A discussion of a recent Scientific Advances paper by Katz et al comparing aptamer- and antibody-based protein detection methods using genetics and clinical traits, just in time for the American Society for Human Genetics conference (#ASHG22) in Los Angeles.
Proteomics in Proximity discusses 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. Co-hosted by Olink's Dale Yuzuki, Cindy Lawley and Sarantis Chlamydas.
Welcome to the
Proteomics in Proximity podcast, where
your co hosts, Dale Yuzuki, Cindy
Lawley and Sarantis Clamydas from Olink
Proteomics talk about the intersection of
proteomics with genomics for drug target
discovery to the application of
proteomics to reveal disease
biomarkers and current trends in using
proteomics to unlock
biological mechanisms. Here we have
your hosts Dale, Cindy, and
Sarantis.
Welcome to the Proteomics and
Proximity podcast. This is your host,
Dale Yuzuki, along with
Cindy Lawley and
Sarantis Clamydas.
We want to welcome you today. We've got
some pretty exciting news,
and by news, I mean a recent paper.
And Cindy, would you like to introduce it?
Sure. So this is a, uh, paper that came
from Daniel Katz and Rob Gertszten and a
series, uh, of co authors as well.
It's in Science Advances, so
very, uh, prestigious
journal. And we've known
that the work had been done for
quite a while and we knew that it was going
through peer review. And so we've just been
very excited to see it come out and
we're excited to talk about it today.
Uh, go ahead.
Yeah, sure. It's a comparison paper among
proteomic profiling platforms.
It looks at antibody, uh,
methods versus aptamer
methods. The pros and cons
pluses and minuses. I think
that, uh, both platforms are
incredibly valuable. So the two
platforms that were compared to
the SomaLogic platform, their
also their 5K.
Uh, and the OLink platform, this was
our previous product, the Explorer
which uh, they called the Olink 1.5K
So yeah. Dale, do
you want to give us a little bit of
background on Soma?
Sure, happy to. I was
a protein product manager for QIAGEN back in
the early 2000's. And
this was maybe
Protein Science Conference and I
met a very interesting individual named
Larry Gold. He was the
founder of SomaLogic, a
company in Colorado. He's a
genius, really. And he says, yeah, he's a
genius. A uh, remarkable individual.
He made quite an impression on me because he
was working on a very unusual
approach to protein detection. Instead of
using, uh,
recombinant antigens and
inoculating
goat or mouse or rat to
develop antibodies right. Or develop
hybridomas or
monoclonals, he actually
was using a method of in vitro
selection.
it's in organisms like
microorganisms, a series of
selective processes,
to find out particular
synthetic nucleotides,
either RNA or DNA, that would
bind very specifically to
proteins. This is called aptamers. These
are synthetic stretches
of DNA and RNA
that bind to proteins, you might say. Well,
we have DNA binding proteins,
RNA binding proteins all the time. Well,
yeah, transcription factors, what have you.
But these are proteins that have been
evolved to bind to DNA sequence. Now
we're doing the reverse. We're actually
finding out what sequence can
bind to a particular protein
that doesn't normally bind to
DNA. Well, they've evolved,
the technology evolved,
uh, not only
to use natural
oligonucleotides, they actually have,
uh, developed something called Somamers,
which are unnatural
nucleotides, uh, because if
you think about the positive charges
of DNA, etc, etc, etc, etc. All
you're going to have certain limitations in
terms of how the DNA and RNA can
fold, using natural
nucleotides. So they've
developed some chemicals they call
Somamers to
expand the kind of,
uh, synthetic aptamers,
what it's called, to bind
to more and more proteins. So
they've scaled the technology in
earlier. They published some really high
profile papers. And what these papers
did was look
at 1300 different
proteins out of the circulation
and connect it to the genomics. So what
they would do is, frankly, since you
had a million genotypes from,
um, an Infinium array,
a genotyping array, and you have all these
genotypes from individuals, they can do
GWAS to protein level using
the SomaLogic, um,
readout. Again, they're doing
GWAS to the circulating protein
level and then connect that to
phenotypes. And the
first high-impact papers in Nature and
in Science were pretty remarkable
because you're talking about genomics
and you're tying it in to the
circulating proteome, which you then can
tie into disease. Yeah. And
I'll ask, I'll also add
is, you know prevailing
technologies and mass spectrometry. Of
course, looking at proteomics,
um, amazing advances there
and a lot of,
um, transition to the clinic, some of the
discoveries there. The idea of
being able to, um,
hook out of, um,
a plasma sample, for example,
uh, to hook with an
affinity-based method like a
Somamer allows uh, you to
do the low abundant proteins, the ones that
just may not show up very often or in very
high abundance in plasma
allows you to start to look
at patterns of
those proteins as well with
health and disease. And so I think this
is amazing innovation
is that in Mass Spec, of
course you can do this, but it takes
a lot more sample. And doing it in large
numbers of samples can be challenging simply
because of, uh, what it takes
to put a service wrapper around
running many samples through a mass
spec. Uh, I think you bring up a really good
point, Cindy, which is people just
say, well, why can't you just use mass
spec? It's a mature technology, it's
been around a long time. HDL, LDL,
we've got a lot of great clinical
labs from it. Exactly.
But it's the level of
abundance, right, in that there's a
number of really important
molecules that aren't
very prevalent in the
circulation that both, uh, the
Somamer technology and the Olink technology
can pick up that Mass Spec
simply cannot. You also bring up the other
point, which is Mass Spec,
uh, has a lot of, uh, sort of
upfront steps, right? If you're doing
liquid chromatography, tandem
mass spec, there's a whole bunch of
sequential... steps uh,
you have to do that is just not high
throughput is that correct? Yes. And
managing the
variability. Yeah. Sarantis? Sorry. Go
ahead, please. No, I'm sorry, I was talking
about Mass Spec and plasma. We know
that in plasma there
are like 40 or 50 proteins that are super
abundant. And when you try to run a
Mass Spec, you let's say mask
all the other proteins. And for this is for
uh, low abundant proteins. I think affinity,
capture assays,
like Olink assays, can
help to identify this, because
you overcome this problem with Mass Spec.
It's a nice complementary
approach, obviously. Yeah. Sarantis,
it would be great if people could get kind
of an overview of how Olink is different
from the SomaLogic approach, because of
the two platforms being compared in this
paper. Would you mind tackling
that? Yeah, I mean,
in this paper, actually, they have
uh, used Olink Explore
our first Explore based on the
NGS platform. And
SomaScan
That's the most expanded version, with more,
let's say, reagents. And they
profile Jackson Heart study,
around
500 people, and
Heritage Family Study
Uh, in the first case, when they
see, they try to see overlap in between.
SomaScan 1.3K and Olink they
see like roughly 500
proteins being overlapped. When uh,
they switch to the
expanded version of SomaScan, they were
able to see like
uh, actually
proteins. The nice advantage of Olink,
of course, is uh, NGS based
approach and uh, the
antibody capturing that gives
obviously more
specificity compared to
others.
But... yes, please. So if I
understand correctly, you mentioned the
Jackson Heart Study of what, some
Yeah, uh, 500
individuals, yes. And I have to say
here, for all of these
individuals, they have like
whole genome sequencing data.
Yeah. Mmm hmm, so really nice genetic
background information. As far
as a comparison study.
Right. Actually, uh, before we talk about
the study itself, Cindy, did you do some
research on Jackson Heart? Can you tell me a
little bit more about that? Yeah, so Jackson
Heart Study is a community-based, it's
a longitudinal cohort study. So uh,
essentially they're looking at
understanding, cardiovascular disease
primarily, but also renal respiratory
diseases. The nice thing. And what
I love about the Jackson Heart study is
that it's African Americans.
So, it's really
helping us get a view into
proteomic variability, not just
within uh, the northern European
populations that have been characterized so
well, genetically as well as now,
um, quite a few of them have done
proteomically, like the UK
Biobank. But um, it allows us to get
some sense of the diversity in
African Diaspora.
So... the
Jackson and Jackson Heart study, then, is
referring to Jackson, Mississippi? Is
that right? That's right, exactly.
The majority of people then are from
Mississippi area. That's right, yeah.
And when you say longitudinal, are you
meaning that what they're followed up
over time? Exactly.
So they were
recruited, and
then followed,
over time. The nice uh,
thing about that is as people
evolve, as they
get older and they have health
challenges, those health challenges
can be better understood by
looking back at samples
before they had diagnosis of
disease. And that's going to help
us develop more preventative
approaches... for
diseases. Today, our healthcare
system is a
diagnosis-based system,
not uh, only within the US.
But also worldwide. Really, the
funding for healthcare,
revolves
around diagnoses. And
so this concept or this ability, and
I think I would argue this
might be, um, one of Larry Gold's
big motivations in
developing uh, of the SomaLogic technology.
I think we're really excited about this for
Olink as well, is the ability
to be more preventative and
understand risk, not only
from the genetic perspective
that has been enabled over the last 20
years, but also from the proteomics
perspective. And so understanding proteomic
risk at any given moment,
um, appears to be
providing a little bit more of a window into
more real-time health. And so I think that's
the important aspect of having longitudinal
data like this, especially in this
underrepresented population. So this
is a pretty expensive undertaking because
we're talking about whole genome sequence
out of these 568
individuals. Right? And then you're also
talking about,
Olink 1.5,
Olink Explore 1536 plus,
SomaLogic 1.3K on all
the same samples. Do I understand
that correctly? That's right.
And as far as
then, what can you tell me about the
Heritage study? Yeah. So this is
health risk factors and
exercise training and genetics. So the
HERITAGE stands for
literally, that: HEalth RIsk factors.
exercise, Training And GEnetics.
And it's, ah,
um, a
partnership among I think it's seven
universities, actually. I don't have those
seven universities. Maybe it's five
universities. Sorry. Yeah, I think it's
five universities and I don't have them
off the top of my head, but they're across
the US. As well as Canada. So,
um, really trying
to get information
across a large segment of
the population and
again, whole genome sequencing
information. Is that right, Sarantis?
That there was whole genome sequencing
within the Heritage as well? Yeah,
which like you said,
Dale, it blows my mind because I think
about the exome
um, sequencing consortium from the UK
Biobank, which is a massive undertaking,
but still exome sequencing is only
about 1 to 3% of a whole genome
sequence. So we're talking a lot
of sequencing to your point. And this is
really important, especially for health
equity, because we really have an
underrepresentation of African
Diaspora in um,
in sequencing data and
you know being biased
by a chip that might not have a lot
of content that was
designed, a genotyping chip, I'm thinking
about the comparison between a genotyping
chip versus whole genome, um, sequencing.
It's sort of like getting a
satellite view of a
population census.
You can take a picture from the satellite
and you can estimate the representation
in those,
those buildings that are in an eight
block area. For example. Or you can go
house to house and you can knock on the
doors of every one of those
residences. Which one is going to give you a
more accurate representation of the
population? The knocking on the doors.
But... it's going to be a lot more
expensive as well. Right. So whole
genome sequencing goes base to base
to the extent that our sequencing methods
allow that. And we're going to see diversity
that we might not know about
before... when we developed that
genotyping chip initially. Yeah, that's such
an excellent point because you just assume
all, you just get the genotypes and then
you capture the majority of the
variation. But what you're saying is
yes, the whole exome sequencing doesn't
capture a fraction of
that variation because right. These SNPs
are in non coding regions.
That's right.
And to be able to get them the whole
genome sequence, we can get a very
fine grained look, right, at the
variation within the
population and then the connection to
risk and disease. Is that right? Yes. And
you don't know what you don't know. Right?
Uh, so the fact that this is whole
genome sequence, I just think
Rob's team did a
phenomenal job of bringing together
important data to really,
advance our understanding not only of
the two platforms, but also of
advancing health equity.
Yeah. Yeah. There was something
else. I'm sure it'll come up
again if I think of it, but
um,
it's essential
that we characterize diversity in these
populations. Sure.
So Sarantis, what can you tell me then about
the primary findings?
Actually, really uh,
nicely, they try to
see if there's a correlation with
cis-pQTLs. It's something that comes
again, we have seen it in other papers
as well, that Olink assays based on
antibodies has a nice correlation with
cis-pQTLs. And actually
Olink Panel is uh,
associated with new
pQTLs. And I think that is a
really important, um,
finding, especially if you want to identify
new biomarkers and drug targets. Right.
Do you agree? Yeah. And
we have seen a lot of cohort studies
actually, uh, Cindy you can have some from
your side and from your experience as well
on that. I was just going to define cis-
pQTLs again. I know we've talked about
it previously, uh,
on episodes of this
podcast, but as a reminder,
a cis-pQTL is a correlation
in genotypes at a locus,
with protein levels.
Again, just a correlation, but
um, it's something you can detect through
statistical analysis and large data sets.
And of course, the larger your data set, the more
your power to detect any
association. Right. So a cis-
pQTL is when a variant is correlated
with protein levels. If that variant is
within a million bases or
itself that is coding
for that protein, that's what we call a
cis-pQTL.
So it makes good logical sense, makes
us feel good about the measurement of
the protein. If you actually see
a correlation between a region that
is coding for the protein and that protein
itself. The...
other thing to say about that is there are
good biological reasons why sometimes you
might not see a cis-pQTL
there's protein-protein interactions that
might knock that protein
level off of a, uh, correlation
directly with that region.
But... it is a nice feel-good
measure that you're measuring the right
protein when you do see a cis-pQTL
association. And that's a tool that this
team used.
And to be absolutely clear, "cis"
means it's within that one
megabase... close. And
P... Yes. And "p" stands
for protein, and QTL
stands for "Quantitative Trait
Loci". So you're saying that
a particular SNP genotype,
the loci, is actually
controlling the level
of protein as a
quantitative trait. I'd say appears to be controlling it.
Appears to be controlling. Associated,
yeah. Right.
Statistically associated, yes. You can't
say causality at that point where you're
just looking at correlations. So we're
associating the presence of a SNP
within 500,000,
bases of a particular
gene, and that
SNP is positively
associated with the level of
that protein in terms of
Olink-quantitated or SomaLogic-
quantitated understanding.
Associated. Yes, that's right.
Okay. And the value of
these pQTLs
is?... So the cis-
pQTLs in this paper were used as a...
a surrogate measure of
specificity. But in
general, cis-
pQTLs as well as trans-pqtls. And those
are ones that are
correlation with protein level that's
outside of the gene that's coding for that
protein. And that means outside of
that one megabase region, uh,
around that gene,
that those are valuable because they
help us understand the
pathways that may be
important for diseases
that are associated with not
only the proteins, but also diseases that
we've identified in the past through
GWAS. This catalog of
amazing GWAS
associations, uh, it helps us understand
what protein pathways are
involved in those diseases. And then, of
course, if we have a sense of protein
pathways important in diseases, that
gives us the ability to start
to propose therapeutic
targets or ways that we may
develop therapies
to go after these proteins or
to go after the mRNAs that
are translating to
proteins. ...
To then have an
approach to either nudge people
back into health away from disease that's
that preventative side or
as what our health care system
pays for today. Which is drugs to
treat diseases once they've been
diagnosed. So here we get the
payoff of the Human Genome Project
right? Which is new
drugs, new diagnostics,
new therapies potential
for cures, is that correct?
Potential for cures. Right! Which
is right
now, what do we say, 90% of
clinical trials?
I think that's the latest number
that I've heard. 90% of
clinical trials are
failing. And the
ones that you have genetic information
going into the clinical trial, uh, have been
reported to be twice as successful,
so twice as likely to be
successful. So the question is, what
can proteomic signatures from
SomaLogic or Olink. What can these
approaches do to help
improve the success of clinical trials? I
think that's yet to be seen, but that's
certainly the hope
of the future.
And using leveraging large data sets like
these important studies like Jackson Heart
study and the Heritage Family study.
Uh,
Sarantis, you mentioned 40%
in terms of cis-pQTLs. Sort of
getting to that, right? What did you
mean by that? That means
that from the old thousand
five hundred proteins,
uh, that they check from Olink
platform, more than 40%, they
are correlated with new cis-
pQTLs. And I think that was really
amazing. That's a really amazing number,
because it gives the possibility to identify
new biomarkers, for example, as you
mentioned before, a new drug
targets. And,
and the nice thing of the
Olink assay is that not
only, uh, were
correlated proteins, they are having
cis-pQTLs, but also when they don't have
correlation with the Soma assay, we have,
cis-pQTLs. That means
they have a really nice
capability in the Olink
Explore to identify this
cis-pQTLs. That's the take home message
from this. And so what was the
percentage relative to
SomaLogic? I think Soma if I'm not
mistaken, it's like roughly
So the higher percentage than the
overall numbers, were they also
the overall numbers are
different? Yes,
of course.
I think I'm just looking at the paper
as you're talking, Sarantis. I think for Soma
there were 370
of 1301,
cis-pQTLs detected for
Olink, it was 575
of 1472 total
measurements, uh, where they
detected cis-pQTLs. But like I
said, there's good reasons why we
might not sometimes detect a cis-
pQTL. So I think one of the
interesting aspects that I didn't see
them, um, I saw a little
bit of this, but the ones that they have in
common between the two platforms, if
you can see a cis-pQTL on
one that would suggest that there should
be a detectability of a cis-
pQTL, then you should be seeing that
on the other. And in fact, I think the
comparison between the two I
think the median comparison was
about 41% between the two
platforms. Am I remembering that right? I
didn't bring that pull that figure out.
Um, and so it's
compelling, right, to
wonder, is one platform
actually pulling in a, um,
phosphorylated version of the protein,
as well as the protein, uh,
without the phosphorylation,
which may
be good information to have. If you
map those epitopes, then,
um, you can determine that, I
think. But, uh, I think that's the value of
being able to look at both
technologies together and the
complementarity of them. And I think David,
uh, does a nice job of
characterizing that. And I will also
point to something you showed me. I think it
was you, Serantis. You showed me the
tweet, um, that David put
up on Twitter that has a
beautiful walks us through
his primary findings which maybe we can put
a, ah, link to that Tweet in the show notes.
I see. Yeah. You're referring to Daniel
Katz, the first author. I know it's hard
to visualize no
worries. It's hard to visualize large
numbers. Right. We're talking about roughly,
what, 370 out of
Um, and so, yeah, these numbers,
they're hard to remember, but nonetheless,
I think the take home message,
right, is that when you compare
both side by side on these particular
platforms, sort of the
findings, uh, of cis-
pQTLs is really
important. It can be useful as a
discovery tool. The
overlap is what you're saying? The
overlap between the 370 and
pretty low, is what you're saying.
Yeah. Is that right? I think this
specificity analysis, uh, that
they did was, um,
super important. I think another aspect
where we weren't showing
up as,
um,
beneficial, as I would love for our
assay to show up, was in
precision, in what they assess as
precision and repeated measurements. So
Sarantis, you had a really good explanation
of that. Do you want to go over what they
talked about in the paper? There
actually, um, authors, they have seen
that Olink has bigger
CV's than, the Soma platform.
There could be a lot of reasons, but they
speculate that one explanation
could come from the fact that Olink, we are
using small sample volume for,
our assays. Another
explanation could be for the fact that
Olink antibodies are
polyclonals. This could
affect precision, but may also
make them more resistant, making them more
resistant to binding interference.
That means that it will capture some, uh,
complexes, some protein complexes, that
aptamers could not see or could not capture,
because their interface are
covered by proteins.
So one of the advantages of using
higher volumes in an assay, certainly I
think, might be that your
coefficient of variation are these
CV's, which are surrogate
measures. We've got these surrogate measures
of precision, repeated
measures being right spot-on
top of each other.
Um, that might be
a good reason to
have higher volumes. There are trade offs,
right.
They have seen also that if we pool
sample plasma, then we
improve our measured CVs. I have seen this happen.
One of the very strengths of
Olink, which is using minimal sample
volumes, what Explore only needs like
six microlitres is actually a
weakness, which is interesting
in terms of the variability. But I think
we can say that the pQTL
results - right - do speak for
themselves. But there's another angle in the
paper that I think is some of the
strongest data. And
this is regard to phenotypes,
right? Yeah. And it speaks to
just to touch back on the precision. So, in
a vacuum, or when only
thinking about precision alone, when you
have higher CVS or higher,
um, variability, you need
bigger sample sizes to
detect a difference between, say,
cases and controls. In a biomarker study,
for example, uh, and
so if you're just thinking
about precision in that way, it's really
important um, as a
consideration for power. And
so then, Dale, this uh, is
where the rubber hits the road, right - is trying
to make a phenotypic
association, in the real
world with disease.
Do you want to summarize
that for us
from the paper?
The phenotypic, uh,
results? Yeah, those
phenotypic results were really interesting
because they pulled out some
eight different phenotypes
from total cholesterol to
EGFR to body mass index.
And they show this bar chart where. The
Olink
pQTLs right, compared to the
SomaLogic pQTLs, there's
this huge difference on a
phenotype by phenotype
basis. And thinking about
it, it's well this is really what
you care about, which is phenotypic
associations
between the genetics and
the particular thing you're
measuring. If it's uh,
hemoglobin A1c, if it's
systolic blood pressure,
these are biomarkers, these are
phenotypes from the population that
they really care about. Why?
They even have an association with
ASCVD risk score. And if
you've taken a physical
recently, your doctor will actually
have you calculate your
ASCVD risk score. And I was
really surprised in my last physical where
I'm punching in the numbers and they're
saying, okay Dale, you've got an elevated
risk of 4% and we need to keep an eye
on this. Uh, but
nonetheless, those phenotypes are real
world, everyday rubber meets the
road, like you mentioned, Cindy.
Yeah, it's
exciting because this is really what Larry
Gold had in mind. I think this is
what Ulf Lungren had in mind
in terms of being able
to broaden uh, a
discovery platform for
proteomics.
Sorry,
I'm sorry. Go ahead, please go. Sarantis,
please. No as an example,
brings up the HSP-70, it's
really uh, well-known and
famous
heat-shock protein 70
and connected
correlated with a BMI. And there are a lot
of studies nowadays for,
drugs against uh, the activity of
this protein. Actually, that's uh,
really exciting finding.
And regarding heat-shock protein
a handful of Elisa's at the very end of
the paper. Sarantis do you want to
comment on that? Yeah, I
mean they try to see
which of the
Soma and Olink proteins correlate
better uh, with uh,
Eliza. Of course they use Eliza that they
are, let's say, commercial available for
this. They focus on these
five targets, let's
say. And overall it's really
striking how Olink
data, correlates really nicely
with Eliza
data. And uh, again,
they focus with HSP-70
and a handful of other proteins
that really nicely
correlate uh, the two assays.
Giving, again a
bonus to Olink for
specificity, I think you agree on that case.
Well, and I think, I think the ability to
translate to a clinical
tool and to be fair, Eliza
is immunoassay based, right? It's
immuno-absorbent based and
we're an immunoassay.
Olink uses two antibodies for
each protein, whereas Soma
has this novel aptamer
technology, this synthetic...
aptamer technology, uh, that
they've innovated. And so,
yeah, it's something
I like about antibody-based
is that so many of our, uh,
therapeutic targets that
we use
that have passed clinical trials are
antibody-based.
Well,
thank you both for really
excellent analysis of a side by side
comparison paper. For those
interested in the reference, this is
Katz DH. This is Daniel Katz,
the first author. The senior author
is Rob Gerszten. The
title is "Proteomic profiling platforms head
to head: leveraging
genetics and clinical traits to compare
after an antibody based methods" Yeah.
Thank you very much for joining us today.
Thank you very much. "Go Beth Deaconness!"
right? "Beth Israel Deaconness"
Exactly. Thank you very
much. Thank, uh, you thank you. Thank you.
Bye bye. Thank
you for listening to the Proteomics in
Proximity podcast brought to you by
OLink Proteomics. To contact the
host or for further information,
simply email
info@olink.com.