October 3, 2023 • 42 mins
Welcome to the Olink® Proteomics in Proximity podcast!
Below are some useful resources from this episode:
Published study of primary focus
Álvez MB, Edfors F, von Feilitzen K, Zwahlen M, Mardinoglu A, Edqvist PH, Sjöblom T, Lundin E, Rameika N, Enblad G, Lindman H, Höglund M, Hesselager G, Stålberg K, Enblad M, Simonson OE, Häggman M, Axelsson T, Åberg M, Nordlund J, Zhong W, Karlsson M, Gyllensten U, Ponten F, Fagerberg L, Uhlén M. Next generation pan-cancer blood proteome profiling using proximity extension assay. Nat Commun. 2023 Jul 18;14(1):4308. doi: 10.1038/s41467-023-39765-y. PMID: 37463882; PMCID: PMC10354027. https://pubmed.ncbi.nlm.nih.gov/37463882/
Laboratory, first author, and corresponding author of the study
· SciLifeLab, a collaborative resource for life scientists located in Sweden: https://www.scilifelab.se/
· María Bueno Álvez (first author), PhD student, Science for Life Laboratory (SciLifeLab): https://www.linkedin.com/in/mar%C3%ADa-bueno-%C3%A1lvez-33395b192/
· Dr. Mathias Uhlén (corresponding author), Professor of Microbiology, Royal Institute of Technology (KTH), Leader of the Human Protein Atlas, Founding director of the Science for Life Laboratory (SciLifeLab): https://www.kth.se/pro/sysbio/uhlen-group/researchers/mathias-uhlen-1.67763
Olink tools and software
· Olink® Explore 1536, the platform that measured proteins in this study with a next-generation sequencing (NGS) readout: https://olink.com/products-services/explore/
· Olink® Explore HT, Olink’s newest solution for high-throughput biomarker discovery that measures 5300+ proteins simultaneously with minimal sample consumption: https://olink.com/products-services/exploreht/
· Olink® Insight, an open-access resource to accelerate protein biomarker discovery: https://insight.olink.com/
UK Biobank Pharma Proteomics Project (UKB-PPP), one of the world’s largest scientific studies of blood protein biomarkers conducted to date
· Published article: Styrkarsdottir U, Lund SH, Thorleifsson G, Saevarsdottir S, Gudbjartsson DF, Thorsteinsdottir U, Stefansson K. Cartilage Acidic Protein 1 in Plasma Associates With Prevalent Osteoarthritis and Predicts Future Risk as Well as Progression to Joint Replacements: Results From the UK Biobank Resource. Arthritis Rheumatol. 2023 Apr;75(4):544-552. doi: 10.1002/art.42376. Epub 2022 Dec 28. PMID: 36239377. https://pubmed.ncbi.nlm.nih.gov/36239377/
· UKB-PPP website: https://www.ukbiobank.ac.uk/learn-more-about-uk-biobank/news/uk-biobank-launches-one-of-the-largest-scientific-studies
Analysis of UK Biobank proteomics data from cancer patients, with co-authors including Ruth C. Travis, Karl Smith-Byrne, and Joshua R. Atkins
Preprint article: Papier K, Atkins JR, Tong TYN, Gaitskell K, Desai T, Ogamba CF, Parsaeian M, Reeves GK, Mills IG, Key TJ, Smith-Byrne K, Travis RC. Identifying proteomic risk factors for cancer using prospective and exome analyses: 1,463 circulating proteins and risk of 19 cancers in the UK Biobank. medRxiv 2023.07.28.23293330; doi: ht
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
[Music]
(00:05):
Welcome to the Proteomics and Proximity Podcast
where co-hosts, Dale Yuzuki, Cindy Lawley and Sarantis Chlamydas from Olink Proteomics
talk about the intersection of proteomics with genomics for drug target discovery,
the application of proteomics to reveal disease biomarkers,
and current trends in using proteomics to unlock biological mechanisms.
(00:27):
Here we have your hosts, Dale, Cindy and Sarantis.
Welcome to Proteomics and Proximity.
I'm Dale Yuzuki, your host, along with Cindy Lawley and Sarantis Chlamydas.
And here we have a special treat today. Sarantis?
Yeah, thank you, thank you, Dale. I'm really happy to be back from holidays
and then having the first host, Maria.
(00:48):
And Maria is a student, she's a PhD student in Mathias Uhlén's lab,
and she had recently a great paper, she was the first author of a great paper that
we'll discuss a little bit later in more detail.
Maria started her studies in Barcelona in Autònoma University of Barcelona,
then she moved to Stockholm [Sweden], and she gained a Master's degree,
(01:11):
joined the Protein Atlas, the Human Protein Atlas project in 2020,
and, since 2022, she's been a part of Mathias Uhlén's lab
where she's working on an amazing project around proteomics in different diseases and cancer
to try to identify signatures, protein signatures.
(01:33):
Maria's background is around, of course, proteomics, transcriptomics,
and she's also a bioinformatician and wet lab scientist.
That means she has a complete package of a full experience and expertise in the field,
also in the field of multiomics - that's something I'm really interested also to hear.
And I will start my question, actually, Maria, I mean, we are really happy to have you,
(01:55):
and it is always a pleasure to discuss with you. I would like to know a little bit more
how you see your transition from transcriptomics to proteomics,
and how you see the match between these two omics approaches from your perspective?
Thank you, Sarantis, it's really great to be here today.
So, I think for me it was a very smooth transition,
(02:18):
basically because at the Human Protein Atlas, that's been kind of a very connected thing.
It's never been like, transcriptomics is one thing and then proteomics is something
completely independent. It's always been quite linked, and that's because, of course, the Human
Protein Atlas has been about understanding proteins, but we have not always been able to just
(02:39):
look at them right away in such a scale, right? I mean, but now with Olink, it seems like that has been
there forever, but in the very early days, transcriptomics was a really, really useful tool
to look at the proteins - indirectly, of course - but we learned a lot and it's
a resource that is still used to this day in many, many groups. So, I was really lucky to work
(03:02):
with transcriptomics first, I think, so that I could understand ... I sometimes say
that that's the simple part where everything is clean and you don't have so much
noise of post-translational modifications and so on. So, I think it's more of the straightforward
tissue transcriptomics analysis. But now, of course, it's really great to go to more toward plasma
(03:24):
proteomics, which is kind of the other extreme. To ask you a little bit about the transcriptomics piece,
there at Science for Life Laboratories, right? And was the interest there
in transcriptomics around a cancer indication, or was it across tissues?
Yes, so the very early transcriptomics, that was published around 2015 in Science, and that was
(03:46):
healthy tissue. So, that was kind of a really big study across different tissues, try to just
characterize which proteins are specific, and that was kind of the flagship of the Protein
Atlas, try to just see what's there. But then, of course, there's been other, more
integration with GTEx data and trying to understand, of course, a little bit more about disease.
(04:08):
And so, this transcriptomic data set was from the tissues from Science for Life Laboratories,
right? Like you mentioned, there are other projects that are characterizing tissues like GTEx.
What made the Science for Life effort unique or different?
So, nowadays, at the Human Protein Atlas (HPA), we have both data from HPA and GTEx integrated
(04:30):
into the database. When I say HPA, I mean Human Protein Atlas, it's just a little bit too long.
So, our own samples, for example, we have a really detailed brain collection of samples, so it's
not only brain, but we have tons of different regions in the brain. So,
for example, we have a ton of detail on that stage that you can't really have with GTEx data.
(04:57):
So, it was more trying to complement, and also, of course, if you have two different independent
sources of data that agree on the same transcriptome levels, that's, of course, very reassuring.
So, that's why we try to combine both in the same database.
Amazing. So, hey, Maria, this is my first chance to ask a question, and I'm actually going to
(05:21):
click back to ask about your interest in getting into science. So usually we say,
you don't have to start with elementary school, but if you want to, I mean, whatever,
you know, sparked this interest in you, I'd love to hear about that. And then
what nurtured that interest to end up in Mathias's lab? I mean, what a remarkable
(05:47):
place to be. Thank you for question. It's, of course, always difficult to know when it all
started, but I think you will probably click with this (05:54):
that
biology is just fascinating. Like, you see people getting old, people getting sick.
I don't know why, I see the effect of epigenetics as well. I have a twin system myself,
so I was always surrounded by what I thought was biology. And then I thought,
(06:15):
studying genetics, which was my bachelor's [degree], that would allow me to see why this is happening,
why we're getting old, why we're getting different, why we're getting sick. So, I think that was for
me what raised my interest in science. But I think also, back then I didn't know how science really
looked like. So, I didn't know about scientific research and so on. It was more of a dream, right? Like,
(06:39):
let's work on something fun. Yeah, but then, of course, about Mathias's group, that was when I moved
here to Sweden. Of course, I had heard about the Protein Atlas before. We used it in our Bachelor's
and during the Master's [programs]. But then, I think, moving to Sweden was really putting me very close to
(07:00):
his group and to his research. So, you learn a lot during the Master's [graduate program] about the database. And I also
got to know people in the lab. So, that's really how I ended up there. It sounds
like you committed to doing a Master's [degree] first and then that evolved into doing the PhD. How did
(07:23):
they keep the carrot in front of you, that you kept moving through that?
Because I love the stories behind making these decisions. I think students
are hungry for other people's stories about this. Yeah, I think for me, it was moving here to
(07:44):
the SciLifeLab, or starting my Master's [degree] here. I had not connected so much with research and like
academic life at my previous university. Not because of the university, I think it was more of the
like, how bachelor's [programs] are structured in Spain and in many other countries. You don't really get
to feel how it is to be a researcher. But then I moved here to Sweden and I started this program
(08:08):
and they had a really nice mentorship initiative where you could get a PhD student as a mentor.
And for me, that was really what changed my perspective. I was paired with Max
Karlsson of PhD student here at Mathias Uhlén's [lab]. And that's really when I saw, okay, this is
what I want to do. This is what I want to be. That's amazing! I love that story, the mentorship,
(08:33):
and the fact that they structured that with the mentorship. That's amazing. So then this publication
that Sarantis hinted at and that Dale talked about in the intro (08:40):
this is essentially a pretty
remarkable set of data looking at over 1400 patients that had 12 different cancers.
This was our Explore 1536 Olink tool. And so I think 1463 proteins were included in
(09:06):
this study. Tell us a little bit about this study because it's a remarkable paper.
For me, I was already working at this group when they started the planning and
and everything was being run. So of course hearing about the total number of samples for
for the first phase was 10,000. So this is only the the the cancers that we're talking about.
(09:31):
So I was hearing about this and I was like this is unbelievable. I had already worked with Olink data,
but it was only a few panels. So also moving to the Explore [platform], I was really, really interested.
And the data was delivered in very early 2022 or maybe even in
(09:51):
December 2021. And that was exactly when I started my PhD. So you can't imagine how it felt like.
This is, first of all, a Christmas present and then a welcoming present. So yeah,
that felt really, really exciting also because we didn't know anything back then. Like, is there
going to be any signal in these 10,000 samples? So that was really exciting times. And you already had
(10:16):
pretty strong experience in bioinformatics, right? So I imagine it really was.
Exactly yes. Otherwise, it would have been a bit scary maybe. Jump right into the deep
end of the pool. That's right. And so the 10,000 samples, is that
including the U-CAN Biobank? Are all those within the U-CAN Biobank? You also in this study
(10:41):
used the SCAPIS Wellness cohort that's also under the auspices of the SciLife Lab as I understand it,
is that right? Yes, so these 10,000 samples included U-CAN Biobank, which is this Swedish
initiative which collects samples from many different cancers. And that's the study that
(11:02):
is not published. So that's around 1,500 samples in total. And then we also had other cohorts. Like,
you mentioned the SCAPIS, you mentioned Wellness. So this is other Swedish and also non-Swedish cohorts
from other different diseases. We also have infectious data, infectious disease,
autoimmune disease, from Karolinska Institute. So it's really a lot. And of course, we had to
(11:26):
break it down. And that's kind of the first published study, but we have a lot of lines of research open
right now. Can I ask you some technical, I'm sorry if it's very naive question. If I understand
correctly, the controls were from wellness cohorts, from a different cohort in a way.
Do you think that (11:50):
is it easy or how easy it is to just have a control for different cohorts in order to
to have a well-studied project from your perspective.
Yes, so ideally you would always run case-control, right? Like you would have for every single disease
and every single cohort, you would have matched controls. That's usually possible, but then of course,
(12:15):
if you run 10,000 samples, it's a real loss if you just need to spend 50% of that on control. So
we've been trying to include different sets of healthy cohorts, so not only wellness, but we also have,
for example, we talked about the SCAPIS, then we have also health individuals also from Turkey. So
(12:35):
just try to have many different ones so we can have a look at the preanalytical variation
and we can see that everything makes sense across cohorts. But of course, that's always a
limitation, right? In this big studies that you can't ... you have to live with maybe a
couple of healthy cohorts, but not much for a disease. No, and I think also the fact that you have different
(13:00):
type of counts is a kind of control in the way, right? Because you expect to see this classification
in a specific type of cancer than the other that's kind of, you know, there's specific effects
happening in one like the other, which is a kind of control. This is what makes the study really,
really amazing. Can you give a comment? Because when I start reading a little bit ... can you give a
(13:21):
comment about the lung and colorectal cancer? There's something common between these cancers. I
haven't heard about. What is your feeling? And what do you think is happening in these
types of cancer? There are so many commonalities rather than differences. And just for context,
I'll just say those were two of the 12 cancers for which the team characterized the ability
(13:45):
to differentiate among these cancers within this study. So that was the primary focus and the
exciting result, which allowed for leveraging machine learning and characterizing the signature.
So I think these particular cancers were a little challenging perhaps, but also an opportunity
for advancing our understanding. So yeah, Maria, please. You know, I'm happy just about that,
(14:11):
because, as a bioinformatician without a very strong medical background, that was
my first question (14:16):
like, what is this? So that's a shoutout to our doctors that we are very
likely to have and call in anytime we have these kind of questions. So we had very strong discussions
about this because it's kind of the biggest overlap that we saw. So there is an overlap between
the immune cancers. So they have some shared signals, but also these lung and colorectum cancers. And I was
really wondering (14:38):
why could this be? And after meeting with doctors, I mean, of course, we
would have to look farther into these, probably get more like samples with more detail,
like with histology and so on. But their guess is that we are looking at two cancers that are mostly
adenocarcinomas, and it's probably a common signature in the pathway in the development of this
(15:04):
type of cancer. So it was mostly apoptotic signal, and I think also like cellular stress, which is
quite a general thing, but it makes sense that we don't really see that in, for example, immune
cell cancers. For them, it was quite obvious, but I had exactly the same reaction
you had now like, why would that be? So yeah, that's a nice thing about collaborating with the doctors.
(15:29):
It's almost like I pictured as a Venn diagram of these. What
seem like quite disparate diseases, the pathways that are overlapping in these diseases, I think,
are giving us a sense of that mechanistic biology. So I think you're turning yourself into
someone who understands that as well as the bioinformatics like Sarantis said, quite a full package of
(15:51):
someone who can work with these data and their complexity. I mean on that note on
data complexity, if I understand correctly, it was in review for many, many months, right? I think
the pre-print to showed maybe December, January, and it wasn't published until late July.
(16:12):
Can you comment on that in terms of the review process in terms of what you can share? Was it because
the data with the machine learning in terms of those algorithms, was that where a lot of the
review work entailed? Yeah, so I think you understood a little bit of the complexity. So it's really a
(16:33):
huge amount of data and when you're dealing with this kind of data and machine learning,
it's a little bit dangerous sometimes. So it's kind of you need machine learning to understand it,
but then you also, you should be skeptical about it. So I understand 100% why it took maybe a
bit longer than expected because you need to make sure that the pipeline is three years and that
(16:54):
there's enough controls. That it's not noise that you're capturing. Of course,
that we will never know until we replicate or validate our findings in different cohorts
to be sure as much as possible that this is a really well-structured study. And I'm happy
that we went through these revisions so that everything is in place. When I saw
(17:19):
some of the first reactions on social media, namely over LinkedIn, it was interesting, right? One
of the - not criticisms, or may have been a slight criticism - was that these were post-diagnostic
samples. By that I mean these were taken from untreated individuals at the time of diagnosis, these 1400
samples. What can you say about that in terms of could you have gotten pre-diagnostic samples or
(17:44):
what have you for an early detection? I really understand. And I mean, I would call it also a
limitation, right? Of course, you need like many different kinds of studies. I think
this one is also very interesting, but then of course, knowing that these proteins are also up
before diagnostics, that's of course really, really interesting. But in our case, we had access to this
(18:08):
biobank, so we couldn't start now. We would have to wait 10 years, right? To get all these data. So
we still think it's valuable to know what happens at the time of diagnosis.
We also, for example, we had early-stage and late-stage [cancer] samples, so we also think it's relevant to
see that these proteins are also up in early stage. And I can comment maybe that now with all
(18:34):
these UK Biobank, new data that has been released. There was, for example, a pre-print, I think it was
last week, also on different kinds of cancers. And I was, for example, very happy to see that
a lot of our proteins that we found upregulated, they also see seven years prior to diagnostics. So I
(18:55):
think it's valuable and complementary, maybe. So we don't have to wait so many years to see
when we already have samples, but then of course, if we also have that data, that's definitely great.
I think that was one of the most fascinating aspects of your paper in that you found in your
signatures across the 12 cancer types, some well-known markers for well-known cancers have been
(19:19):
researched extensively, and then other new markers that imply new mechanisms. And what can
you comment on that? Yeah, for me, that was the most fascinating. Because if, I mean, it's great when
you find the top marker and that's already been found, because then you're sure that you're doing
things, right? Exactly, but if everything was known, then why am I even working with this, right? So it's
(19:45):
I think it's good to have a combination of really well-known markers, unknown, and maybe some of
my favorite are those that they have been described somewhere in 2015 in a random paper, but no one
has ever looked at those, and they have been forgotten. And then you find it, and you dig in the
(20:06):
literature and you realize, "Wow, this was, this was linked to melanoma many years ago, and I'm seeing
it again. These are my favorites." So here it is. You heard it here first from a bioinformaticist
that she has favorite markers. Big surprise. I do want to just mention that pre-print,
I'm so happy you talked about it. I think that was Ruth Travis and Karl Smith-Byrne and Joshua (Atkins), and
(20:32):
I just have a little tiny story when we were at ESHG, Joshua came up to our booth, and he told us how
excited he was about the UKB data being available that he's been sifting through it. And he was the one,
Dale and Sarantis, I think I told you this story before, who just kept telling me I can't tell you
anything about it, because we're coming out with a pre-print, but Holy crap, Holy crap, Holy crap,
(20:54):
he was so excited about mining those data. And so if he's a representative of the kind of excitement
that a resource like that can generate, I think we're going to have some really fun conversations
in the next few months about what people are seeing there. So it's, yeah, thank you for highlighting
(21:16):
that pre-print. Truly pioneering work, right? There is only one first, and like you mentioned,
you're looking at some random paper in 2015 that had some association with one of those 12
cancer types, and all of a sudden it pops up, right? As being an important, I thought that was very
interesting too, and the analysis where you actually look at the weightings and its influence on the
(21:41):
overall power in terms of that signature. Yeah, and I think something that is also,
for me, a take-home from this study, and we probably haven't talked about it yet, is that
not all cancers have the same amount of proteins that are important, right? So you can have,
(22:02):
maybe you need 20 proteins to characterize a cancer, or you might need 200. And I think that was not
so evident for all of us a few years ago, like we were hoping to find one protein.
Ah, the smoking gun, we want the smoking gun. The one single biomarker, right? It's so much easier
to get clinical utility past the FDA, and they love having one single test, one single HDL, LDL,
(22:28):
you know. But the body's so good at
repurposing all these proteins for all these different tissues. I think it's a
good lesson. I will say I'm just, I love this figure, and having Mathias talk through this figure,
(22:50):
which really characterizes some of the signatures that are showing up stronger in some cancers than
others. And then the connective diagram that you have, sorry, we don't use slides on this
podcast, but I will say that the images are gorgeous that allow us to just talk about what you've done,
(23:11):
and help folks understand the significance of the approach across multiple different diseases,
right? So, and then I think you capture it well around the importance of early cancer detection,
and these tools around genetic detection of cell-free DNA, for example, in blood, they'd suffer
(23:33):
from false positives, and that's a big concern that folks will be given a potential diagnosis
that isn't real because we want to make sure we capture some of these cancers
in early stages that are real. And so being able to add value to that approach, do you have any
(23:59):
thoughts about how this might complement that, how it might be leveraged in combination
with some of these innovative new approaches? Yeah, that's a very, very relevant
comment, and we talk about it a lot here because I mean, when we see the false
positive rates, and it does look so, so bad when you look at the numbers, but then if you think about
(24:23):
a population scale, then that would be a disaster, right? If you could call in so many patients.
And an individual patient experience, too? How that feels, how that would feel to an individual,
exactly what you say. Yeah, and I mean, what we've discussed before is that
it's, of course, really nice with cancers that you can have external means of
(24:46):
getting a validation that the person has the cancer, so instead of just running the blood test,
and saying, "Oh, yeah, you're diagnosed with the cervical cancer." If you could have a way to
verify that, like for example, the mammography or some other test. Then it's kind of
easy to convince the population that this is only a screening, that it's not that sensitive, and then
(25:10):
if you get the false alarm, then you of course get sent back home when you get the negative
second test. But when I think about, for example, cell-free DNA and all the other ways of
finding cancer that are very promising, I think that if we could maybe combine
cell-free DNA testing with proteomics in plasma, that would probably filter out some
(25:36):
of the false positives, some people that would have maybe DNA signal, but are not so strong on the proteomics
side or the other way around. So I think it would be at least two layers where it's more difficult to
be a false positive or, on the other hand, a false negative. Yeah, but so I think maybe, multi-omics,
multi-modal, both words. Yeah, what a lot of people found out at AACR in Orlando was the
(26:02):
readout for the Phase 3 clinical trial for the Galleri GRAIL test, and the fact is it just replicated
the Phase 2 and Phase 1, pivotal trial, or clinical trials, in that there is about a 50% false positive
rate, even with 99% specificity, just because of the prevalence of cancers, about 1% in the general
(26:23):
population of normal risk, and that is a problem, right? But then to think,
the harms that happen where people are saying, well, we have the
signal [that you're positive for cancer], but we're not sure, right? If you have cancer, you have to look for regular diagnosis,
and naturally for colorectal, for lung adenocarcinoma, it's all very straightforward in terms
(26:47):
of diagnosing those. But the Galleri test is looking at 48 other cancer types, and some of them really
do not have gold standard diagnostic methods. And so patients can be sent, right? You might say, well,
it must be terrible to be that person. Yeah, right? Where they're told, you might have cancer,
and we can't really find it using our best diagnostic tools. A real, genuine harm
(27:14):
to individuals, and yet the same time, the value of early detection, right? The incredible power
to prevent death? To catch it in stage one. Yeah. If you know about
this false positive rate and you can't validate that this is cancer, then you're
also not going to treat it, I guess. So, yeah, I mean, for me, I think that in the ideal case
(27:39):
scenario, you would just be able to sample everyone longitudinally through their lives, and I think
then it would be really easy. I'm saying easy, but it would be easier to find differences in your
proteome and like actual pathological states. But now, I mean, you're sampling a person comparing it
to some threshold, some other data that has been collected, there's going to be a lot of noise.
(28:04):
But of course, that's daydreaming, like, what would I like to have? Yeah, for sure. And next
steps, right? Sort of what's in the future? In this study, you were able to do sort of a 70-30 split
to be able to train your machine learning around a set of the samples and then validate with a
(28:25):
subset of those samples. And I think your areas under the curve for across all cancers were like
0.8 to 1, but really, really high in AML, CLL and myeloma, as I wrote down in my notes. I think that
next steps are for validation. Anything you can say about what you all are planning or what you hope
(28:50):
others will do to help validate this so that we understand its power? Yeah, so I mean, ideally, as I said,
it would be nice to just work with these cancers for many, many years and get additional
cohorts and buy data these findings. But our plan for now was to share these data, share
all the protein levels in the database. So for example, I love the preprint that we
(29:15):
just referred to, of course, they found exactly the same proteins in myeloma and that was really,
okay, we haven't validated ourselves, but these people have found it in Sweden, right? So it's not
only up to us to maybe validate these findings, but we also hope that other research groups will
refer to this and realize that they have the same thing in their own data, maybe even with other
(29:38):
methods. So just, it was more to share a list and to share our results and hope that also other people
will do the same. Yeah, and I think this ability to, because I can imagine, you know, sort of these
three to 12 to, however many proteins are needed for each cancer, those tests, but then you did
(30:00):
this beautiful thing, which is combine the proteins that were across all of these into an 83-protein
panel that was, I think - I'll let you describe that. I think it's a great part of the paper that I
appreciated and wanted to highlight. Yeah, so of course, I mean, the most important is to
(30:24):
learn about the proteins that are important for a disease or that was the, the aim, but what we wanted
to see also is if we can go down. So instead of looking at the 1400 proteins, what happens if we
just pick, would we think are the most representative? Do we still get good classifications?
Are we still able to separate all these samples into their specific cancer of origin?
(30:47):
And I think, I mean, of course, this is only the first study. We haven't validated,
but it looks quite nice that only using 80 proteins instead of 1,000, you can really guess what
that sample is. And I mean, in the future that I think would be a really nice way to
look at diseases, to just have a list of markers for not only cancers, but maybe other diseases that
(31:15):
could be related and then just try to see where your sample falls. So based on these markers,
you're most likely a cervix cancer patient. Yeah, I have a few questions. The first question is:
how do you see ... for sure, there are other bases of disease, right? How would you
(31:37):
see this protein signature of 83 correlates with other diseases? I mean, how unique? Because you haven't
done this correlation, how unique are you stuck between cancer, but how unique is compared to other
biological disorders, for example? Yeah, so for that, you will have to wait a little bit.
I mean, as I said, we have many, many different diseases now. I think we have around 88 in
(32:01):
phase one, and we are planning for a second phase. So you expect us to write about this.
I think it's also been highlighted in other pieces of work that there is a few proteins that are found
kind of everywhere. I think GDF-15 is one of them. It's some proteins that are just up, also
(32:22):
immune proteins. There is probably inflammation everywhere, but there's also quite many that are
very, very specific. So I'm always reading Olink papers because it's really fun to see, okay,
this protein that popped up for colorectal is also maybe related to HIV, I don't know. So it's
quite nice. That's so cool. And expanding the protein panels, right, we've just announced the
(32:44):
[Olink Explore] expansion to 5,300 proteins. Sorry, go ahead, Sarantis, you have another question.
No, no, it's perfectly fine. We think the same. I think
about what you just said in terms of proteins that are common
to disease that go beyond just cancer or just autoimmune or just infectious and to think, wow,
(33:10):
to have a certain set of markers that yeah, there's something going on here or yeah, you better
get this checked out. I guess, you could start monitoring these more general
markers since it is probably expected that more specific markers will pop up later on in the disease.
And then once you know something is wrong, you could go deeper and run another panel maybe.
(33:34):
And I just want to comment on this new release of the Explore HT, do you call it?
Yeah, absolutely. I think that's really exciting at least from my side because when, I don't know,
when you work in in biomarker discovery, I don't think a lot of people are used to this really high
(33:56):
number of targets. And for me already 1,500, I felt like that was amazing, right? Like eventually,
you start getting to know some proteins by heart, but it's still a lot of proteins to look at.
But now, I mean, from, of course, you had 3000 in between, but for me, going from 1,500 to
more than 5,000 proteins that we will have in the next phase, I'm really looking forward
(34:21):
for that. That's going to be a really nice, well, it's a really nice list. It's a good chance to
mention that just this week, we've integrated that entire list into our Insight app. So this is
a freely available software that has a web-based interface at insight.olink.com. And you can
browse pathways, see how many of the total proteins in those pathways we have on that HT [platform] and our
(34:46):
other panels. I am in there all the time. I was surprised that I'm a super user of it. So clearly,
it's underutilized because it's relatively a new set of tools that we've provided in the last
year. And the data stories in there include the Cancer Protein Atlas that we're talking
(35:09):
about here today. So folks can browse some of the results that you found in that exciting study,
which by the way, is in Nature Communications. You're a PhD student! Look at you, Maria. Amazing.
Amazing. I mean, on that note, Maria, there is an idea for a Disease Protein Atlas. Is that
(35:31):
correct? Not just cancer, but other diseases. What can you comment on that? So right now,
we have published this story. So that's the data that's available now on the Atlas. It's still called
Human Disease Blood Atlas, although now it's more like a cancer atlas. But the idea is we're going to
keep on releasing data. So I mentioned phase one that has infectious diseases. It has autoimmune
(35:56):
disease and many others that's going to be soon, relatively soon part of the database.
And then of course, we have plans to expand to a phase two and and and have more targets. So that's
also going to be part of the Atlas and hopefully on Olink Insight as well. So that it can reach
as many researchers as possible. That's an amazing resource. Does Mathias [Uhlen] talk about how many
(36:21):
millions of hits and how many thousands of pages of data are freely available? Yeah. I mean, it's
really a unique effort, I'd say. And also, I kind of really like the history of focusing on
proteins and now going a little bit more. We have this health study and now we have this disease study.
So I think it's quite a natural course of how you should look at biology. From the
(36:45):
healthy to the disease and look at those differences. Maria, talking about a little bit biomarkers
and I know that the scope of the proteins is around identifying biomarkers for prognosis.
But for sure, among these biomarkers, there will be targets, right? For drugs. I mean, are you
considering to take a look more to the drug development perspective? I'm sure that you are. But
(37:08):
have you found some really cool new targets for example, that pharma companis could take advantage
of? I mean, if you want to share or is it too early? It's very tough for you. Do you need
more cohorts to validate those drug targets? Yeah, both of those. Great question. So I think
you're both on the right path. We, of course, understand how how valuable this data is
(37:32):
and we are of course starting to see some proteins that look very interesting. But as you both
mentioned, it's early in the sense that you need lots of validation. So we want to be careful with that.
We are of course looking at different markers. We have an eye on them. But it
takes a while before you can say anything about the disease. But crowdsourcing the data
(37:57):
like this, making it publicly available, allowing different folks with different ideas about how to
analyze these data, have them debate over social media as Dale was alluding to, really move things
long quickly in my experience. I think the genetic history is a great testament to that. I think
we're seeing that now too. So I'm excited that you're making these data publicly available
(38:23):
and Mathias's commitment to that throughout his career. Yeah, for me, that's a really important
thing that of course it's great to move forward in your own science. But what you said, for example,
about drug drug development, maybe some people doing their own research, they will see this resource,
they will find okay, this is really a good target and that will inspire them. So I think Mathias's
(38:48):
spirit has always been to share and to inspire and I think that's a great way to do science
at least I'm very happy I'm here. think that's an important point, right? Because
wasn't there a decision made on the on the analysis to focus on the upregulated proteins in the
disease samples? Yes, and that's also been a long debate of course because typically
(39:14):
upregulated proteins are very interesting because it's kind of easy to detect
them when they are high compared to the healthy population, right? But also sometimes like
for example, we saw with AML, with leukemia, that you had this really good marker that was upregulated
and this really good marker that was downregulated and you could think maybe let's take both but
(39:37):
many times they are just correlated because one is the receptor of the other, right? So in a way
sometimes up- and down-regulated proteins, they are still in the same pathway. So maybe it's not
that interesting to focus on both at the same time since they are leading to the same story.
And then the other piece though is that if you can modulate the upregulated-ness
(40:08):
of the diseased protein. There are only a handful of drugs that upregulate proteins. Yeah, I guess when
we talk about it being more classical from the clinical perspective, it's related to exactly that.
It's kind of easier to suppress than it is maybe to just somehow make this protein be produced or
provide. Yeah, I've never thought about that. That's a really interesting point. Yeah, thanks.
(40:31):
So therefore, we have drug targets that you need to lower in order to lower whatever
incidence or whatever mechanism that's driving the cancers. And what I think is fascinating about
cancer biology, after all we've learned in the cancer genomics and transcriptomics,
here it is, you're looking at protein level now. Maybe we're circling back all the way back to the
(40:53):
beginning of the conversation, right? It started with transcriptomics. And wow, once you look at the
proteome, everything is illuminated in a new way. Yeah, that's the real thing, right? I mean,
it is, we are proteins. So it's really, really nice to be able now to look at proteins directly.
And see, also in blood, like you don't even have to go to your pancreas to study pancreatic cancer,
(41:18):
you can hopefully just look at your blood. I never thought of that before. I am protein. Yeah,
we are protein. I think that because you haven't talked to Mathias.
Hard to get on his schedule, but we've got a surrogate for Mathias ight here. So I think this will
be our new new mantra here. We are protein. I think you made an impact, Maria. Every one of us
(41:41):
is protein. That's right. Yeah, what a thought. Yeah, thank you for your generosity with
your time today. We really enjoyed this conversation. Yeah. Thank you for inviting me. It's been really
great to discuss with you. Yeah. And then not just not to forget to thank all of these patients
that consented to allow, you know, now and in the future, all the work that you and others do, I mean,
(42:03):
what a wonderful generous act and without it, we wouldn't be where we are today. For sure.
All right. Thank you. Thank you. Thanks so much.
Thank you for listening to the Proteomics in Proximity podcast brought to you by Olink Proteomics.
(42:27):
To contact the hosts or for further information, simply email info@olink.com.
[Music]
[Music]
[Music]
(00:05):
Welcome to the Proteomics and Proximity Podcast
where co-hosts, Dale Yuzuki, Cindy Lawley and Sarantis Chlamydas from Olink Proteomics
talk about the intersection of proteomics with genomics for drug target discovery,
the application of proteomics to reveal disease biomarkers,
and current trends in using proteomics to unlock biological mechanisms.
(00:27):
Here we have your hosts, Dale, Cindy and Sarantis.
Welcome to Proteomics and Proximity.
I'm Dale Yuzuki, your host, along with Cindy Lawley and Sarantis Chlamydas.
And here we have a special treat today. Sarantis?
Yeah, thank you, thank you, Dale. I'm really happy to be back from holidays
and then having the first host, Maria.
(00:48):
And Maria is a student, she's a PhD student in Mathias Uhlén's lab,
and she had recently a great paper, she was the first author of a great paper that
we'll discuss a little bit later in more detail.
Maria started her studies in Barcelona in Autònoma University of Barcelona,
then she moved to Stockholm [Sweden], and she gained a Master's degree,
(01:11):
joined the Protein Atlas, the Human Protein Atlas project in 2020,
and, since 2022, she's been a part of Mathias Uhlén's lab
where she's working on an amazing project around proteomics in different diseases and cancer
to try to identify signatures, protein signatures.
(01:33):
Maria's background is around, of course, proteomics, transcriptomics,
and she's also a bioinformatician and wet lab scientist.
That means she has a complete package of a full experience and expertise in the field,
also in the field of multiomics - that's something I'm really interested also to hear.
And I will start my question, actually, Maria, I mean, we are really happy to have you,
(01:55):
and it is always a pleasure to discuss with you. I would like to know a little bit more
how you see your transition from transcriptomics to proteomics,
and how you see the match between these two omics approaches from your perspective?
Thank you, Sarantis, it's really great to be here today.
So, I think for me it was a very smooth transition,
(02:18):
basically because at the Human Protein Atlas, that's been kind of a very connected thing.
It's never been like, transcriptomics is one thing and then proteomics is something
completely independent. It's always been quite linked, and that's because, of course, the Human
Protein Atlas has been about understanding proteins, but we have not always been able to just
(02:39):
look at them right away in such a scale, right? I mean, but now with Olink, it seems like that has been
there forever, but in the very early days, transcriptomics was a really, really useful tool
to look at the proteins - indirectly, of course - but we learned a lot and it's
a resource that is still used to this day in many, many groups. So, I was really lucky to work
(03:02):
with transcriptomics first, I think, so that I could understand ... I sometimes say
that that's the simple part where everything is clean and you don't have so much
noise of post-translational modifications and so on. So, I think it's more of the straightforward
tissue transcriptomics analysis. But now, of course, it's really great to go to more toward plasma
(03:24):
proteomics, which is kind of the other extreme. To ask you a little bit about the transcriptomics piece,
there at Science for Life Laboratories, right? And was the interest there
in transcriptomics around a cancer indication, or was it across tissues?
Yes, so the very early transcriptomics, that was published around 2015 in Science, and that was
(03:46):
healthy tissue. So, that was kind of a really big study across different tissues, try to just
characterize which proteins are specific, and that was kind of the flagship of the Protein
Atlas, try to just see what's there. But then, of course, there's been other, more
integration with GTEx data and trying to understand, of course, a little bit more about disease.
(04:08):
And so, this transcriptomic data set was from the tissues from Science for Life Laboratories,
right? Like you mentioned, there are other projects that are characterizing tissues like GTEx.
What made the Science for Life effort unique or different?
So, nowadays, at the Human Protein Atlas (HPA), we have both data from HPA and GTEx integrated
(04:30):
into the database. When I say HPA, I mean Human Protein Atlas, it's just a little bit too long.
So, our own samples, for example, we have a really detailed brain collection of samples, so it's
not only brain, but we have tons of different regions in the brain. So,
for example, we have a ton of detail on that stage that you can't really have with GTEx data.
(04:57):
So, it was more trying to complement, and also, of course, if you have two different independent
sources of data that agree on the same transcriptome levels, that's, of course, very reassuring.
So, that's why we try to combine both in the same database.
Amazing. So, hey, Maria, this is my first chance to ask a question, and I'm actually going to
(05:21):
click back to ask about your interest in getting into science. So usually we say,
you don't have to start with elementary school, but if you want to, I mean, whatever,
you know, sparked this interest in you, I'd love to hear about that. And then
what nurtured that interest to end up in Mathias's lab? I mean, what a remarkable
(05:47):
place to be. Thank you for question. It's, of course, always difficult to know when it all
started, but I think you will probably click with this (05:54):
that
biology is just fascinating. Like, you see people getting old, people getting sick.
I don't know why, I see the effect of epigenetics as well. I have a twin system myself,
so I was always surrounded by what I thought was biology. And then I thought,
(06:15):
studying genetics, which was my bachelor's [degree], that would allow me to see why this is happening,
why we're getting old, why we're getting different, why we're getting sick. So, I think that was for
me what raised my interest in science. But I think also, back then I didn't know how science really
looked like. So, I didn't know about scientific research and so on. It was more of a dream, right? Like,
(06:39):
let's work on something fun. Yeah, but then, of course, about Mathias's group, that was when I moved
here to Sweden. Of course, I had heard about the Protein Atlas before. We used it in our Bachelor's
and during the Master's [programs]. But then, I think, moving to Sweden was really putting me very close to
(07:00):
his group and to his research. So, you learn a lot during the Master's [graduate program] about the database. And I also
got to know people in the lab. So, that's really how I ended up there. It sounds
like you committed to doing a Master's [degree] first and then that evolved into doing the PhD. How did
(07:23):
they keep the carrot in front of you, that you kept moving through that?
Because I love the stories behind making these decisions. I think students
are hungry for other people's stories about this. Yeah, I think for me, it was moving here to
(07:44):
the SciLifeLab, or starting my Master's [degree] here. I had not connected so much with research and like
academic life at my previous university. Not because of the university, I think it was more of the
like, how bachelor's [programs] are structured in Spain and in many other countries. You don't really get
to feel how it is to be a researcher. But then I moved here to Sweden and I started this program
(08:08):
and they had a really nice mentorship initiative where you could get a PhD student as a mentor.
And for me, that was really what changed my perspective. I was paired with Max
Karlsson of PhD student here at Mathias Uhlén's [lab]. And that's really when I saw, okay, this is
what I want to do. This is what I want to be. That's amazing! I love that story, the mentorship,
(08:33):
and the fact that they structured that with the mentorship. That's amazing. So then this publication
that Sarantis hinted at and that Dale talked about in the intro (08:40):
this is essentially a pretty
remarkable set of data looking at over 1400 patients that had 12 different cancers.
This was our Explore 1536 Olink tool. And so I think 1463 proteins were included in
(09:06):
this study. Tell us a little bit about this study because it's a remarkable paper.
For me, I was already working at this group when they started the planning and
and everything was being run. So of course hearing about the total number of samples for
for the first phase was 10,000. So this is only the the the cancers that we're talking about.
(09:31):
So I was hearing about this and I was like this is unbelievable. I had already worked with Olink data,
but it was only a few panels. So also moving to the Explore [platform], I was really, really interested.
And the data was delivered in very early 2022 or maybe even in
(09:51):
December 2021. And that was exactly when I started my PhD. So you can't imagine how it felt like.
This is, first of all, a Christmas present and then a welcoming present. So yeah,
that felt really, really exciting also because we didn't know anything back then. Like, is there
going to be any signal in these 10,000 samples? So that was really exciting times. And you already had
(10:16):
pretty strong experience in bioinformatics, right? So I imagine it really was.
Exactly yes. Otherwise, it would have been a bit scary maybe. Jump right into the deep
end of the pool. That's right. And so the 10,000 samples, is that
including the U-CAN Biobank? Are all those within the U-CAN Biobank? You also in this study
(10:41):
used the SCAPIS Wellness cohort that's also under the auspices of the SciLife Lab as I understand it,
is that right? Yes, so these 10,000 samples included U-CAN Biobank, which is this Swedish
initiative which collects samples from many different cancers. And that's the study that
(11:02):
is not published. So that's around 1,500 samples in total. And then we also had other cohorts. Like,
you mentioned the SCAPIS, you mentioned Wellness. So this is other Swedish and also non-Swedish cohorts
from other different diseases. We also have infectious data, infectious disease,
autoimmune disease, from Karolinska Institute. So it's really a lot. And of course, we had to
(11:26):
break it down. And that's kind of the first published study, but we have a lot of lines of research open
right now. Can I ask you some technical, I'm sorry if it's very naive question. If I understand
correctly, the controls were from wellness cohorts, from a different cohort in a way.
Do you think that (11:50):
is it easy or how easy it is to just have a control for different cohorts in order to
to have a well-studied project from your perspective.
Yes, so ideally you would always run case-control, right? Like you would have for every single disease
and every single cohort, you would have matched controls. That's usually possible, but then of course,
(12:15):
if you run 10,000 samples, it's a real loss if you just need to spend 50% of that on control. So
we've been trying to include different sets of healthy cohorts, so not only wellness, but we also have,
for example, we talked about the SCAPIS, then we have also health individuals also from Turkey. So
(12:35):
just try to have many different ones so we can have a look at the preanalytical variation
and we can see that everything makes sense across cohorts. But of course, that's always a
limitation, right? In this big studies that you can't ... you have to live with maybe a
couple of healthy cohorts, but not much for a disease. No, and I think also the fact that you have different
(13:00):
type of counts is a kind of control in the way, right? Because you expect to see this classification
in a specific type of cancer than the other that's kind of, you know, there's specific effects
happening in one like the other, which is a kind of control. This is what makes the study really,
really amazing. Can you give a comment? Because when I start reading a little bit ... can you give a
(13:21):
comment about the lung and colorectal cancer? There's something common between these cancers. I
haven't heard about. What is your feeling? And what do you think is happening in these
types of cancer? There are so many commonalities rather than differences. And just for context,
I'll just say those were two of the 12 cancers for which the team characterized the ability
(13:45):
to differentiate among these cancers within this study. So that was the primary focus and the
exciting result, which allowed for leveraging machine learning and characterizing the signature.
So I think these particular cancers were a little challenging perhaps, but also an opportunity
for advancing our understanding. So yeah, Maria, please. You know, I'm happy just about that,
(14:11):
because, as a bioinformatician without a very strong medical background, that was
my first question (14:16):
like, what is this? So that's a shoutout to our doctors that we are very
likely to have and call in anytime we have these kind of questions. So we had very strong discussions
about this because it's kind of the biggest overlap that we saw. So there is an overlap between
the immune cancers. So they have some shared signals, but also these lung and colorectum cancers. And I was
really wondering (14:38):
why could this be? And after meeting with doctors, I mean, of course, we
would have to look farther into these, probably get more like samples with more detail,
like with histology and so on. But their guess is that we are looking at two cancers that are mostly
adenocarcinomas, and it's probably a common signature in the pathway in the development of this
(15:04):
type of cancer. So it was mostly apoptotic signal, and I think also like cellular stress, which is
quite a general thing, but it makes sense that we don't really see that in, for example, immune
cell cancers. For them, it was quite obvious, but I had exactly the same reaction
you had now like, why would that be? So yeah, that's a nice thing about collaborating with the doctors.
(15:29):
It's almost like I pictured as a Venn diagram of these. What
seem like quite disparate diseases, the pathways that are overlapping in these diseases, I think,
are giving us a sense of that mechanistic biology. So I think you're turning yourself into
someone who understands that as well as the bioinformatics like Sarantis said, quite a full package of
(15:51):
someone who can work with these data and their complexity. I mean on that note on
data complexity, if I understand correctly, it was in review for many, many months, right? I think
the pre-print to showed maybe December, January, and it wasn't published until late July.
(16:12):
Can you comment on that in terms of the review process in terms of what you can share? Was it because
the data with the machine learning in terms of those algorithms, was that where a lot of the
review work entailed? Yeah, so I think you understood a little bit of the complexity. So it's really a
(16:33):
huge amount of data and when you're dealing with this kind of data and machine learning,
it's a little bit dangerous sometimes. So it's kind of you need machine learning to understand it,
but then you also, you should be skeptical about it. So I understand 100% why it took maybe a
bit longer than expected because you need to make sure that the pipeline is three years and that
(16:54):
there's enough controls. That it's not noise that you're capturing. Of course,
that we will never know until we replicate or validate our findings in different cohorts
to be sure as much as possible that this is a really well-structured study. And I'm happy
that we went through these revisions so that everything is in place. When I saw
(17:19):
some of the first reactions on social media, namely over LinkedIn, it was interesting, right? One
of the - not criticisms, or may have been a slight criticism - was that these were post-diagnostic
samples. By that I mean these were taken from untreated individuals at the time of diagnosis, these 1400
samples. What can you say about that in terms of could you have gotten pre-diagnostic samples or
(17:44):
what have you for an early detection? I really understand. And I mean, I would call it also a
limitation, right? Of course, you need like many different kinds of studies. I think
this one is also very interesting, but then of course, knowing that these proteins are also up
before diagnostics, that's of course really, really interesting. But in our case, we had access to this
(18:08):
biobank, so we couldn't start now. We would have to wait 10 years, right? To get all these data. So
we still think it's valuable to know what happens at the time of diagnosis.
We also, for example, we had early-stage and late-stage [cancer] samples, so we also think it's relevant to
see that these proteins are also up in early stage. And I can comment maybe that now with all
(18:34):
these UK Biobank, new data that has been released. There was, for example, a pre-print, I think it was
last week, also on different kinds of cancers. And I was, for example, very happy to see that
a lot of our proteins that we found upregulated, they also see seven years prior to diagnostics. So I
(18:55):
think it's valuable and complementary, maybe. So we don't have to wait so many years to see
when we already have samples, but then of course, if we also have that data, that's definitely great.
I think that was one of the most fascinating aspects of your paper in that you found in your
signatures across the 12 cancer types, some well-known markers for well-known cancers have been
(19:19):
researched extensively, and then other new markers that imply new mechanisms. And what can
you comment on that? Yeah, for me, that was the most fascinating. Because if, I mean, it's great when
you find the top marker and that's already been found, because then you're sure that you're doing
things, right? Exactly, but if everything was known, then why am I even working with this, right? So it's
(19:45):
I think it's good to have a combination of really well-known markers, unknown, and maybe some of
my favorite are those that they have been described somewhere in 2015 in a random paper, but no one
has ever looked at those, and they have been forgotten. And then you find it, and you dig in the
(20:06):
literature and you realize, "Wow, this was, this was linked to melanoma many years ago, and I'm seeing
it again. These are my favorites." So here it is. You heard it here first from a bioinformaticist
that she has favorite markers. Big surprise. I do want to just mention that pre-print,
I'm so happy you talked about it. I think that was Ruth Travis and Karl Smith-Byrne and Joshua (Atkins), and
(20:32):
I just have a little tiny story when we were at ESHG, Joshua came up to our booth, and he told us how
excited he was about the UKB data being available that he's been sifting through it. And he was the one,
Dale and Sarantis, I think I told you this story before, who just kept telling me I can't tell you
anything about it, because we're coming out with a pre-print, but Holy crap, Holy crap, Holy crap,
(20:54):
he was so excited about mining those data. And so if he's a representative of the kind of excitement
that a resource like that can generate, I think we're going to have some really fun conversations
in the next few months about what people are seeing there. So it's, yeah, thank you for highlighting
(21:16):
that pre-print. Truly pioneering work, right? There is only one first, and like you mentioned,
you're looking at some random paper in 2015 that had some association with one of those 12
cancer types, and all of a sudden it pops up, right? As being an important, I thought that was very
interesting too, and the analysis where you actually look at the weightings and its influence on the
(21:41):
overall power in terms of that signature. Yeah, and I think something that is also,
for me, a take-home from this study, and we probably haven't talked about it yet, is that
not all cancers have the same amount of proteins that are important, right? So you can have,
(22:02):
maybe you need 20 proteins to characterize a cancer, or you might need 200. And I think that was not
so evident for all of us a few years ago, like we were hoping to find one protein.
Ah, the smoking gun, we want the smoking gun. The one single biomarker, right? It's so much easier
to get clinical utility past the FDA, and they love having one single test, one single HDL, LDL,
(22:28):
you know. But the body's so good at
repurposing all these proteins for all these different tissues. I think it's a
good lesson. I will say I'm just, I love this figure, and having Mathias talk through this figure,
(22:50):
which really characterizes some of the signatures that are showing up stronger in some cancers than
others. And then the connective diagram that you have, sorry, we don't use slides on this
podcast, but I will say that the images are gorgeous that allow us to just talk about what you've done,
(23:11):
and help folks understand the significance of the approach across multiple different diseases,
right? So, and then I think you capture it well around the importance of early cancer detection,
and these tools around genetic detection of cell-free DNA, for example, in blood, they'd suffer
(23:33):
from false positives, and that's a big concern that folks will be given a potential diagnosis
that isn't real because we want to make sure we capture some of these cancers
in early stages that are real. And so being able to add value to that approach, do you have any
(23:59):
thoughts about how this might complement that, how it might be leveraged in combination
with some of these innovative new approaches? Yeah, that's a very, very relevant
comment, and we talk about it a lot here because I mean, when we see the false
positive rates, and it does look so, so bad when you look at the numbers, but then if you think about
(24:23):
a population scale, then that would be a disaster, right? If you could call in so many patients.
And an individual patient experience, too? How that feels, how that would feel to an individual,
exactly what you say. Yeah, and I mean, what we've discussed before is that
it's, of course, really nice with cancers that you can have external means of
(24:46):
getting a validation that the person has the cancer, so instead of just running the blood test,
and saying, "Oh, yeah, you're diagnosed with the cervical cancer." If you could have a way to
verify that, like for example, the mammography or some other test. Then it's kind of
easy to convince the population that this is only a screening, that it's not that sensitive, and then
(25:10):
if you get the false alarm, then you of course get sent back home when you get the negative
second test. But when I think about, for example, cell-free DNA and all the other ways of
finding cancer that are very promising, I think that if we could maybe combine
cell-free DNA testing with proteomics in plasma, that would probably filter out some
(25:36):
of the false positives, some people that would have maybe DNA signal, but are not so strong on the proteomics
side or the other way around. So I think it would be at least two layers where it's more difficult to
be a false positive or, on the other hand, a false negative. Yeah, but so I think maybe, multi-omics,
multi-modal, both words. Yeah, what a lot of people found out at AACR in Orlando was the
(26:02):
readout for the Phase 3 clinical trial for the Galleri GRAIL test, and the fact is it just replicated
the Phase 2 and Phase 1, pivotal trial, or clinical trials, in that there is about a 50% false positive
rate, even with 99% specificity, just because of the prevalence of cancers, about 1% in the general
(26:23):
population of normal risk, and that is a problem, right? But then to think,
the harms that happen where people are saying, well, we have the
signal [that you're positive for cancer], but we're not sure, right? If you have cancer, you have to look for regular diagnosis,
and naturally for colorectal, for lung adenocarcinoma, it's all very straightforward in terms
(26:47):
of diagnosing those. But the Galleri test is looking at 48 other cancer types, and some of them really
do not have gold standard diagnostic methods. And so patients can be sent, right? You might say, well,
it must be terrible to be that person. Yeah, right? Where they're told, you might have cancer,
and we can't really find it using our best diagnostic tools. A real, genuine harm
(27:14):
to individuals, and yet the same time, the value of early detection, right? The incredible power
to prevent death? To catch it in stage one. Yeah. If you know about
this false positive rate and you can't validate that this is cancer, then you're
also not going to treat it, I guess. So, yeah, I mean, for me, I think that in the ideal case
(27:39):
scenario, you would just be able to sample everyone longitudinally through their lives, and I think
then it would be really easy. I'm saying easy, but it would be easier to find differences in your
proteome and like actual pathological states. But now, I mean, you're sampling a person comparing it
to some threshold, some other data that has been collected, there's going to be a lot of noise.
(28:04):
But of course, that's daydreaming, like, what would I like to have? Yeah, for sure. And next
steps, right? Sort of what's in the future? In this study, you were able to do sort of a 70-30 split
to be able to train your machine learning around a set of the samples and then validate with a
(28:25):
subset of those samples. And I think your areas under the curve for across all cancers were like
0.8 to 1, but really, really high in AML, CLL and myeloma, as I wrote down in my notes. I think that
next steps are for validation. Anything you can say about what you all are planning or what you hope
(28:50):
others will do to help validate this so that we understand its power? Yeah, so I mean, ideally, as I said,
it would be nice to just work with these cancers for many, many years and get additional
cohorts and buy data these findings. But our plan for now was to share these data, share
all the protein levels in the database. So for example, I love the preprint that we
(29:15):
just referred to, of course, they found exactly the same proteins in myeloma and that was really,
okay, we haven't validated ourselves, but these people have found it in Sweden, right? So it's not
only up to us to maybe validate these findings, but we also hope that other research groups will
refer to this and realize that they have the same thing in their own data, maybe even with other
(29:38):
methods. So just, it was more to share a list and to share our results and hope that also other people
will do the same. Yeah, and I think this ability to, because I can imagine, you know, sort of these
three to 12 to, however many proteins are needed for each cancer, those tests, but then you did
(30:00):
this beautiful thing, which is combine the proteins that were across all of these into an 83-protein
panel that was, I think - I'll let you describe that. I think it's a great part of the paper that I
appreciated and wanted to highlight. Yeah, so of course, I mean, the most important is to
(30:24):
learn about the proteins that are important for a disease or that was the, the aim, but what we wanted
to see also is if we can go down. So instead of looking at the 1400 proteins, what happens if we
just pick, would we think are the most representative? Do we still get good classifications?
Are we still able to separate all these samples into their specific cancer of origin?
(30:47):
And I think, I mean, of course, this is only the first study. We haven't validated,
but it looks quite nice that only using 80 proteins instead of 1,000, you can really guess what
that sample is. And I mean, in the future that I think would be a really nice way to
look at diseases, to just have a list of markers for not only cancers, but maybe other diseases that
(31:15):
could be related and then just try to see where your sample falls. So based on these markers,
you're most likely a cervix cancer patient. Yeah, I have a few questions. The first question is:
how do you see ... for sure, there are other bases of disease, right? How would you
(31:37):
see this protein signature of 83 correlates with other diseases? I mean, how unique? Because you haven't
done this correlation, how unique are you stuck between cancer, but how unique is compared to other
biological disorders, for example? Yeah, so for that, you will have to wait a little bit.
I mean, as I said, we have many, many different diseases now. I think we have around 88 in
(32:01):
phase one, and we are planning for a second phase. So you expect us to write about this.
I think it's also been highlighted in other pieces of work that there is a few proteins that are found
kind of everywhere. I think GDF-15 is one of them. It's some proteins that are just up, also
(32:22):
immune proteins. There is probably inflammation everywhere, but there's also quite many that are
very, very specific. So I'm always reading Olink papers because it's really fun to see, okay,
this protein that popped up for colorectal is also maybe related to HIV, I don't know. So it's
quite nice. That's so cool. And expanding the protein panels, right, we've just announced the
(32:44):
[Olink Explore] expansion to 5,300 proteins. Sorry, go ahead, Sarantis, you have another question.
No, no, it's perfectly fine. We think the same. I think
about what you just said in terms of proteins that are common
to disease that go beyond just cancer or just autoimmune or just infectious and to think, wow,
(33:10):
to have a certain set of markers that yeah, there's something going on here or yeah, you better
get this checked out. I guess, you could start monitoring these more general
markers since it is probably expected that more specific markers will pop up later on in the disease.
And then once you know something is wrong, you could go deeper and run another panel maybe.
(33:34):
And I just want to comment on this new release of the Explore HT, do you call it?
Yeah, absolutely. I think that's really exciting at least from my side because when, I don't know,
when you work in in biomarker discovery, I don't think a lot of people are used to this really high
(33:56):
number of targets. And for me already 1,500, I felt like that was amazing, right? Like eventually,
you start getting to know some proteins by heart, but it's still a lot of proteins to look at.
But now, I mean, from, of course, you had 3000 in between, but for me, going from 1,500 to
more than 5,000 proteins that we will have in the next phase, I'm really looking forward
(34:21):
for that. That's going to be a really nice, well, it's a really nice list. It's a good chance to
mention that just this week, we've integrated that entire list into our Insight app. So this is
a freely available software that has a web-based interface at insight.olink.com. And you can
browse pathways, see how many of the total proteins in those pathways we have on that HT [platform] and our
(34:46):
other panels. I am in there all the time. I was surprised that I'm a super user of it. So clearly,
it's underutilized because it's relatively a new set of tools that we've provided in the last
year. And the data stories in there include the Cancer Protein Atlas that we're talking
(35:09):
about here today. So folks can browse some of the results that you found in that exciting study,
which by the way, is in Nature Communications. You're a PhD student! Look at you, Maria. Amazing.
Amazing. I mean, on that note, Maria, there is an idea for a Disease Protein Atlas. Is that
(35:31):
correct? Not just cancer, but other diseases. What can you comment on that? So right now,
we have published this story. So that's the data that's available now on the Atlas. It's still called
Human Disease Blood Atlas, although now it's more like a cancer atlas. But the idea is we're going to
keep on releasing data. So I mentioned phase one that has infectious diseases. It has autoimmune
(35:56):
disease and many others that's going to be soon, relatively soon part of the database.
And then of course, we have plans to expand to a phase two and and and have more targets. So that's
also going to be part of the Atlas and hopefully on Olink Insight as well. So that it can reach
as many researchers as possible. That's an amazing resource. Does Mathias [Uhlen] talk about how many
(36:21):
millions of hits and how many thousands of pages of data are freely available? Yeah. I mean, it's
really a unique effort, I'd say. And also, I kind of really like the history of focusing on
proteins and now going a little bit more. We have this health study and now we have this disease study.
So I think it's quite a natural course of how you should look at biology. From the
(36:45):
healthy to the disease and look at those differences. Maria, talking about a little bit biomarkers
and I know that the scope of the proteins is around identifying biomarkers for prognosis.
But for sure, among these biomarkers, there will be targets, right? For drugs. I mean, are you
considering to take a look more to the drug development perspective? I'm sure that you are. But
(37:08):
have you found some really cool new targets for example, that pharma companis could take advantage
of? I mean, if you want to share or is it too early? It's very tough for you. Do you need
more cohorts to validate those drug targets? Yeah, both of those. Great question. So I think
you're both on the right path. We, of course, understand how how valuable this data is
(37:32):
and we are of course starting to see some proteins that look very interesting. But as you both
mentioned, it's early in the sense that you need lots of validation. So we want to be careful with that.
We are of course looking at different markers. We have an eye on them. But it
takes a while before you can say anything about the disease. But crowdsourcing the data
(37:57):
like this, making it publicly available, allowing different folks with different ideas about how to
analyze these data, have them debate over social media as Dale was alluding to, really move things
long quickly in my experience. I think the genetic history is a great testament to that. I think
we're seeing that now too. So I'm excited that you're making these data publicly available
(38:23):
and Mathias's commitment to that throughout his career. Yeah, for me, that's a really important
thing that of course it's great to move forward in your own science. But what you said, for example,
about drug drug development, maybe some people doing their own research, they will see this resource,
they will find okay, this is really a good target and that will inspire them. So I think Mathias's
(38:48):
spirit has always been to share and to inspire and I think that's a great way to do science
at least I'm very happy I'm here. think that's an important point, right? Because
wasn't there a decision made on the on the analysis to focus on the upregulated proteins in the
disease samples? Yes, and that's also been a long debate of course because typically
(39:14):
upregulated proteins are very interesting because it's kind of easy to detect
them when they are high compared to the healthy population, right? But also sometimes like
for example, we saw with AML, with leukemia, that you had this really good marker that was upregulated
and this really good marker that was downregulated and you could think maybe let's take both but
(39:37):
many times they are just correlated because one is the receptor of the other, right? So in a way
sometimes up- and down-regulated proteins, they are still in the same pathway. So maybe it's not
that interesting to focus on both at the same time since they are leading to the same story.
And then the other piece though is that if you can modulate the upregulated-ness
(40:08):
of the diseased protein. There are only a handful of drugs that upregulate proteins. Yeah, I guess when
we talk about it being more classical from the clinical perspective, it's related to exactly that.
It's kind of easier to suppress than it is maybe to just somehow make this protein be produced or
provide. Yeah, I've never thought about that. That's a really interesting point. Yeah, thanks.
(40:31):
So therefore, we have drug targets that you need to lower in order to lower whatever
incidence or whatever mechanism that's driving the cancers. And what I think is fascinating about
cancer biology, after all we've learned in the cancer genomics and transcriptomics,
here it is, you're looking at protein level now. Maybe we're circling back all the way back to the
(40:53):
beginning of the conversation, right? It started with transcriptomics. And wow, once you look at the
proteome, everything is illuminated in a new way. Yeah, that's the real thing, right? I mean,
it is, we are proteins. So it's really, really nice to be able now to look at proteins directly.
And see, also in blood, like you don't even have to go to your pancreas to study pancreatic cancer,
(41:18):
you can hopefully just look at your blood. I never thought of that before. I am protein. Yeah,
we are protein. I think that because you haven't talked to Mathias.
Hard to get on his schedule, but we've got a surrogate for Mathias ight here. So I think this will
be our new new mantra here. We are protein. I think you made an impact, Maria. Every one of us
(41:41):
is protein. That's right. Yeah, what a thought. Yeah, thank you for your generosity with
your time today. We really enjoyed this conversation. Yeah. Thank you for inviting me. It's been really
great to discuss with you. Yeah. And then not just not to forget to thank all of these patients
that consented to allow, you know, now and in the future, all the work that you and others do, I mean,
(42:03):
what a wonderful generous act and without it, we wouldn't be where we are today. For sure.
All right. Thank you. Thank you. Thanks so much.
Thank you for listening to the Proteomics in Proximity podcast brought to you by Olink Proteomics.
(42:27):
To contact the hosts or for further information, simply email info@olink.com.
[Music]
[Music]
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.