Introduction to Mass Cytometry

Top

00:00:13.00 Hello.
00:00:13.18 My name is Susanne Heck.
00:00:15.22 I'm leading the flow cytometry core facility
00:00:17.24 at the Biomedical Research Center of Guy's and St Thomas' Hospital
00:00:19.29 and King's College in London.
00:00:21.19 And I'm here to give you an introduction
00:00:23.25 into mass cytometry today.
00:00:25.11 Mass cytometry is a version of flow cytometry,
00:00:28.05 which we know has been used for
00:00:31.08 phenotype analysis for many years.
00:00:32.26 However, mass cytometry has moved phenotype analysis
00:00:37.18 into the high-dimensional arena
00:00:39.29 and made it a true "-omics" tool.
00:00:41.26 So, let's have a look why that is.
00:00:44.02 When we normally phenotype complex mixtures
00:00:46.23 such as blood or bone marrow,
00:00:48.16 we use a number of fluorescently conjugated probes
00:00:51.11 to serve as intracellular markers.
00:00:53.12 When we just look at the surface, it's called phenotyping.
00:00:56.00 If we are looking at their functions,
00:00:58.05 we call it functional phenotyping.
00:01:00.00 But what you'll also notice is
00:01:02.00 our fluorochromes have some spectral overlap,
00:01:04.10 and that limits the number of parameters
00:01:06.13 that we can analyze in one panel.
00:01:08.29 In traditional panels and cytometers,
00:01:12.00 that's normally 12-14.
00:01:14.10 And in the very highest end instruments,
00:01:16.12 we have 28 detection channels.
00:01:18.22 However, in complex mixtures,
00:01:20.25 like bone marrow for example,
00:01:22.10 we have many different cell types,
00:01:23.26 and therefore researchers are faced with a choice.
00:01:26.23 Either they look broadly across many different cell types,
00:01:29.13 but then do not have many markers left
00:01:32.13 to evaluate their function states,
00:01:35.25 or they decide to very deeply dig
00:01:38.17 into one particular cell type,
00:01:40.19 but at the expense of all the other players
00:01:43.08 in this mixture.
00:01:44.23 And this is exactly where mass cytometry come in
00:01:46.26 and truly makes a difference.
00:01:48.02 Here, we are no longer using fluorochromes,
00:01:50.29 but we are labeling our probes with metal isotopes,
00:01:53.10 meaning with metals that are conjugated to our antibodies.
00:01:57.04 And we are detecting those in a mass spectrometer,
00:01:59.18 which separates these by time of flight.
00:02:02.07 The advantage being masses
00:02:05.01 do not have a spectral overlap,
00:02:07.01 and that allows us to put many more markers
00:02:09.12 into one particular panel,
00:02:11.17 and that means you can now go for very deep phenotyping.
00:02:14.24 And that also results in
00:02:18.02 high-dimensional data coming out of
00:02:20.08 each one of those experiments.
00:02:21.25 And it will allow us to analyze those data
00:02:25.18 with high-dimensional data analysis tools,
00:02:27.13 as we will see in the end.
00:02:28.26 So, let's have a look at this technology in a bit more detail.
00:02:31.22 And I will give you, first, an overview of the presentation.
00:02:34.24 We will look at the mass cytometry platform and the workflow.
00:02:38.02 We will spend some time on the reagents and the instrument.
00:02:41.26 Then we will briefly have a look
00:02:44.12 at mass cytometry data
00:02:46.11 and how they are analyzed in high-dimensional space.
00:02:49.04 And we will also look at a bit of literature.
00:02:52.00 So, the mass cytometry platform consists of three parts,
00:02:55.01 the first one being our reagent system.
00:02:57.14 We are actually using the same antibodies
00:02:59.18 that we are also using
00:03:02.02 for fluorescent flow cytometry,
00:03:03.23 only we are tagging them with metal isotopes.
00:03:06.02 And this results in metal-conjugated antibodies.
00:03:11.04 And we are also using metal isotopes to label DNA, RNA,
00:03:13.23 and, for example,
00:03:16.05 also if we want to barcode our sample that is possible.
00:03:19.12 And currently, we have around 50 tags available at our disposal,
00:03:24.19 most of them coming from the lanthanide family.
00:03:26.24 At the moment, there is only one worldwide supplier
00:03:30.00 of such reagents.
00:03:31.20 And despite the fact that their catalog
00:03:33.11 now contains hundreds of different reagents,
00:03:35.01 the repertoire is nonetheless limited,
00:03:37.00 and you very often times end up
00:03:39.16 making your own metal-conjugated antibodies.
00:03:41.24 And I'll show you in a bit more...
00:03:44.06 a bit further down the line how that is done.
00:03:46.24 The second part of our mass cytometry platform
00:03:49.20 is obviously the instrument.
00:03:51.05 And what you are seeing here
00:03:53.04 is a picture of the latest generation of mass cytometer
00:03:55.12 called a Helios.
00:03:56.29 And it's based on a mass spectrometer.
00:04:01.01 Mass specs have been around for a long time,
00:04:03.04 and they are used to, for example,
00:04:05.01 analyze trace element contaminations
00:04:07.23 in complex mixtures.
00:04:09.11 But this particular instrument
00:04:11.11 has obtained a modified front end
00:04:13.05 and allows us to introduce cells.
00:04:15.05 When the technology came out the first time,
00:04:17.05 the instrument was called a CyTOF,
00:04:20.13 meaning a cytometer running by time of flight -- TOF --
00:04:25.04 and many people use this term, also, nowadays.
00:04:28.05 Mass cytometry works with an argon plasma.
00:04:31.11 This particular instrument has 135 detection channels,
00:04:34.28 so we can add quite a few more tags
00:04:37.00 if we want to and can.
00:04:38.29 But there are a few differences
00:04:41.05 to a flow cytometry, for example.
00:04:43.00 Given that it runs on a very hot argon plasma,
00:04:45.11 it destroys the cells that are introduced into it,
00:04:47.11 so it also means you can't sort them;
00:04:49.11 you can't isolate them afterward.
00:04:51.01 And we also are not getting a so-called scatter profile
00:04:53.26 that allows us to see the cells.
00:04:56.07 We need to apply a few tricks to make them visible.
00:04:59.01 The good news is we are obtaining
00:05:01.01 a data format called the fcs standard format
00:05:03.08 that we are used to from flow cytometry.
00:05:05.23 And that also means we can apply
00:05:08.02 the same kind of data analysis tools, overall,
00:05:10.11 as we do to fluorescent flow data.
00:05:13.05 But we are now looking at marker panels
00:05:15.24 of 30-40.
00:05:17.11 And that means traditional Boolean gating
00:05:19.10 is not really giving us sufficient information
00:05:22.14 out of our data set.
00:05:24.04 And therefore the third part of our platform
00:05:26.25 is the high-dimensional data analysis tools,
00:05:28.22 where we can have unsupervised clustering
00:05:31.10 or dimensionality reduction approaches
00:05:33.13 to analyze these data.
00:05:35.08 I have listed a few of them here.
00:05:37.04 There are many more of them.
00:05:38.26 And in the end, we will have a closer look as viSNE and SPADE.
00:05:41.04 For you as a user, it means that to analyze such an experiment
00:05:44.22 you oftentimes want to engage
00:05:46.25 with a bioinformatician
00:05:48.22 or learn how to do this yourself,
00:05:50.14 and there are training programs available now.
00:05:52.24 So, before you as a user
00:05:54.28 start to run a mass cytometry experiment,
00:05:57.06 you will have made a clear experimental plan,
00:05:59.20 have designed your panel,
00:06:01.11 have assigned methods to each probe,
00:06:03.11 have eventually conjugated yourself,
00:06:05.18 had a chat with a bioinformatician
00:06:07.24 on how to analyze your data.
00:06:09.20 And once all of that is clear and present,
00:06:11.14 you will start to label your cells.
00:06:13.13 So, on the left-hand side, you will see...
00:06:16.17 you will see our labeled cells,
00:06:18.23 the gray balls with the sticking-out
00:06:21.25 metal-conjugated antibodies.
00:06:23.13 And these are then introduced into our apparatus.
00:06:25.24 The first thing that brings the cells
00:06:28.10 into the mass cytometry is called a nebulizer.
00:06:31.23 That looks like a... it's a glass structure
00:06:33.22 that looks like a little spray pistol.
00:06:35.20 And our cells, in water,
00:06:37.20 are introduced via the nebulizer into the apparatus,
00:06:39.28 and the nebulizer produces an aerosol of our cells.
00:06:43.04 They are then migrating down the line
00:06:45.16 in our interconnected structures
00:06:48.10 until they end up in an argon plasma.
00:06:51.01 The argon plasma has an average temperature of 7,500 Kelvin.
00:06:55.13 And it atomizes and ionizes
00:06:58.01 all our cell components.
00:06:59.20 So, what we are having at the end here
00:07:01.25 is an ion cloud.
00:07:03.05 The ion cloud consists of everything that we have put in
00:07:05.22 as tags in the beginning,
00:07:07.13 as well as all the elements
00:07:09.12 that our biological sample is made of.
00:07:11.18 And at first, of course, in the apparatus,
00:07:14.07 meaning the cytometer,
00:07:16.19 these ions are filtered further.
00:07:19.16 Some low masses and some unwanted elements are removed.
00:07:22.14 And at the end, we are left with the tags
00:07:24.02 that we have added to label our cells
00:07:26.10 from our panel of antibodies.
00:07:28.17 And these are then introduced into the detection chamber
00:07:31.13 -- the time of flight, or TOF, chamber --
00:07:33.11 where they are separated by their mass-to-charge ratio.
00:07:36.18 The lightest masses arrive at the detector first;
00:07:39.04 the heaviest arrive last.
00:07:40.20 And from each cell that is introduced,
00:07:43.19 we are obtaining a so-called
00:07:46.01 mass fingerprint or mass spectrum.
00:07:48.05 So, you see, down here on the left-hand side,
00:07:51.08 three different color peaks,
00:07:53.04 corresponding to different masses,
00:07:54.25 and each mass corresponds to one particular antibody probe.
00:07:57.16 And these events are then integrated up,
00:08:00.06 put into an fcs data file that we can then, further on, analyze.
00:08:04.20 Okay. Let's have a look at the reagents
00:08:07.04 that we are working with.
00:08:08.23 We are working with isotopes.
00:08:10.17 And this is a periodic table,
00:08:12.12 and highlighted in orange is a...
00:08:14.17 is the main tag group.
00:08:16.11 So, these are our lanthanides,
00:08:18.05 or rare earth metals.
00:08:20.25 And when you look at the highlighted elements...
00:08:23.01 they have names on it, obviously,
00:08:25.06 or abbreviations to be more precise.
00:08:27.01 Let's say Gd for gadolinium,
00:08:29.20 Tb for terbium,
00:08:31.12 Er for erbium.
00:08:32.23 And there are numbers on them.
00:08:34.12 Each number corresponds to the number of naturally occurring isotopes.
00:08:37.23 So, when you, for example,
00:08:39.16 isolate terbium in nature,
00:08:41.13 there is one single isotope available.
00:08:43.06 When you isolate gadolinium,
00:08:45.04 there are five different isotopes available.
00:08:47.13 And for mass cytometry,
00:08:49.18 each one of those isotopes
00:08:52.14 has been separated by... in their mass,
00:08:55.04 so we have five different gadoliniums with five different masses
00:08:58.29 available that we can use.
00:09:00.18 And therefore, out of those 13 elements,
00:09:02.14 we are ending up with 37 different usable tags.
00:09:05.25 You see up here in the periodic table of elements,
00:09:08.25 we have made use of a couple of other elements
00:09:11.04 that also are good for us to use.
00:09:14.00 And we start with the lowest mass of yttrium, right now, at a mass of 89,
00:09:20.07 and the highest mass of bismuth, at 209.
00:09:23.16 So, this is basically giving us the mass range
00:09:26.05 that we are currently analyzing.
00:09:27.25 And these tags can all be added to your probes.
00:09:31.06 So, why lanthanides?
00:09:33.13 Lanthanides have the advantage
00:09:35.20 that they are absent from biological metal.
00:09:37.02 They are not radioactive.
00:09:39.08 And they are also heavier
00:09:41.11 than whatever composes biological metal,
00:09:43.29 and they are in a mass range that we can detect
00:09:46.11 on our instrument.
00:09:47.28 The additional tags
00:09:51.08 add up our toolkit to the 50
00:09:53.25 that I already mentioned.
00:09:55.10 And all of those probes are added to our most of the time
00:09:59.00 antibody probes by chelating agents.
00:10:01.22 Of the chelating agents,
00:10:03.28 there are two different larger groups.
00:10:05.29 The commercial antibodies that are lanthanide-tagged
00:10:10.10 are all using a polymeric linker.
00:10:12.05 So, we are seeing an antibody molecule
00:10:14.17 that is covalently bound to a polymer chain
00:10:17.12 that is loaded with different metals.
00:10:20.02 Each one of those metals is basically
00:10:22.01 one pure metal isotope with one defined mass.
00:10:24.10 So, let's say this has mass 150,
00:10:26.19 conjugated to CD45.
00:10:28.24 And shown here is the structure of such an element,
00:10:32.14 and I would like to point out two particular entities.
00:10:35.12 There's a so-called maleimide linker.
00:10:37.18 This maleimide linker is later on used
00:10:40.14 to conjugate this loaded polymer to your antibody.
00:10:44.07 And this is basically one part of the chelator.
00:10:47.19 And there are a number of repeats to that.
00:10:50.06 Commercial antibodies
00:10:52.10 use a polymer called the X8 polymer,
00:10:54.22 which has 22 such repeats.
00:10:56.11 That means you can at most
00:10:58.24 pack 22 isotopes per polymer chain.
00:11:02.11 And on each antibody,
00:11:04.11 they can conjugate 3-4 such polymer chains,
00:11:06.24 and that leads us to something around
00:11:09.01 80 different isotopes per antibody
00:11:11.14 as our label.
00:11:13.20 This particular polymer works for all the lanthanides,
00:11:16.12 but it doesn't work for some other metals that we can also use.
00:11:19.28 And for those, we are using a different chelator,
00:11:22.04 DOTA.
00:11:24.08 Or, we can use it.
00:11:26.00 DOTA has the difference
00:11:29.11 that we can bind one single metal isotope per molecule,
00:11:32.13 and we can normally bind
00:11:35.06 3-5 such mDOTA molecules per antibody.
00:11:37.25 That means we have less metal labeling density,
00:11:40.07 meaning the reagents that are resulting from those
00:11:43.20 are less sensitive,
00:11:45.16 and they are good for highly expressed antigens
00:11:47.25 that we're interested in coding.
00:11:50.24 In addition to that,
00:11:52.14 we have reagents for cell identification.
00:11:54.23 To identify a cell,
00:11:57.01 we don't have a scatter profile.
00:11:58.26 But to identify a cell in mass cytometry,
00:12:01.01 we are using a natural iridium that has
00:12:04.14 been conjugated, also, to an intercalator,
00:12:06.28 which can basically intercalate into your DNA.
00:12:09.14 Every cell that has a cell nucleus and is intact
00:12:12.25 will have iridium intercalating in it,
00:12:15.02 and this allows us later on to identify a cell event.
00:12:18.08 In addition to that,
00:12:20.08 we need to label dead cells.
00:12:21.29 Dead cell removal is very important in cytometry in general,
00:12:25.16 and absolutely essential in mass cytometry.
00:12:27.11 Because we are looking at 30+ dimensions,
00:12:29.26 and dead cells bind anything unspecifically,
00:12:32.26 so we would get lots of false positive events.
00:12:35.11 For dead cell labeling,
00:12:37.05 we have two metals that are predominantly used.
00:12:41.04 One of them is rhodium,
00:12:43.13 of a mass of 103,
00:12:45.22 or cisplatin.
00:12:47.07 Both of those can bind to dead cells.
00:12:49.04 And before you are labeling your cells
00:12:51.22 with your antibody cocktail,
00:12:53.26 you are tagging the cells that are dead
00:12:56.18 so that they are later on excluded from your analysis
00:12:58.25 and will not give you false positive results.
00:13:01.24 So, once you have your probes all available,
00:13:05.25 you can stain your cells.
00:13:07.20 I have produced us a small panel, here.
00:13:10.02 A panel of
00:13:12.28 rhodium-103 to remove dead cells,
00:13:14.11 iridium to label our DNA nucleus,
00:13:16.16 and three different antibodies
00:13:18.17 -- masses 89, 145, and 170 --
00:13:21.25 tagging CD45 for all lymphocytes,
00:13:25.04 CD33 for all T cells,
00:13:26.29 and CD4 for all T-helper cells.
00:13:30.10 So, once you have this cocktail,
00:13:32.13 you will have your cell solution.
00:13:34.00 We have four cells here, as an example.
00:13:36.17 One of them, on the top left, is dead.
00:13:39.01 It has a porous membrane,
00:13:40.24 so when we are tagging the sample with rhodium,
00:13:42.19 this cell will bind rhodium.
00:13:44.05 All the others won't.
00:13:45.16 After that, we will stain with our antibody cocktail,
00:13:48.00 and they will bind to the cells
00:13:50.24 according to the presence of the specific epitopes
00:13:52.23 we want to label.
00:13:54.01 So now we have a labeled sample.
00:13:56.22 Then we will fix and stain.
00:13:58.17 It's very important to fix and stain with paraformaldehyde,
00:14:01.11 because later these cells will go into water,
00:14:03.14 so we need to structurally preserve them really well.
00:14:05.20 Finally, we will add iridium
00:14:08.13 to label the DNA,
00:14:10.10 and this is actually our cell label.
00:14:11.24 And the last thing that we are adding
00:14:13.18 is something called EQ beads.
00:14:15.11 EQ beads is a mixture of four different beads
00:14:18.26 of different masses ranging from 140 to 175.
00:14:23.04 And they are used to normalize our data.
00:14:25.29 During the day, the instrument shows
00:14:29.07 a certain instrument drift.
00:14:30.13 And to not have this impacting on the data...
00:14:32.07 intensity of your actual data points,
00:14:34.10 we run these beads alongside,
00:14:36.10 and we normalize the data later on,
00:14:38.14 before we put them into high-dimensional data analysis.
00:14:42.02 The next thing that we then need to look at is
00:14:45.06 how our cell migrates through the instrument.
00:14:48.24 So, I'm showing you the passages...
00:14:50.14 the stages of passage in a mass cytometer.
00:14:52.22 We are starting with the sample introduction
00:14:55.01 -- we have already seen the nebulizer --
00:14:57.08 and ending up with the torch at the very end.
00:14:59.14 These are parts that are still outside,
00:15:01.25 and you can see them when you come into a lab.
00:15:03.20 And our cells are at this stage
00:15:05.28 at atmospheric pressure.
00:15:07.10 We are now moving them into the instrument,
00:15:09.24 and we are moving them into deep vacuum in that process.
00:15:12.11 Initially, they come, via a number of cones,
00:15:15.28 into the focusing and ion filter part of the instrument.
00:15:21.26 Here, we remove some unwanted particles
00:15:24.02 and also focus them.
00:15:25.17 Once that has happened,
00:15:26.26 they come into the TOF chamber.
00:15:28.19 And finally, the data that have been detected at the detector
00:15:32.07 are amplified and stored as an fcs data file.
00:15:35.25 Okay.
00:15:36.29 So, let's have a look at all of them in a bit more detail.
00:15:38.29 First, we are looking at the sample introduction part.
00:15:42.04 In the modern cytometers,
00:15:44.16 the modern mass cytometers,
00:15:46.08 we have a sample introduction port.
00:15:48.29 Our samples are put into water,
00:15:51.22 and then they are pressurized by argon gas.
00:15:54.25 It migrates through a sample line into the nebulizer
00:15:57.04 at a relatively low speed
00:15:59.16 of only 30 microliters a minute.
00:16:01.06 And that also runs at a sp...
00:16:04.19 at a density of half a million cells, roughly,
00:16:06.19 so we do not introduce many events per time.
00:16:09.14 This is one other very big difference to flow cytometry.
00:16:12.16 Mass cytometry is quite slow.
00:16:14.22 In flow cytometers, you normally run
00:16:17.22 dozens or hundreds of samples within a day.
00:16:19.10 Here, we are rather talking about
00:16:21.07 8 to maybe 12 samples a day.
00:16:24.00 Once our cells are migrating into the sample lines,
00:16:26.17 they come into the nebulizer,
00:16:28.11 this little spray chamber that produces an aerosol.
00:16:31.10 And they are produced...
00:16:33.04 that aerosol is produced also by using argon gas,
00:16:35.11 as is through the entire apparatus that we are seeing here.
00:16:38.07 And then they move into the spray chamber.
00:16:40.22 In the spray chamber,
00:16:42.04 the cells are dried down a little bit,
00:16:44.17 and our cloud is also reduced
00:16:47.00 and narrowed down a bit more,
00:16:48.29 before they move on into the torch
00:16:51.13 and into the argon plasma.
00:16:52.26 It needs to be said that this introduction process
00:16:55.04 is not extremely efficient.
00:16:56.20 You lose around 50% of the cells that you are introducing,
00:16:59.25 so for every million events that you are introducing,
00:17:03.14 you should expect more than half a million data points.
00:17:06.00 Finally, when our cells have reached the argon plasma,
00:17:09.28 they are very quickly vaporized, atomized, and ionized.
00:17:14.19 So, within the torch,
00:17:16.08 we have an argon plasma,
00:17:17.22 and I will show you in a minute
00:17:19.13 a real life picture of how that works.
00:17:20.22 Before you actually run your sample,
00:17:23.01 the plasma needs to be started and stabilized.
00:17:24.22 So, don't get shocked about too much of the detail.
00:17:27.22 So, what you are seeing here is a real life picture
00:17:30.24 of the torch.
00:17:32.11 The very end of this torch will be enclosed
00:17:34.25 by a so-called RF load coil.
00:17:37.08 And when you start the plasma,
00:17:39.01 argon gas begins to flow.
00:17:40.24 There's an ignition pin that produces an initial ignition,
00:17:44.06 but then the plasma itself containing...
00:17:47.01 because this RF load coil produces a high radio frequency field
00:17:51.05 that basically induces the argon gas
00:17:54.02 to form a plasma, a stable plasma.
00:17:57.16 In real life, it looks like that.
00:17:59.17 So, we are seeing here the end of the torch.
00:18:02.00 This is the argon plasma.
00:18:04.17 And you can see, here,
00:18:06.10 how this argon plasma and also your ion cloud moves down,
00:18:08.14 because what you are seeing here is already the sec...
00:18:11.06 the next part.
00:18:12.18 This is the so-called sampler cone,
00:18:14.02 and this is what sucks in your ion cloud
00:18:16.00 into the instrument,
00:18:17.14 and brings it then into...
00:18:19.11 begins to bring it into a vacuum.
00:18:21.19 So, the ion passage... to orientate you, again,
00:18:25.07 we are, here, having your RF load coil.
00:18:26.20 You are having, here, our ion cloud.
00:18:29.18 And this basically the first part,
00:18:32.10 basically, that takes your cell in.
00:18:33.24 That's the so-called sampler cone,
00:18:35.12 and behind that are two more cones,
00:18:37.18 meaning metal structures that suck in your sample.
00:18:39.29 They are called the skimmer and reducer.
00:18:42.19 In real life, they look like this,
00:18:44.25 and like that.
00:18:46.12 They are basically right next to each other.
00:18:47.25 And together they cool down the sample
00:18:51.01 and they reduce the pressure from atmospheric pressure,
00:18:53.20 where it comes from,
00:18:55.13 to 10^-4 bar, about,
00:18:57.29 and bring your samples that way into the apparatus.
00:19:01.20 Why are we producing a vacuum?
00:19:03.25 Later on, we want to measure
00:19:06.22 the masses of each of our particles,
00:19:08.22 and we would like to be sure that the trajectory of these particles
00:19:11.28 through our detector is reproducible.
00:19:14.25 So, what must not happen is
00:19:17.02 there must not be any other molecules about,
00:19:19.24 where our ions can actually bump into them,
00:19:22.07 and such other things could eventually
00:19:23.29 be the molecules coming out of air.
00:19:25.18 So, we are working in a vacuum
00:19:27.24 so we can have an actual...
00:19:29.28 accurate TOF measurements.
00:19:33.20 Finally, after we have moved through all of our cones...
00:19:38.02 again, to orientate yourself,
00:19:39.22 we are now entering the ion filter optics.
00:19:43.12 The first thing is our optical lenses.
00:19:46.19 They will turn all the charged particles
00:19:49.06 in a 90-degree angle,
00:19:51.10 and all the uncharged items...
00:19:53.25 uncharged ions as well as any photons
00:19:58.02 that have entered are removed,
00:19:59.26 so they are not producing a signal.
00:20:01.17 Whatever is left of our cloud
00:20:03.12 is now making it into the high pass ion optic,
00:20:06.11 also called a quadrupole filter.
00:20:08.06 Right now, we have all the ions
00:20:10.22 that basically make our cells, our antibodies,
00:20:12.22 as well as the tags.
00:20:14.01 They have, however, different masses.
00:20:17.05 In this quadrupole filter, we have four metal rods.
00:20:20.00 And between those metal rods
00:20:22.01 we have applied an electromagnetic field.
00:20:24.00 And this field induces our ions to oscillate,
00:20:27.11 meaning they will begin to swing.
00:20:30.00 Heavier ions make it through this high pass ion optic
00:20:33.01 to the other side.
00:20:34.07 Lighter ones swing a little bit more,
00:20:35.20 and at some point in time they are
00:20:37.13 making contact with the oppositely charged pole
00:20:39.25 and are removed.
00:20:41.04 After that, we only have certain masses left,
00:20:43.20 and we can actually...
00:20:45.07 by tuning this electromagnetic field,
00:20:47.04 you can actually choose what masses make it through.
00:20:50.06 In our case, we are starting to detect masses 89,
00:20:53.18 and that... our measurement range is a bit wider.
00:20:57.05 So, now we have basically left
00:20:59.29 only the tags that we have added to our probes.
00:21:02.23 These tags are then introduced into the TOF chamber.
00:21:05.25 They enter via a slit...
00:21:07.26 which is also one other function
00:21:10.22 of the quadrupole filter.
00:21:12.00 It basically narrows our ion cloud in...
00:21:14.16 or our ion swing in such a way
00:21:16.22 that they fit through this slit.
00:21:18.14 And all the ions that are basically coming
00:21:20.10 from our previously tagged probe
00:21:22.19 are now making it through this particular part of our instrument.
00:21:25.25 Here, we have a very high vacuum.
00:21:28.08 Every 13 microseconds,
00:21:29.28 a portion of ions is thrown in.
00:21:32.29 And then it basically migrates along the trajectory,
00:21:36.08 first down and then up,
00:21:38.07 all regulated by charge,
00:21:40.00 until they are hitting the detector.
00:21:42.07 Initially, we have a mixture.
00:21:44.08 At the detector, they arrive in the order of their masses,
00:21:46.28 because the ions that we are bringing in have the same charge,
00:21:50.06 so the difference between them is their weight,
00:21:53.23 and the lightest ions travel the fastest,
00:21:55.18 so they arrive at the detector first.
00:21:57.16 In our example, that would be the 89,
00:22:00.05 corresponding to CD45.
00:22:02.14 And the heaviest mass would be
00:22:05.11 iridium-191/193,
00:22:07.16 so that would be our nucleus.
00:22:09.18 The detector counts,
00:22:11.28 for a given amount of time,
00:22:13.07 each one of those masses.
00:22:14.19 The next one arrives, and it integrates those counts up,
00:22:16.28 and so on and so on.
00:22:18.17 So, when we have, for example,
00:22:20.08 two different cells...
00:22:22.24 let's say the first one has bound CD45,
00:22:24.25 iridium and rhodium,
00:22:26.14 so this is a dead lymphocyte...
00:22:31.01 a dead leukocyte, sorry.
00:22:33.18 This is a cell that we later on
00:22:35.18 will exclude from our analysis.
00:22:36.26 The next one has positive signals for
00:22:39.25 DNA, CD45, 3, and 4,
00:22:42.07 so this is a live T cell,
00:22:44.13 a live helper T cell, to be more precise.
00:22:46.28 And this something that we later on want to analyze.
00:22:49.28 So, at the end of this,
00:22:51.20 we have mass fingerprints,
00:22:53.06 and all the intensities of these peaks are integrated up
00:22:55.13 and put into the fcs data file.
00:22:57.18 So, when you come into the mass cytometry facility
00:22:59.28 and your data are running,
00:23:01.23 you will not see any dot plots as you might see
00:23:03.28 in a normal traditional flow core,
00:23:05.14 but you will see something like this.
00:23:07.01 This is referred to as a rain plot.
00:23:09.08 On the x-axis, which is the one on top,
00:23:13.11 you will see all the masses that have been acquired.
00:23:17.05 If we have had a CD marker
00:23:19.28 in one of our channels...
00:23:22.01 attached, it will be noted,
00:23:23.16 as is done by our facility.
00:23:25.01 And on the y-axis you have the number of pushes,
00:23:27.07 meaning how many ions have been
00:23:29.06 pushed into the TOF chamber
00:23:31.02 in a precisely given amount of time.
00:23:33.05 You will also notice that there are some areas
00:23:35.23 where you have densities showing up.
00:23:37.18 And each one of those areas
00:23:39.24 actually corresponds to a cell event.
00:23:42.10 Down here, we are seeing
00:23:44.24 the traces for iridium-191 and -193,
00:23:46.29 so that means there has been a nucleus,
00:23:48.14 and that means this has been a cell.
00:23:50.07 Everything that lines up with the iridium
00:23:52.22 is considered the markers of this particular cell event.
00:23:56.10 In here, you will notice that we have
00:23:59.19 CD45, CD3, and CD4,
00:24:02.26 so this has been a T-helper cell.
00:24:05.28 In the line below, you notice, for example,
00:24:08.18 there is no CD4 expression,
00:24:10.07 and there is no signal for CD4,
00:24:11.20 so this hasn't been a T-helper cell,
00:24:13.06 but this might have been something else.
00:24:15.13 In between,
00:24:17.18 you will also see that there is a little bit of signal,
00:24:19.23 for example, in this channel, 138 barium.
00:24:23.02 This is background, and barium is present in almost all samples.
00:24:26.08 It comes from many different sources.
00:24:28.01 It shows up in our mass range.
00:24:30.08 Barium can come from soap.
00:24:32.08 It can come from glassware.
00:24:34.00 It can come from any of the buffers
00:24:36.07 or your cell-freezing medium.
00:24:37.18 But this amount of background is completely acceptable,
00:24:39.16 and it will later on not have an impact on our data.
00:24:43.04 So, once we are done,
00:24:45.18 we will have as raw data a text file and an fcs file.
00:24:50.19 And what we are then needing to do is
00:24:53.28 we need to clean up our data.
00:24:55.20 And we do that always, for everything,
00:24:57.18 but in particular when you want to do high-dimensional data analysis
00:25:00.20 you need to remove certain things.
00:25:02.19 The first thing we are doing is
00:25:05.08 gating on DNA, meaning,
00:25:07.06 has there been a nucleus?
00:25:09.02 Versus the event length, which corresponds to,
00:25:10.29 how long has it taken for this particular event to pass through?
00:25:13.23 Or basically to finish...
00:25:16.00 be in an ion cloud?
00:25:17.18 And things that are stuck together
00:25:19.23 have longer event lengths,
00:25:21.18 and things that are...
00:25:24.12 and also usually have higher DNA content.
00:25:27.06 And things that are broken apart
00:25:29.11 but might have bound some DNA, still,
00:25:31.10 will have a lower iridium label,
00:25:32.29 and we will remove all of these.
00:25:34.19 These are normally giving you your intact cells.
00:25:37.02 Then, in our mixture,
00:25:38.19 remember we still have the EQ beads,
00:25:40.07 but we don't want to analyze them.
00:25:41.23 They are just bead particles.
00:25:43.10 So, we will remove those,
00:25:45.17 and the channel that is exclusive to the beads is channel 140,
00:25:48.14 so we will gate out all events
00:25:50.25 that are DNA-positive and not positive for beads.
00:25:53.26 So, we have cleared out the beads.
00:25:55.20 Then the next thing is all events
00:25:58.06 that are staining positive for our dead cell marker
00:26:00.10 -- in this particular example, that was cisplatin --
00:26:02.13 are also removed.
00:26:04.05 Positive... cells...
00:26:06.14 positive events are dead cells;
00:26:07.28 we don't want those in our analysis.
00:26:09.14 And then finally, we will see
00:26:11.20 which of those cells are CD45-positive.
00:26:14.16 So, we have cleaned up our data.
00:26:16.11 So, they are live, not beads,
00:26:19.06 hopefully intact CD45-positive cells.
00:26:21.23 And these pre-cleaned up cells
00:26:23.20 are now moving into high-dimensional data analysis,
00:26:27.01 as we can see.
00:26:28.24 So, now that we have our cleaned up data,
00:26:30.14 we are ready to analyze them.
00:26:32.18 But remember, we have
00:26:35.02 35 or 40 or even more dimensions, now,
00:26:37.22 to analyze.
00:26:39.13 And this poses... gives us a bit of a challenge.
00:26:42.05 So, we need different ways of analyzing
00:26:44.14 our data to illustrate that.
00:26:46.02 When you are analyzing
00:26:48.17 two parameters in a sample,
00:26:50.05 you are good with one plot that looks at both dimensions.
00:26:53.18 If we are looking at a staining
00:26:55.23 that has three parameters,
00:26:57.04 we need three plots in the end.
00:26:58.26 If we are moving already to a 9-parameter panel,
00:27:01.10 as many flow cytometry panels do,
00:27:03.24 we would theoretically need
00:27:05.22 36 different plots,
00:27:07.06 and many people actually don't look at all of them at once.
00:27:09.08 But we are now looking at 30+ markers,
00:27:12.02 and a 32-parameter panel
00:27:14.22 will result in 496 plots.
00:27:16.26 So, close to 500 different plots,
00:27:18.24 and nobody is able to analyze data like that.
00:27:21.16 Now that we have data with such high dimensionality,
00:27:25.17 we are going to use algorithms to analyze our data.
00:27:29.07 The very first one that has been used to analyze mass cytometry data
00:27:32.24 is called SPADE.
00:27:34.07 It was also part of the seminal publication
00:27:36.06 of the mass cytometry technology.
00:27:38.07 And SPADE does two things.
00:27:39.23 It clusters your data into groups,
00:27:42.12 as you can see here,
00:27:44.13 and produces those groups of clustered events
00:27:46.26 into a so-called SPADE or progression tree.
00:27:50.17 Before doing so, it does something called downsampling.
00:27:54.04 So, let us assume you have a quarter of a million events
00:27:56.16 of a complex mixture with 33 markers.
00:27:58.22 There will be some populations that will be very frequent,
00:28:01.23 and others, like for example stem cells in bone marrow,
00:28:04.05 that are very rare.
00:28:05.23 To not drown out the rare cell populations
00:28:08.11 in the big ones,
00:28:10.02 we are removing from those a couple of events
00:28:12.22 and putting them to the side, in a basket.
00:28:15.06 The downsampled part of the sample
00:28:16.29 is then used to look at,
00:28:19.00 how similar are the expression patterns
00:28:21.08 in the remaining cells
00:28:23.16 using all of our 30-some markers
00:28:26.29 that we have in our panel?
00:28:28.29 This will result in a distribution
00:28:33.22 that looks like that.
00:28:35.06 So, let us assume this bit here, this branch here,
00:28:37.23 are our T cells.
00:28:39.11 They are relatively far away,
00:28:41.20 let's say, from our B cells,
00:28:43.04 which are also not related to each other.
00:28:45.23 And inside the T cell branch,
00:28:48.02 we have a population over here.
00:28:50.00 Let's assume these are our
00:28:51.28 CD4 T cells, our T-helper cells.
00:28:54.05 And a subset thereof also expresses a marker called CD45RA,
00:28:58.18 making those into a group of naive T cells.
00:29:02.11 So, when you spread this further with 30+ dimensions,
00:29:06.25 you will get quite a number of different groups of cells
00:29:09.28 that are then depicted in a SPADE tree
00:29:12.01 in the form of a node.
00:29:13.28 And the closer these nodes are to each other,
00:29:17.01 the more related the cells in these individual nodes
00:29:19.23 are to each other.
00:29:21.04 The further way, the less related they are.
00:29:24.26 Finally, we take all the data that we had
00:29:26.17 initially put to the side back,
00:29:28.10 and put them back into our now-produced SPADE tree
00:29:30.14 so they can distribute according to their density.
00:29:33.07 You are ending up with a SPADE tree looking like that.
00:29:35.24 So, you have your populations distributed in different nodes.
00:29:39.12 The size of the node depicts
00:29:41.08 how many events are in this particular group.
00:29:43.10 And when you look at the next one,
00:29:45.11 the intensity of the SPADE node
00:29:47.17 depicts the expression level of a particular marker.
00:29:50.23 So, in this first paper that came out,
00:29:52.28 the group around Sean Bendall
00:29:57.11 has taken human bone marrow,
00:29:59.20 has stained it with a total of 33 different markers,
00:30:02.22 out of which 13 are surface markers,
00:30:04.29 and on top of that the bone marrow
00:30:07.07 had been stimulated in the presence or absence
00:30:10.10 of a number of inhibitors to elicit signaling,
00:30:13.18 in particular here down the phosphosignaling pathway.
00:30:16.19 So, we have 18 functional and 13 surface markers.
00:30:21.04 This number of markers was absolutely impossible
00:30:23.26 until that moment,
00:30:25.04 and it will allow us to do a few things
00:30:27.19 that were not present... possible before.
00:30:29.14 So, let's first of all look at the surface markers.
00:30:32.10 The surface markers have been used
00:30:34.11 to reproduce the known lineage relationships
00:30:36.12 in bone marrow.
00:30:37.29 And we know that there are stem cells.
00:30:39.25 They are sitting quite high up in the tree.
00:30:41.29 And we have our lymphoid branch:
00:30:44.29 T cells, B cells, and K cells.
00:30:47.16 We have our myeloid branch
00:30:50.01 with the monocytes, the dendritic cells.
00:30:51.24 And down here, as an almost separate group,
00:30:53.22 we have our red blood cells.
00:30:55.16 What we are seeing here is
00:30:58.04 basically the reconstructed bone marrow
00:31:01.00 developmental hierarchy,
00:31:02.23 highlighted by the marker CD45.
00:31:05.06 CD45 is expressed on
00:31:08.11 very much all of those cells,
00:31:09.19 but at different expression levels.
00:31:11.20 Except for our red cells;
00:31:13.16 they are negative for CD45.
00:31:15.08 So, blue colored means low expression.
00:31:16.19 Red, hot, means high expression.
00:31:19.01 If you are looking a different marker,
00:31:21.20 for example CD3, you will highlight...
00:31:24.17 see this highlighting of the T cell part
00:31:27.04 of this branch,
00:31:28.08 and it is pretty much absent almost everywhere else.
00:31:31.22 So, this is what a SPADE tree looks like.
00:31:33.19 Then... this has been 13 markers,
00:31:36.13 so theoretically you could have
00:31:38.29 done the same thing with normal fluorescence flow cytometry,
00:31:40.26 but we still have lots of channels left.
00:31:44.28 So, the next thing we have been doing is looking at
00:31:48.01 18 different signaling markers,
00:31:50.06 and I'm showing you one example here.
00:31:52.12 We are looking at the signal
00:31:56.05 for a phosphorylation event on phospho-ERK,
00:31:59.23 an intracellular signaling molecule,
00:32:01.25 which you can stimulate
00:32:04.25 with PMA/ionomycin.
00:32:06.10 PMA/ionomycin is a stimulus that activates
00:32:08.10 pretty much every single signaling pathway in your cell,
00:32:11.07 so all of the cells are highlighting
00:32:13.14 after stimulation with PMA/ionomycin.
00:32:15.28 When you, on the other hand,
00:32:17.11 stimulate via the B cell receptor,
00:32:19.00 BCR,
00:32:20.22 only the B cell branch will light up,
00:32:23.12 while all the other cells in your bone marrow
00:32:26.13 stay silent.
00:32:27.28 If you then add an inhibitor, dasatinib,
00:32:31.08 this will inhibit back the signal on your B cells
00:32:34.18 while it doesn't do anything for PMA/ionomycin,
00:32:37.04 because it's not a specific inhibitor
00:32:39.04 for this particular branch,
00:32:41.04 but it inhibits the B cell receptor.
00:32:42.26 So, why is this so great?
00:32:44.26 This is only one example,
00:32:46.26 only one signaling molecule,
00:32:48.08 but you can look at all the different cell types in bone marrow
00:32:51.02 at the same time,
00:32:52.17 so you can do systems analysis for the very first time.
00:32:55.11 And if you are imagining that you want to
00:32:57.07 stratify patients or you want to develop a drug,
00:32:59.19 you have for the very first time
00:33:02.00 the chance to do an unbiased analysis
00:33:04.11 and look at every player in the field all at once.
00:33:07.13 And maybe then, later on,
00:33:09.28 narrow down to the few markers
00:33:11.19 that you want to characterize further.
00:33:13.24 So, this was done with SPADE.
00:33:16.05 The second algorithm that is fairly widespread
00:33:18.07 is called viSNE.
00:33:19.28 And viSNE does not do clustering;
00:33:22.07 it does dimensionality reduction.
00:33:24.08 So, every cell basically stays in the mixture.
00:33:26.11 It's not formed... basically, put into small groups.
00:33:29.11 But the very first thing, here,
00:33:32.12 that happens is it looks at all the markers,
00:33:34.05 and it calculates how similar
00:33:36.23 is each cell to the other neighboring cells,
00:33:38.08 and then puts them down into a two-dimensional space.
00:33:40.26 So, what you are getting
00:33:43.06 is something that looks very much like
00:33:44.26 a very big two-dimensional dot plot.
00:33:48.00 To give you an idea of how a viSNE plot looks like,
00:33:50.16 let's go a bit further.
00:33:52.08 So, first of all, we start with something familiar.
00:33:54.12 This is a two-dimensional dot plot,
00:33:57.18 and we have gated on our CD8-positive cells,
00:34:00.17 over here.
00:34:02.05 If you depict this alongside all of the other events
00:34:04.29 and markers in a viSNE plot,
00:34:06.19 you get a map that looks like that.
00:34:08.10 You have tSNE1 and tSNE2
00:34:10.19 as axis markers.
00:34:12.00 And you will see groups or islands of cells
00:34:14.24 that are more or less close to each other.
00:34:17.16 You can highlight the expression of CD8,
00:34:19.25 again, as you could do it on the SPADE plot.
00:34:22.07 And you can see this group, here,
00:34:24.22 is positive for CD8,
00:34:26.07 and they form a separate island
00:34:28.24 in this particular sample.
00:34:31.19 Different other islands are not positive for CD8,
00:34:34.11 so they won't light up.
00:34:36.08 You see all of the cells in your mixture
00:34:38.25 depicted into different islands,
00:34:40.17 and how these islands look like
00:34:42.12 and how many there are depend on the sample,
00:34:44.19 and also how many markers you have used.
00:34:46.14 It has to be said for SPADE and viSNE
00:34:49.10 that every time you run these algorithms
00:34:52.29 they produce a different tree
00:34:55.25 or a different SPADE map.
00:34:57.04 So, this is not reproducible in its outlook.
00:35:00.05 It is reproducible in the
00:35:02.24 readout of the total frequencies,
00:35:04.07 but how these islands are distributed
00:35:05.28 or how your tree looks like is not reproducible.
00:35:08.14 In the case of viSNE,
00:35:10.00 as long as you run 10-20 samples
00:35:12.20 together at the same point in time,
00:35:14.17 all of those will fall into the same viSNE map,
00:35:18.07 so you can at least compare alongside the other.
00:35:21.26 But to be honest,
00:35:23.13 most people that are wanting to extract real data out of that
00:35:26.05 take the data coming out of the viSNE algorithm
00:35:28.04 and put them into a heatmap,
00:35:30.00 where they compare expression levels
00:35:32.02 of certain parameters versus
00:35:34.12 particular patients or particular treatments,
00:35:36.10 and this makes it much more manageable.
00:35:38.07 To give you an example of how this has been used,
00:35:40.07 this particular group has been
00:35:42.14 analyzing patients suffering from aplastic anemia,
00:35:45.12 and these people get treatment with immunosuppressive drugs.
00:35:51.20 So, they have found a marker in the T-regulatory cells...
00:35:55.04 and T-regulatory cells
00:35:56.24 are characterized by expressing CD25,
00:35:59.08 the transcription factor FOXP3,
00:36:02.05 and not expressing CD127.
00:36:05.12 So, this is just the proof
00:36:07.12 that we are truly looking at the T-regulatory cells
00:36:08.26 in this sample.
00:36:10.22 And then we have compared
00:36:13.18 this in our different patient groups.
00:36:14.22 So, our control group were healthy donor cells.
00:36:17.07 You can see on the last column, here...
00:36:21.20 this is focusing on our regulatory T cells,
00:36:24.07 and you might see that there are two different nodes,
00:36:26.20 or groups.
00:36:28.06 Healthy donor cells have a lot of
00:36:30.06 what was referred to as T-regulatory cells B.
00:36:32.11 They had all of the markers in here
00:36:35.02 and then a couple more.
00:36:36.20 And they have very little
00:36:38.29 of what is referred to as T-regulatory cells group A.
00:36:42.13 Respondents that are not responding...
00:36:45.11 patients that are not responding to treatment
00:36:47.23 have literally no T-reg Bs,
00:36:50.14 lot of T-reg As.
00:36:52.07 While those patients that originally also
00:36:55.10 looked like that
00:36:56.20 but are responding to the drug
00:36:58.22 begin to shift more and more towards the healthy phenotype,
00:37:01.10 and develop more and more of the T-regulatory cells B.
00:37:03.11 The panel to analyze this
00:37:06.11 contained a total of 37 markers.
00:37:09.02 And this phenotype, or also this biomarker,
00:37:12.12 has been stable.
00:37:14.11 So, this has been reproduced.
00:37:16.00 And the group could identify
00:37:17.25 a new biomarker, a predictive biomarker,
00:37:21.04 for treatment of their patients.
00:37:23.14 This is something that has been
00:37:25.25 also a hallmark of high-dimensional cytometry
00:37:28.02 or mass cytometry.
00:37:29.16 It has allowed us to identify
00:37:32.14 biomarkers much faster,
00:37:34.29 because we take an unbiased view on our entire data set,
00:37:37.13 and we can then hone down on those that are really specific.
00:37:40.26 To identify, finally,
00:37:43.01 the cells that are truly of interest,
00:37:44.23 not even 10 markers were needed out of the mixture of 40.
00:37:48.19 With this, I hope I could explain to you
00:37:51.05 how mass cytometry works,
00:37:52.14 how to set up an experiment,
00:37:54.07 and also what the power of this technology
00:37:57.07 can do for your research.
00:37:59.17 If you would like to learn a bit more about it,
00:38:01.07 there is a wealth of literature,
00:38:02.29 many review articles but also a good book,
00:38:05.06 coming out of the group of Garry Nolan at Stanford University,
00:38:08.29 where mass cytometry has actually started.
00:38:11.18 And below that, I have listed
00:38:14.01 one of the quite a few review articles
00:38:17.20 on high-dimensional data analysis,
00:38:20.02 and it gives you an idea of how to
00:38:23.29 process high-dimensional cytometry data,
00:38:26.05 which is quite important, as you might agree.
00:38:28.22 With that, I would like to thank you,
00:38:30.28 and goodbye.

This Talk

Speaker: Susanne Heck

Audience:

Educators of H. School / Intro Undergrad
Student
Educators of Adv. Undergrad / Grad
Researcher
Educators

Recorded: February 2019

Talk Overview

Dr. Susanne Heck begins her talk by explaining why we might choose to use mass cytometry rather than other types of flow cytometry. Traditional flow cytometry is typically limited to the detection of about a dozen parameters in one sample due to overlap between the emission spectra of fluorochromes used to label antibodies. Mass cytometry, on the other hand, allows for the detection of up to 50 parameters in one sample because antibodies are labelled with metal isotopes and separated based on their mass. Heck goes on to explain which metal isotopes are typically used for mass cytometry and why, and she describes how a mass cytometer functions. She finishes by running through an example of using mass cytometry to perform functional phenotyping on human bone marrow cells.

Speaker Bio

Susanne Heck

Dr. Susanne Heck received her PhD in molecular biology from the University of Bremen, Germany, in 1997. After a postdoc in molecular and cellular biology at Albert Einstein College in New York, Heck joined Cellular Genomics Inc., USA, to work on preclinical models for small molecule kinase inhibitors. In 2004, she moved to the Lindsey… Continue Reading

Comments

Ediri E Metitiri says

May 18, 2021 at 2:08 pm

Thank you for this thorough introduction into mass cytometry! I come away now able to recognize concepts and participate in discussions where mass cytometry is being discussed on my campus.

Reply

Talk Overview

Speaker Bio

Susanne Heck

Comments

Leave a Reply Cancel reply

Find a Video

Topics

For Educators

Educators Recommend

Techniques

Free Courses

Career Development

Our Favorites

Connect with Us

Talk Overview

Speaker Bio

Susanne Heck

More Talks in Techniques

Related Resources

Reader Interactions

Comments

Leave a Reply Cancel reply

Footer

Find a Video

Topics

For Educators

Educators Recommend

Techniques

Free Courses

Career Development

Our Favorites

Connect with Us