Single-cell multiplexed imaging methods to understand tissue and tumor architecture

Video Transcript

• 00:00 - 00:13: Hi, my name is Subham Basu and I’m the Director of Strategy for Immuno-Oncology here at Abcam.
• 00:13 - 00:18: I’d like to welcome you to our talk today, entitled Single-Cell Multiplex Imaging Methods to Understand
• 00:18 - 00:22: Tissue and Tumor Architecture, brought to you by Dr. Hartland Jackson.
• 00:22 - 00:27: Dr. Hartland Jackson is an investigator at the Lunenfeld-Tanenbaum Research Institute
• 00:27 - 00:31: in the Sinai Health System and is part of the Ontario Institute for Cancer Research,
• 00:31 - 00:36: as well as an assistant professor in the Department of Molecular Genetics at the University of Toronto.
• 00:36 - 00:41: His research involves the use of mass cytometry for highly multiplexed imaging of tumor tissues,
• 00:41 - 00:46: and he is developing the methods for the analysis of spatially resolved single-cell data.
• 00:47 - 00:52: He obtained his PhD with Dr. Rama Kothakara at the University of Toronto and undertook postdoctoral
• 00:52 - 00:58: training in Bert Bodenmiller’s lab in Zurich. His work with high-dimension pathology imaging
• 00:58 - 01:03: of breast cancer patient samples has revealed the organization of the tumor microenvironment
• 01:03 - 01:07: and identified single-cell-defined patient subgroups with distinct clinical outcomes.
• 01:08 - 01:12: The recently established Jackson Lab in Toronto strives to understand tissues
• 01:12 - 01:16: and tumors as the integrated outcome of their single-cell components.
• 01:16 - 01:22: To do so, they are utilizing multiplex imaging to simultaneously quantify single-cell phenotypes
• 01:22 - 01:27: and markers of their functional state, as well as interaction over organization
• 01:27 - 01:31: and contribution to tissue architecture for what we hope is downstream clinical benefit.
• 01:32 - 01:37: Today, Dr. Jackson will provide an overview of various epitope-based multiplex imaging techniques
• 01:37 - 01:41: and methods and discuss the benefits and challenges of each technique.
• 01:41 - 01:46: He will discuss his recently published work using imaging mass cytometry to study the
• 01:46 - 01:51: single-cell architecture of biobank clinical tissue sections and detail potential uses
• 01:51 - 01:55: of spatially resolved single-cell methods and analyses. Over to you, Hart.
• 01:57 - 02:04: Thank you, and thank you very much for having me here today. As you mentioned, I’m going to be
• 02:04 - 02:10: speaking about multiple methods today, giving you an overview, a simplified view of the different
• 02:10 - 02:15: techniques that are available for you to use. I’m going to highlight the potential of multiplex
• 02:15 - 02:21: imaging using a recently published work of mine from my postdoctoral fellowship in the
• 02:21 - 02:30: Bodenmiller Lab using imaging mass cytometry. To date, my work has mostly focused on breast
• 02:30 - 02:36: cancer, which, like many diseases, arises within normal tissues and is diagnosed using tissue
• 02:36 - 02:42: imaging when a pathologist identifies changes in the organization of cells or in cell morphology.
• 02:43 - 02:50: Malignant cells are identified when tissues change their organization and invade out into
• 02:50 - 02:56: the surrounding area. Inflammation can be seen by the presence of leukocytes with very small nuclei,
• 02:58 - 03:03: and different types of tumors or different types of cells can be identified using different markers.
• 03:03 - 03:07: But no individual marker can really address the complexity of a tumor or any tissue,
• 03:07 - 03:13: as these are made up of many cells working together in a coordinated fashion. These
• 03:13 - 03:19: phenotypes that we see can have a major impact on clinical response, with all cell types likely being
• 03:19 - 03:25: needed to be treated for successful cancer therapies. These phenotypes themselves can
• 03:25 - 03:30: actually change at different locations within a tumor, resulting in intratumor heterogeneity.
• 03:32 - 03:37: To actually quantify what’s happening in any tissue or in any tumor, we need technologies
• 03:38 - 03:43: that are able to quantify all the single-cell components of a tissue. They need to have single
• 03:43 - 03:47: cell resolution, they need to be spatially resolved to see their organization, and they
• 03:47 - 03:53: need to be quantitative to be able to measure multiple markers, multiple cell types, and the
• 03:53 - 04:00: cell states of these. If we want to do this in human samples or in patient samples, this needs
• 04:00 - 04:06: to be compatible with clinical procedures. Here I’m giving you a very simplified overview
• 04:06 - 04:12: of epitope-based multiplexed imaging methods that are able to do this in tissue samples.
• 04:13 - 04:19: These can be broken down into two main groups, those that are based on fluorescent or colorimetric
• 04:19 - 04:26: tags and use microscopy, and those that use mass-tagged antibodies with a mass-spec-based
• 04:26 - 04:32: readout. First, we’ll look specifically at the fluorescence or colorimetric methods.
• 04:33 - 04:37: The one that’s most similar to classic immunohistochemistry are those that use
• 04:37 - 04:44: enzyme-based signaling for staining and amplification. These will use a primary
• 04:44 - 04:50: antibody, which is bound by a secondary antibody linked to horseradish peroxidase.
• 04:51 - 04:57: When substrates are added, these will then deposit colorimetric signals onto a sample,
• 04:57 - 05:02: exactly like immunohistochemistry. These are then imaged, and to make this multiplexed,
• 05:03 - 05:09: the original antibodies from the first stain are then enzymatically or chemically stripped away
• 05:09 - 05:15: before a second round of staining takes place. This process of staining and then imaging will
• 05:15 - 05:22: happen over and over again, and the resulting images computationally combined into multiplexed
• 05:22 - 05:30: images. So, a popular method that’s commonly used is the OPAL system, which uses similar
• 05:30 - 05:35: antibody staining with a primary antibody bound to a secondary that’s enzymatically linked.
• 05:35 - 05:41: The difference here is that the substrates used are tyramides linked to a fluorophore,
• 05:41 - 05:45: and the fluorophore is then activated upon the enzymatic reaction and deposited
• 05:45 - 05:52: in the areas surrounding the antigen and antibody combination. These are then removed,
• 05:53 - 05:58: the antibodies are then removed chemically, but the fluorophores are left behind.
• 05:58 - 06:03: So, iterative stains of antibodies will then continue to deposit different fluorophores onto
• 06:03 - 06:09: the tissue, and this is done in up to seven rounds of staining, and these are then imaged all at the
• 06:09 - 06:15: same time using multispectral imaging to then deconvolute the different antibody stains.
• 06:16 - 06:22: The benefits of these enzyme-based immunostaining methods are that some of them will use standard
• 06:22 - 06:28: pathology equipment and reagents. This allows you to look at large areas of whole slides of tissue.
• 06:28 - 06:33: You can do many slides in parallel, and with the enzymatic reaction, you get excellent signal
• 06:33 - 06:39: amplification per target. Seven markers are commonly shown with the OPAL system, and up to
• 06:39 - 06:45: 12 have been shown with serial immunohistochemistry staining. Challenges that can arise include that
• 06:45 - 06:51: with increased number of cycles and tissue processing in each cycle, you can get degradation
• 06:51 - 06:57: or changes in antigens with each cycle. It’s possible with tyramide deposition that the
• 06:57 - 07:03: tyramide with the fluorophores can mask or cover an antigen of interest that you wanted to look at
• 07:03 - 07:09: in a later cycle, and each of these cycles itself can actually take a long time, up to one day,
• 07:09 - 07:14: similar to a normal immunohistochemistry experiment. So doing more cycles can lead to very long
• 07:14 - 07:20: experiments times to get a multiplexed image, and even though you get great signal amplification,
• 07:20 - 07:26: enzymatic methods can lead to a decreased dynamic range compared to some of the other methods.
• 07:27 - 07:34: So additional methods that lead to even higher multiplexing are serial immunofluorescence methods.
• 07:36 - 07:45: These include 4I, MXIF, CySIF, and multiple others, and these kind of have two approaches
• 07:45 - 07:50: with serial imaging. One will use both primary antibodies unlabeled and secondary antibodies
• 07:51 - 07:58: with attached fluorophores, while others use primary antibodies, all of which have been
• 07:58 - 08:04: conjugated to a fluorophore themselves. These are then individually imaged per cycle,
• 08:05 - 08:12: at which point after the first imaging, the antibodies are then either removed in the
• 08:12 - 08:19: presence of buffers that have prevented cross-linking that can be formed when the
• 08:19 - 08:24: fluorophore reactions are exposed to light, or the fluorophores themselves are chemically
• 08:24 - 08:30: inactivated or inactivated by the use of ultraviolet light. Then an additional round
• 08:30 - 08:38: of antibodies will be added with additional signaling and imaging, and this will happen
• 08:38 - 08:44: over and over again in iterative cycles, each cycle resulting in an additional channel being measured.
• 08:45 - 08:50: So again, the benefit of this is that you can use standard immunofluorescence microscopy
• 08:50 - 08:56: equipment that will be available to you. There is the potential here for unlimited readouts as you
• 08:56 - 09:02: do more and more cycles. Using fluorescent slide scanners, you’d be able to look at large
• 09:02 - 09:08: areas for whole slides, and up to 90 markers have been shown using these types of methods.
• 09:09 - 09:14: But again, with more markers is more time, and each cycle can result in changes in the tissue
• 09:14 - 09:20: that you’re looking at, and also the challenges that surround using fluorescence microscopy in
• 09:20 - 09:28: general, such as the presence of autofluorescence in some tissues, changes in dynamic range that
• 09:28 - 09:33: need to be controlled by properly titrating your antibodies and altering the gain on your
• 09:34 - 09:40: microscope, and even tiling effects that can happen as the edge of one image may have been
• 09:40 - 09:49: impacted by excitation in a previous imaging round. So newer methods now for multiplexed
• 09:49 - 09:54: imaging are no longer using fluorophores directly on the antibodies, but they’re using antibodies
• 09:54 - 10:02: that have been bound to DNA oligos. One of those methods, CODEX, uses nucleotides that are each
• 10:02 - 10:09: bound to a fluorophore. In each round of imaging, a cocktail of nucleotides is
• 10:10 - 10:15: mixed in with the antibodies, and these will then extend the oligo on every single antibody.
• 10:15 - 10:21: Some of the nucleotides, according to a specific sequence, will have a fluorophore, and some will
• 10:21 - 10:29: be blanks, so that the fluorophore signal will correlate with a very specific antibody,
• 10:30 - 10:33: whereas the others, the oligo will be extended but not give a signal.
• 10:34 - 10:42: Then iterative imaging, each round of imaging will result in the indexing of a specific location in
• 10:42 - 10:48: the oligo having the fluorophore bound, and this allows you to look at many different antibodies
• 10:48 - 10:55: at the same time, up to 56 being shown at the moment. A different method, such as ImmunoSaber,
• 10:55 - 11:01: combines both signal amplification and multiplexed imaging by using a primer exchange
• 11:01 - 11:08: reaction to extend the oligo that’s tagged to each antibody, and then each individual cycle
• 11:08 - 11:15: of imaging and staining will attach a sequence-specific fluorophore-tagged oligo to the
• 11:15 - 11:23: antibody of interest, resulting in each signaling round imaging one antibody of interest.
• 11:24 - 11:29: These methods can also use standard fluorescence equipment and have the potential to measure many,
• 11:29 - 11:35: many numbers of antibodies. A benefit versus the serial immunofluorescence is that, in this case,
• 11:35 - 11:40: you bind all your antibodies at once, at the beginning and the first stain, and don’t have
• 11:40 - 11:47: to worry about antigen changes over the different cycles. Saber or ImmunoSaber provides amplification,
• 11:48 - 11:54: and the CODEX system can come with an automated sample introduction that can control the
• 11:54 - 12:00: cycling rounds, thereby speeding up the time for each cycle. Being fluorescence-based, there’s
• 12:00 - 12:06: still the presence of autofluorescence and things that need to be controlled for in your
• 12:06 - 12:12: microscopy experiment. Another challenge can be that some of these antibodies, you need to have
• 12:12 - 12:18: the oligos conjugated, which can be a challenge for some labs. So that is kind of an overview of
• 12:18 - 12:24: the fluorescence-based methods. You’ll see for that a lot of those, in every round of imaging,
• 12:24 - 12:29: you’re only able to look at one, two, or maybe three markers at the same time, and these need
• 12:29 - 12:36: to be done in additional cycles. Now, with mass-tagged imaging, this issue has been overcome,
• 12:36 - 12:40: as there’s very little overlap between the different channels or the different signals
• 12:40 - 12:46: that we’re looking at, and this allows each antibody to have a specific metal isotope of a
• 12:46 - 12:50: specific mass, and these can be easily distinguished with a mass spec-based reading.
• 12:51 - 12:57: So the two common methods here are imaging mass cytometry and multiplexed ion beam imaging.
• 12:57 - 13:04: Both of these stain the tissues with a cocktail of antibodies, each with its own metal isotope.
• 13:04 - 13:12: Imaging mass cytometry then uses laser ablation, which will aerosolize a one-micron spot of tissue,
• 13:12 - 13:18: which is then introduced into the mass spec for measurement, resulting in one pixel in your image.
• 13:18 - 13:25: The laser then rasterizes across the whole tissue, resulting in individual pixel measurements
• 13:25 - 13:35: and an outgoing multiplexed image. Ion beam imaging actually uses a high-energy ion beam,
• 13:35 - 13:41: which releases secondary ions from the isotope and rasterizes the tissues as well,
• 13:41 - 13:46: but in a non-destructive manner, allowing you to go back over the tissue over and over again.
• 13:47 - 13:53: These methods have a very high linear dynamic range, and because these isotopes are not ever
• 13:53 - 13:59: present in biological samples, there’s little background stain and no autofluorescence.
• 14:00 - 14:04: Very low signal spillover allows multiple channels to be measured at the same time,
• 14:05 - 14:11: and imaging mass cytometry has very high quantitative capacity, as you know, you’re
• 14:11 - 14:17: measuring all the antibodies in that one spot every single time, whereas the ion beam imaging
• 14:17 - 14:24: with MIBI allows you to change the size of your focal region that is releasing the secondary ions,
• 14:24 - 14:30: tuning the resolution of your measurements. All of these do simultaneous antibody staining,
• 14:30 - 14:35: and because these are not based on a chemical reaction or a fluorophore, but just the presence
• 14:35 - 14:40: of metal isotopes, these stains are actually quite stable, so you can stain your tissues
• 14:40 - 14:45: and actually leave them even just at room temperature for weeks, months, or years before
• 14:45 - 14:51: your actual measurements. Challenges are that these require dedicated expensive equipment,
• 14:52 - 14:56: and the number of markers is limited to the number of metal isotopes that we have access to,
• 14:56 - 15:03: and can conjugate to antibodies, and these can be slower to acquire large areas of tissue,
• 15:03 - 15:09: as they have to progressively rasterize across the tissue. Tiling effects similar to
• 15:10 - 15:17: microscopy can be seen occasionally with MIBI, as there could have been release of ions from the
• 15:17 - 15:22: edge of a specific region of interest, and in tiling that will be seen in the next tile,
• 15:22 - 15:28: whereas imaging mass cytometry has a resolution that’s locked at one micron
• 15:29 - 15:33: per pixel, which is slightly less than standard microscopy.
• 15:34 - 15:40: So, to actually see what multiplex images look like, the challenge with visualizing them is not
• 15:40 - 15:48: the technologies, but it’s actually us. We’re not able to see the differences between seven colors
• 15:48 - 15:53: in an image, let alone up to 90 different markers, so what you’re looking at here is
• 15:53 - 16:00: a breast cancer sample that has been imaged using mass cytometry. On the left of your screen is
• 16:01 - 16:06: a breast tumor, and on the right, the black circles are the healthy fat cells of the stroma.
• 16:07 - 16:12: We visualize them similar to standard immunofluorescence, just in RGB,
• 16:12 - 16:17: as that is what we’re used to looking at, but every combination or every image that you see
• 16:17 - 16:21: here is actually coming from the same piece of tissue and is present at the same time.
• 16:22 - 16:27: So, this allows us to look at the general overall architecture of the samples,
• 16:27 - 16:32: and even the activity of individual cells. For example, the red cells are undergoing mitosis in
• 16:32 - 16:38: this case. We can look at different cell populations in the tissue, and when tumor
• 16:38 - 16:44: cells have invaded outside of an encapsulating layer, such as right here. We can also look at
• 16:44 - 16:50: the immune response in these samples to see the different areas and the different immune
• 16:50 - 16:55: infiltration in these areas, and simultaneously look at the heterogeneity of different samples
• 16:55 - 17:00: and the different phenotypes that are present within them. To analyze these images, a standard
• 17:00 - 17:06: approach is to segment the images into single cells. So, this then defines the area that
• 17:06 - 17:13: corresponds to every individual cell in the image. Then, all the signal from that area, or from that
• 17:13 - 17:21: mass of a single cell, will be quantified to make single-cell data. So, we will quantify the average
• 17:21 - 17:25: signal for every single one of the antibodies that you’ve measured, and also look at some of
• 17:25 - 17:30: the spatial features for each of these, and incorporate these into a single-cell file that
• 17:30 - 17:37: can then be analyzed using methods that are common for suspension-based single-cell methods, such as
• 17:37 - 17:44: clustering or dimension reduction tools. The way that segmentation takes place is based on
• 17:45 - 17:51: very standard methods that are used for microscopy, and then some adjustments
• 17:51 - 17:57: that are made for multiplexed imaging. So, in a microscopy image, the first step would be to separate
• 17:57 - 18:03: out your marker that corresponds to every nuclei, and those that can identify the edge
• 18:03 - 18:08: of the cell, or the membrane of the cell. We then use the nuclei signal to identify all the individual
• 18:08 - 18:15: cells in the image, and then extend from the nuclei out to the membrane in order to identify
• 18:15 - 18:20: the total area of the cell. This then identifies the region corresponding to every individual cell,
• 18:20 - 18:26: and we quantify the markers in there in order to produce the single-cell data. The issue here is
• 18:26 - 18:33: that this doesn’t take advantage of the multiplexed capacity of the tissues. You can sum up all your
• 18:33 - 18:38: membrane channels and all your nuclei channels and do this in a quick manner, so you have multiplexed
• 18:38 - 18:43: images, but then you lose some of the cellular heterogeneity, and you don’t see that two neighboring
• 18:43 - 18:49: cells actually have different markers. So, for this purpose, what we use is a tool called
• 18:49 - 18:55: Elastic, which allows us to train a random forest classifier or deep learning classifier to identify
• 18:56 - 19:03: every pixel, whether it’s the nuclei, the membrane, or background, and this will use all the high
• 19:03 - 19:08: dimension data to identify the identity of that pixel and do the segmentation. So, on a multiplexed
• 19:08 - 19:14: image, you have to train to identify which are nuclei, which are membranes, and Elastic will then
• 19:14 - 19:20: provide you the probability that each pixel is a nuclei or a membrane, and this has been segmented
• 19:20 - 19:25: in a similar manner to identify all the nuclei and extend this to the edge of every single cell,
• 19:25 - 19:31: you can then quantify your single-cell data across all channels, all dimensions of your multiplexed
• 19:31 - 19:37: image. So, the importance of single-cell segmentation really comes out when you can
• 19:37 - 19:45: compare quantification within an individual image. So, this data is from a paper from Daniel Schultz,
• 19:45 - 19:50: where he was able to show that with imaging mass cytometry, you can measure both RNA and protein
• 19:50 - 19:54: in the same samples, and he could see that looking at lots of different tumors,
• 19:54 - 19:59: we knew there’s differences in the tumors, but across these was very high correlation
• 19:59 - 20:06: between the protein and RNA expression for HER2, as would be expected, but if these are analyzed
• 20:06 - 20:13: per pixel, any of these relationships are completely lost, whereas after segmentation, we could then
• 20:13 - 20:20: return to definitely an understanding where both the RNA and protein for a specific target
• 20:20 - 20:26: are related. So, this really solidified to us that the segmentation and the measurement of single
• 20:26 - 20:30: cells from these images are the biologically relevant units, and this is what we should be
• 20:30 - 20:36: measuring in all images. So, the current standard for multiplexed imaging based on different methods
• 20:36 - 20:43: allows many different markers to be measured. With signal amplification and enzymatic reactions,
• 20:43 - 20:51: up to 12 markers have now actually been shown. With serial fluorescence, 90 have been shown,
• 20:51 - 20:57: and reporter cycling in CODEX 56 and mass spec-based methods are both showing 40 different
• 20:57 - 21:03: antibodies at the same time. A big benefit of doing imaging versus other single-cell methods
• 21:03 - 21:08: is that there’s actually no sample loss in any of these images that we’re looking at. You don’t have
• 21:08 - 21:14: to enzymatically dissociate your samples into single cells, as we know that this can result
• 21:14 - 21:19: in changes in expression in those single cells and also the loss of different cell types during
• 21:19 - 21:26: that process. This allows us to look and quantify millions of cells in biobanked samples, as we can
• 21:26 - 21:32: do formalin-fixed paraffin-embedded, but also, as there’s no loss, it’s a very valuable way to
• 21:32 - 21:36: look at very valuable samples that are small or rare and get rich amounts of information from
• 21:36 - 21:42: these. A lot of systems now are commercially available, making them available to many more
• 21:42 - 21:48: labs, such as the multi-omics, multiplex serial immunofluorescence from GE, and both the OPAL
• 21:48 - 21:57: tyramide amplification and reporter cycling with CODEX from Akoya, imaging mass cytometry from
• 21:57 - 22:04: Fluidigm, and maybe from IonPath. One of the things we’re interested in is now multimodal
• 22:04 - 22:09: measurements. As I just showed you, we’re able to do both RNA and antibodies at the same time
• 22:09 - 22:15: with imaging mass cytometry. We’ve also seen that we can combine immunofluorescence and imaging mass
• 22:15 - 22:20: cytometry as well. This allowed us to really validate the technology in the first place
• 22:20 - 22:25: and show that using an antibody strategy that would be used for standard immunofluorescence,
• 22:26 - 22:30: we were able to see identical signals with imaging mass cytometry and immunofluorescence.
• 22:30 - 22:37: In this case, we had a primary antibody with a metal tag, and then we mixed secondary antibodies
• 22:37 - 22:44: with both fluorescence and a second metal tag. This allowed us to get three independent signals
• 22:44 - 22:49: from one primary antibody. We did this at the capacity of standard immunofluorescence,
• 22:49 - 22:56: so we were able to measure HER2, pancytokeratin, and combine a metal DNA intercalator
• 22:56 - 23:02: and DAPI to look at the DNA. If we overlaid the three different methods to compare them,
• 23:02 - 23:07: we could see by the white signal that these very highly overlapped. The differences that came out
• 23:07 - 23:13: were that there was some autofluorescence in this channel that can be seen in the red at the bottom.
• 23:15 - 23:20: All the methods really nicely overlapped, and any individual method also could produce
• 23:20 - 23:26: very similar images. To zoom in on these small areas, we could then compare the resolution of
• 23:26 - 23:33: standard immunofluorescence with the one-micron resolution of imaging mass cytometry and see that,
• 23:33 - 23:37: while it’s not quite as crisp as the immunofluorescence, it’s more than good
• 23:37 - 23:42: enough for us to identify single cells and even subcellular features in these tumors.
• 23:43 - 23:49: Then with Jana Fischer, we moved on to see the capacity of the technology. For this,
• 23:49 - 23:55: we looked at over 300 different breast cancers. In each sample, we’re able to quantify all the
• 23:55 - 24:01: different cell types using a high-dimension panel of antibodies targeted for different
• 24:01 - 24:06: clinical markers or different cell types of the breast. We didn’t just quantify them in the sample.
• 24:06 - 24:11: We could also localize these different phenotypes by the different colors that are shown here
• 24:11 - 24:16: on specific images. Because we know their organization, we were then able to look at
• 24:16 - 24:21: additional features of these tumors in the organization of individual cells. We’re able to
• 24:21 - 24:28: look at which cells were close to each other and use a permutation-based test to compare a tumor
• 24:28 - 24:34: to a randomized version of itself and identify statistically significant interactions. We were
• 24:34 - 24:40: also able to use low-variance community detection, which identifies modules of cells within the
• 24:40 - 24:47: overall structure of a tumor. Further, we were able to look at these cell populations that we’ve
• 24:47 - 24:52: seen and look at their distances to certain features. We’re able to look at cells of the
• 24:52 - 24:58: tumor microenvironment or epithelial cells and look at their distance to the boundary
• 24:58 - 25:05: of the tumor mass. Anything directly at zero in the middle is right on the edge of the tumor,
• 25:05 - 25:10: whereas cells on the right of the plot are those that have been excluded and outside of
• 25:10 - 25:15: the tumor. Simultaneously, we could quantify infiltration of the immune cells into the tumor
• 25:15 - 25:22: and their distance inside by moving to the left. Importantly, all of these samples that we’re
• 25:22 - 25:28: looking at have been biobanked for over 20 years. We had clinical outcomes. We’re able to identify
• 25:28 - 25:32: that these spatial features that you can pull out of multiplexed imaging, some of them were
• 25:32 - 25:37: actually related to different outcomes for the patients. We could identify features such as blood
• 25:37 - 25:44: vessels that contained immune cells being related to poor prognosis or different tumor cell structures
• 25:44 - 25:50: that were related to good prognosis. Using this single-cell data, we were able to identify groups
• 25:50 - 25:55: of patients that had distinct outcomes different than what can be identified in the clinic.
• 25:55 - 26:01: This identified both subsets of patients with very high risk and those of which actually no
• 26:01 - 26:07: patients succumbed to disease. Further, we were able to look at these different types of tumors
• 26:07 - 26:13: and score their spatial heterogeneity. By imaging different regions of one tumor, we could compare
• 26:13 - 26:21: the phenotypes of these tumors to the tumor average and see that many tumors were quite homogeneous,
• 26:21 - 26:26: but some of these tumors had very large regions, had very large amounts of spatial heterogeneity,
• 26:26 - 26:32: with specific images being very different from the tumor average. We can see that this was related
• 26:32 - 26:38: to phenotype, with some phenotypes being specifically homogeneous, whereas other tumor types could be
• 26:38 - 26:46: found mixing with different tumors across space in a specific tumor. We then went on in a separate
• 26:46 - 26:53: study with Raza Ali. We looked at banked samples that had also been highly classified by the
• 26:53 - 27:00: METABRIC team, where they had looked at sequencing and transcriptomics of these same samples.
• 27:00 - 27:06: We then used multiplex imaging to identify the single-cell phenotypes. In this plot, you’re
• 27:06 - 27:13: looking at each row are cell types that we identified using imaging mass cytometry. Each
• 27:13 - 27:18: column is a different mutation or amplification in the breast cancer samples. The black circles
• 27:18 - 27:24: identify significant associations linking a single-cell phenotype to a specific mutation or
• 27:24 - 27:30: amplification in breast cancer. This was able to identify known associations and also reveal new
• 27:30 - 27:38: associations between specific mutations and cellular phenotype. The common uses for multiplex
• 27:38 - 27:46: imaging now are a big use is to be able to use this to relate to single-cell RNA-seq phenotypes
• 27:46 - 27:52: and to identify the locations and organization of those phenotypes that were identified with
• 27:52 - 27:57: other methods, and also to look at single-cell interactions. Further, a big benefit is that
• 27:57 - 28:02: these can actually be used on formalin-fixed paraffin-embedded tissues, so biobank samples
• 28:02 - 28:09: with known outcomes can be investigated using multiplex imaging. And because there’s no sample
• 28:09 - 28:14: loss in the processing, numerous studies are also looking at using multiplex imaging to identify
• 28:15 - 28:20: quantified phenotype single cells from small samples such as patient biopsies.
• 28:21 - 28:27: Further, my lab in Toronto is now looking at the ability to extend these technologies to use them
• 28:27 - 28:34: in modifiable systems. So, can we use mouse model systems to actually alter cellular interactions
• 28:35 - 28:41: and the organization of different cellular communities using mouse models and measure this outcome with
• 28:41 - 28:48: multiplex imaging? In using these methods, we’re investigating three different questions
• 28:48 - 28:54: in terms of what happens with cell-cell interactions as tumors develop, how do the
• 28:54 - 29:00: early lesions actually alter the tumor microenvironment to lead to tumor progression,
• 29:00 - 29:04: and can these specific markers be used as biomarkers in the future for precision
• 29:04 - 29:10: medicine applications? So, I also need to thank collaborators and funding for these projects,
• 29:10 - 29:15: specifically my colleagues in the Bodenmiller Lab, Jana Fischer, who I worked with closely on
• 29:15 - 29:20: the large clinical cohorts, and Vito Zanotelli, who established our analysis pipelines,
• 29:20 - 29:27: and the supervision and guidance of Bernd Bodenmiller for many years. We had a great time
• 29:27 - 29:34: also working with Raza Ali, who joined our lab for a while from CRUK Cambridge, and Carlos Caldas
• 29:35 - 29:41: and the METABRIC team. A special thanks also goes to the pathologists we worked with at the
• 29:41 - 29:48: University Hospitals in Zurich and Basel, and the patients who donated their samples upwards of 20
• 29:48 - 29:53: years ago, who have allowed us to make these new discoveries. Finally, I also want to thank the
• 29:53 - 29:58: Jackson Lab members here in Toronto, Jennifer Gorman, Somi Afiuni, who are now producing
• 29:58 - 30:05: new mouse data that you’ve seen today. I need to thank many funding sources for my time
• 30:05 - 30:12: in Zurich, and also for funding, startup funding for the Jackson Lab in Toronto.
• 30:13 - 30:16: Thank you very much. I look forward to questions.
• 30:16 - 30:23: Thank you very much for that enlightening and very exciting talk, Hart. My first questions I
• 30:23 - 30:28: would have is, obviously, you went through a number of different methodologies and showed
• 30:28 - 30:35: the pros and cons for each. In your own lab going forward, what’s going to be, let’s say,
• 30:35 - 30:42: maybe your focus for clinical versus your focus for possible digital pathology routine versus for
• 30:42 - 30:46: discovery with your colleagues in Toronto? Where do you see that breaking?
• 30:47 - 30:53: I think there’s definitely a mix of imaging methods that we’re using, and each of these has
• 30:53 - 30:58: their benefits and drawbacks. I think the combination of them is also an interesting
• 30:58 - 31:04: new field that will provide the benefits and compensate for each other going forward.
• 31:06 - 31:12: We’re viewing the digital analysis as a big step in this. Some of the things we’re doing would be
• 31:12 - 31:21: standard immunohistochemistry with clinicians. Then, moving more into discovery, we’re focused
• 31:21 - 31:27: on imaging mass cytometry, which is a very robust system allowing us to measure lots of samples at
• 31:27 - 31:33: the same time. That’s the major workhorse in the lab at the moment, for sure.
• 31:33 - 31:39: Okay. In terms of all the different methodologies, is there…
• 31:42 - 31:45: What types of antibodies you’re going to use come into play?
• 31:52 - 31:57: I would say this. There’s a standard assumption, sometimes, that the antibodies they would use
• 31:57 - 32:02: for single-cell methods, such as flow cytometry, are also used here, whereas that’s not the case.
• 32:03 - 32:08: For multiplex imaging, we’re looking for antibodies that have been validated for
• 32:08 - 32:12: imaging methods specifically, for immunohistochemistry and immunofluorescence.
• 32:12 - 32:19: This is likely because the antigen retrieval process is completely different in a suspension
• 32:19 - 32:26: dissociated cell versus a tissue section, which is undergoing heat-induced antigen retrieval,
• 32:26 - 32:32: and you need to break the cross-links from fixation. For this purpose, we will then shock
• 32:32 - 32:36: a little differently, where we’re looking for antibodies that have been well-validated in
• 32:36 - 32:42: immunohistochemistry or immunofluorescence to show strong specific signals for imaging.
• 32:43 - 32:47: Then, to keep in mind that if you need to do your own conjugation reactions,
• 32:48 - 32:53: depending on the chemistry, you likely need specific purifications for those antibodies.
• 32:53 - 32:58: We find that antibodies that have good validation data and look quite good for immunohistochemistry
• 32:58 - 33:03: or immunofluorescence imaging of tissues are highly successful in multiplex
• 33:04 - 33:08: imaging methods as well. Greater than 90 percent of those antibodies make it
• 33:08 - 33:12: through the pipeline if they have great validation data in the first place.
• 33:13 - 33:18: That’s good to know. I think many people in the field who are not experts, such as you,
• 33:19 - 33:23: would want to know where is the start, where they can get, once they’ve selected a system,
• 33:23 - 33:28: or perhaps they have collaborated with various systems, you want to know what antibodies to get
• 33:28 - 33:35: and I think the fact that the ones are validated. Obviously, you go to the references and citations
• 33:35 - 33:39: of what’s already been published, but I think looking at what antibodies work in immunohistochemistry
• 33:40 - 33:42: and does it matter if it’s for paraffin or frozen?
• 33:43 - 33:50: So, in general, the application that you want to do with multiplexed imaging, you would want
• 33:50 - 33:55: the validation to be similar. That can be for your specific tissue of interest or for your
• 33:59 - 34:06: fixation of interest. So, the methods can work on frozen samples and OCT embedded and FFPE,
• 34:06 - 34:11: but then you want to do your validation using single-plex imaging in the same manner. So,
• 34:11 - 34:18: keeping that in mind, for multiplexed imaging, a lot of the time we use one quite standard antigen
• 34:18 - 34:24: retrieval method. The idea being that you want a method that’s very broad and will work for many
• 34:24 - 34:30: antibodies. So, if you were doing a single antibody experiment, you might validate all
• 34:30 - 34:35: the different steps of your pipeline for that specific antibody. But what we found is that
• 34:35 - 34:41: if you optimize for this antibody, then you lose the antigen for that antibody. And so, it’s almost
• 34:41 - 34:50: not worth the effort or the challenges of balancing these things all out and instead we can provide a
• 34:50 - 34:58: very general antibody antigen retrieval method and identify antibodies that work within that
• 34:58 - 35:07: method. And that allows us to build a catalog of antibodies that work together in a high-plex
• 35:07 - 35:12: manner. So, the antibodies are best if they’re validated in immunochemistry, whether paraffin
• 35:12 - 35:17: or frozen, depending on what kind of tissue you’re looking at. Have you found that, I don’t know,
• 35:17 - 35:22: like do monoclonals work better than polyclonals? Or is it a very target-specific thing? Do you know
• 35:22 - 35:28: like recombinant antibodies working better than not than other types of monoclonals?
• 35:30 - 35:34: Does this come into the thinking? So, it definitely does, but more as
• 35:36 - 35:40: standardization of the tool. So, we do find that it’s very antibody-specific and very
• 35:40 - 35:46: antigen-specific and that we find monoclonals that are fantastic and we find polyclonals that
• 35:46 - 35:51: are fantastic. The issue with the polyclonals is that you then need to revalidate every batch.
• 35:52 - 35:59: As you do, we have definitely seen drifts in the specificity of polyclonals over time and
• 35:59 - 36:06: suddenly the polyclonal that was working for you fantastically last year is not working anymore.
• 36:06 - 36:13: So, this is a major reason to go after monoclonals. Also, with the multiplex capacity,
• 36:13 - 36:18: people want to do multiple antigens on the same target or to look at post-translational
• 36:18 - 36:24: modifications on the same target, in which case monoclonals are definitely the way to go
• 36:24 - 36:32: or combining different targets against the same antibody. As per previous comment, I should clarify
• 36:32 - 36:37: a lot of the validation that we will do for our multiplexed imaging method, we will do with
• 36:37 - 36:44: immunofluorescence because we are also interested in the amplification and the sensitivity of each
• 36:44 - 36:48: antibody. And if you validate with immunohistochemistry, you get massive signal
• 36:48 - 36:54: amplification, which is great, but not all of those will translate through to an antibody
• 36:54 - 36:59: stain that doesn’t have any amplification at all. So, a lot of our validation will do with
• 37:00 - 37:05: secondary antibodies in immunofluorescence or even secondary antibodies in imaging mass cytometry.
• 37:06 - 37:10: As you rightly mentioned, and as we’ve also heard from other leading lights in the field,
• 37:10 - 37:17: antigen retrieval is obviously a very important issue here, as well as the selection of antibodies.
• 37:17 - 37:22: And are there retrieval methods that you use, you know, let’s say one for imaging mass cytometry,
• 37:22 - 37:29: one for if you’re doing MIBI, one for if you’re doing, you know, CODEX, you know, does it matter
• 37:29 - 37:32: if it’s metal versus oligo, does it break that way, or does it really break persistent?
• 37:33 - 37:38: So, most of my disclaimer would be that the majority of my experience is with
• 37:38 - 37:43: standard immunofluorescence and with imaging mass cytometry and isotope-based methods.
• 37:43 - 37:49: But it really goes back to you want the antigen retrieval method that works best for your
• 37:49 - 37:56: tissue type is what we see. All of these are using very similar antibodies and all seem to work.
• 37:57 - 38:01: If the antibody works with your antigen retrieval, it will work with the different
• 38:01 - 38:10: multiplex methods is my general experience and likely assumption with methods that I haven’t used.
• 38:13 - 38:19: So, it really comes down to having a protocol that is general for different antigens,
• 38:19 - 38:22: but works quite well for your specific tissue of interest.
• 38:23 - 38:27: Okay. And have you found a difference? Yeah. So, the tissue of interest really matters. So,
• 38:27 - 38:32: if you’re doing non-small cell versus, you know, you’re doing things in from, I don’t know,
• 38:32 - 38:38: breast cancer, you really have to go by the tissue that you’re looking at.
• 38:40 - 38:48: Yes. So, in general, yes. But also, we’ve found kind of our method of choice for tissue sections
• 38:48 - 38:57: that have been formed with fixed paraffin embedded is to use heat-induced antigen retrieval
• 38:57 - 39:04: with a basic buffer and a pressure cooker for antigen retrieval. And that’s something that
• 39:04 - 39:10: we’ve looked at for multiple different clones, what works best. And yes, it’s true there’s
• 39:10 - 39:16: some antigens that would prefer an acidic buffer, but a lot of those will still work
• 39:17 - 39:24: in this basic buffer. I think there’s other literature out there showing that this basic
• 39:24 - 39:30: buffer is kind of more general and will be used for multiple antigens, works well for multiple
• 39:30 - 39:35: antigens. So, that’s what we’ve gone through. But people have had success with all sorts of
• 39:35 - 39:42: other antigen retrieval processes as well with the multiplexed image. Great. I think, you know,
• 39:42 - 39:47: you’ve touched upon the leading methodologies at the moment. You mentioned what’s important
• 39:47 - 39:52: for antibody selection and validation, and also the fact that antigen retrieval method is quite
• 39:52 - 39:58: important for this. And I’m assuming just to let our audience know, people can do these studies
• 39:58 - 40:04: both in sort of archived tissue as well as fresh tissue. So, you know, you can go back and look at
• 40:04 - 40:09: clinical trial material. But also, I’m assuming this can be done for, let’s say, experimental
• 40:09 - 40:16: models like mouse and rat as long as antibodies are there. And even in vitro as well. Cell lines
• 40:16 - 40:24: grown on glass can also be used quite nicely for all these methods. And as always, it’s cleaner.
• 40:24 - 40:30: The in vitro systems are always nice and crisp and clean compared to the complexity and difficulties
• 40:30 - 40:35: that can be coming from processed tissues or tissues from the clinic that the researchers
• 40:35 - 40:39: themselves can’t control all the steps of the processing before you receive.
• 40:40 - 40:45: Well, that’s great. I just leave it for me to thank you again for this great presentation.
• 40:45 - 40:52: And we look forward to seeing, you know, papers from your own lab, the Jackson Lab in Toronto.
• 40:52 - 41:06: That will be at least as, at least as exciting as the ones from the Bodenmiller Lab. Thank you very much.