0:03
Perfect.
0:04
Yeah, I changed the slide, the top title slightly, but the focus is going to be the same.
0:10
So the next 15 or 20 minutes or so, I want to talk to you about the mass spec based platform we've set up in house as a CRO service provider to analyse tissue and the plasma samples, mostly focusing in this case with some case study focusing on IO.
0:30
Just quick words, you know, to us.
0:32
So we are a Swiss company based in Shirin, which is a small town next right next to Zurich.
0:40
That's where you know, one of our major lab site is recently.
0:45
So since 2 years, we also operate a lab in the US in Newton, MA in the last more than 15 years.
0:53
So we have really been dedicated in the developments and optimisation of mass spectrometry based platforms.
1:00
And with that what we're trying to do is really to, you know, to bring to democratise these technologies and to bring this technology to the drug developer and drug counters.
1:11
And as of 2023, we also became part of Bruker and this of course also helped us a lot in the growth and in the entire scope of the company.
1:22
What we would like to address here.
1:24
I think so very, you know, everyone has seen this.
1:26
I don't need to you know, go into the detail.
1:29
So genomics or genome sequencing, the cost has dropped really significantly, you know, from the start already 25 years ago, which, you know, I haven't been able to experience myself, but I think many of you probably have seen this change.
1:42
And one of the real challenges for proteomics being, you know, a routine technology to be applied in clinical sample analysis when dropped discovery general is typically the cost and the time it takes to complete and to execute such an analysis.
2:01
And adding to that, I think everyone can appreciate this plot is that with the translation to the proteome level, the complexity increases dramatically.
2:13
And here, you know, we're just, you know, talking about on the protein level.
2:17
And as you many of you know, they're isoform, they're PTM, they're what we call proteoforms.
2:22
And of course, protein also have their defined tertiary structures.
2:26
And this really makes proteomics way more complex in a sense than genomics and way more dynamic.
2:32
And this again really requires a technology that is as high throughput is cost efficient.
2:38
And this can allow us really to characterise the proteome because it's not really a static picture, but it's really a dynamic picture that needs to be sampled again and again to actually gain any useful insights from a biological sample.
2:54
So what has happened really to mass spec-based proteomics?
2:58
I think a very good example which is not even on this bar plot is if you look back at 2008, there's a landmark paper by Matthias Mann's lab where they could publish the proteome of haploid versus diploid yeast in Nature on the cover of Nature, right?
3:16
Which is not thinkable nowadays where you can do it probably in 10 to 15 minutes.
3:21
And I think what you can see really from this plot is with the advancement and instrumentation and technology.
3:30
Basically the coverage of the proteome by mass spec has steadily increased for the last 10 or 12 years where I think you know based on the extrapolation by next year or a year later, mass spec will be able to cover the expressed proteome in a given cell line or even in the given tissue samples obey the complexity of these type of samples.
3:55
I think what's even more important than the coverage is really depicted on this slide.
4:01
So this is really showcasing one of the challenges why mass spec hasn't been really used routinely in the analysis and you know, applied in, you know, either analysis of clinical samples or in preclinical even, you know, early discovery in the entire pipeline is first, really the speed of analysis and second the cost.
4:22
And all these has to do with how much maintenance you need for an instrument.
4:26
It is a robust process.
4:28
Can you really run it, you know, without having, you know, few skilled scientists really taking care of it.
4:34
And it was not the case even like 5 or 6 years ago where you need several FTE, right to really work on an instrument to make sure it can deliver reproducible data and outcome.
4:46
And this really has changed I would say really rapidly in the last 2-3 years with the newer generation of both liquid chromatography systems, so LC and mass spectrometer.
4:57
And this really has, I would say really transformed, right, the entire mass spec-based platform.
5:04
And this kind of puts mass spec really into the realm of routine analysis, right, because it can think about it.
5:12
So SPD stands for samples per day.
5:14
So if we can analyse nowadays maybe 80 to 100 samples per day at a cost around $300 to $400 per sample, can imagine this would enable you to analyse small to medium, let's say FFPE cohort, which I will show us a case study later.
5:31
And I think given the steady increase of the speed and decrease of cost, this could also enable really large-scale population scale biofluid or plasma proteomics in the future.
5:48
So you know, just going back really to the basics of mass-spec based proteomics.
5:54
So what happens is samples are denatured no matter what you know where they come from, they might be tissue, might be biofluid, can come from human, but also different type of species.
6:03
They're digested into tryptic peptides using trypsin and those peptides are subjected to the LCMS analysis.
6:11
So LC helps to separate those peptides and then before they get ionised and analysed using the mass spectrometer.
6:19
And typically mass spectrometer produces raw data that are search using dedicated software that are mapped to the proteome of the species you would like to analyse.
6:30
And then what we typically use it for, right.
6:33
So it can be used to generate hypothesis in an unbiased way.
6:36
It can be used to validate certain biological hypothesis that you have.
6:40
Of course this applies to things like biomarkers discovery, but also many specialised application I mentioned before you can use it to study PTM study, you know, specific population of peptides such as immuno peptide which are presented by MHC complexes.
6:55
In this case, what I would like to, you know, tell you maybe in the next 10 minutes are, you know, 2 case studies.
7:03
The first one is a collaboration with clinical colleagues at Gustav Roussy in Paris, where we carried out a small pilot study analysing FFP samples from endometrial patients from an immune-oncology trial.
7:21
So in this case, because it's a fairly rare type of cancer, especially if you're taking the mismatch deficiency background into it.
7:30
So our colleagues managed to collect around 20 to 25 samples in the past 10 years.
7:37
So these are, you know, bank samples that were stored in the clinic.
7:41
And the question they asked really to identify a potential biomarkers or pathway that are associated with response or resistance.
7:51
So what basically happens?
7:53
So these samples were clinically characterised and with metadata provided by our collaborators.
8:00
And what we did is to take, you know, these in total 25 FFPE samples.
8:05
We process them with our developed proteomics workflow in house.
8:10
We generate you know raw data and then we together with the collaborator we interpret this data to get to certain hypothesis and also followed up some with orthogonal methods.
8:25
Just a very quick technical run.
8:26
So these are endometrial FFPE samples and you can see that with mass spectrometry, you can fairly easily quantify around 10,000 protein across all these samples.
8:39
And since these are in total 25 samples, the total analysis time is less than a day.
8:45
Together with bioinformatics, maybe that will go take another two or three days.
8:49
But this really allows them also, you know, to get to the data pretty fast, right?
8:54
It's not another two or three months where they have to wait for it.
8:57
So basically they get the data really in a very short time frame without, you know, really going much into the exploratory data analysis since, you know, it's not a bioinformatics focused talk.
9:09
But some of the tools, you know, one can use, right, are for a very simple, you know, looking at, you know, regulated proteins, right between responder and non responder and use different, you know, in this case, machine learning to basically find a certain panel or a certain signature panel that can differentiate in this case responders from non-responders.
9:34
And then once we have this panel, we can then ask ourselves the question together with the clinical collaborator, from which pathway are these markers actually from?
9:45
And since, you know, these guys have been working decades, right, with these indications, they know they certainly have certain hypothesis in mind and they know exactly what they're looking for, right?
9:55
So from these analyses, these are some of the hypothesis they come up with, you know, things including non specific inflammation mediators.
10:03
And most importantly, what really caught their attention that they've seen a lot of extracellular matrix components being upregulated in the resistant cases.
10:15
And this is something they select to follow up with also because they have in their hand, you know, microscopy tools to really follow up on collagen.
10:26
So the two markers we found as you can see here, 2 collagen proteins from the family being significantly up regulated in the known responder.
10:38
And what our clinical collaborator then did in this case is the Masson trichrome staining to really showcase that, you know, based on these samples, they have that in the non-responder, you have a higher density of collagen.
10:55
So this is, you know, an orthogonal methods in this case to really validate the hypothesis they found using mass like proteomics.
11:03
Another thing, another, you know, fancy way of doing a microscopy.
11:06
Don't ask me the details there.
11:08
I'm an analytical chemist by training.
11:10
Absolutely no idea how this works, but I know this picture look very different.
11:14
So one is organised.
11:15
This means, you know, this correspond to the responder and the disorganised one that is pretty, you know, obvious on the right side of the screen are coming from the non-responders.
11:31
And based on those picture, they could also quantify, of course the amount of collagen and that correlates very well with the findings we have when we analyse the data using mass spec on the bulk tissue.
11:44
And interestingly, what they did is they used T cell markers to correlate immune cell infiltration with collagen content.
11:54
And this is also what they could observe is basically if the collagen content is high, of course they would create, you know, let's say stearic hindrance, right?
12:03
That's prevents those immune cells from entering the tumour environment.
12:08
And this is also a behaviour that has observation that has been previously reported.
12:13
And right now what they're really interested in, because they're really in the clinic training patient, they're trying really to develop a method based on microscopy, which they could potentially later on even use on site to decide, right?
12:26
Whether you know, based on the collagen environment of a patient sample, whether they should treat the patient with such a treatment selection.
12:37
Because of course, if they know with high probability that the treatment is not going to be effective, you can of course save all the adverse effects for the patients.
12:45
So this is really something for us, it's also a lot of learning, right?
12:49
Because we are far from the clinics and it's nice to see how our clinical colleagues actually use this information in a real world scenario.
12:59
In the next in the second part right of the talk, I want to, you know, move a bit from tissue proteomics to biofluid O2 plasma proteomic.
13:10
And the reason for that is very simple.
13:11
I think I have to mention in many meetings here and everyone knows the answer, right?
13:15
So it's much easier to collect, you know, blood than getting FFPE tissue from a bank or even collect, you know, any biopsy from, you know, from a patient.
13:26
But of course it comes also with a lot of challenges.
13:28
So as you know, in tissue, it's fairly easy to interpret the data because the composition of the tissue is known.
13:36
They're, they come, you know, they're basically made out of different cells, right.
13:40
And also there the dynamic range of those samples are fairly low compared to plasma.
13:47
So I think these are two very, let's say, informative graphs that really showcase why it is very difficult to do plasma proteomics, apart from all the pre analytical issues that everyone is of course aware of from collection from, you know, anything that could happen between the collection side and the testing lab, so on the far side.
14:14
So one of the challenge in plasma is really there's no ground truth, right?
14:17
If you look at a cell line, it's very clear which pathway should be there and you know, how what protein should be like expressed a certain level in plasma.
14:27
This is really unknown apart from the functional plasma protein like R Bloomin or RFR 1 to trypsin or the, you know, immunoglobulin families, right?
14:35
Because there's a lot of leakage going on, a lot of shedding going on, you know, from different cells and tissue in the body.
14:43
So basically that the plasma proteome will always be dynamic and we're always just looking at a single time point and a very narrow window, right, instead of, you know, getting the whole picture.
14:56
And then the second challenge is really the broad dynamic range.
14:59
So as I mentioned already, our boom in, of course, it's a very abundant protein that makes up, you know, and the top 22 proteins make up 99% of the entire protein mass.
15:10
And that of course, mask a lot of signals, especially for mass-spec, this is challenging, but it also can also be challenging for other platforms as well.
15:18
And I think what's not even on this plot, right?
15:21
They are lower abundant proteins such as cytokines and chemokines that can be addressed with different technologies.
15:27
But what can be even lower abundance is the polio form of many proteins.
15:32
So for example, P Tau is A subpopulation of Tau, right, that it's way lower bound and you have different P Tau options, right, on different sides and different combination of those, right.
15:44
So basically this creates really the much lower part that is not actually shown on this graph and they might be a valuable also for biomarker discovery.
15:57
So one of the advantages of, you know, mass spectrometry is really it is completely unbiased.
16:01
So before we look at this sample, we have no idea what's going on.
16:04
We don't have any reagents targeting certain proteins.
16:07
So there's no there's the concept of panel doesn't exist, right?
16:11
So what we're really trying to do is to develop method that can basically somehow compress the dynamic range so we can analyse plasma sample in a much simpler way.
16:24
So in this case, what we did is we developed a platform called P2 enrichment system.
16:29
So these are particles where we can incubate with plasma.
16:32
And then what happens is the higher abundant protein gets basically kind of removed from the sample where we enrich for the lower abundant protein.
16:42
So in general, the dynamic range of plasma from around maybe 12 and 13 orders of magnitude get compressed to probably 7 to 8 orders of magnitude.
16:51
And this helps a lot the follow up analysis by mass spectrometry.
16:56
And basically what this can do is, you know, instead of just looking at a few 100 proteins, now we can quantify around five or six thousand proteins it depends of course always on the cohort and on the indication we can easily quantify to 6000 protein.
17:14
So as a showcase here, so we collaborated with a phase two clinical trial happening in Switzerland on NFCLC.
17:24
And the question asked there is also very simple, can we use baseline sample for predictive biomarker discovery and can we use the follow up time points, you know, to make any type of prognostic biomarker discovery attempt.
17:42
So what we did here, so we analysed the same plasma sample collected from this cohort with our P2 platform and to cover the cytokine chemokine we use on our Biosciences article on the NULISA platform, basically what you have here.
17:58
So from 35 patients, three time points, we unbiasedly quantify 6000 protein using spec and complement this with around 250 inflammatory markers.
18:11
So when we came back with this data, you know, to our clinical collaborator, the first question we get is, oh, this is a lot of thing we have to look at and they're really not used to it, right?
18:19
Because they typically deal with patient and treatment selections.
18:23
So the first question they ask.
18:25
So can we somehow simplify this in a way that, you know, we can at least get a feeling of what can be achieved with the data before you do all these fancy, you know, machine learning or AI based predictor, etcetera, right.
18:38
But then what we try to do as we try to turn this, you know, this resources of data into hallmarks and pathway that our clinical collaborators suggested based on their, you know, previous work and experience with an SLC and immune oncology treatment and resistance.
18:58
And we turn basically this gigantic data matrices into something much simpler.
19:04
So we summarise of course all this data and you know, this is some bioinformatics work in the back end, right to normalise and to really try to condense this really into much simpler matrices and representing different hallmarks, you know, of either lung cancer or of IO treatment.
19:23
So in this case we're looking at the responder.
19:26
So if you are looking at the pre dose, right, we see, for example, just, you know, the glycolysis there.
19:33
If so, blue is rather low, right, red is rather high.
19:36
So the light red means, OK, it's probably medium high.
19:41
But if we again, looking at a non responder here in red, right, we see that glycolysis is lower in this case, compared to the responder.
19:54
And for our clinic collaborator, right, each don't responder non responder, right is a patient for them, right.
20:01
So it's nice to have statistics of course, but also nice for them to relate to each of the patient, right, Because they have of course a lot of other clinical data and follow up data on each of them.
20:12
And it's good to correlate all these back with the clinical records, etc.
20:19
To see how let's say in this case plasma proteomics can better support, you know, clinical decision making or at least use it as a resource later on for the planning and design of the next trial.
20:31
So without really going to much detail on that and the interest of time, really the take home message here first is I think you know, really with the development in instrumentation and tools, mass spec is mass spec based party.
20:49
Oh makes really becoming a more routine way to analyse the party ohm of different type of samples.
20:55
I think the really key part there is first the speed and the robustness, but also the cost.
21:01
So these things, you know, because you know, you're generating A hypothesis or testing, you're approving A hypothesis, right?
21:07
So of course, we really understood from our, let's say, experience interacting with collaborator that, you know, we have also to the risk there uphold, right?
21:18
But because it's very hot, you know, it's very probable that you won't find any biomarker just from 30 samples, right?
21:24
This is just the truth.
21:26
So what we were you trying to do is trying to use the data in the most efficient way to help biologists and clinicians to really get to any actionable insights in this case.
21:40
And of course, what we observe all during the years, right, it could be FFPE, it could be any bio fluid samples pre analytic.
21:47
So always very important.
21:48
So they, of course, there are ways to deal with, let's say heterogeneous cohort, but of course, this of already, let's say poses a challenge right, to any biomarker discovery exercise.
22:05
Yeah.
22:05
With that, we'd like to thank our collaborators, of course, the patients who donated the samples of research and the team at diagnosis who carried all this great work and thank you for your attention.
