Hi Everybody, and thanks so much Maria for the invite to give this presentation today. My name is Tom Villani and I’m going to be giving an overview of Cytometric Analysis of Immune Cell Populations in Archived clinical FFPE Biopsies for Immuno-oncology Research.
So, before we dive in, I wanted to give a little bit of an introduction about the company that I work for, Visikol. We are an industry leader in advanced models and tissue imaging, we’re focused on improving and accelerating drug discovery and development through contract research services.
We are an innovative and profitable company based in the US and offer research services for drug discovery, development, and clinical research. We have a worldwide customer base of startups, small, medium, and large pharma, and a number of prestigious research institutions.
Our team is composed of a number of experts in drug discovery, assay design, oncology, immunology, inflammation, liver disease, and many other therapeutic areas. We leverage that expertise to do several hundred drug-screening, preclinical and clinical campaigns per year.
We have dozens of unique technologies and patents related to histology, tissue imaging, 3D cell culture, and other kinds of advanced techniques for biological research and our 2 main pillars of our business are our advanced tissue imaging and our advanced cell culture and tissue model assay services.
So, a brief overview of the presentation today. So, this presentation is going to be focused on how we extract cytometric data from FFPE tissue and towards that end I wanted to start with a brief overview of immuno-oncology, which is an emerging, one of the leading paradigms in cancer research these days. Touch base with a brief overview of flow cytometry and its applications to cancer research. We’ll talk about some of the limitations of flow cytometry applied to clinical specimens, especially archived specimens. We’ll dive into some applications of our FFPE technic for doing cytometry on tissue sections and conclude with an overview of some of the limitations and the future direction we’re going wit this technique.
To start off, Immuno-oncology has been a rapidly emerging paradigm in cancer research over the last decade or so. And this is of course potentially unlocking the power of the immune system to fight cancer. Over the past 10 years there’s been an explosion of interest in understanding the immune system and how it responds when a tumor is growing. And there’s been a huge amount of development towards pharmaceutical strategy to leverage to immune system to help out in fighting cancer. A lot of this started in the 70s and early 80s, as we were learning more about the immune system to try to combat AIDS, it was understood that these cells were able to attack cancer cells in some clinical experiments. By the late 80s it was shown that we could genetically engineer T-cells to recognize and attack tumor cells directly. Into the 90s there was a lot of research done looking at adoptive cell therapy and understanding immune checkpoint inhibition. And by the early 2000s there was already the first trial to test the immune checkpoint inhibiting drugs. The first one was tested in 2000 and it was approved in 2011, Ipilumab also known as Yervoy. And by 2012 the first child with leukemia had been treated with CAR-T cell therapy. Over the last 5-10 years there’s been a huge amount of development-Keytruda was approved in 2016. The FDA granted accelerated approval to avelumab, a PD-L1 checkpoint inhibitor for Merkel cell carcinoma. With all the success in a very short period of time there’s been an explosion of interest in research in this space and there’s a huge number of companies right now seeking to modulate the immune response to try to treat cancer.
So, let’s step back for a second and talk a little bit about Flow Cytometry. Flow Cytometry is a critical tool for understanding immunology and really for diving into the population of cells. Flow Cytometry is used to quantify cells, but they need to be suspended so that you can run them through the flow cytometer. In a flow cytometer we measure the intensity of the fluorescent signal of the cell, we typically get them to be fluorescent by immunolabeling, we also measure Diameter and the refraction/reflection. And this allows us to analyze the population of cells by selective parameters, such as intensity ranges, size and shape. And this gating allows us to do multiparametric analysis on complex cell populations to really quickly count cells, understand changes in phenotype, protein expression. To understand the composition of the immune cells that are located inside a tissue, understand the difference in cell cycle, apoptosis, activation of immune cells-there’s obviously a wide variety of applications that flow cytometry is critical for and it’s routinely used in research, obviously, and in clinical practice such as hematology.
So, flow cytometry has been a critical tool for a long time. It allows us to take extremely complex population of cells and essentially come up with a systematic way to evaluate differences in those populations and how they may change or shift with treatment or the progression of a disease. In immuno-oncology it’s incredibly important to understand the immune cell population within the tumor, so this is typically done in research by homogenization of a culture tumor or a fresh biopsy sample. They essentially disincorporate the cell, de-cellularize the tissue and then label it and put it into the flow cytometer and then can do this multi-parametric analysis to better understand the different immune cell populations in the specimens they’re looking at. Determining the relative receptor expression in the immune cell subpopulations is critical to understanding the mechanisms of oncogenesis, resistance, and drug effects and so at the end of the day the more data you can get about these cell populations the more information you can then uncover about the different factors that lead to the progression of the tumor itself.
So, a brief overview of flow cytometry and cancer research. Obviously, there’s a lot of pros. It’s very high throughput and highly automatable. You can do high-plex labeling, so it’s easy to do 7, 10, 15 labels on a single population cell. It’s extremely quantitative and you get very high statistical power due to the sheer number of cells you are measuring. It’s got widespread adoption and familiarity; most immunologists are familiar with how to read these scatter plots. And of course, the major advantage is that you can recover viable, sorted cells for further experimentation. Now, there are some limitations to flow cytometry. Flow cytometry can only measure a total of signal, it can’t really distinguish between the specificity of staining of internal structures, and sometimes this is important for different phenotypes. Another major con is that you lose all the spatial information in the tissue. Since you have to dis-incorporate the tissue you end up with just a mixture of cells and you don’t know which ones were near which others, and that can be a difficulty when you’re trying to quantify the changes in the immune cell population in terms of how close the immune cells are to interacting with one another or with the tumor cell. Furthermore, only suspended cells can be put through a flow cytometer, and as I mentioned you have to disaggregate these tissues using enzymes and so because of this you can’t use flow cytometry on archived Formalin Fixed Paraffin Embedded tissue specimens, it’s just not an option. Now, of course, a huge amount of the tissue specimens stored in the world from former clinical trials in research are stored as FFPE tissue blocks, and so enabling a technique that would allow us to do this type of immune cell population research on FFPE tissues will be very valuable to cancer research and to research in general.
So, to that end, we thought to develop a technique to do exactly that. We wanted to characterize the immune cell population in these archival tissues to take advantage of the millions of archived tissue specimens that pharma clients, researcher repositories, universities, research hospitals, etc. have at their disposal. The most critical aspect is that these stored clinical archived samples have a huge amount of information about the progression of the tissue, disease, etc. so this represents an extremely large repository of information about the needs that we can get access to. So how are we going to do this? So the way that we accomplish this is to adapt multiplex immunofluorescent techniques that we had developed for doing high plex immune-histochemistry using digital slide scanning and we developed some open-source data mining and visualization tools, that allowed us to develop what we call Tissue Cytometry. This is essentially doing the same kind of analysis, where you’re measuring the intensity and count of the different cells in your sample, but we are doing it in tissue. So, directly from the FFPE slide.
So how does this work? The approach is essentially sequential, multi-round immuno-labeling. So, the way that this works is by taking a tissue section, labeling with 3-5 antibodies, then we do digital fluorescent slide scanning. After that we strip the antibodies off the tissue using a proprietary reagent that we developed here. We repeat this process over and over until we’ve covered all the targets on that same tissue. Then we use a computational co-registration algorithm to assemble this into one, big mega stack of an image that contains all of different channels and that allows us to conduct quantitative analysis, where we conduct cell counting , colocalization, intensity-gating, etc.
This is a brief overview of how our approach works, and some of the preliminary information to demonstrate why we can use it this way. So the approach that we developed involves using a chemical technique to essentially denature the antibodies from the tissue, and this is a very gentle technique that we developed. It’s very rapid, and this allows us to do multi-round immuno-labeling. It’s essentially like an antibody eraser. And so we can take a tissue, you can see on the far left panel we have a tissue that is before staining, this is obviously nothing. Then we stain it with our selected antibodies. We can strip the tissue, which will get us back to the tissue as it was with no signal, and now to demonstrate that we have fully removed all the primary, and secondary, antibodies from that tissue we add just the secondary antibody here, and this will demonstrate that indeed all the primary has been removed as we get no background signal. Then we can repeat with new markers. In this case we just demonstrated that the antigens is preserved, we can stain the same markers and we get an almost indistinguishable image. One of the coolest advantages of this technique in the eyes of many pathologists is that after you’re done collecting all this fluorescent data you can still go back and get an H&E slide or cross-validate with traditional immunohistochemistry techniques, and so it makes it very easy to cross-validate this technique with existing pipeline.
So, the critical instrument that allows us to do this, in our lab, is the Aperio VERSA. And so this allows us to do multiple channel slide scanning at 40x on the samples extremely rapidly, it usually takes between 15 minutes to an hour for these samples depending on the size. We get extremely high image quality, they’re extremely well suited for qualitative image analysis, and this allows us to move directly into the next part of our pipeline.
The most important part of this process, at least from a data science standpoint, is assembling these images together into one continuous image. If you’re going to do colocalization of these different biomarkers in the cell population that is required for tissue cytometry, then we have to make sure that the different rounds of imaging have perfectly aligning cells. The way that we accomplish this is by using a technique that’s known as computational elastic co-registration. This is a technique that was originally developed to essentially stitch satellite images together way back, several decades ago, but has been re-adapted for use in bioimaging. And so, we actually used, for this particular project, an ImageJ plugin that’s publicly available-it’s free, which essentially takes the DAPI channels of the different rounds of imaging, uses those as a guide, and manages to align these images by slightly deforming the space between the cells to account for the slight deformation of cells that happen when you remove the cover slip and repeat the process of staining. This allows us to get one continuous image set out of our multiplex image run, and that way we can get a single image with 10,12,15 channels which represent the different markers that we’re trying to look at in the panel.
So how does this all fit together? In the next couple of slides we’ll show you an example of how we have applied this technique to take a look at a couple of patients, and understand the differences between the immune cell population within these patients. What we thought to do was develop a panel of 12 targets, you can see these targets listed on this slide here. They’re mostly for T-cells, cytotoxic T-cells, we have memory T-Cell markers in here, we’re looking at proliferating markers, we have Pan-CK in there to distinguish between the tumor regions and the non-tumor stroma regions, as well as some checkpoint markers such as PD-1 and PD-L1.
And so we put together this panel and labeled the 3 different patients who are all suffering from the same type of breast cancer, and you can see the representative images here displayed with a zoomed in cut-out down below. And as you can see from a quick glance, there’s not a huge amount of difference between these patients here, and so moving along here.
For each of these patients we get a 13 plex image stack, where we have 12 antibody targets- a channel for each one, as well as the nuclear space. Here is just a representation of that. The composite image is shown on the left, and in the middle you’re seeing a sequential, essentially a video, of the different channels going one after the other on the same image. And also on the right you can see just a montage.
So, this enables us to get really beautiful multi-channel composite images that look really cool and they can go up on the wall, but now what, right? This doesn’t really tell us a lot. Here we’re looking at 12 different channels and at the end of the day there are so many different colors your eyes can see. These images are pretty, but they don’t really tell you much on their own.
The next step of this process involves computational analysis and so we utilize image analysis software to do cell segmentation and marker colocalization. So, we utilize one of several pieces of software. We have several pieces of software that we have developed internally, we also use ImageJ as well as commercially available software HALO to process these image sets, depending on the need. From that image analysis we get huge data, we get usually more than a million rows per sample, because there are so many cells that we are detecting in these patients’ tumors. We take that huge amount of data and we feed it through the multi-source data analysis pipeline that allows us to generate the conventional scatter plots that are typically produced when doing flow cytometry analysis. Just to point out, the kind of data that we get for this analysis is also compatible with FlowJo and other commercially available flow cytometry software, for those who are more comfortable with it. We just wanted to emphasize that you don’t need commercial software to do this analysis this is very straightforward and can be reproduced with a basic understanding of Python coding and some of these open-source libraries.
So, what is this overall process of doing the data analysis for tissue cytometry? So, first off with the raw data we filter by size and shape and colocalize by signals. Then, we do processing, just background subtraction, normalization, log-scaling. Finally we will apply our actual analysis, this is where we are doing the gating and counting, the statistical analyses, and one of the advantages that we have of doing tissue cytometry is that we can do spatial analysis. So after we are done looking at cell population’s direct colocalizations we can dive into which cell populations were located near one another in tissue, and that is critically important for understanding the roles the multiple immune cell phenotypes play in the progression of cancer.
Here is just a representative of example of some selected tissue cytometry scatter plots here. Just for a point of emphasizing how much data we are really dealing with, we’ve got 12 markers, we have 132 one-to-one primary marker comparisons, and if we want to compare the coincidence of every marker to every other marker, that’s 12 factorial possible phenotypes. That’s about five hundred million possible combinations to look at if you want to be complete. And so obviously that’s a lot of data. And so the best way to do this is obviously systematically, and drive by hypothesis.
And so we were particularly interested in understanding the differences in T cell exhaustion between these patients. And so you can see here on the top we have the scatter plots representing the PD-1 intensity is the Y-axis and the CD3 is the X-axis. And you can see here that Patient 2 displays a particular shift in its phenotype, compared to Patient 1, and we see that same trend with Patient 3. Patient 1 has a significantly higher relative percentage of CD3 and PD-1 positive cells, this would be exhausted T cells, compared to Patient 2 and Patient 3. Patient 2 and Patient 3 we see a much smaller percentage of the total T cell population, especially in the PD-1 marker. This goes hand in hand with what we see when we look at the PD-L1 tumor, and so the figure on the bottom we see PD-L1 intensity in the Y-axis and PanCK intensity in the X-axis, and here we see 2 distinct populations of cells in our samples. We have a number of PD-L1 PanCK positive cells, and we also have a number of PD-L1 that are not PanCK expressing. And so you can see when we look at the bar graph that is out on the right, patient 1 had a significant decrease, I guess significantly fewer, PD-L1/PanCK negative cells compared to Patient 2 or Patient 3.
Moving along here, we were interested in understanding the tumor associated macrophages, and so this is where we had started to look at the advantages of spatial analysis, and so in our study we can correlate the CD68 positive cells with the nearby PanCK positive cells and do a scatter plot here to look at the relative expression of the CD68 in the region where we see PanCK positive cells. So you can see on the figure at the top we have a kind of bi-modal distribution of between our 3 patients. We have in Patient 1 a significantly higher concentration of our macrophages contained within the tumor comparison to the stroma, whereas Patient 2 and 3 show very similar relative percentage of macrophages in the tumor region compared to the stroma. And so to summarize in Patient 1 we are seeing a significant decrease in the tumor associated macrophages in the stroma compared to Patients 2 and 3.
Now finally, the spatial analysis is really what brings this tissue cytometry to the forefront interest to immuno-oncology because of the way that we do this analysis from the actual immunohisto fluorescence images. We can take the data analysis and start to bucket these cells into the distance from cell types of interest. And so we were particularly interested here in understanding the engagement between T cells and tumor associated macrophages in Patient 1, we saw such distinct difference in some of the other t cell exhaustion and some of the other factors that we measured here in Patient 1, and so we generated a histogram of the distribution of the distance between CD3 cells and the nearest CD68 positive cell. We see these figures that are shown in the middle here. And so look at the top one, we are looking at the distance between CD3+ cells and CD68+ cells you can see that CD68 has considerably higher percentage of its t cell population in close proximity to CD68+ cells. And on the graph at the bottom we are looking at a the histogram of distances between CD3/PD-1+ cells and the nearest CD68+ cell. And you can see that trend continues here-we have a much higher percentage of the total population of exhausted T cells in close proximity to a CD68 cell, withing the tumor in Patient 1 compared to Patient 2 and Patient 3.
So, to summarize, Tissue Cytometry is possible by combining the multiplex immunohistochemistry, digital slide scanning, and open-source data processing packages. The sequential multi-round immunohistochemistry technique that we use allowed for 10-20 targets to be labeled in a single sample, and this allows rapid multi-parametric data analysis so that we can explore the cell populations within archived FFPE tissue specimens. The differential analysis of cell populations in archived patient tumor biopsies is rapid, it’s quite inexpensive, and you get a huge matrix of quantitative data for analysis. We end up with an extremely high statistical power with this type of analysis, since we usually end up measuring millions of cells in every single sample and it’s at relatively high throughput. With our technique we can do 50-100 slides per day per scanner that we have, we have 2 VERSAs pretty much running around the clock here. But it’s a fairly high throughput technique for histopathology. Most importantly, in my opinion, is the ability to then do the spatial analysis and understand the network of immune cells and how they’re interacting in contributing to the progression of the disease.
So, there are a few limitations to this technique. So, compared to flow cytometry we have a much lower dynamic range, the imagers typically take 8-bit images so we’re dealing with between 0-255 as our possible intensity values. That can make it a little bit trickier to get separation between your cell populations, unless you’re careful in how you do your data processing. Because a lot of tissues have Autofluorescence this requires either background correction or some kind of chemical processing ahead of time to reduce the background. Of course since these tissues are fixed you cannot isolate the cells easily after this analysis, and because these tissues are often archived for decades they all require very careful antigen retrieval and it’s well known that some antigens are destroyed by FFPE targeting, and so not every target will remain available in archived specimens.
The next steps, at least on the Visikol side, we are currently expanding this for the analysis of tissue microarrays so that we can offer more complex services for patient stratification. We’re next incorporating a number of panels to measure aspects about the extra cellular matrix and tumor micro environment. We’re currently in the process of validating a wide library of standard panels for immune markers and other disease markers, and we are continuing to improve our techniques by incorporating AI where it fits best.
So, with that I will draw our presentation to a close.