0:00
My name is Przemek Fleszar.
0:01
I'm Leica’s workload specialist.
0:04
I'm dealing with the spatial biology technologies that we developed at the Leica.
0:10
Today I would like to talk about the imaging, image analysis, everything in the scope of the spatial biology projects.
0:20
Let's consider the samples that we are currently using in the Cancer Research and immunology.
0:25
Depending on a kind of questions you are trying to answer, probably you'll be working with a different kind of a samples.
0:31
So obviously both 2D and 3D live and fixed samples and models are allowed.
0:38
Depending on the model or sample you pick, obviously you will have to use different kind of instrumentation.
0:44
So for instance, if we consider 3D live organoids or spheroids, systems like light sheet microscopy would be preferred.
0:53
So for example, what you see on the left hand side.
0:56
On the other hand, trying to characterise the cell hundreds of microns deep into the tissue, probably you're going to need to use point scanning confocal that I would like to introduce the Stellaris Spectra Plex, the confocal microscope that we introduced to the spatial biology field enables imaging of 15 biomarkers at the same time maintaining the ability to look deep into the sample.
1:22
Probably the 2D samples, the thin tissue sections are the ones that are giving you the broadest choice of ways to check on them.
1:33
And by broad I mean both the area of the sample but also the number of the antibodies you can introduce to characterise individual cells and also see the complex patterns within the tissue.
1:48
This can be covered by a cell DIVE system.
1:50
This is like US hyperplexing instrument that allows you to simultaneously or sequentially, sorry, acquire 60 unique biomarkers.
2:00
And such experiments can be followed up by laser microdissection that allows you to isolate specific targets and run downstream analysis on them.
2:12
So with that, let's start with the multiplexing.
2:14
This is the first topic of my talk today.
2:17
And as I mentioned, we are using the cell DIVE for that.
2:21
The key to understanding the precision oncology and that includes stratification of the patients, checking the outcomes of the treatment is comprehensive understanding of the tumour microenvironment, we can all agree on that.
2:37
So here you see the image acquired with the cell DIVE with just 13 biomarkers and already this image gives you the idea how complex this micro environment can be, right?
2:47
So we are looking at the, I don't know, carcinoma sample with the tumour sitting on the bottom right part of the picture and then the regular tissue in the top left corner of it.
2:58
So even to the untrained eye like myself, I can see the huge differences in the structures looking at both parts of the image.
3:07
So let's try to get a little bit deeper in the regular tissue, the yellow cells are stroma, right?
3:12
So they are kind of promoting the growth of the chronic pits or crypts visible here in purple.
3:18
And exactly same cells are visible in the tumour area, but they look completely different, right?
3:24
They don't form this nice boundary anymore, instead of this have been hijacked by the tumour and promoting the growth of the cancerous tissue.
3:32
On top of that, we also see here the orange and blue immune cells, and they are visible pretty much everywhere.
3:38
You see them in the regular tissue, you see them in cancerous one and that already gives us like pretty critical spatial information, right, as we're trying to qualify the patients for the immunotherapy.
3:50
We would like to see this immune cells also within the tumour area because then we know they will have a chance to act on the unhealthy cells.
4:01
So here is another example.
4:03
This one is taken from the early cell deck paper that was published back in 2013.
4:07
In this particular case, we are looking at the colon TMA.
4:12
We have 60 individual biomarkers introduced to the sample, right?
4:16
As you can see, even though we're working with relatively small images, you should be able to see individual cells and resolve them on the sub cellular level to draw some conclusions.
4:29
So this is how the workflow looks like.
4:32
We are starting with FFPE tissue that is placed on the slide and then we are starting with the autofluorescence imaging of the sample.
4:40
So this is the baseline for the rest of our experiment, right?
4:44
We need to subtract this at some point from the actual image of the antibodies we're trying to image.
4:49
The next step is the biomarker staining.
4:51
So here you can introduce up to 4 di-conjugated antibodies.
4:56
It will be imaged subsequently by the imager that we'll be talking about in a second.
5:01
The next step is the dye in activation.
5:03
So this is critical part.
5:04
This is how we are healing or quenching all the fluorescence that we see in the tissue, and we are able to introduce the new combination of [dyson] antibodies for another round of imaging.
5:16
Before we do that, we still acquire the auto fluorescent imaging again to kind of subtract it from the resulting images.
5:23
So this process can be repeated as many times as necessary to cover all the biomarkers that we are trying to visualise in that experiment.
5:32
So what we did validation for 60 of them.
5:34
Obviously there is no technical limitation in terms of larger numbers.
5:40
We just never tried, right.
5:42
But it should be possible to work with the larger numbers of those.
5:46
After all the biomarkers are imaged, we are moving to the last step of the workflow which is image analysis, right.
5:52
So here we, excuse me, segment the image, we classify the data trying to identify specific phenotypes and then draw the conclusions from whatever results we achieve.
6:07
So the image here is taking all the images automatically, right?
6:10
But all the wet work that we have to do in between rounds can be either done manually or can be fully automated using this by assembly bot from Advanced Solutions.
6:23
So here it means that biomarker staining, inactivation, all the washing steps will be done by the robot and virtually the entire workflow can run without human intervention.
6:34
So you're essentially putting your slides in and after hours, days, weeks, depending on how many of the samples and antibodies you want to image, you can get your results that are ready to analyse them.
6:50
Here we have a nice video of the robot.
6:52
So obviously the idea is to increase the throughput of the whole experiment since the imager can only hold 1 slide at a time.
7:00
But with the robot you can actually put up to 15 slides though we processed in parallel.
7:07
So it means there is no dead time essentially.
7:09
If we consider the experiment in which we have 20 Plex and 15 slides, it's going to take us about 5 days with relatively large piece of tissue.
7:19
So this is thanks to the robot.
7:21
But even the imagery itself is contributing to this increased throughput, right?
7:26
So the fluorescence we can image here, we can go with five channels at the same time.
7:32
Typically 1 channel is reserved for the nuclear staining as we are using this nuclei in every single step to be able to register the images between different rounds.
7:40
But then all four channels are actually used by the biomarkers, right?
7:45
So then it means for this 20 Plex that you see down below, we're going to need 5 different rounds of staining, inactivating and imaging.
7:57
OK, So this is the last step of the workflow, which is in our case done with the Aivia software.
8:03
So Aivia is an in house software that Leica has been developing for quite a while.
8:08
It's utilising nice tools for AI segmentation and classification.
8:13
Anybody whoever did the image analysis will be familiar with the workflow.
8:17
So you're typically starting with the input.
8:19
In this case we are looking at 15 Plex 3D image, but obviously this is capable of running any other images in 60 Plex three samples.
8:29
We are segmenting the image to detect all the targets that we want to detect.
8:35
We classify them and then visualising all the results using nice tools that we have incorporated inside the software with some extra optional quality control tests for confidence if necessary.
8:50
So here we have the example of exactly the same sample.
8:52
This is still 15 Plex and we use the deep learning cell segmentation model to identify the cells.
9:00
So on top of the models that are already incorporated inside the software, you as a user always have the ability to train all models and apply them to the samples you're working with, right?
9:12
So we are not restricting it to anything that we provide you with, OK.
9:17
So far, we've been talking about the individual cells or maybe larger areas in context of the surrounding tissue, right.
9:26
But what about the case that we would like to isolate individual targets and run downstream analysis on them like mass spec for example?
9:35
Well, for that you can use laser micro dissection that is integral part of our portfolio.
9:41
So here we have like a nice video highlighting how it is done.
9:45
Essentially we are marking individual cells, we are using the laser to isolate it and then it's free falling into the collector known below that can be actually checked whether the collection was done properly.
9:55
So all this samples that we collect now be analysed by means of some molecular methods.
10:03
This is available in both brightened and fluorescence.
10:05
So depending what kind of a sample you're working with, you always cover.
10:11
So this is what it looks like from the technical standpoint.
10:16
We are operating the laser through the objective, cutting the membrane that it's that the sample is attached to and then the sample is free falling into the collector, mentioning collector.
10:26
I will explain that in a second.
10:29
So you can do this as many times as you want depending on the experiment you are trying to sustain here.
10:36
OK, so this is the microscope itself.
10:38
So it looks like pretty regular motorised upright microscope.
10:41
It has some extras, right?
10:42
So you probably notice the laser box that it's sitting at top of the instrument.
10:47
So here is where all the laser optics and electronics is located.
10:52
We also using the gavel to manipulate the laser directly on the sample plane.
10:55
Therefore we don't need to move the stage during the cutting.
10:59
What is really interesting here is the stage.
11:03
So the stage is actually built from 2 independent parts.
11:08
The other one is holding the samples that we'll be dissecting.
11:11
So you can navigate.
11:12
You can select the areas while the one down below is holding the collector.
11:17
So I mentioned the collector before.
11:20
Typically this will be a PCR tube or recently multi well plate, right?
11:24
So you can use 96 well plates.
11:26
You can use 384 well plates and collect all the dissected material into the individual wells.
11:31
If you want to run some higher throughput experiments later on using the mass spec, for example, this is probably a desired solution for the collection.
11:42
All right, so the microdissections on the market for quite a while.
11:52
So this year marks 25th anniversary of Leica launching the first microdissector.
11:58
But couple of years ago back in 2022, thanks to the group of people on the right-hand side, it became massively popular again, right?
12:07
And this is thanks to this deep visual proteomics workflow that they described.
12:11
So you may ask what the deep visual proteomics is essentially what they did, they scanned the entire surface of the sample.
12:19
They use the software to identify the targets they are interested with.
12:25
This information is then transferred to the dissector software and automatically we are isolating individual cells into desired wells on the 96 or 384 well plates.
12:40
So I broke down this whole process to make it more digestible.
12:44
So here we have the first step of imaging.
12:47
In this particular case, we'll be using the Leica microscope with the imaging software.
12:52
We are taking the overview of the sample with the lower magnification objective and then we can select either region of interest of number of these regions that will be scanned with the desired magnification that will be later on using for cutting the individual cells as well.
13:08
So after we collect this data, the images are automatically stitched and saved for the next step of the workflow, which is segmentation again.
13:23
And for that we'll be using the Aivia software, so exactly the same one as we did with the multiplex data.
13:29
So we are opening the image that we just acquired.
13:31
This can be done in bright field, can be done in fluorescence of course.
13:36
So here we'll be cropping the part of the image to speed up the whole process of training the segmentation.
13:44
And in this particular case, the operator is using the machine learning based random forest Pixel Classifier segmentation.
13:51
But obviously you can use the other tools like a deep learning here as well.
13:55
For that particular segmentation, we have to highlight the background and the areas of interest, right?
14:01
So it's an iterative process.
14:03
You can fix it if you don't like the resulting masks.
14:07
They are popping up automatically as soon as you're done with the marking and after fine tuning, you can actually save this model to use it on the entire section that we previously acquired, which is happening right now.
14:30
So here we have all these masks.
14:31
Now we need to transfer them to the micro dissector.
14:34
For that, we're going to need to mark reference points.
14:37
So in this particular example, it's happening the more manual fashion.
14:42
Essentially we are just highlighting the points that we will be able to recognise later on when working with the actual sample.
14:49
But recently this process was automated.
14:51
So currently the slides developed for the microdissection and specifically for the DVP workflow got some marks that can be recognised automatically by the software.
15:01
So this step can be eliminated.
15:04
All right.
15:04
So we have three reference points that were saved.
15:08
Then like with DNA image analysis software, you can still filter out any unwanted data that you acquired, right.
15:16
So for example, smaller masks and adjacent cells, anything that you don't want to dissect in the final process of this workflow.
15:25
So obviously we're kind of using the measurements that were applied automatically when the data was segmented.
15:33
So here they are using the area, but can be intensity, it can be nearest neighbour, it can be pretty much anything that we want it to be.
15:42
And then in the last step, we were using the built in feature to import all these masks, all these shapes in the XML format, right?
15:50
So essentially what we are doing, we're saving the coordinates of these masks that has to be now transferred into the next step of the workflow which is microdissection itself.
16:03
So here we are opening the shapes, importing them.
16:08
Now we have to mark or calibrate all the reference points we've been marking a second ago, right?
16:13
So we kind of indicate where are these marking points and after a few seconds we should be able to see the masks being applied directly on the sample.
16:25
This is actually visualised light on the laser micro dissection microscope.
16:37
There you go.
16:38
So now it's pretty accurate, but obviously you can still adjust it if necessary, right?
16:43
So you can work individually with the shapes that are not really imposed properly, or you can select the entire groups of them.
16:50
And so thanks to that, you're always sure that you will be able to dissect and isolate the material that you want without cutting it in half, for example.
16:59
After that step, we are ready for cutting.
17:01
So I got the conversation with the people behind the original digital proteomics paper.
17:07
So they told me this is actually quite a time-consuming step.
17:10
The optimised parameters, they were able to dissect 1200 cells per hour, but then the whole tissue section is obviously larger than that, right?
17:19
So it's relatively quick when it comes to imaging, relatively quick when it comes to image analysis.
17:26
Dissection takes some time.
17:27
And then depending on what kind of experiment they are trying to set up with the downstream analysis, it can again take it a little bit, can take a little bit longer.
17:38
OK.
17:39
So as I mentioned earlier, that method was chosen to be the method of the year 2024 by Nature methods.
17:49
And currently there is a huge interest around these techniques.
17:53
So what would be the next thing with these technologies I just shown you currently with the DVP you're probably introducing just couple of markets using typically regular fluorescent microscopes, but they're already researchers who are trying to combine multiplexing with dissection, right.
18:11
So imagine introducing 15 biomarkers based on that you can identify specific targets and then in the next step you can actually dissect them, put them in a separate wells in the multi well plate and get pretty comprehensive understanding on what is happening with the single targets in the molecular level.
18:26
So this is where we see as the next step the whole workflow.
18:32
But let's see, there are new things popping up almost every day in spatial biology field.
18:37
OK, so before I finish that, I just want you to visit our website.
18:42
We have like a dedicated special biology website where you can get some more information about the system, about the methods, about the white papers behind the technology that we are providing.
18:52
And also there is a nice link to the spatial biology book that you can download and have a little bit of a flavour of what we are doing and like and how we can help you with the technology with that.
19:05
Thank you so much.
19:06
I'm happy to answer any questions if there are any.