Unravelling tissue pathology by multiplexed imaging

Video transcript

• 00:00 - 00:12: Hi, everyone.
• 00:12 - 00:17: My name is Leeat Keren, and I’m from the Department of Molecular Cell Biology at the Weizmann
• 00:17 - 00:18: Institute of Science.
• 00:18 - 00:25: I will be talking today about how we use multiplexed imaging to look at tissue pathology, and I’m
• 00:25 - 00:30: very excited to be presenting here before you today.
• 00:30 - 00:35: So my lab is very interested in the tumor microenvironment, and the tumor microenvironment
• 00:35 - 00:37: is incredibly complex.
• 00:37 - 00:44: It is comprised of many different types of cells, and these include not only the transformed
• 00:44 - 00:50: tumor cells, but also different kinds of vasculature, fibroblasts, and different types of immune
• 00:50 - 00:51: cells.
• 00:52 - 00:57: And it’s really the interactions between all of these cell types working together that
• 00:57 - 01:03: drive complicated phenotypes that we’re interested in understanding, such as tumor progression
• 01:03 - 01:08: or metastases or response to treatment.
• 01:08 - 01:11: So the way people have been looking at the tumor microenvironment for a while has kind
• 01:11 - 01:16: of been you can divide it into, let’s say, two different types of technologies.
• 01:16 - 01:21: The first one is using imaging approaches, and this is what is routinely used in the
• 01:21 - 01:22: clinic.
• 01:22 - 01:26: And imaging is really great because it gives you a lot of information about the tissue
• 01:26 - 01:28: and what is happening in it.
• 01:28 - 01:33: However, you can only visualize the expression of a few proteins at a time.
• 01:33 - 01:37: On the other hand of the spectrum, we have a lot of omics approaches, and these give
• 01:37 - 01:43: you a lot of molecular details about what’s happening in the tissue.
• 01:43 - 01:49: However, these all require tissue dissociation, and so you really lose the context for what
• 01:49 - 01:55: the cells were experiencing when they had a particular expression profile.
• 01:55 - 01:59: In recent years, there have been a couple of attempts to bring these technologies closer
• 01:59 - 02:05: together by developing new types of multiplexed imaging approaches, where the idea is to be
• 02:05 - 02:12: able to visualize many proteins in situ in intact tissue sections.
• 02:12 - 02:17: The specific technology that I work with is called MIBI-TOF, which stands for Multiplexed
• 02:17 - 02:22: Ion Beam Imaging by Time of Flight, and it was developed by my postdoctoral advisors,
• 02:22 - 02:28: Michael Angelo and Sean Bendel from Stanford University.
• 02:28 - 02:33: What MIBI-TOF allows us to do is really to look at a lot of proteins at the same time,
• 02:33 - 02:37: and the protocol that we use for staining is actually very much reminiscent of how one
• 02:37 - 02:40: would conduct regular IHC.
• 02:40 - 02:42: So what we do is as follows.
• 02:42 - 02:47: We start with a tissue biopsy, and this can be any tissue biopsy, even clinical specimens,
• 02:47 - 02:52: and then we stain it with a mixture of 40 different antibodies, very similar to how
• 02:52 - 02:58: one would conjugate antibodies to fluorophores to visualize them using immunofluorescence.
• 02:58 - 03:04: Here we conjugate our antibodies to metal tags and then visualize them by time of flight
• 03:04 - 03:05: mass spec.
• 03:05 - 03:10: The way this works is that we use secondary ionization mass spec.
• 03:10 - 03:15: We basically shoot at the sample with a primary ion beam, and when these primary ions hit
• 03:15 - 03:22: the sample, secondary ions come flying out, and this will include a lot of organic material
• 03:22 - 03:27: that is present in the tissue, things such as carbon and nitrogen and oxygen, but in
• 03:27 - 03:31: addition to that, it will also include these metal tags that we have conjugated to the
• 03:31 - 03:33: antibodies.
• 03:33 - 03:39: So now what we can do is we can actually raster the tissue, going over it pixel by pixel,
• 03:39 - 03:44: repeat the process for each pixel, and so for each pixel, what we will get is a mass
• 03:44 - 03:50: spectrum depicting the elements in that pixel, and from that, we can then reconstruct the
• 03:50 - 03:54: n-dimensional image.
• 03:54 - 03:58: Here are a couple of validations and examples to show you how the data looks like.
• 03:58 - 04:05: This is specifically in human tonsil, and here I’ve just outlined in this circle the
• 04:05 - 04:10: germinal center, and we’re staining here for two antibodies, two targets.
• 04:10 - 04:16: You can see staining for CD3, which marks T-cells in red, and CD20, which marks B-cells
• 04:16 - 04:20: in green, and you can appreciate very nicely a couple of features.
• 04:20 - 04:26: The first one is that the CD20 staining, the B-cells are primarily located within the germinal
• 04:26 - 04:32: center and that the T-cells are located outside in the interfollicular regions.
• 04:32 - 04:37: Also if we zoom in at a specific region and show it in high magnification, you can really
• 04:37 - 04:43: appreciate the membranous staining of both CD3 and CD20, which is exactly what one would
• 04:43 - 04:45: expect.
• 04:45 - 04:50: You can also appreciate the fact that we don’t see any cells which are expressing both CD3
• 04:50 - 04:56: and CD20, right, all the cells here are either green or red, which again matches what we
• 04:56 - 05:00: know about the biology of these cells.
• 05:00 - 05:01: Here is another example.
• 05:01 - 05:08: This time I’m showing staining for CD3, CD4, and CD8, CD3 is in red, CD4 in green, and
• 05:08 - 05:09: CD8 in blue.
• 05:09 - 05:13: And it’s important to say that all of these stains, also the one that I showed you before,
• 05:13 - 05:17: they were all acquired together, but I’m just now showing them to you one by one because
• 05:17 - 05:22: it’s difficult to visualize 40 proteins at the same time.
• 05:22 - 05:26: And here again, what you can appreciate quite nicely is that all of these different types
• 05:26 - 05:31: of T cells are really localized outside of the germinal center.
• 05:31 - 05:36: You can again appreciate the fact that these stainings are membranous and not nuclear.
• 05:36 - 05:40: And finally, you can also appreciate the fact that we see here two kinds of cells.
• 05:40 - 05:44: So we see here yellow cells and magenta cells.
• 05:44 - 05:50: So these yellow cells are cells that are expressing both CD3 and CD4, these are T-helper cells,
• 05:50 - 05:55: and the magenta cells are cells that are expressing both CD3 and CD8.
• 05:55 - 05:57: These are cytotoxic T cells.
• 05:57 - 06:03: Notably, you don’t find here any cyan cells, so you don’t see co-expression of CD4 and
• 06:03 - 06:08: CD8, again matching what we know about the biology of these cells.
• 06:08 - 06:12: Here you can see another example where I’m just quickly going to show you some nuclear
• 06:12 - 06:13: staining.
• 06:13 - 06:19: So here we’re staining in red for FOXP3, FOXP3 is a transcription factor, which is expressed
• 06:19 - 06:24: by a certain type of T regulatory T cells, and you can see here quite nicely these cells
• 06:25 - 06:30: where FOXP3 is really staining in the nucleus, and it is surrounded on the membrane by expression
• 06:30 - 06:31: of CD4.
• 06:31 - 06:36: Again, what we would expect for these cells, and it’s important to say that both these
• 06:36 - 06:44: targets that are shown here, FOXP3 and PD-1, shown here in blue, are low abundant targets,
• 06:44 - 06:48: which really illustrate the sensitivity of the approach.
• 06:48 - 06:54: So we’re quite confident that, you know, what we’re measuring is indeed what one would expect.
• 06:54 - 06:59: Some of the properties of mass-based imaging and why we think it’s great.
• 06:59 - 07:04: First thing to explain is that really the multiplexing of 40 different channels comes
• 07:04 - 07:11: from the low spectral overlap when we look at mass spectrometry data, and this is unlike
• 07:11 - 07:13: the case in fluorescence.
• 07:13 - 07:18: So that means that the entire reaction is performed together and there are no cyclical
• 07:18 - 07:19: stages.
• 07:19 - 07:25: So we create a panel of 40 antibodies, everything is stained at once, and everything is read
• 07:25 - 07:26: at once.
• 07:26 - 07:31: In terms of resolution, we very routinely acquire the data at resolutions of 400 to
• 07:31 - 07:32: 500 nanometers.
• 07:32 - 07:39: However, it is possible with the instrument to go up to resolutions of about 260 nanometers,
• 07:39 - 07:43: which really gives you very nice subcellular detail.
• 07:43 - 07:48: And the field of view size that we routinely image is about one millimeter squared, and
• 07:48 - 07:53: of course, it is possible to increase beyond that if you now tile several such squares
• 07:53 - 07:56: adjacent to each other.
• 07:56 - 08:01: Importantly, this methodology is compatible with FFPE, which stands for formalin-fixed
• 08:01 - 08:07: paraffin-embedded tissue, which is really the routine way in which samples are stored
• 08:07 - 08:13: in the clinic, which allow us to go back to the hospitals and retrieve archival samples,
• 08:13 - 08:21: which give us, you know, very long-term and meaningful clinical follow-up.
• 08:21 - 08:25: Some of the analytical properties of the system which we have validated.
• 08:25 - 08:32: So first of all, MIBI-TOF is incredibly sensitive, and we think that we can detect down to single-digit
• 08:32 - 08:35: molecules of different antibodies.
• 08:35 - 08:40: Also the dynamic range is quite nice, and you can see here that we get very linear dynamic
• 08:40 - 08:45: range spanning five orders of magnitude.
• 08:45 - 08:50: Another nice property about MIBI-TOF is that it can actually be customized for a variety
• 08:50 - 08:51: of applications.
• 08:51 - 08:56: This is because, as I told you, we really acquire the data pixel by pixel, so you can
• 08:56 - 09:02: play with the size of the pixels, and you can play with how densely you want to space
• 09:02 - 09:07: these pixels, and therefore customize both the size of the field of view that you’re
• 09:08 - 09:10: visualizing, as well as the resolution.
• 09:10 - 09:16: So here, for example, you can see the same data being acquired at four different resolutions,
• 09:16 - 09:19: and you can appreciate the increase in detail.
• 09:19 - 09:24: This is, for example, for double-stranded DNA as we increase the resolution, and here
• 09:24 - 09:29: you can also see a lamin A/C, which is located at the nuclear envelope.
• 09:29 - 09:32: Again, we’re in a very quick survey stand.
• 09:32 - 09:37: We can’t really differentiate the details, but as we go into higher and higher resolutions,
• 09:37 - 09:43: you can really start to appreciate these fine details of the nuclear envelope.
• 09:43 - 09:47: Of course, there is a trade-off between the resolution in which you image and the time
• 09:47 - 09:50: that it takes you to acquire the data.
• 09:50 - 09:54: So the higher the resolution, the more pixels you need to acquire, and the longer it will
• 09:54 - 09:58: take to acquire the image.
• 09:58 - 10:02: Another nice property about MIBI is that we can actually use this property of being able
• 10:02 - 10:07: to perform rescans in order to acquire 3D data.
• 10:07 - 10:12: So here I’m just going to flash quickly a movie, and I would like you to focus on these
• 10:12 - 10:13: two arrows.
• 10:13 - 10:17: And what you’re going to see here as the movie plays is you’re going to see that we’re going
• 10:17 - 10:24: down in Z within the tissue, and over here in this arrow, you’ll see a cell really coming
• 10:24 - 10:28: into the field of view, into the plane, sorry, a cell leaving the plane, and here you’ll
• 10:28 - 10:31: see a nucleus coming into the plane.
• 10:31 - 10:33: So let me play that quickly again for you.
• 10:33 - 10:38: So you can see this nucleus here disappearing as we go down in Z, whereas here you can see
• 10:38 - 10:40: this nucleus appearing.
• 10:40 - 10:43: So we can acquire 3D data.
• 10:43 - 10:48: So now I would like to talk about really how we start to go about these experiments, and
• 10:48 - 10:53: very specifically, if we now want, we have a specific cohort that we would like to interrogate
• 10:53 - 10:59: using multiplexed imaging, where it’s the MIBI-TOF or another multiplexed imaging approach.
• 10:59 - 11:04: So how do we choose and validate the reagents that we’re going to use?
• 11:04 - 11:10: And so I think the most important part is, of course, the antibodies, because this approach
• 11:10 - 11:14: relies on antibodies, and they’re really at the basis of everything, and an antibody that
• 11:14 - 11:22: is not good in IHC will also not work well in MIBI very naturally.
• 11:22 - 11:26: And so the first things that we do is we really look for antibodies which are compatible for
• 11:26 - 11:33: IHC, and it is important to note that these antibodies need to be compatible for IHC using
• 11:33 - 11:38: the conditions that you’re going to be using in your own experiment.
• 11:38 - 11:42: The second thing that we do is we look at these different antibodies shown in the company’s
• 11:42 - 11:48: websites and see which of these have been used in previous publications and whether
• 11:48 - 11:56: really they give you the kind of robust and consistent and well-defined staining that
• 11:56 - 11:57: you’re interested in seeing.
• 11:57 - 12:01: And it is, of course, very important to note that usually these images look better than
• 12:01 - 12:04: what you would expect for clinical samples.
• 12:04 - 12:09: So if the image doesn’t look well, then you definitely don’t want to use this antibody,
• 12:09 - 12:14: and you really want to look for the ones that you think are going to perform well.
• 12:14 - 12:21: An important part of choosing antibodies for multiplexed imaging using mass spectrometry
• 12:21 - 12:23: is the formulation.
• 12:23 - 12:27: So we require all of our antibodies to be BSA-free.
• 12:27 - 12:32: The reason for that is that when we perform the conjugation, we don’t want our metals
• 12:32 - 12:38: to be conjugated to the BSA, but rather we want them to be conjugated to the antibody
• 12:38 - 12:44: so that the metals will then be good reporters for the antibody, and so we always need to
• 12:44 - 12:50: check that the antibody is available in a BSA-free format.
• 12:51 - 12:56: We need to check, as I’ve said before, that it matches our staining protocol.
• 12:56 - 13:00: And here, very importantly, it’s the antigen retrieval conditions.
• 13:00 - 13:07: So different antibodies are usually validated and work best at slightly different antigen
• 13:07 - 13:09: retrieval conditions.
• 13:09 - 13:15: And when you do single-plex IHC or immunofluorescence, you have the option to play with your antigen
• 13:15 - 13:21: retrieval conditions and really tailor these to the specific antibody that you’re using.
• 13:21 - 13:26: However, when you use a large panel of 40 different antibodies, you can’t tailor the
• 13:26 - 13:32: protocol to the antibodies, and you have to tailor the antibodies to the protocol.
• 13:32 - 13:38: And so it’s very important both to check online and also then to validate experimentally that
• 13:38 - 13:43: these antibodies work well with the staining protocol that you will eventually be using
• 13:43 - 13:46: for your assay.
• 13:46 - 13:53: And of course, you know, it’s very important to look at images showing the staining of
• 13:53 - 13:56: this antibody in different tissues.
• 13:56 - 14:00: One thing that I would like to note here is that some companies, and this is including
• 14:00 - 14:06: Abcam, have this option of recombinant antibodies.
• 14:06 - 14:12: And in the event that we can, we really like using these recombinant antibodies for a variety
• 14:12 - 14:13: of reasons.
• 14:13 - 14:18: First, you know, with them, it’s very easy to control the formulation that we need.
• 14:18 - 14:25: And we’ve also found that these tend to give us a higher consistency and less batch-to-batch
• 14:25 - 14:29: variability over time and just overall better performance.
• 14:29 - 14:34: And so, you know, when we try and evaluate different reagents, this is definitely something
• 14:34 - 14:38: that we take into consideration.
• 14:38 - 14:42: Once we purchase an antibody, the next thing that we do is we actually perform a pretty
• 14:42 - 14:48: extensive validation procedure in-house within the lab.
• 14:48 - 14:53: This includes the first thing that we do is we take the antibody and we run it on a gel.
• 14:53 - 14:58: This is really to validate that the antibody is in good condition and that it really doesn’t
• 14:58 - 14:59: contain any BSA.
• 14:59 - 15:04: So here you can see an example for such a gel where we have in this column the control
• 15:04 - 15:07: where we run just BSA.
• 15:07 - 15:10: In the second column, another control where we just run GOAT IgG.
• 15:10 - 15:14: And here we have two different antibodies that we’re running.
• 15:14 - 15:19: And we really want to see that we’re seeing nice bands and that we don’t see any staining
• 15:19 - 15:24: here for BSA.
• 15:24 - 15:30: Also in this list of validations, we will start by doing an IHC using our MIBI staining
• 15:30 - 15:31: protocol.
• 15:31 - 15:35: This is really to validate that, you know, using the conditions that we will ultimately
• 15:35 - 15:42: be using, that this antibody performs well and gives us the nice kind of crisp staining
• 15:42 - 15:45: that we would want to see.
• 15:45 - 15:50: After we’ve performed IHC with the antibody, we will then conjugate it to the metal tag
• 15:50 - 15:52: that we’re going to be using.
• 15:52 - 15:59: And I’ll talk a little bit about how to choose the appropriate metal tag for each antibody.
• 15:59 - 16:03: But once we have this antibody conjugated to a metal tag, we then like to perform an
• 16:03 - 16:07: additional IHC, this time with the conjugate.
• 16:07 - 16:12: And this is really to validate that the conjugation reaction did not alter anything about the
• 16:12 - 16:16: specificity or the performance of the antibody.
• 16:16 - 16:21: And once we’re convinced that the conjugate really gives us nice staining with IHC, only
• 16:21 - 16:29: then do we take it for mass-based multiplexed imaging.
• 16:29 - 16:34: A few more steps in the validation that we do in order to test that the antibody really
• 16:34 - 16:36: works and that we’re seeing what we like.
• 16:36 - 16:40: The first thing that we do is we stain in control tissues.
• 16:40 - 16:43: So here, these are similar images to what I’ve shown you before.
• 16:43 - 16:51: These are different antibodies marking targets in immune cells, CD20, CD3, CD4, and PD-1.
• 16:52 - 16:55: And on the top row, I’m showing you images that were acquired by MIBI.
• 16:55 - 17:02: And on the bottom row, I’m showing you the equivalent staining for these antibodies by
• 17:02 - 17:03: IHC.
• 17:03 - 17:08: And so, you can really appreciate here, you know, that you really want to see that these
• 17:08 - 17:13: antibodies are expressed in the regions that you would expect them to be expressed.
• 17:13 - 17:18: So, for example, CD20, as I’ve shown you before, should be expressed in the germinal centers
• 17:18 - 17:19: of tonsils.
• 17:19 - 17:24: And this is exactly what we see both by the MIBI image and in IHC.
• 17:24 - 17:30: The CD3, on the contrary, needs to be excluded from the germinal center except for individual
• 17:30 - 17:33: cells and really localize around it.
• 17:33 - 17:38: And again, this is clearly what we see both for the MIBI and for the IHC images.
• 17:38 - 17:44: And so, this is kind of a good control to validate that the antibody is working where
• 17:44 - 17:48: you’ve tested in control tissues where you have good expectations of what you would like
• 17:48 - 17:50: to see.
• 17:50 - 17:56: We also use the colocalization of markers as something that will tell us if the antibody
• 17:56 - 17:57: is performing well or not.
• 17:57 - 18:02: And this is really using the advantage of the fact that we can do multiplexed imaging
• 18:02 - 18:06: and look at simultaneously at many different antibodies.
• 18:06 - 18:10: So, here for an example, this is the same example that I showed you before.
• 18:10 - 18:17: If T cells, all T cells will express CD3 according to textbooks, and then, you know, they come
• 18:17 - 18:22: in these two different flavors, T helper cells that also express CD4 and cytotoxic
• 18:22 - 18:29: T cells, which also express CD8, then I really would expect now that, you know, if I stain
• 18:29 - 18:35: CD3 in green, CD4 in blue, and CD8 in red, I will expect to see cyan cells and yellow
• 18:35 - 18:40: cells, but no magenta cells because I don’t expect to see co-expression of CD8 and
• 18:40 - 18:42: CD4 on the same cell.
• 18:42 - 18:47: And this is exactly what I want to see when I do these kinds of stainings in my control
• 18:47 - 18:48: tissues.
• 18:48 - 18:54: And here you can see this quantified quite nicely.
• 18:54 - 18:59: Another thing that we use in order to validate our antibodies is really the subcellular localization
• 18:59 - 19:01: of the staining that we see.
• 19:01 - 19:05: So, we know that for some markers, we expect to see membranous staining.
• 19:05 - 19:11: Here you can see an example for CD45 or for HLA-1, whereas for other markers, we would
• 19:11 - 19:12: expect nuclear staining.
• 19:12 - 19:15: This is, for example, for FOXP3 or for KI67.
• 19:15 - 19:20: And so, this is, of course, you know, something that you can really use in order to evaluate
• 19:20 - 19:24: whether the reagent is working properly.
• 19:24 - 19:27: Finally, we also like to perform replicates.
• 19:27 - 19:33: So, as I told you in our multiplexed imaging, we really can’t perform exactly the same replicates
• 19:33 - 19:38: because once we image a tissue, then it no longer exists, we ablated it, but we can actually
• 19:38 - 19:45: take serial sections, and we like to perform stainings of serial sections really to see
• 19:45 - 19:56: the consistency of the performance of the reagents across time and space.
• 19:56 - 20:02: After you’ve chosen appropriate reagents for your panel, the next big thing is to then
• 20:02 - 20:07: construct all of them together and bring them to generate your panel.
• 20:07 - 20:10: And so, when you’re thinking about putting all these reagents together to generate the
• 20:10 - 20:16: panel, there are a few points that you would probably like to consider when deciding, you
• 20:16 - 20:20: know, which metal you’re going to conjugate to which antibody.
• 20:20 - 20:23: The first point to think about is channel sensitivity.
• 20:23 - 20:28: So, according to the modality that you’re going to use, different channels may have
• 20:28 - 20:29: different sensitivities.
• 20:29 - 20:34: This is true both for fluorescence and also for mass-based imaging.
• 20:34 - 20:39: And so, generally, you know, you would like to save your more sensitive channels for targets
• 20:39 - 20:46: that are more difficult or lower abundance and, you know, kind of give your more dirty,
• 20:46 - 20:52: less sensitive channels to antibodies that are really, you know, very high staining and
• 20:52 - 20:55: always perform very well.
• 20:55 - 21:00: Another important point to consider when you’re thinking about how to build your panel together
• 21:00 - 21:03: is this issue of bleed over.
• 21:03 - 21:09: So, although, you know, our mass peaks are much better separated than one usually finds
• 21:09 - 21:14: in fluorescence, we still have some bleed over between channels.
• 21:14 - 21:19: This can be either from adjacent channels because of just, you know, tailing of the
• 21:19 - 21:23: signal if it’s a very strong signal, for example.
• 21:23 - 21:30: It can also be due to the formation of hydrides and oxides and hydroxides in the tissue.
• 21:30 - 21:35: So, that means that, you know, when we ionize the tissue and we have the secondary ions
• 21:35 - 21:40: coming out, we expect them to measure the mass of the metal that we conjugated to the
• 21:40 - 21:41: antibody.
• 21:41 - 21:47: Sometimes we will measure a mass which is plus one, which happens, let’s say, if the
• 21:47 - 21:53: metal came out of the tissue, you know, with a hydride attack, with a hydrogen atom
• 21:53 - 21:54: attached to it.
• 21:54 - 22:00: And so, you can expect some bleed over into the channels of plus one, plus 16, and plus
• 22:00 - 22:01: 17.
• 22:01 - 22:03: And this is, again, a thing to consider.
• 22:03 - 22:06: So, let’s say you have a very strong marker.
• 22:06 - 22:12: Maybe it’s not a good idea to put on the plus one, on the mass directly adjacent to it,
• 22:12 - 22:17: a very weak antibody.
• 22:17 - 22:21: Another thing to consider when you’re building your panel is that sometimes we like to divide
• 22:21 - 22:28: our panels into two different panels, actually, and perform two staining reactions.
• 22:28 - 22:33: What this allows us to do is it allows us to play a little bit with the incubation time
• 22:33 - 22:35: of the antibodies on the tissue.
• 22:35 - 22:40: So, in general, for the most part, we perform overnight incubation, where we allow the antibodies
• 22:40 - 22:43: to bind the target overnight.
• 22:43 - 22:48: But for some antibodies, this will result in very, very, very strong staining, which
• 22:48 - 22:53: can then lead to issues of bleed over or problems with channel sensitivity.
• 22:53 - 22:58: And so, sometimes some of our strong markers, for example, antibodies, let’s say, against
• 22:58 - 23:05: histone or DNA, we actually leave for a second round of staining, which we perform on the
• 23:05 - 23:06: second day.
• 23:06 - 23:11: And then, we just incubate the antibodies for an hour with the sample.
• 23:11 - 23:17: Finally, the last point to consider is that, you know, although with mass-based imaging,
• 23:17 - 23:19: we don’t normally use secondaries.
• 23:19 - 23:24: We just conjugate the metal directly to the antibodies and measure them in the tissue.
• 23:24 - 23:26: Using secondaries is, of course, optional.
• 23:26 - 23:32: So, you know, one could, for example, use an antibody conjugated to biotin, and then
• 23:32 - 23:36: come with a secondary antibiotin, which is conjugated to the metal, which then allows
• 23:36 - 23:39: you to amplify the signal.
• 23:39 - 23:44: And this is something that could be used, you know, for these very low abundant, very
• 23:44 - 23:46: difficult targets.
• 23:47 - 23:53: And the last thing that we do is, you know, once you’ve kind of compiled all of your different
• 23:53 - 23:59: agents into a panel, a very good idea is to validate, you know, that you don’t suffer
• 23:59 - 24:02: from all of these issues with partial panels.
• 24:02 - 24:07: So, what I like to do is I would like to take my panel and then divide it into a couple
• 24:07 - 24:08: of groups.
• 24:08 - 24:13: You know, let’s say panel one will include the first two out of these five different
• 24:13 - 24:14: antibodies.
• 24:14 - 24:16: Panel two will include the first, the third, and the fourth.
• 24:16 - 24:20: And panel three will include the first and the fifth.
• 24:20 - 24:25: And just to really make sure that, you know, in this panel, you’re really seeing staining
• 24:25 - 24:31: where you inserted the antibodies, and you really should not expect any signal in channels
• 24:31 - 24:33: for which you have not put the antibodies.
• 24:33 - 24:38: So, you can see this very nicely over here in this serial section of the lymph node.
• 24:38 - 24:44: So, you can see that all these three panels had CD3, and you can see nicely this CD3 staining,
• 24:44 - 24:51: but only the first panel had LAG-3, the second had CD4 and PD-1, and the last one had KI67.
• 24:51 - 24:54: And this is, of course, also visualized here also in the mass spectrum, so you can see
• 24:54 - 25:01: very nicely here this LAG-3 peak that exists only in the first panel, whereas, for example,
• 25:01 - 25:05: this CD4 peak exists only in the second panel.
• 25:06 - 25:07: Okay.
• 25:07 - 25:13: While this webinar is mainly dedicated to the assay and to the reagents, I’m just going
• 25:13 - 25:19: to close by two minutes talking very, very briefly about data analysis.
• 25:19 - 25:25: And so, of course, all these multiplexed imaging modalities have a huge challenge of data analysis.
• 25:25 - 25:32: So, you know, this is an example from a cohort where we looked at 41 different triple negative
• 25:32 - 25:38: breast cancer patients, and we stained them using 36 different antibodies.
• 25:38 - 25:44: And so, you know, multiplying 41 by 36, this results in nearly 1,500 images.
• 25:44 - 25:49: I know this is really an overwhelming amount of data from which we then want to deduce
• 25:49 - 25:54: principles in the organization of the tumor immune microenvironment.
• 25:54 - 25:58: So, I’m just going to outline one approach that we’ve used and found successful in the
• 25:58 - 26:03: past in order to look at this data, but, you know, this is, of course, it’s important
• 26:03 - 26:08: to say a very active field, which is still developing.
• 26:08 - 26:13: And so, what we usually do with our data, once, you know, while we’ve stained the samples
• 26:13 - 26:19: and we visualize them and we image them, we start with these, you know, n-dimensional
• 26:19 - 26:24: images, right, where each n is a different channel, a different antibody that we stain
• 26:24 - 26:26: with.
• 26:26 - 26:29: The first thing that we usually do is we perform segmentation.
• 26:29 - 26:35: This is the task of identifying individual cells in the sample.
• 26:35 - 26:41: Here I would like to give a huge shout out to DeepCell, which is a segmentation platform
• 26:41 - 26:46: developed by our good collaborator, David VanValen from Caltech, and we really found
• 26:46 - 26:54: it to be much superior in terms of segmentation to anything else that we’ve tried.
• 26:54 - 26:59: So, once we have cells segmented in the tissue, we then annotate for each cell how much it
• 26:59 - 27:04: expressed of the different proteins that we stained first, so for each cell we’ll have
• 27:04 - 27:09: how much it expressed of, let’s say, FOXP3 and CD4 and CD8, and we then can construct
• 27:09 - 27:15: this kind of matrix where we have all of the different cells and how much they expressed
• 27:15 - 27:16: of the different proteins.
• 27:16 - 27:20: This now looks very similar to the type of data that one would get, let’s say, from a
• 27:20 - 27:26: flow cytometry experiment or a SIDOF experiment or a single cell RNA sequencing experiment,
• 27:26 - 27:31: and so we can take these single cells, we can cluster them into different cell types,
• 27:31 - 27:36: but then the nice thing is that we can go back to the image and overlay these cell types
• 27:36 - 27:41: on top of the image to start asking interesting questions about how these cell types relate
• 27:41 - 27:42: to one another.
• 27:42 - 27:49: And so, some of the things that we use this data to ask is we ask about spatial enrichment,
• 27:49 - 27:54: so let’s say I have here green cells and purple cells, so do green cells always sit
• 27:54 - 27:59: next to purple cells, do they sit next to purple cells only in patients in which therapy
• 27:59 - 28:02: succeeded or didn’t succeed?
• 28:02 - 28:06: We also try and identify multicellular structure.
• 28:06 - 28:11: These are things, as illustrated here, for example, like the tumor immune boundary, and
• 28:11 - 28:15: we use all these different features, so, you know, the cell types that are present in the
• 28:15 - 28:20: tissue, the interactions between them, and the multicellular structures that we identify
• 28:20 - 28:26: to then stratify different environments that we find in different patients and then try
• 28:26 - 28:32: and correlate all these different attributes to their survival or any other clinical attributes
• 28:32 - 28:36: that we’re interested in identifying.
• 28:36 - 28:41: And so here I’m just going to very, very briefly show you one example from work that we did
• 28:41 - 28:48: in triple negative breast cancer, where really we identify here this border between the immune
• 28:48 - 28:54: region and the tumor region, and we could really tease out this very nice organization
• 28:54 - 28:59: whereby next to these tumor cells that are colored here in gray, we find this layer of
• 28:59 - 29:06: MDSCs colored here in this cyan and green colors, which are really separating between
• 29:06 - 29:11: the tumor cells, and this layer, which is enriched in lymphocytes, colored here in these
• 29:11 - 29:13: purple and pink colors.
• 29:13 - 29:22: So this is, you know, the types of organization that one can find using these kinds of approaches.
• 29:22 - 29:26: And I think really, you know, taking a step backwards, the nice thing about these kinds
• 29:26 - 29:33: of approaches is that they really allow you to put together three different types of data
• 29:33 - 29:37: that I think have been measured in the past, but individually, and now we can actually
• 29:37 - 29:39: combine them together.
• 29:39 - 29:43: So you know, in the same image, you can really start asking questions about the histology
• 29:43 - 29:50: of the tissue, really the grand organization of how this tissue behaves, connect that with
• 29:50 - 29:55: the different cell types that are infiltrating into the tissue, and really go down to the
• 29:55 - 30:00: molecular level to start interrogating, you know, really who are the different pathways
• 30:00 - 30:06: who are mediating these interactions between these different cell types and driving perhaps
• 30:06 - 30:10: these histologies that we observe in different patients.
• 30:10 - 30:14: We’ve now applied these kinds of approaches across many, many disciplines.
• 30:14 - 30:20: So we have now examples looking at this in oncology, so in breast cancer and colorectal
• 30:20 - 30:21: cancer.
• 30:21 - 30:28: We’ve also used it to look at infectious diseases, such as tuberculosis, where we’re using it
• 30:28 - 30:32: to look at neurodegenerative diseases, such as Alzheimer’s.
• 30:32 - 30:38: And as I said, we’re very much actively developing the computational pipelines to deal with these
• 30:38 - 30:40: types of data.
• 30:40 - 30:44: And with that, I’d like to thank a lot of people who really contributed to a lot of
• 30:44 - 30:47: the things that I’ve shown you today.
• 30:47 - 30:52: MIBI-TOF was, as I said, developed by Mike Angelo together with Sean Bendall.
• 30:52 - 30:56: There’s a lot of people from their lab who really worked very hard to put together these
• 30:56 - 30:58: protocols that I’ve been talking about today.
• 30:58 - 31:08: So Mark Bosse and Diana, Steve Thompson, Roshan, Erin, Noah, Tyler, and Robert West, who was
• 31:08 - 31:13: our pathologist collaborator on the breast cancer work that I’ve shown you, and also
• 31:13 - 31:21: from Caltech, David VanValen, who is a great collaborator working with us on image segmentation.
• 31:21 - 31:26: And this is my newly established lab at Weizmann.
• 31:26 - 31:33: Tomer is working in the core facility and is in charge of our multiplex imaging technology.
• 31:33 - 31:35: And these are my funding sources.
• 31:35 - 31:36: And thank you for listening.