Imaging Flow Cytometry

Top

00:00:13.04 Hello, and welcome to iBiology.
00:00:14.24 My name is Andrew Filby.
00:00:16.19 I'm the Director of the Flow Cytometry Core Facility
00:00:18.21 at Newcastle University in the United Kingdom.
00:00:20.21 Today, I'd like to tell you about a very powerful
00:00:23.03 analytical single cell technology called imaging flow cytometry.
00:00:27.20 In this presentation, I'd like to cover the following subjects.
00:00:32.00 I'd like to begin by giving an intro to cytometry and cytometers.
00:00:35.10 I'd then like to move on to talk specifically about what imaging flow cytometry is;
00:00:39.11 why we might choose this as a technology to measure our cells;
00:00:42.23 what kind of data we get out of it,
00:00:45.16 and so, particularly, what we can derive from a digital image.
00:00:48.07 I'd then like to give a few examples of some data analysis.
00:00:51.12 And then finish with a little bit about the future of imaging flow cytometry
00:00:54.21 and a summary.
00:00:57.14 Okay, so let's begin with the word cytometry.
00:00:59.20 The word cytometry is made from two words.
00:01:05.02 The first word is the Greek word Kytos,
00:01:08.08 which means "hollow basket".
00:01:11.04 In this case, we're using it to refer to the word cell.
00:01:14.06 The other half of the word comes from another Greek word,
00:01:16.13 which is Metria, which means "the process of measurement".
00:01:19.12 So, we can broadly define cytometry as the process of cell measurement.
00:01:24.02 But to expand on this definition more,
00:01:26.07 I would say that cytometry is very much the measurement of phenotype of cells
00:01:30.11 always done at the single cell level.
00:01:32.15 But it's always conducted on a population scale.
00:01:35.08 This means that we're not just looking at a few cells;
00:01:37.03 we're always trying to measure as many as possible.
00:01:38.23 And the reason we do this, of course,
00:01:40.18 is because every biological system, no matter what, is heterogeneous.
00:01:44.09 And so, cytometry allows us to measure and decode this heterogeneity.
00:01:49.02 So, in this cartoon, very simply put,
00:01:51.04 we take our cell biology, we perform cytometry,
00:01:55.00 and we generate, hopefully, a set of meaningful numbers from it.
00:01:57.24 And with these numbers, we may ask these following questions.
00:02:02.14 So, for example, we may ask
00:02:05.08 what type of cells there are within our sample,
00:02:07.05 and how many there are of each phenotype.
00:02:09.15 We may want to know what they do functionally.
00:02:11.14 And then we may want to put this into some kind of context,
00:02:14.09 so, particularly, human development or human disease.
00:02:18.19 So, again, in this cartoon I just want to reiterate this point.
00:02:22.01 Cytometry is very much the measurement of everything single cell.
00:02:25.24 So, within this cartoon here,
00:02:28.11 we can see that using cytometry we can measure
00:02:31.10 the whole flow of information, right through the single cell,
00:02:34.01 from looking at receptor expression
00:02:36.23 to the binding of a ligand
00:02:38.19 to measuring signaling cascades, looking at phosphorylation.
00:02:41.02 We can look at DNA modification.
00:02:43.07 We can also look at RNA transcription.
00:02:46.02 And then, finally, we can also look again at things like post-translational modification.
00:02:50.11 And again, the important thing to say about cytometry is that we're taking
00:02:53.21 our unknown cell population,
00:02:56.01 we're potentially labeling it with something,
00:02:57.21 and we're able to then ask all these different questions,
00:03:00.08 and finally put this back into the context of the wider population.
00:03:05.16 So, cytometry itself is very much kind of formed on three different pillars,
00:03:09.12 three different principles.
00:03:11.10 So, the first pillar, or the first principle, of cytometry is that
00:03:15.00 whatever we do we need to take our measurements in a very controlled fashion.
00:03:18.15 The next thing as well, if possible,
00:03:21.19 is that cytometry should be at minimum semi-quantitative.
00:03:25.18 Then, if we follow this controlled and quantitative approach,
00:03:29.04 we can then say that our measurements are comparable.
00:03:31.24 And what I mean by that is that for every cell that we measure within a sample,
00:03:35.20 we can make a fair measurement comparison between them.
00:03:38.17 We can even make measurements across samples.
00:03:41.00 And hopefully, we can even make comparisons, fair comparisons,
00:03:44.05 between samples across experiments done on different days.
00:03:47.11 All of this, of course, is built on a foundation that cytometry
00:03:50.15 is often multi-parameter,
00:03:53.04 meaning that we're not just looking, again, at one cell and one measurement.
00:03:55.20 We're looking at lots of measurements on lots of cells.
00:03:57.20 And in order to give the scalability,
00:03:59.23 cytometry also tends to be very high-throughput in nature,
00:04:02.07 so that we can sample as many cells as possible as quickly as possible.
00:04:06.22 Okay. So, in order to do our cytometry,
00:04:09.23 we need to have a specialized system,
00:04:12.01 a measurement device we call a cytometer.
00:04:14.16 So, what I'm showing here is just a very quick output
00:04:17.10 from doing a google image search on the word cytometer.
00:04:20.13 And all I really want to make...
00:04:23.08 the point on this slide is that there are many different types of cytometers
00:04:26.05 and many, many different ways of doing cytometry,
00:04:28.09 but of course they still follow the three principles I just introduced.
00:04:31.19 So, again, just to reiterate this point, a cytometer is an instrument,
00:04:36.09 a specialized instrument, which generates and measure signals
00:04:39.14 from single cells, often in a high-throughput manner,
00:04:41.24 using semiquantitative measurement capabilities
00:04:44.11 and often in a multi-parameter fashion,
00:04:47.02 which, again, forms the foundation of these three pillars of all cytometry.
00:04:51.12 So, to be a little bit more specific for the purposes of this presentation,
00:04:55.00 I want to introduce five different cytometry systems to you.
00:04:59.07 So, I'm gonna start right at the far end here,
00:05:01.16 and I'm going to talk a little bit about conventional flow cytometry.
00:05:05.00 This is fluorescence-based.
00:05:07.04 Then we have mass-based cytometry, also flow-based.
00:05:11.06 Then we also have flow cytometry
00:05:14.12 done with the genomic measurement, a genomic endpoint,
00:05:16.22 often looking at DNA and RNA within the cells.
00:05:19.24 And then, very much the focus of this presentation,
00:05:22.17 we have imaging flow cytometry,
00:05:24.15 and then another technology called imaging mass cytometry.
00:05:27.08 Now, the way we can kind of rank these different technologies is,
00:05:29.12 firstly, we can rank them by the number of parameters
00:05:31.23 that they're able to measure on each individual cell.
00:05:34.01 So, typically speaking, conventional fluorescence-based flow cytometry
00:05:37.06 counts as kind of the low-parameter space.
00:05:39.18 As we move up in this direction, though, to mass-based cytometry,
00:05:42.09 we can make more measurements per cell.
00:05:44.04 And at this stage, I'd like to refer you to another talk in the iBiology cytometry series
00:05:47.24 by Dr. Susanne Heck on this subject.
00:05:51.00 If we move up to this particular system, here,
00:05:53.10 we can look at RNA within our cells.
00:05:55.07 And again, we move the parameter space up
00:05:57.21 and increase the number of measurements we can make per cell.
00:06:00.15 When we then work into the area where we're actually looking at images of cells, now,
00:06:04.05 we increase the parameter space infinite... infinitely almost,
00:06:07.16 and particularly when we're looking at tissue imaging as well.
00:06:11.12 But the trade-off here is as we increase the number of parameters is,
00:06:14.08 unfortunately, we do tend to then concomitantly reduce the throughput.
00:06:18.06 So, what this means is that we're not as able to make as many measurements
00:06:22.04 on as many cells as possible.
00:06:24.18 So, we can also subdivide these systems
00:06:27.19 into two different classes.
00:06:30.02 These three systems tend to be described as zero-resolution systems,
00:06:33.03 and what this means is that we are not
00:06:35.15 deriving any kind of imaging information from the cells we're measuring.
00:06:39.09 These two cytometers, though, are image-based
00:06:42.16 and will give us images of the cells that we're looking at.
00:06:45.02 And again, particularly for this talk, this will be the focus of my talk.
00:06:49.04 Okay.
00:06:51.11 So, let's go back to our definitions here.
00:06:53.02 Here we have the word cytometry again.
00:06:54.20 We've already defined this as the process of cell measurement.
00:06:57.03 What happens now if I put the word flow in front of it?
00:07:01.03 Well, to define the word flow,
00:07:02.24 we know that this defines as moving along in a stream.
00:07:05.15 So, this straightaway gives us a clue about how we're making our cell measurement.
00:07:10.03 So, what we're doing in flow cytometry
00:07:12.12 is we're making all our cell or particle measurements in a liquid suspension.
00:07:17.07 So, this is very easy for samples
00:07:19.02 that are derived from liquid biopsies,
00:07:20.18 things like blood or liquor or even suspension cell lines.
00:07:25.06 But as you can imagine, if you're working with tissue,
00:07:27.05 you have to then firstly be able to disaggregate the tissue
00:07:29.16 down into a single cell suspension.
00:07:32.13 Okay. So, what I want to do very briefly is give
00:07:35.07 a very quick overview of what we call conventional fluorescence-based flow cytometry.
00:07:38.06 And again, at this stage, I'd like to refer you to another iBiology series cytometry talk
00:07:42.12 by Malte Paulsen on this subject.
00:07:45.12 So, what we have here in a conventional flow cytometer... we...
00:07:48.06 we tend to have a laser.
00:07:50.02 We tend to have some kind of optical setup,
00:07:52.00 where we have some mirrors.
00:07:53.21 And then we have some kind of detector.
00:07:56.00 In this case, it's a photomultiplier tube,
00:07:58.06 a detector that is not spatially registered.
00:08:00.13 So, over here we have our cell.
00:08:03.13 We introduce some kind of targeted fluorescence to our cells.
00:08:07.12 So, in this case, I'm using an example of some antibodies,
00:08:09.09 each one binding a different surface receptor,
00:08:12.04 each one with a different fluorescent color on the back of it.
00:08:15.13 So now, as we flow these cells through the laser beam,
00:08:19.05 as they make contact with the laser,
00:08:21.17 these fluorochromes become excited,
00:08:23.19 they emit photons at their given wavelength,
00:08:27.23 this light then hits a series of mirrors,
00:08:30.24 and each mirror then is responsible for
00:08:34.24 subdividing the light up into different wavelengths of light,
00:08:37.19 and this hits our detector.
00:08:40.04 So, what do we get from the detector?
00:08:41.21 What we get from the detector is this.
00:08:44.24 It's a pulse profile.
00:08:46.19 So, what we're seeing here is that as the cell enters the laser beam,
00:08:50.04 we start to see a measurement of signal appearing.
00:08:54.15 As the cell then moves through the laser beam,
00:08:57.13 we find the pulse forming.
00:09:00.12 And then as the cell exits the laser beam,
00:09:02.18 our pulse profile completes.
00:09:05.09 Now, we're able to measure three parameters
00:09:08.04 from this pulse profile.
00:09:09.18 We can measure the area under the pulse curve,
00:09:12.10 we can measure the height of the pulse,
00:09:14.10 and we can also measure the width.
00:09:16.01 And I'll talk a little bit about those on the next slide.
00:09:19.13 But let's go back to our cells.
00:09:21.07 Let's think about this now.
00:09:22.24 So, cells have structure.
00:09:24.20 Cells have morphology.
00:09:28.03 When we add fluorescent signals to our cells,
00:09:30.14 those signals are located somewhere
00:09:34.01 within or outside of the cell.
00:09:36.10 The problem we have is that when we use conventional flow cytometry,
00:09:39.08 we lose all that spatial resolution.
00:09:42.06 So, literally, all this lovely cell biology and structure
00:09:45.11 is collapsed down into this pulse profile,
00:09:48.14 where, again, we can derive the area under the curve,
00:09:51.10 which gives us the measure of total integrated fluorescence
00:09:54.08 as that cell's gone through that laser beam;
00:09:56.12 we can also look at the height,
00:09:58.12 which is the maximum signal;
00:10:00.06 and we can also measure the width,
00:10:02.00 which gives us an idea of the time of flight of the cell or particle
00:10:04.19 through the laser beam.
00:10:06.07 But the important thing to say is that we've lost all that spatial resolution.
00:10:10.08 So, as I said, for every single cell, then,
00:10:13.16 that goes through a conventional fluorescence-based flow cytometer,
00:10:15.21 we have a cell ID or an event ID.
00:10:17.22 And then, for each measurement we make of the pulse curve
00:10:20.08 -- the area under the pulse curve, this is what the A refers to --
00:10:23.05 we have a numerical value, a single value.
00:10:25.11 So, again, all that morphology, cell shape,
00:10:27.13 structure, is collapsed down into one value.
00:10:30.12 And then what we do when we look at the data is we take these values
00:10:34.16 and we plot them against different parameters.
00:10:37.07 In this example, it's a scatter graph.
00:10:41.05 And this is some real data.
00:10:43.01 This is some real conventional flow cytometry data.
00:10:45.03 And we can see that we have different cell events
00:10:48.06 appearing in this biaxial plot.
00:10:51.08 And of course, we can get a lot of information.
00:10:53.24 We know a little bit about the cells,
00:10:55.21 and we can derive some measurements here.
00:10:57.23 So, in this example, just using two parameters
00:10:59.11 -- forward-scattered light and side-scattered light --
00:11:02.02 we can get a very good idea about where our lymphocytes would be,
00:11:05.19 where our monocytes would be,
00:11:08.06 and finally where our granulocytes would be
00:11:11.05 that have a very, very high scatter value.
00:11:12.19 But again, as I keep stressing to you,
00:11:14.16 we've lost all the morphology of our cells.
00:11:17.19 So, what are we missing?
00:11:19.11 In this example, here,
00:11:21.11 what I'm showing you are two samples,
00:11:23.02 two cell samples.
00:11:24.22 This is a cell line
00:11:27.07 which has been engineered to express a fluorescently tagged
00:11:29.13 nuclear transcription factor.
00:11:32.16 And what we're looking at here is
00:11:34.24 we're looking at the data as though we would look at it
00:11:36.24 on a conventional flow cytometer.
00:11:38.15 So, we have no spatial information.
00:11:40.01 We're just looking at the total integrated fluorescence signal.
00:11:42.20 And of course, what we can do here is we can say,
00:11:45.03 well, of my unstimulated population and my stimulated population,
00:11:51.05 what is the level of fluorescence of this transcription factor?
00:11:54.19 And we can report this using something called the median fluorescence intensity
00:11:57.10 of the GFP signal.
00:12:00.19 And hopefully you can see from these data that we don't really have a lot of difference.
00:12:04.20 So, we'd be disappointed. We'd be confused.
00:12:07.04 We wouldn't see any difference, really,
00:12:08.24 between an unstimulated cell population and a stimulated cell population.
00:12:13.04 But what if we have the imagery?
00:12:15.07 So, let me show you this.
00:12:17.03 This is actually data from an imaging flow cytometry experiment.
00:12:20.06 Now, when we're able to look at our signals here,
00:12:23.09 we can see two very different morphological states.
00:12:27.07 In the unstimulated sample,
00:12:29.13 we can see that our GFP, against our nucleus stain,
00:12:32.20 is not translocated.
00:12:35.15 The two signals are separate from one another,
00:12:37.10 telling us the cells are in an unstimulated state.
00:12:40.07 When we add our stimulus, though -- when we activate them --
00:12:43.02 we can see that while the overall fluorescence barely changes,
00:12:45.06 the spatial localization of the two signals
00:12:47.21 are very different.
00:12:49.09 Now we start to see our GFP-tagged NFAT transcription factor
00:12:52.23 moving into the nucleus of our cell.
00:12:56.00 So, the point here to make is that,
00:12:57.23 in many situations in cell biology,
00:12:59.17 the location and the spatial context
00:13:02.01 is essential to the underlying biological question.
00:13:05.03 Okay.
00:13:06.22 So, let's now begin to talk a little bit, specifically,
00:13:09.01 about the technique of imaging flow cytometry.
00:13:11.05 And at this stage, I just want to make a point here
00:13:14.18 that, at the time of recording this seminar,
00:13:17.11 the only commercially available imaging flow cytometer
00:13:19.19 is from a company that was originally called Amnis,
00:13:22.07 who developed a system called the ImageStream X.
00:13:25.09 This is now owned by the Luminex Corporation.
00:13:30.04 So, let's take a little step back, again.
00:13:32.14 Before we get into this,
00:13:35.07 I want to provide another definition to you about what imaging cytometry is.
00:13:37.17 So, again, it's very much related to our definition of what cytometry
00:13:40.21 or flow cytometry is.
00:13:42.09 So, again, it's the measurement of our phenotype at a single cell level.
00:13:45.09 Again, we've got scale and throughput,
00:13:47.09 so we're looking at a population level.
00:13:49.07 We're not looking at just one cell;
00:13:51.03 we're looking at as many as possible.
00:13:52.20 But the important thing to hear...
00:13:55.05 say now is that the single cell level data is going to be mathematically derived
00:13:57.20 from a digital image.
00:13:59.10 So, again, we're taking our cell biology
00:14:02.16 and we're converting into a set of numbers.
00:14:04.16 But now the numbers are advanced,
00:14:06.24 because we're not just looking at a total integrated signal;
00:14:09.19 we're looking at morphology and spatial context as well.
00:14:13.02 So, I want to just quickly go over
00:14:16.15 the optical setup of the Amnis Image…
00:14:18.19 Amnis ImageStream system to try and give some overview
00:14:21.14 about how the system works
00:14:23.12 and how it takes the images.
00:14:25.06 So, here is a schematic diagram.
00:14:27.22 For those of you who are familiar with conventional flow cytometers,
00:14:30.12 you will see a number of familiar features.
00:14:32.23 I will point those out,
00:14:34.15 and I will also point out some of the differences.
00:14:36.13 So, let's begin down here.
00:14:38.19 So, as I said, it's a flow cytometer.
00:14:40.21 Therefore, we're looking at our cells in a liquid suspension phase.
00:14:44.04 We will be adding antibodies with fluorescent tags
00:14:46.14 against different markers of interest.
00:14:48.13 And we will be flowing them through the system.
00:14:51.06 So, as you'd expect to find,
00:14:53.16 you have a flow cell,
00:14:55.06 and this is kind of where the magic happens.
00:14:56.15 This is where the measurement occurs.
00:14:58.00 In the flow cell, the cells or particles are ordered nicely
00:15:00.02 to then go through the excitation lasers.
00:15:02.03 So, again you can see...
00:15:04.03 very similar kinds of excitation lasers as you would see in conventional flow cytometers.
00:15:07.16 But then this is where it gets different.
00:15:09.12 What we have is we also have LEDs
00:15:11.24 that produce transmitted light.
00:15:13.18 So, we have a bright field image of our cells.
00:15:15.16 This is very, very beneficial because,
00:15:17.16 unlike things like forward-scatter,
00:15:19.12 it will give a good correlation and a good idea about the size of your cell.
00:15:23.21 And then, after the flow cell,
00:15:26.23 when the cells travel through and get excited
00:15:29.02 and also are exposed to transmitted light as well,
00:15:32.12 we have a numerical aperture, so, a magnification, basically.
00:15:36.04 And the system has the ability to change the magnification
00:15:39.02 from 20 to 40 to 60X.
00:15:42.07 So, now, all the lights and information
00:15:45.02 that we've produced as the cells go through this area, here,
00:15:47.16 is passed on to another area that would be quite familiar
00:15:50.12 to conventional flow cytometrists.
00:15:53.00 We have here a stack of filters.
00:15:56.04 Now, just like the mirrors I showed you in the cartoon on the slide previously,
00:15:58.09 these mirrors are here to chop the light up
00:16:01.09 that we've been... produced here, through our excitation,
00:16:04.17 into the different emission wavelengths that then correlate
00:16:07.04 to the different dyes and the different markers that we're looking at on our cells.
00:16:11.00 Now, these mirrors are angled in such a way
00:16:14.04 as to focus light of a particular wavelength onto the detector.
00:16:18.01 So, in this case, the detector is a CCD camera.
00:16:21.06 And this CCD camera obviously is what gives us
00:16:23.24 the spatial, registered output,
00:16:26.02 unlike the PMT in the conventional systems.
00:16:28.19 And this is what we get.
00:16:30.09 So, here we can see, for each channel,
00:16:32.03 we have a set of spectrally decomposed,
00:16:34.15 spatially registered images being projected.
00:16:36.22 Now, on the more advanced systems,
00:16:39.17 it operates with two cameras working simultaneously.
00:16:42.22 This then allows us to get up to 12 images per cell
00:16:46.04 in a high-throughput manner.
00:16:49.05 So, how does this work?
00:16:50.17 It's very challenging when you think about it.
00:16:52.08 We're able to image a very fast-moving object.
00:16:55.09 We want to have a blur-free object.
00:16:57.07 And we want to get as much sensitivity and measurement from that object
00:16:59.17 as we can.
00:17:01.17 So, the way that the system I've described works
00:17:04.06 is using a process we call time-delay integration,
00:17:06.21 or TDI.
00:17:08.14 And in this cartoon, I'm gonna endeavor to try and explain
00:17:11.03 how that works to you.
00:17:12.22 So, these are the components of the cartoon, here.
00:17:14.16 Over here, we have our excitation laser.
00:17:16.24 We have our flow cell.
00:17:22.17 Here we have our mirrors, our spectral decomposition element.
00:17:26.04 And then over on this part, here,
00:17:28.00 we just have an example of just 3 channels from the CCD camera.
00:17:32.08 Here we have our cell of interest.
00:17:34.14 Our cell of interest has been labeled up with different types of dyes:
00:17:36.16 one on the membrane, one in the nucleus,
00:17:40.04 and then something else, maybe looking at an organelle structure.
00:17:43.14 So, as we flow this labeled cell through the laser beam,
00:17:46.20 through the flow cell,
00:17:48.21 these dyes become excited.
00:17:50.07 That information is then passed to the spectral decomposition units.
00:17:52.17 And what happens is that
00:17:55.16 each one of these mirrors, then,
00:17:57.10 is responsible for transmitting or projecting
00:17:59.20 a portion of the light onto each one of these camera channels.
00:18:03.01 Now, what happens is that each one of these camera channels
00:18:06.04 is a CCD detector,
00:18:08.18 and the information from the cell
00:18:11.05 is passed down at the same rate
00:18:13.15 as the cell then moves down through the excitation.
00:18:16.01 And at this stage, the information is being continuously integrated
00:18:18.22 onto the camera,
00:18:20.18 so again this increases our sensitivity of measurement.
00:18:23.03 So, then, at the end, as the cells move down
00:18:25.09 and the information is integrated,
00:18:26.21 what we end up with is a spatially registered output image of the cell.
00:18:32.23 And this is an example of some data.
00:18:34.23 So, this is what we get out of it.
00:18:36.12 Very, very powerful.
00:18:38.12 Every single row, here, is an individual cell,
00:18:42.09 and every single column is an imaging feature
00:18:45.10 or an imaging channel
00:18:47.05 from the multispectral data that we've pulled
00:18:49.10 out. So, you can see that we have
00:18:52.02 both transmitted light images and then targeted fluorescence,
00:18:55.02 scatter image, even.
00:18:57.09 And we have all of this with literally hundreds of thousands,
00:18:59.22 if not millions, of cells.
00:19:02.02 So, we have the same scalability and throughput of conventional flow cytometry,
00:19:05.09 but with the added extra of knowing exactly
00:19:07.19 where those signals are in or on the cell.
00:19:12.00 So, how then do we analyze such incredibly rich data?
00:19:15.01 Well, it's very much all about the fact
00:19:17.19 that we're collecting digital imagery.
00:19:20.00 So, if we take a cell and we project it onto the CCD camera,
00:19:23.24 as I'm doing here,
00:19:25.24 the CCD camera is basically an array of pixels.
00:19:30.03 And each one of these pixels is a detector, a photon counter.
00:19:32.00 So, these pixels therefore have
00:19:34.11 X and Y coordinates.
00:19:37.14 But of course, that's not the whole story.
00:19:39.13 At every X and Y pixel coordinate
00:19:41.17 -- depending on the detector,
00:19:43.12 depending on the dynamic range of the detector --
00:19:45.10 there will be a numerical value associated with them.
00:19:48.03 So, in the case of the systems that we're describing here,
00:19:51.07 they have a 12-bit CCD detector.
00:19:53.03 Therefore, every pixel is digitized between...
00:19:55.16 on the range of 0 to 4096.
00:19:58.02 So, what this thing effectively does is that...
00:20:00.06 you don't just have your X and Y coordinate for a pixel,
00:20:03.22 but you have a pseudo intensity value at that position as well.
00:20:06.04 So, if we took this image, here, and flipped it through this axis,
00:20:08.23 this is what we would have.
00:20:11.14 It would be like a mountain range, a textured mountain range of pixel values.
00:20:14.03 And this is really the fundamental basis of what our data analysis involves.
00:20:21.09 So, what can we derive from a single digital image?
00:20:23.22 So, we're not even talking here about 12 digital images.
00:20:25.16 We're not talking about hundreds of thousands of images.
00:20:27.22 We're talking about just one.
00:20:30.00 So, from this spatially registered fluorescence image
00:20:32.20 projected onto a CCD array,
00:20:34.22 I can derive almost limitless numbers of features from it.
00:20:38.18 So, typically speaking, with conventional flow,
00:20:41.16 we're obviously restricted to intensity-based measurements,
00:20:43.23 as we measure the area under the pulse curve.
00:20:46.00 We can do similar things here.
00:20:47.15 We can take all the pixel values above a certain threshold,
00:20:50.14 sum them together,
00:20:52.14 and plot the event as a single number.
00:20:54.10 But then we have a limitless number
00:20:56.18 of other intensity features that can be derived as well.
00:20:59.19 But what really makes the difference here, of course,
00:21:02.04 is that we can look at the shape of our cells.
00:21:04.04 There are a whole suite of features that can be derived based on shape,
00:21:07.13 such as circularity, elongatedness, perimeter.
00:21:11.07 And because, as well, we have this lovely pixelized texture,
00:21:14.15 we can also use that as a measurement too.
00:21:16.10 We can look at how signals are modulated.
00:21:18.10 We can look at texture features and that kind of thing.
00:21:21.09 And then very importantly, as well,
00:21:23.12 we can use the pixel information to look at location of stuff inside the cell.
00:21:27.17 So, we can ask the question, for example, is my receptor globally distributed around the cell?
00:21:34.00 Is my transcription factor inside the nucleus?
00:21:36.08 These are things, again, that we cannot do if we do not have an image.
00:21:39.22 So, why would you want to use imaging flow cytometry
00:21:42.17 as opposed to a zero-resolution cytometry technology?
00:21:46.08 Well, I think really the two answers to that question
00:21:48.11 are very straightforward.
00:21:50.18 If you have a question that really does rely on knowing about cell morphology,
00:21:54.00 having images is gonna really help you answer that question.
00:21:57.09 If you're interested, of course,
00:22:00.00 as well in the spatial context of a signal within the cell,
00:22:03.06 again, imaging flow cytometry is your thing.
00:22:06.01 And I guess if I was to add to either of these two,
00:22:08.07 one other thing I might say is that
00:22:10.22 if you have a very rare cell event or a very short-lived transitional state
00:22:13.07 that would be very hard to find
00:22:15.21 by something like manual microscopy,
00:22:18.06 the ability to analyze hundreds of thousands if not millions of single cells,
00:22:20.21 quite rapidly,
00:22:22.12 will allow you to find enough of those events to be able to analyze them effectively.
00:22:26.12 Okay. So, now I want to move on to a worked example of some imaging flow cytometry data,
00:22:31.00 to try and give you an idea about how we interact with the data
00:22:33.23 and what we can derive from it.
00:22:36.08 So, in this example I thought I would show you is...
00:22:39.05 we're gonna look at measuring exosome uptake by murine T cells,
00:22:42.12 again, using imaging flow cytometry.
00:22:45.10 So, what we want to do is we really want to be able to
00:22:48.14 distinguish between these three different cellular states.
00:22:51.01 So, let's look at this first state, here.
00:22:53.24 Here we have a cell which is clearly not taking exosomes
00:22:57.17 onto its surface or inside of it,
00:22:59.11 so it's negative.
00:23:01.15 Then we have two examples of cells
00:23:04.14 that have become associated with exosomes.
00:23:08.24 Now, we don't need imagery to be able to distinguish
00:23:11.05 between this cell state and these two.
00:23:15.12 But if we also want to distinguish between cells
00:23:18.16 that have the exosome on the surface
00:23:21.02 and cells that have the exosome internalized,
00:23:23.21 we need to have an image-based assay.
00:23:27.09 So, let's begin here.
00:23:29.09 What we've done is we've acquired lots of data on a...
00:23:33.16 on a... on an imaging flow cytometer.
00:23:35.13 And one of the first things we would do when we come to analyze the data
00:23:38.02 is we'd want to basically eliminate events or cells from the file,
00:23:40.15 from the data,
00:23:42.08 that we're not interested in.
00:23:43.16 This is exactly what we would do with other forms of cytometry as well.
00:23:45.15 And we tend to call this gating.
00:23:48.00 So, gating is where we take some kind of data plot
00:23:50.08 and we make an educated and controlled decision
00:23:54.24 about the events or cells that we want to take forward for analysis
00:23:58.24 and eliminate the events that we don't.
00:24:00.22 So, one of the first things we tend to do is we tend to make sure
00:24:04.15 that we're eliminating any kind of debris from our data,
00:24:07.10 and also two cells, or three cells, or aggregates,
00:24:11.08 stuck together.
00:24:13.08 So, the way we do this is very powerful.
00:24:14.22 We use the bright field -- the transmitted image -- of the cell,
00:24:18.13 and we simply use two very basic imaging-based metrics.
00:24:20.20 One is the area,
00:24:24.01 and the other one is basically a measure of circularity.
00:24:25.22 Now, the beauty of this system is that every one of these dots
00:24:28.16 in this measurement space is a cell,
00:24:31.01 and we can check this by clicking on it and seeing the image of the cell.
00:24:34.12 So, what we've done here is we've identified
00:24:37.03 our objects or events with a particular area
00:24:40.09 and a very high circularity as our single cells.
00:24:43.07 So, we gate on those, and then we eliminate our doublets and our aggregates.
00:24:47.24 So, now we take that population forward,
00:24:49.22 and we do a further refining of it.
00:24:51.10 We want to eliminate events
00:24:54.02 that have been collected that are not in good focus.
00:24:55.19 So, in order to do this, again
00:24:57.23
00:25:01.00 we go back to our brightfield imagery.
00:25:02.24 And what we do here is we use one of these texture features.
00:25:05.00 And this texture feature is basically looking at the variation of the pixel values across an area of the object.
00:25:08.03 And anything with a value higher than about maybe 50 units,
00:25:11.13 we tend to score as being in good focus.
00:25:13.17 And then values or cells
00:25:16.06 that are below this threshold tend to be in very poor focus.
00:25:21.03 Okay.
00:25:23.05 So, the third step of this analysis example I want to give you
00:25:26.00 is where we take our single in focus cells
00:25:29.07 and we identify the cells within that population
00:25:32.12 that are either positive or negative for the signal we're interested in
00:25:35.21 looking at the spatial location of.
00:25:37.19 So, in this case, of course, it's the exosome signal.
00:25:40.04 So, our exosomes have been labeled with a particular dye, a fluorescent dye.
00:25:44.18 And what we do here is we effectively treat our imaging data
00:25:47.15 just like we would conventional flow cytometry data.
00:25:49.20 We plot it in such a way that is akin to the pulse area,
00:25:53.18 looking at the integrated fluorescence intensity.
00:25:56.10 And we place, again, a region of interest,
00:25:59.17 a gate,
00:26:01.07 on this distribution to identify cells that have become associated
00:26:04.23 with an exosome signal
00:26:08.00 and distinguish the ones that have not.
00:26:11.17 But then, now, the next thing we want to do, of course,
00:26:14.20 is we want to look at the spatial localization.
00:26:16.08 This is the stage, now,
00:26:18.09 where we really utilize our imagery to the fullest,
00:26:20.22 and we want to identify the exosome-positive events that either have the exosome on the majority on the outside of the cell
00:26:27.05 versus the ones that have it on the inside.
00:26:29.09 And the way we do this is a little bit more complicated.
00:26:32.02 It uses principles of classical image analysis,
00:26:37.02 which are often called either masking or segmentation.
00:26:40.00 So, what I'm showing here in this...
00:26:42.00 in this diagram is I'm showing one individual cell
00:26:44.21 from our population that we've collected.
00:26:46.14 I'm showing a couple of different multispectral images that we've got.
00:26:49.19 Here's the brightfield image,
00:26:53.04 and here is an image of a membrane marker.
00:26:54.18 And what I'm doing here is I'm then showing...
00:26:56.12 overlaying over the image of the membrane marker
00:26:58.22 something called the mask.
00:27:00.16 Now, this mask is what the system -- the software -- is using
00:27:04.13 to define the pixels of interest.
00:27:06.23 So, hopefully you can see in this green overlay
00:27:09.23 that the default mask is very generous
00:27:12.06 and is encapsulating all of the cell.
00:27:15.17 Now, what we want to do is we want to measure how much of our exosome fluorescence
00:27:19.00 resides inside the cell.
00:27:21.12 So, what we've done is we've modified the default mask
00:27:24.12 to then only consider the area of the cell that is inside.
00:27:29.07 And very simply, what we do then is then we say,
00:27:31.12 okay, please tell me what is the intensity of the signal here
00:27:36.06 versus the entire cell?
00:27:38.16 And of course, once we do this,
00:27:41.01 we have the ability to plot this measurement
00:27:44.21 for our whole cell population.
00:27:48.00 And now we have a feature and a metric
00:27:49.19 where we can distinguish between these two different states.
00:27:51.18 So, in this area, here, of our distribution of our cells,
00:27:54.19 we have the majority of events which do not have
00:27:58.12 the exosome internalized within the cell body.
00:28:01.19 Whereas once we've passed this threshold that we've set here...
00:28:04.20 and again, just to mention, we've set this threshold
00:28:08.00 using a control where we kept our samples at 4 degrees
00:28:10.20 so we would minimize or eliminate any exosome internalization.
00:28:14.08 Well, what we can see now in this area of our plot is that we have the majority of cells
00:28:17.23 which have internalized the exosome signal.
00:28:21.12 And the beauty of this approach is that we can obviously, then,
00:28:24.20 derive population-wide metrics that we can plot and analyze.
00:28:29.11 And what I want to give you is the final kind of...
00:28:31.15 the prestige of this data analysis
00:28:34.08 is what we actually derive from it.
00:28:36.24 So, what we have here is we have two different conditions:
00:28:38.18 we have a wild type and a knockout condition.
00:28:41.15 This is where the cells were lacking
00:28:44.16 a particular key protein of interest.
00:28:46.21 And what we find is quite interesting.
00:28:49.00 If we treat our data in a non-image-based way,
00:28:51.24 and simply ask,
00:28:54.10 "what percentage of our cells in these two different genetic conditions
00:28:57.05 have become associated with an exosome?",
00:28:59.15 we get one particular conclusion from the data.
00:29:02.11 We would look and say that the knockout cells have more association.
00:29:06.09 But if we use our imaging data
00:29:09.05 and we look at the spatial localization of the signal,
00:29:11.10 we get a very different picture.
00:29:13.15 We actually find, now,
00:29:16.00 that while one cell type has possibly bound more of them,
00:29:20.04 it does not translate into an increased level of internalization.
00:29:23.10 So, again, what I want to kind of give you in this example
00:29:26.02 is the idea that if we don't have the spatial information,
00:29:30.11 we could possibly come up with a conclusion that is wrong,
00:29:33.14 or only half the picture of the biology we're interested in.
00:29:37.14 Okay. So, I think this is a very important point.
00:29:40.22 One of the last points I want to make in this section
00:29:43.00 is that the beauty of image cytometry
00:29:45.14 is that we are attempting to capture our data
00:29:50.05 in a way which again, is comparable, quantitative, and controlled.
00:29:55.10 Now, this is very important when we draw conclusions
00:29:58.16 from what our eye can interpret.
00:30:01.10 And in this example, I want to give you... again, I want to give you...
00:30:04.06 here is an image of a cell which is expressing,
00:30:08.16 again, a fluorescently tagged transcription factor.
00:30:12.08 And I think, you know, it's fair to say here that
00:30:15.03 this is not a cell where the transcription factor
00:30:17.12 seems to be within the nuclear area.
00:30:19.11 But what we've done here, of course,
00:30:21.03 is that we have made sure that we display the data
00:30:23.20 that we've derived from our camera
00:30:25.23 in an optimal fashion.
00:30:27.22 So, in this diagram here,
00:30:29.13 what we're doing is we're basically showing how we scale the camera data on this axis
00:30:33.04 on a 12-bit scale...
00:30:34.18 how we scale that data to a visible display in 8 bits.
00:30:38.08 And this is an optimal display setting.
00:30:41.07 Because what we're doing is... in this green line,
00:30:43.11 we're making the very best use of our available 8-bit display
00:30:46.20 for the data that is actually contained
00:30:49.05 in this image from the camera.
00:30:53.02 If we draw a bisecting line through this image,
00:30:56.05 you can see, of course, that the pixel values have this pattern.
00:30:59.11 They start by being quite high
00:31:02.02 as they go through this portion,
00:31:03.13 then drop as they go through here,
00:31:04.20 and then they come back up again as they go through
00:31:07.14 the other part of the cytoplasm.
00:31:09.09 Now, what happens now if I take exactly the same cell,
00:31:13.15 and I change the way that I display it visually,
00:31:18.03 on an 8-bit scale.
00:31:19.15 So, what I've done here is, again,
00:31:21.05 I've taken the data from this image
00:31:23.11 and I have altered the scaling
00:31:26.06 so that now I have significantly gained up the display
00:31:29.08 of the dimmer pixels
00:31:32.15 so that, visually, this cell has changed.
00:31:36.21 But it's the same cell.
00:31:38.08 And I think... what I want to stress here is that
00:31:40.12 the way we display our data
00:31:42.14 can influence the way that we would interpret it
00:31:45.20 without a numerical value behind it.
00:31:47.02 Again, if we take a measurement tool
00:31:49.11 and bisect through this image
00:31:51.17 -- it's the same image --
00:31:53.01 you can see hopefully that the pattern of pixels
00:31:55.07 is identical here and here.
00:31:56.18 So, while we've changed the display of the data,
00:31:58.10 we have not changed the data
00:32:01.09 that came from the 12-bit CCD camera.
00:32:05.19 And importantly, when we make our measurement
00:32:09.05 on this cell of our feature,
00:32:11.12 which again is derived from the camera values,
00:32:13.10 it's identical.
00:32:15.14 So, no matter how we present this image,
00:32:17.08 the data underlying it
00:32:19.07 -- because it's an image cytometry experiment --
00:32:21.01 doesn't change.
00:32:22.14 So, to kind of conclude on this part,
00:32:24.21 it's very important to say image cytometry
00:32:27.19 should eliminate the possibility to manipulate
00:32:30.08 image interpretation through display.
00:32:32.12 It's very important, though,
00:32:35.01 that we pay more attention to the numbers
00:32:37.01 and in some cases less to the images.
00:32:38.22 And with great images comes great responsibility.
00:32:41.09 So, again, it's very important also to focus on those numbers too.
00:32:45.15 Okay.
00:32:47.09 So, what does the future hold for imaging flow cytometry?
00:32:48.23 It's a very exciting area at the moment.
00:32:50.18 Again, as I record this seminar,
00:32:52.12 there have been two very big advancements in the field.
00:32:55.01 The first one comes from a group led by... or...
00:33:02.09 the first author is Sadao Ota.
00:33:04.16 This work was published in Science in 2018.
00:33:07.01 It's called ghost cytometry.
00:33:09.01 And again, at this point, I'd like to make another reference
00:33:10.22 to an iBiology seminar by Steffen Schmitt on cell sorting.
00:33:14.09 And again, just briefly, cell sorting
00:33:16.13 is an area of cytometry
00:33:18.22 where we don't just measure our cell heterogeneity;
00:33:21.15 we actually can physically separate our cells
00:33:24.05 based on that heterogeneity profile.
00:33:26.04 Until recently, this could only be done
00:33:28.04 using zero-resolution, conventional fluorescence flow cytometry.
00:33:32.03 But this group here, in Japan,
00:33:34.11 have developed an approach which allows us
00:33:37.07 to sort our cells based on an image.
00:33:39.17 So, what happens here is that you have your cells,
00:33:42.02 again, being introduced into a... into a sheath stream.
00:33:44.14 You have a measurement, here,
00:33:46.20 using a PMT.
00:33:48.20 But the way they've done it is that they've set the PMT's up now
00:33:50.12 where we do get some kind of morphological and spatial information.
00:33:54.02 And the important thing here is now you can select cells
00:33:56.24 not just on what they express and what level they express,
00:33:59.03 but actually how they look
00:34:01.01 and where that signal is within the cell.
00:34:03.11 And then another group as well...
00:34:05.17 last year publishing in Cell... Nao Nitta’s group...
00:34:08.06 they also developed another approach
00:34:11.05 for doing pretty much exactly the same thing.
00:34:13.03 Where you can take cells, you can image them,
00:34:15.22 you can then put it through an intelligent deep learning network,
00:34:18.09 and you can formulate a very rapid sort decision,
00:34:20.20 which then allows us to physically isolate
00:34:23.18 and separate cells of a particular morphology
00:34:25.24 or spatial localization.
00:34:29.23 So, to give a final summary,
00:34:32.22 there are definitely certain experimental questions that absolutely require imagery.
00:34:36.08 Imaging flow cytometry, though,
00:34:38.04 is essentially, in essence,
00:34:40.06 very similar to our zero-resolution flow cytometry
00:34:42.10 in how the cells are processed,
00:34:44.13 how they're stained,
00:34:46.02 and how the measurement occurs using lasers and fluorescence.
00:34:48.12 But, of course, our output is a spatially registered image.
00:34:52.20 And one of the challenges, really,
00:34:54.16 with this kind of output
00:34:56.21 is image analysis makes it very difficult.
00:34:59.19 The measurements are not just based on our total intensity,
00:35:03.07 and as such they can be almost infinite.
00:35:05.21 So, one of the challenges we find, of course,
00:35:08.07 is then, how do you select the best feature or the best parameter to be able
00:35:11.09 to resolve two cell populations from one another?
00:35:14.04 But imaging cytometry... imaging flow cytometry
00:35:16.04 is a very, very powerful discipline that will only become more important
00:35:18.15 to cytometry-based research
00:35:20.16 and all other fields of cell biology, particularly now as we move
00:35:24.10 into the dawn of imaging sorting.
00:35:26.22 Okay. So, I'd just like to finish by acknowledging
00:35:29.05 my team, and also my collaborator
00:35:32.11 who worked with me on the exosome work.
00:35:34.02 Thank you very much for listening.

This Talk

Speaker: Andrew Filby

Audience:

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

Recorded: February 2019

Talk Overview

In this talk, Dr. Andrew Filby provides an overview of imaging flow cytometry, a powerful technique used to measure the phenotype of cells using image-based metrics. Compared with traditional flow cytometry, imaging flow cytometry increases the number of parameters one can measure by providing morphological and spatial information in a high-throughput, controlled manner. Filby explains how the imaging flow cytometer works and describes the benefits of using this technique to better answer biological questions, as well as giving us a glimpse into the future of this exciting field.

Speaker Bio

Andrew Filby

Dr. Andrew Filby is the Director of the flow cytometry core facility at Newcastle University in the UK. He received his bachelor’s degree in biochemistry from the University of Huddersfield, and his Ph.D. in molecular and cellular immunology from the National Institute for Medical Research (NIMR) in Mill Hill, London. Filby joined the laboratory of… Continue Reading