously distributed or not? So cytokeratin, just as a layer mask for you to visualize where is the cancer, shows that the CD3, CD8 are mostly kind of homogeneously distributed within the tumor epithelial areas and around it, and have this one very very dense high spot.

So to understand the legend is that the yellow areas mean that we have a density of around three hundred cell per square millimeter. When we add the function markers, we see that we only have down here in this spot of the CK, we have, PD-L1 expressing cytokeratin positive tumor cells, while we have in the same area also a hotspot for CD68 PD-L1 positive cells, and we have all the tumor epithelium seems to be delineated with these, M2 like macrophages.

The PD-1 expressing cells are less than the total CD8, but that was also expected from me.

And you see that the hotspot for the PD-1 is the same as for the total CD8 on this tumor, and that the distribution of the PD-1 double triple positive CD8s is, more on the tumor epithelium than in the surrounding.

Another very nice feature of these heatmaps is that you can combine outputs. And here we combine the total PD-L1 with CD-8, CD3, PD-1 triple positives.

And what we see is that we have, three clear hotspots for, the co localization of these cells.

To deep dive even further into into your data, is that you export with all of these segmentation areas, all your numbers, into a TSV file, and then we have imported this into Excel here in this case, as you see from the visualization, and did a very basic analysis of this case.

So what we have already seen is this this case is rather a hot one, but it has definitely more t cells in the stroma than on the tumor epithelium.

You also see that what we have seen within these heatmaps is also true within real numbers, that you have more CD8 PD-1 expressing cells on the tumor epithelium than in the stroma.

While you have more immune suppressive cells like regulatory T cells or, PD-L1 expressing macrophages in the stroma than within the tumor epithelial areas.

Another very nice way to visualize this is using the radar plot functionality and where you then clearly see this shift of the CD3s, CD8 towards this, with PD-1 expression towards the tumor. And if you put on the other axis the CD68 PD-L1, so the opponent of the PD-1, then you see that these are clearly located in the stroma.

We can also look at the less abundant cells within these regions of the tumor epithelium with the stroma, and there we see, for example, that we find CD8s that express FOXP3, but more inside the stroma than on the tumor epithelium.

We also find CD4, CD8 double positives and even triple positives with FOXP3.

Now after looking at the global densities in my stroma and, tumor epithelial compartments, we asked the question like how do, the immune cells spread around the tumor in the invasive margin?

The invasive margin around tumor epithelial nests and buds is of high importance to understand where are your, immune cells that are attacking the cancer are located, and can they even move close to it?

So here we added two fifty micron margin zones to the epithelial delineation, one inner and one outer. And when you look at these tumor epithelial margin zones, you can, for example, ask yourself where are the CK positive cells? Are they only within the the tumor, epithelial area, or are they even tumor buds that are migrating away from the tumor mass, and therefore leading to a much more aggressive and invasive phenotype of the tumor.

And what you see here clearly is that the further away you get from this tumor, the less CK positive cells you find, but you find them. We then exported also this area and did the data analysis of the main phenotypes within these regions, And what we clearly see is that the closer you get to the tumor, the more CD8 PD-1 double expressing cells you find.

And it’s a bit different for the CD68 PD-L1, so the inhibitory macrophages.

So these are more abundant in the outer margin and the inner margin, and less abundant in the tumor epithelial area itself.

As said in the introduction is that one of the things that we are also interested in with this case is where are the cell cell contacts and the direct cell site contacts.

So when you do your tissue segmentation and cell segmentation, you can, for example, then identify all the c d eight’s here shown in pink.

And if you then only subset those cells that are in direct contact, so that are closer than, for example, two or three micron, then you can generate a new output class for these and export them. And then, for example, generate, this MLD file where you see all the all the cells that are in direct contact, and then you can sub phenotype that. So you can say direct contact with CD8 PD-1 expressing cells, for example, and then you will see that most of them are in contact with, CD68 or PD-L1 CD68, a lot are in contact with CKs, and so on and so on.

These were just some of the capabilities that you could do in multiplex image analysis with our platform, and there are more, things that you could do in a deep dive with more complex analysis where we are happy to have a chat about after my talk.

The what I’m now really excited about is, as the product manager, to show you the first time the interactive data insights tool that we have developed for the last year at Visiopharm.

The new interactive data inside our QCT tool allows you to plot your data of multiple images or TMA cores or whatever for which you have, single object readouts in a tSNE plot.

In the tSNE plot, you can then, interact with each and every single, data point as each of these dots that you see here is one cell object. And if you click on it, it will move in the image or to the image and zoom in and show you where it lives.

So we can then in another tSNE, we can, then also click on this orange cluster, and the orange cluster will guide us to another cell. And this cell is, for example, a PD-L1 expressing macrophage.

You can also multi select and then, for example, if you’re interested in one cluster, and see where these cells live and reside in in your samples.

And you can also zoom into the area of the tSNE plot and explore each and every cell dot dot by dot and get to know your data and see where they reside, in which image, in which row, etcetera.

One other feature that you will have at hand when you’re using the Insights tool is that you can filter your data by all the features that you have created or labels and also, tissue segmentation results. Like if you have your total tumor area here in this tSNE, you could then filter by stroma area, and then you see all the cells from this tSNE that reside only in the stroma area.

And if you filter by tumor epithelial area, then you see all the cell objects that live in this area. And this allows you then to deep dive and understand where are your cells, where might you have a misclassification, where do you need to tweak your tissue segmentation, for example?

And as this is interactive, you can always go and see where the cell is and why it was, for example, phenotyped as what, which is a great way to QC multiplex phenotyping.

Another step in the multiplex image analysis capabilities of the Visiopharm platform was the recent implementation of the native reading of the MCD files from Fluidigm from the Hyperion machine.

So you can now directly open these files that you have created in your Hyperion machine and open them in the VIS platform since release o nine 2021.

And upon import, all of these images are automatically normalized for a optimized color display.

And you can also tick on and off all these channels and reduce the number of channels and then look only, for example, here at e cat urine in these scripts in green, their proliferating cells in yellow, alpha smooth muscle actin in red, for example, and nuclei in in in blue. This allows us to move into the hyperplexed image analysis space more, smoothly. With this, I would like to summarize what I’ve shown you in this target, that the Visiopharm platform offers you a flexible toolbox to build apps and workflows for your multiplexing essays.

You can use our deep learning capabilities to, have a easy toolkit at hand for tissue and, cell segmentation by train by example.

All our workflow integrate with our multiplex phenotyping module for an automated phenotyping. Our outputs that you, derive in the apps are fully customizable, and you can answer your questions with the outputs that you would like to set up.

We are now also supporting the Fluidigm, MCD file format as the first company in market besides Fluidigm.

And we are happy to, soon release our data insights tool for all customers, which allows you for an interactive bidirectional TSNE and box plot visualizations.

With this, I’m at the end of my talk, and I would like to thank you for your attention till the end.

I hope I could spark your interest in multiplex image analysis and entering this realm of tissue analysis, and we are happy to talk to you at booth sixteen or if you like to, contact us via email.

Especially, I would like to thank our r and d team headed by, Jeppe Thargard, who made the app for the, IO eight plex, and, also which made the whole, tool development for the data insights tool.

Additionally, I would like to thank, Fluidigm and Ultivue for providing images and also for their partnership.

Thank you, and have a great digital event.