0:00
Just first, I've referred to spatial proteomics, which is a bit of a buzzword at the moment, just to explain what I mean by that term, I've been referring to this technology for a long time as tissue multiplexing, multiplexing the immunohistochemistry.
0:15
So what I mean by spatial proteomics is we're looking at a high number of biomarkers in single tissue sections so that we can look at not only how they correlate with each other within cells and between cells and look at complex biological processes and complex cellular niches.
0:32
But also we can do all of this on a single section, which is really critical in clinical trials where quite often small biopsies and tissue going to a lot of different laboratories or a number of different analyses also including often the diagnostic work.
0:49
There we go.
0:58
OK, so the Akoya services, we're going to start talking about the Akoya PhenoCycler Fusion.
1:04
So if anyone who's not familiar with this platform enables Hiplex up to 100 target protein staining on a single section by a single cocktail and single staining cocktails enabled by barcoding.
1:19
As you can see in the bottom right here, barcoding our antibodies with an organic barcode unique to that clone.
1:26
We then detect these antibodies with complementary oligos conjugated to fluorophores and we cycle through imaging and iteratively staining these complementary barcodes.
1:39
So we stain all the antibodies in one go and then cycle the imaging, the stacked image.
1:49
Our service execution service is again a fully integrated service from panel optimisation.
1:55
We have both pre optimised panels that we can start from, but we can produce fully customised assays or to customise these existing assays.
2:05
And the panel's development includes optimising the tissue for your particular study and we can validate the whole process to [unclear]for clinical trial use.
2:14
This also includes, as I said, upstream and downstream analytics, whether that's the Histology back end or the orthogonal analysis such as the data analytics from the image or the validation assays such as in this chemistry in-situ hybridisation or the parallel sort of standard of care clinical assays that we can use to correlate back to.
2:35
So we might use a chromogenic IHC methods to correlate back to our Multiplex.
2:44
I'm going to be talking today about specifically the Phenocode IO60 assay from Akoya, which is not just a pre optimised panel, but a pre curated immunophenotyping panel looking at immune response to tumours and therapy.
3:00
This shortcuts development time because we have a large number of common immune phenotype markers that we can work from and then we can customise this from here.
3:06
So we can either pick or choose each image that we're cycling through here represents one of these modules of targets.
3:13
Those modules can be recombined in any combination or all included together to make a 60 Plex panel.
3:19
And we can add custom targets to this as well.
3:23
It wouldn't be a Multiplex for essence talk without a nice pretty picture, but here's an example of a multi indication tissue bot that we've used to optimise this panel and an example of some context markers to see the quality of the staining complement our PhenoCycler.
3:43
We also have an Akoya PhenoImager HT, rather than cyclic imaging, sorry, we have cyclic staining of antibodies and a single imaging cycle.
3:57
This is a lower Plex method, but it's really optimised to high throughput, high sensitivity.
4:04
So the staining here is done on your standard in his sketchy platform such as the Bond Rx and the Ventana Discovery Ultra.
4:12
We can do same slide multi omics up to six targets any combination of RNA and protein.
4:19
This is really well suited to signature validation.
4:22
So when we've run these large discovery experiments that we can take a core cellular process or an immune cell niche and we can validate these on large number of samples to the tune of hundreds of samples a week rather than dozens with the other methods.
4:38
It's also wealthy to taking those signatures and moving them towards the clinic and generating CDxs things.
4:44
The utility for platforms like this is for screening large numbers of samples with a key pathway to find groupings and then to take key examples of those groups and then reverse engineer the mechanisms from there.
4:59
And as I said, we also provide a fully integrated data analysis service.
5:06
So as we can see on the image on the right here, starting from the image of this cycle here, we then perform feature identification.
5:15
We'll classify the features.
5:18
And then once we've got the classified features, we can extract data from the image based on these features and then perform downstream data analysis, both relating these biomarkers to each other in space in a sample and then comparing between samples and sample groups.
5:33
All of this is fully customizable and again for validated [unclear] pipeline.
5:40
So we have been working with our suppliers and our biopharma clients to continue to push the robustness of these technologies and continue to drive the quality which we can deliver these and the scale at which we can deliver these to support clinical trials.
6:01
So we've been working with Akoya and the IS60 panel and another part here that I'll mention shortly.
6:09
That's part of an investment in this drive to push quality for us.
6:15
Automation enables both not just scale but also standardisation within a lab and between labs.
6:21
And these are both key requirements for large clinical multi-site studies.
6:26
So as well as optimising this automated method, we are also in the process of performing.
6:33
I don't have the data today, we're sharing it later this year.
6:35
A fit for purpose validation for clinical trial use of this fully automated method both within our site and between us and another site to show that we can enable the scale or if we enable the scale of standardisation we will get from automation, Can we also maintain a high level of quality and enable assay transfer and capacity sharing with clients.
6:58
And I'll shortly describe our data-driven evaluation of that.
7:04
So a partner of the automation of this is here.
7:06
The spatial station is an automated liquor handling platform, a robotic and but this has been optimised for the requirements of these complex spatial assays.
7:20
It's been optimised to handle histology samples, to handle long standing protocols with humidity and temperature requirements.
7:26
And it's also been optimised to have very accurate low volume pipetting because when we're pipetting many reagents into small numbers of samples here.
7:36
So we need a high reproducibility of small bonds as well as being highly flexible to customise these complex assays.
7:45
A strength that this platform brings to us is it's been ground up designed to be FDA CFR 21 Part 11 compliant, which enables us to also create all of the auditing and reagent tracking that we need to provide a really robust clinical trial service.
8:03
This is our standard panel development work where we would start by screening novel antibodies, test their specificity and their sensitivity, compare them against control samples.
8:15
And once we're happy the antibodies are performing well and move them into our pre-existing panel, reoptimise the panel for the tissue with the new markers present and then move to validation.
8:25
By working by pre optimising this fully automated IO60 method, we can shortcut both the screening and the panel optimisation plus or minus a couple of key additional targets and move quickly onto panel validation.
8:39
As I said, our whole pipeline is validated.
8:44
All our technology and all the processes are pre validated.
8:46
So all we have to do at this stage is an effective position study, make sure that the assay that we finalised is reproducible with the variables that we can predict between days, between operators, between instruments and sites, for example.
9:00
Once we've done this, we can set an expected performance of the assay.
9:03
And what I'll talk about next is how we then use this to monitor performance and not just identify quality issues, but prevent them from happening in this place.
9:12
So it's just a quick diagram of our precision study that we've taken multiple serial sections from the same multi tissue indication, multi indication TMA.
9:22
We have a single batch running multiple slides to measure variation within a large run and then multiple smaller runs to measure the difference between time and operators.
9:32
These sections however are alternate sections and the inter interlaced sections have been retained by our pharma partner who is going to repeat the same assay.
9:41
We'll transfer the method to them.
9:43
They are repeating the assay and we're going to quantify not just performance within the two individual labs, but the performance between the labs to show that these methods can be effectively transferred between sites.
9:58
This data, as I said, is currently being generated and we'll share this by webinar later this year.
10:03
So I'm just going to finish off by talking about how we do this quality measurement.
10:07
It's something that we already do and that we're going to be applying to this new method.
10:10
I'm going to start by just defining a couple of straightforward terms.
10:15
We're going to be basing this around a plot called the Levey Jennings plot, which is a way of measuring expected ranges over time.
10:22
We're going to be using mean and standard deviation.
10:25
And just a quick reminder, standard deviation anything outside 2SD represents a less than 5% chance that this is an expected variance.
10:37
So once we've performed the precision study we just described, we come out with a quantifiable metric from our images and these numbers we plot onto the Levey Jennings plot.
10:47
These aren't absolute figures here.
10:48
This isn't intensity levels.
10:49
These are mean and variation plotted.
10:52
This is why they all overlap each other, even though they have very different reference ranges.
10:56
So what we're doing here is monitoring between three runs, the variation that we see and setting our expected variance.
11:05
What this enables us to do is to then monitor performance against this range over time.
11:11
We can also annotate this plot with all the information we need to be able to trace any issues that we see.
11:17
So we've we have instrument number, we have dates, we have marker, we have fluorophore, we have channels.
11:24
So if we see correlating patterns between multiple markers on the same graph, we can trace them down to their common features.
11:35
In order to generate this figure, we there are a number, there are a number of different ways that it can be done.
11:40
It depends on the requirements for the study and the requirements for the markers.
11:43
So we have to extract a numerical feature from the image.
11:47
We're going to select an ROI.
11:49
There's an example ROI on the right here, a region of interest, which represents the staining on the slide.
11:54
We're not looking for true biological variance between slides.
11:57
What we're looking for is a standardised staining area that will not change between sections.
12:02
So it's a different requirement for what you'd be looking for if you're trying to measure a biological process.
12:06
We want to find a stable structure that we can reproduce to be measured across several sections.
12:13
Another issue that can arise is, for example, if we have a weak rare marker as we see here, you might see a couple of small purple dots here.
12:21
Within the same ROI.
12:22
We have a very low signal level and if we take a single numerical average of the intensity from this area, the majority of the contribution to this signal is actually background.
12:31
So a small adjustment in background.
12:33
You can see on this last point on the graph here, a small adjustment to background from a high auto fluorescent slide contributes a huge amount of variance, which when we look back at the image doesn't represent true variance.
12:43
So we can enrich for these signals what we're taking our in regions interest we can enrich by segmenting cells and only looking at cells and not the area between cells, which removes a lot of this background.
12:55
If it's a rare cell, we can classify the cells first and then only look at cells that are classified as positive.
13:00
All of these steps add bias because they're analysis steps where we're restricting the view.
13:06
So we on a marker by marker basis, we'll decide which of these methods most appropriate for monitoring accurately the performance of each marker.
13:16
As you've seen back here, the graph doesn't necessarily tell you there's a problem, it just indicates there might be.
13:23
And when we look at the image, there definitely isn't.
13:25
So there's a lot of visual expertise involved here as well.
13:28
So our combination approach, qualitative, we look at the image, does it look correct?
13:32
Do they look similar between images?
13:33
Is the staining pattern as expected?
13:36
So we have to understand it by markers.
13:37
We have a semi quantitative approach where we're going to look at grading.
13:41
So do we have low, medium or high staining and do the histograms look good?
13:46
And then we'll also have a fully qualified quantifiable method with [unclear] reference ranges.
13:51
And between these three methods, we can identify the performance of each marker.
13:57
Here's an example of 6 serial ROIs looking relatively reproducible.
14:03
Let's check the graphs.
14:05
They also look pretty reproducible.
14:07
Here's an example of three that don't look reproducible.
14:11
Let's check the graphs.
14:12
The graphs also confirm that the 4th, 5th and 6th images are looking brighter.
14:18
This helped us go back, identify what the cause was and fix that before we finalise the method for this assay.
14:26
It also helps identify why difference.
14:27
So here we've identified antibody 3 has poor performance across a lot of different tissues on this multi tissue indication.
14:34
So there's an issue with this antibody across tissues.
14:37
And here a single sample where all the antibodies perform poorly.
14:40
And when we looked into it, the high variance was due to low signal.
14:43
There was low signal because we had no staining.
14:46
So the sample was dead and we excluded this from analysis.
14:50
So once we have the plot, we've set our reference ranges and it's not just a static data set used for a preview method.
14:58
We'll then use this to monitor performance over time.
15:00
So every time we run a batch of patient samples, we include a serial section from the same block and monitor performance.
15:06
What this enables us to do is not only find QC errors and rectify them in a root cause analysis approach, but actually prevents us from finding errors.
15:17
You can see an example here of an antibody of a target that is dropping in performance over a number of different runs.
15:23
And if you look between any 2 given runs, they're not always dropping, they're not necessarily dropping by that much.
15:28
But when we look at the trend over time, it's trending towards the end of our reference range and before it falls outside our reference range and we fail a run and lose all of the data.
15:38
We can correct this.
15:39
We can change the antibody batch, we can change the fluorophore, we can maintain the instrument, and we can prevent the loss of data.
15:44
So it's not just protecting us from including that data in a data set.
15:48
We won't even perform a stain that will fail in the first place.
15:52
It doesn't just save time and money, but for precious clinical samples, we may not have spare tissue.
15:56
So this will ensure that a patient is kept inside the trial data set and not lost due to QC issues.
16:09
So in summary, Propath, we have been collaborating with Akoya and Parhelia to really drive this approach.
16:14
This kind of data-driven quality first approach to delivering Hiplex spatial proteomics in clinical trials. Akoya’s Phenocycler assay and their IO60 assay really help shortcut the development cycle time.
16:30
By having not just pre optimised but pre curated immune environment panel and with the automation provided by Parhelia, we've managed to generate a method that removes a lot of the manual steps.
16:43
Will enable cross site working and enable scale.
16:49
Importantly because we'll be able to run high numbers of slides with the same pair of hands.
16:58
Propath end to end solution then starts before this running everything from the Histology and sample preparation, all the way through the assay development and validation with our pre validated pipeline, through sample staining and all the way into bioinformatics and data analysis.
17:16
Thank you for listening to my talk.