Special Feature

User Panel

My Panel

My Panel

Bookmark Science Articles

Recent News
Bookmark / Share This Science Site

Proteomics Biomarker Discovery

Proteomics Lecture

Please Note: this text is only for quick reference and contains spelling/grammar mistakes. We are not responsible for any errors or ommissions. See our Disclaimer/Terms of Service. If you are interested in this lecture, please watch the full video here:

Progress Toward a Biomarker Discovery-to-Development Pipeline in Clinical Proteomics
Friday, April 11, 2008
Steven Carr, Broad Institute, MIT and Harvard
Total Running Time: 1:11:57

Progress Toward a Biomarker Discovery-to-Development Pipeline in Clinical Proteomics

Good morning.

I'd like to get started so we can give our speaker plenty of time.

The first compile of slides show that today's speaker is Dr. Steven Carr, and the next slide shows that the next talk will be on may rd, not may ninth.

And as i'm plugging the proteomics group, as I have ideas for speakers you would like to see next year, you can see we're near the end of our schedule for this year and we'll be solic so--soliciting nominations for candidates for speakers in the coming year.

So this morning it's our pleasure to here from Steven Carr.

Steve and I have known each other for a very long time, both trace our ancestry back to the same birth place of mass spectrometry of Klau s' lab at MIT. At the same time Steve went on to several different careers that took him first to industry, gsk where he was director of a laboratory that eventually became proteomics laboratory.

And then to moo lenium where he had--millennium where he had three years at millennium with gtsk and then moved back interest academia with the brood institute at MIT. And Harvard where he's been working on biomarker discovery and and clinical proteomics.

So we look forward to hearing from him.

Can everybody hear me so first I'll start by thanking sandy and the interest group for the invitation to come and talk to you today about work that my group is doing at the broad.

My title is kind of self explanatory but i'll begin by saying you are not going to hear a fait accomplie today.

We are not at the point where we know precisely how to do this is there's a notion of what eye pipeline might look like to take us through discovery to verification and i'll define that during the course of the talk and finally through clinical validation which had a very different meaning than verification as we'll come to show you.


So just set the stage, I don't know what the representation of expertise and backgrounds in the audience are but just to put us on the same playing field, i'll begin by saying that vertually--virtually any journal you open up, clinical oriented or chemistry oriented or biology oriented, emphasize that there in principle some be tremendous value coming up with biomarkers, particularly those for early detection of the disease.

And this just makes the point that in the case of probably the deadliest cancer we're faced with, lung cancer, that the five year survival rate is very dependent on the stage at which the disease is diagnosed and which is a stage of diagnosis is shown in the light blue and the background so you'll notice that unfortunately very few cases of lung cancer are identified in an early stage.

If you do, if you are so lucky and I should say that I have a sister-in-law who recently was diagnosed with lung cancer, nonsmoker, age roughly , she's in stage one-b.

Even that's not got long term stats, only % but it's certainly better than where the majority of people are diagnosed which is fivea and the survival rate is extremely poor.

So if you could come up with an early marker for the detection of lung cancer, your chance of survival would be greatly improved and this has been recognized at the centers many peptide and biomarkers which have been approvedot order of thousands.

At the rate at which new protein analytes are being introduced is exceedingly low.

And we have been focusing on this in the sense that the lack of success is beginning to drive us toward new methods of trying to discovery, so there's been genomic based approaches particularly recent success like things like nano prints where one can use microarray expression profiling as a biomarker related approach.

We are trying to do this from a proteomic based strategy.

And it goes beyond early detection.

The impact goes beyond early detection of the disease.

So we've touch on this early detection, but of course there are genomic based methods that could be applied to you know these snips or happen lo types from these individual, you could predict based on their genomic profile that this person is likely to progress to coronary disease early on and we should begin, you know statin therapy at a very early stage or there are selective estrogen reuptake inhibitors on the market.

We're treating breast cancer and osteoporosis, it's possible one could use predisposition markers to begin this quite early.

Screening of course, as a--using either genetic or expression profile for protein related biomarkers from early detection could provide you with a biochemical measures to replace what are right now, imaging based methodologies or phenotype based methodologies and i'm referring in particular to mammography for example, it would be great if there were biochemical tests they did as well as--or in some cases could replace mammography when the breast tissue is too dense in order to get a good mammogram.

And then beyond this early detection fated once clinical symptoms have appeared, you will like to be able to use your markers to measure treatments, monitor the treatment.

Is the drug working are you in a particular subtype that is more of the aggressive than another so this may be a basically a fork where you will make a decision as to which type of therapy you use.

For the drug industry biomarkers can be potentially exceedingly important because you could use the presence of a response marker in a population of individuals to seggregate their initial clinical is its who should should be treating with the compound to see each in the group you think should respond to the drug whether they do or not and this would have enormous impact in terms of decreasing the overall cost of a clinical trial.

So enough on--and I just want to make sure that you're aware that using markers alone, markers come in different flavors.

We talk about the protein related ones but of course images is extremely important in this whole marker paradigm, it for example, to give you a trivial case.

Tell be terrible if you had an early marker detection for lung cancer, so early that it said you have lung cancer, but you didn't know where in the lung in the cancer was.

So what will you do actually you could resort to chemo therapeutic approaches but surgery is typically the best intervention for this but you will know where to operate.

So these go hand in hand.

But the track record for delivery of biopropene markers has been pf--i'm not going to mix words, it's been dismal and this is a compilation of various papers which have come out claiming to have found all sorts of markers, but the fact of the matter is that the rate of introduction of new biomarkers is less than one per year and I then is largely possible by the fact that takes things from discovery and actually demonstrating which has--that discovery paradigm has this dimensionality over end problem.

Where you're dealing with s of thousands of potential markers and your false discovery rate in these experiments using very small numbers of samples, typically what's done is very high.

So you end up with a very high fdr rate and what youny is a way of moving things from discovery with high speed relatively low expense to demonstrate the action we have with promising ending.

Much of my talk will focus on that aspect, moving from discovery into, you know putting credentially markers or discarding.

And right now, the way that credentialing is done is by using what's commercially available in the catalog for the antibodies who are going out and making them.

And this sort of opportunistic approach is insufficient, the content is just not there to do an adequate job.

Discovery work has failed pretty miserably when people have tried to start from plasma.

That's not to say you can't do it and it's note to say the markers aren't there.

In fact the markers presumably are there.

The problem is they're being obscured by the very high abundance of a small number of proteinsot order of perhaps proteins that dominate the protein content of the blood in the high microgram up to the minimilligram per mill range but in this range of interest, presumably the range of interest for things that are markers that are disease specific, you are looking at thing that are in the nano gram from the range and often a picogram from no range and these are representative markers that are in these ranges.

These are commercial markers for various diseases.

Those are the things which are being obscured when we try to do y. So what can you do it's not to say the blood should be given up on but there are in many cases there are other potential sources of markers that you could be lining.

And let's look for a moment at what we referred to as proximal flu i. These are flew is that--fluids that bathe the site of injury, whether it's the sin ovium base of a rheumatoid arthritis on an oftioart ritic event or it could be the site where the cyst fluid bathing the tumor.

The notion here is quite simple, if you sample flu is in close proximity to the site of the injury, the concentration of markers related to the disease are likely to be much higher.

Pretty simple notion, is it born out by analysis this case it really is.

This is just monitoring the levels of cain the variety of histologically typed cancer tissues looking in serum then in the fluid and then in the cyst.

These are in--right.

So, if you look from serum progressed to cyst fluid, it's obvious the levels in each case, in these cancers increase dramatically.

So, this works in that case.

Another example is the detection of this pancreatic ductile carcinoma marker, hip-happen one.

This marker was discoveried in pancreatic juice, again, approx malflu I of the pancreatic cancer and it was confirmed and sterile.

If you look at the markers in the pancreatic juice and it is a thousand times higher than it is in blood and pretty clear why, you have to somehow make it from the site of injury into the peripheral blood flow and you have many obstacles in term of transport as well as the sheer delusion aspect of the marker into the peripheral blood.

So probably what's more or equally important is that the fold ratio of hip-happ one in cancer verses a control, a healthy control, is to one if you look in pancreatic juice.

It's only three to one if you look in serum.

So all this is to say footwork you had start in blood rather than starts in the proximate fluid, your analytical challenge for the hip-happ one would be enormously more difficult.

So why not start in the blood.

So this reiterates everything I said so maybe we can skip on to this other than to say that okay, proximal fluids are good, your markers are more likely detectable, what do you then do do, once have you a marker you found in discovery paradigm.

Ultimately you have to sample an easily sampled biofluid and in most cases that is peripheral blood.

So you have to then define a method to make your discoveries and move them into a verification mode and by verification this is what I mean.

I have a candidate biomarker that i've discovery in cyst fluid, I know am going to try to determine if that marker is detectable in peripheral blood from cases and is clearly and statistically different than the amount of that, the level of that protein in caseis verses control.

--cases verses control.

That verification.

It is not biological validation for clinical validation.

Okay but it is the first step in this process and the at the moment, the huge barrier in terms of moving these long list of candidates forward.

So let's walk you through some examples so this become concrete.

There is a number of biomarker projects going on in my laboratory in breast, lung and ovarian cancer as well as in cardio vascular disease, so i'm going to walk you through two examples, one having to do with breast cancer and the second having to do with cardio vascular.

So in this case, proximal flu is that you have available to us are ripple fluid and ductal lavage fluid.

And neither of one of these are an ideal proximal fluid.  and reason for that is in order to get these flu is to get them noninvasively, it's not without some discomfort to the patient but they are routinely obtained sthat the architecture of the duct work of the breast is really quite complex.

 located here and you are is sampling bio suction on the nipple, or by cannulation in order to get the lavage, nipple or lavage fluid, you are sampling from a large amount of the network, not just the region that's in close prokes proximit.

So even in this case we're getting a high dilution relative to the ideal case, where we just like to sample the vicinity of the tumor, now that said, it's highly enriched in these things than if peripheral blood.

So we have caveat about those.

We have a pseudoproximal fluid that we use that we refer to as tissue interstitial flu I this, is a methodology that julio cellis reported on a number of years ago, where you take the tumor, the solid tumor sample, in this case breast tissue, immediately after resection in the surgical suite and you--while it's fresh--and after the pathologist said this is tumor, this is not tumor, we sample roughly a quarter to half a centimeter of that breast tumor.

It gets minced up and put up in a phosphate buffer say lean in an incubator and we let it sit for an hour.

We are allowing proteins that are being shed or secreted by the tumor to make it out into the solution.

You then spin down the cellular debris, and take off the fluid, that is the tissue interstitial fluid.

The work i'm going to show you right now concentrates on a specific subtype of breast cancer, so I should say that our patient samples are being taken at m. D. Anderson in houston, very close collaboration led by gabriel and his colleagues there.  so in order to get samples like this, sample fluid, tissue interstitial fluid and from the patients this is is close to newcastle here but we have to work closely with our clinical colleagues.

This has been a tremendous learning experience for me and a rewarding one.

So the type we're looking at is the most common form of breast cancer which is the negative buyer the basic ductal carcinoma, and the processing method we use is not for the faint of heart.

Really, autofairly lengthy--it's fairly lengthy.

Just digest and shoot it into the mass spectrometer, there's a fair amount of processing here.

Let me walk you through that.

So here arative, we whirlpool, we do this to our samples, this is a pipe line for sample preparation for analysis.

We have--why are we pooling because an individual patient sample, there's not enough protein from an individual to do this detail level of analysis so right now we are pooling roughly five patient samples, in a case and five healthy controls and we're doing this essentially tent--essentially of these pools are being analyzed in this process.

So those samples are then depleted because much like plasma, with nipple aspiration fluid, it has high concentration %f a small number of proteins which are essentially are the same, very nearly the same as those found in blood.

There are differences, we'll talk about that later.

This is a useful step for taking out a panel of highly abundant proteinss and allowing to you dig deep.

We then do the usual denaturing and the proteins and then cut them up into peptides using proteolytic digestion and then we fractionate by strong cat and exchange chromatography making  or so fractions out of that.

Each one of those fractions, then goes on to a roughly two hour lcms ms experiment.

So it takes essentially two weeks or so, two and half weeks to run through a single pair of samples on this--using this process i--process.

I would like to point out we're doing this with a consortium effort with colleagues at northwest, where the samples are commonly prepared and then split at the point of proteolytic digestion, been fractionated in our have labs and analyzed in our individual labs and all the data can pass through to different data analysis pipe line, so we're comparing apples to apple in terms of the peptides that are identified and the protein that are identified.

And I point out we're using the latest and greatest, you know swiss army knife tool for mass spectrometry.

The orbi-trap.

So I think it's important to ask and answer the question well, how many samples to--do you have to analyze to get any statistical components in what you're measuring and this is a pretty simple analysis and i'm not going to detend this because i'm not the one we did it.

D-arm annie in nigh group went through this.

But we're asking the staof statl question, the pragmatic sentence, geven the fact that we're likely to be only analyze by this complex process, roughly  patient samples, what is the minimum fold change in protein levels that we ought to be focusing on as being stattivityically meaningful when we detect them.

So I highlight this row right here.

We have to make some guess as to what your variance is in any of these levels.

So we take a pretty servative estimate of that thing that we essentially have a cv of %.

So, there are two fold difference essentially meaningless.

But if we--if we assume that cv and a p-value of .  which is the minimum p-value that most are willing to accept, the minimum foal change difference between case and control for protein that we ought to be focusing on is essentially five fold difference.

So of all the data i'm going to show you is only focus on thing that are five fold over express or differential in the express in case verses control.

One of the things--one of the questions that comes up, in fact there is a program which has been really under the sponsorship, leadership of henry rodriguez in the audience funded through the nci, one of the proteomics initiatives where consortium groups are trying to evaluate the reproducibility platform, what's the best way of going about discovery what's the best way of going about evaluation we started this in one instance and i'm quite hopeful about the prospects for discovery becomes fairly reproducible.

So this is that mass sample that was analyzed in specific northwest national labs.

There other than actually, an ab gre combat close to to be identified.

To be conservative we stripped out all that were identified based on a single peptide so it's only two or more and secondly we took out the immunoglobulins which left with is total proteins and the overlap between what we identified in the two labs is % and I can tell you that upon repeat analysis in your own lab, using an orbi trap instrument, the best can you do on a complex mixture like this is %.

So this is very encouraging.

It says that labs running standard--and actually this is not identical between the two lab.

There's actually variance in how we handle the sample at the point of--where we did the digestion and the fractionation down stream.

So I think there's an important lesson here that can you tolerate some differencess and get quite good reproducibility.

Most importantly, if one focuses on the proteins that were overexpress in that, identified in the two labs there were in our data set, in the broad set, of these proteins were identified by tnnl and the ratio of that to is that % of the protein detected in both were observed as being five fold-over express by both labs.

So there's a fairly high degree of concordance.

Now we're in the process of in an n-of one on these pooled sams.

We've repeated these two more times and we are going to have an n-of by the time we're finish.

And one of the things I would point out though is that in these--in the case control in harrison, nearly half of the protein identified are differentially five fold over express and hang on to that number, that should be a-surprising and b-frightening.

I should say the control in this case, your control is not very good.

There's an enormous amount of biological variability if your technical variance is not that high.

Both of which could be true.

Is this a one-off well if you look at tumor interstirrial fluid and the only thing i'm going to diagonal you is you can get higher numbers by analyzing the tissue directly, so now we're at we need protein.

In this case around % are highly differentially express in the tumor verses the control.

The control in this case is tissue that was sampled at distal to the excised tumor piece.

So they don't just cut the tumor out, they cut out quite a bit of tissue and this is tissue taken distal from where the pathologist said as to the boundariess of the tumor.

So is it a perfect control no.

But it does call into question how you use these controls if you really need the control.

So we have a very high number,  proteins here which appear to be differentially expressed.

Now do those make sense if you look at the list, this is a selective look into that list, what you find are--as you always do when you do these is it, a large number of them are reactant proteins.

The s are all over the place.

They're proteins of that.

But more importantly if you look at the list you'll find kinase, transcription factors enzymes and growth factorss and many proteins which if you pulled up their stories individually, using med line you would find a cancer connection and in many cases if not most cases you would find a specific breast cancer connection, so the fact that these lists we're coming up with are highly enriched with things which would be previously credentials as being markers of cancer generally and breast cancer specifically is very encouraging.

I would say that we have I think pushed the boundariess on this a bit on these analysis.

It's not like we were the first to look at nipper asperric flu I but prior to this publication, there's only proteins idebtify and our current work is in excess of .

Tissue interstitial fluid, a publication of proteins and this happen to use two-d gelss and ours is in excess of .

We've done similar work in varying cyst fluid where our studiess are again approaching  proteins in previous study of the same material, there was were only eight differentially regulated proteins.

So the catalog is much larger of potential candidates.

I want to move quickly into this model of my o cardial infarction.

Cardio vascular disease.

What can you do there what is the proximal flu I for heart disease well it is blood.

So we're kind of stuck, I made all this blue-ha ha about not start nothing blood and there's going to be diseases where you have to really use it.

But even in blood, in this particular disease, there are waying of getting to a more proximal fluid for the disease.

I'm going to bach you through that.

This is a model of a planned therapeutic heart attack or planned my o cardial infarct that is used as a treatment.

There's a patient who has severely compromised left ventricular function, usually, they will do a septal alcohol oblation of the tissue in this region of the heart and shown here, which causes then the tissue to die and in the process patient has for what all intents and purposes has what appears to be a heart attack, they come out of it, they're in the operating room.

And the reason this is done is that after the data is kill, measure tissue relax aroup it and blood flow is restored in the left ventricular.

How they discover this process, don't ask me, I don't know i'm not a clinician, but the used therapeutically and we're use thanksgiving as a valuable tool to monitor patient as their own control preceding and during an after this proseure.

So it's a--procedure.

So it's a remarkable set of examples, where we have the patient as their own control and b-a temporal control with the ability to look at a controlled study while this is going on.

So we can sample in the coronary sinus while the patient is having this therapeutic intervention and we can also sample simultaneously in the periphery.

What we find when we do this, interestingly are all the usual suspects or--you find kra tin kinase, components going up, you can look down and it says cardio myop aty associated protein-three.

Many that have been currently in clinical use as markers or which have had some credentially but anythingly, there was a lot of things on this list which are novels.

This isn't the full list.

The full list is in the order of  poo proteins that are differentially expressed or higher.

But we did this temporally.

This is an interesting study for us and we've done this in six patients, we're up to showing you two of these here.

So, for three previously credentialed markers of cardio vascular disease, crt and s--mile o peroxidase, this is baseline, this is minutes and this is minutes.

After minutes we can't sample in the sinus any longer.

You can see, there's interesting variance here.

This patient had relatively high levels of crp at baseline and at  minutes and then it dropped off in an hour.

This patient had increasing levels of crp at minutess and then it fell down.

Each of these in these different patient behaved differently.

And I should say that this data is borne out or supported by ealiza data for these same markers we have in these same samples.

So anyway, I won't walk you through all of this but you can see in each case, can you see some proteins increasing like mile o peroxidase and we have this nice temporal behavior for these proteins which build our confidence in the measurements that we're making.

Therefore novel markers of mi that were pick up, so this is bit-kine, this is apoptosis related protein, and this is p-gam two and this is p-gam-one.

So this is all of which appear to be elevating in response to the intervention.

So, where we are here at the end of this story, part of the story is that we're building these candidate lists, candidate lists are hundreds and many cases up ward of a thousands layer.

How do we do it mow do we deal with that is there a way to move those candidates forward one method which has been proposed and is being done by our collaborator mandy polovich at the cancer research center is to try to build credentially information around our experimentally defined biomarkers by looking in other data types.

So for example, I mentioned we talk about the proteomic information but one can look at expression publicly available expression and profiling data sets and one can look at cgh in human as well as other organisms such as the mouse.

One can do literature mining so there are list now of proteins which have been pulled together through deep literature mining.

So from the fergy lab there's example from the anderson's group.

And one can try to incorporate owl this information--all this information waiting each of the various sources to try to move or prioritize the candidates in our lengthy list.

Again, this doesn't tend to remove things but it does at least change the order in which we try to progress things.

Yeah, this is the question how do we bridge this discovery paradigm to clinical validation we talk about this.

What do we do here there's a way to reliably test to accept, or discard large numbers of candidate proteins in blood for markers will be measured, how do we do this so, well, we talked briefly about the potential of using antibodiess.

And let's look at that in a bit more detail.

So a western blot, typically has sensitivity in reasonable range, can you get into the low nanee gram per mill range for protection and blood but it's not quantitated by any stretch.

To develop a western blot assay could be a short time frame, and it's just a matter of making a polyclonal abet body or in a mouse or bunny.

Can you with the through put of these can be reasonable, do many s per day.

Reagent cost is not--it's inconsiderable and this is on a per analyte basis we're talking.

Ful sample comsumption is reasonable.

And begin an ealiza which is the gold stan ard still for any of these test and which has much, much, better sensitivity.

Ot ord of of pico grams to low nano gram range with good c v's, good cv's, those to be much longer, so it's not just getting an antibody, but getting an antibody that will get interferences and work in the matrix complex background and through put can be very good, the cost is much higher per analyte and interestingly this is kind of news to me.

I thought, oh, eliza is very sensitive, right in they don't use much much sample, well, not true, if you get proponent measure made in the cardio vascular study, compared to validate finding, it was microliters of patient plasma for each measurement.

That's a lot.

So, we're in a position of having insufficient antibody content to satisfy this, until we're willing to set out on a huge endeavor and one can argue that's an important thing to on be doing.

Bake all these antiaboutss and then have you this long and costly option.

There is an ole ternative which is to use mass spectrometry in a targeted manner as a pseudowestern if you will to go looking for proteins of interest who are candidates in the complex fluid.

And for sake of time i'm not going to go into too much detail on this, I just simply want to point out that to distinguish the targeted ms approach from the discovery or unbias ms approach that we have been talking about today.

So here, we're put nothing a mixture of peptide at any given time into the mass spectrometer, one of those selected, collisionally excited and then we're looking at the spectrum of if it's all of the products, all of the praying ments from that form that were selected from a peptide and the system sort of does ptides of interest and modn systems typically do eight to in any given cycle.

So select eight to different signals and fragments and determine sequence.

In a targeted experiment, we're asking and using instruments in a different way.

We're using ms one to select the particular peptide of interest but now it's not the data system that's selecting us.

What is selected is coming from an individual that said, I want that peptide on the list, go after this.

And then what we look at in the tag ments is not the entire spectrum of the peptide but just a few selected transitions.

This allows us to sit on those ions and accumulate signals and therefore have much better sensitivity and selectivity that than we otherwise would.

That's the main advantage of this.

It's highly multiplexible so we being go after hundreds or more peptide signal in any given cycle on the instrument.

And we get anywhere from to  fold higher sensitivity by using this method verses the discovery mode.

So there's a lot of good things going for it.

The would use this in a practical example is illustrated here, have you your list of protein markers your candidates.

You either have a list of observed signature or proatio typic peptides for those candidate protein, or you have predict the one.

So for example, have you candidates that came from literature mining that you never experimentally detected but they're highly credentials want to build an assay to tell whether or not you can observe them.

You do ancill o digestion of that candidate protein and build your list based on the theoretical peptic and bell an analysis.

You won't take every one, but you would say which ones would buildot spectrometer, and which one have the proper weights and various process analysis which I won't go into detail.

But then, after you have your list of pep ties, you will construct your assay synthesize those peptides and both the unlabel form, the wild-type cform and a labeled form.

And this is very straight forward, clinical application of mass spectrometry that's been around for deck's.

It's isotope dilution mass spectrometry, it's been used for small molecules, as is four deck'ss, this is applying that to the peptide world and use calibration curves by measuring the ratio of spiked in c peptide to the endogenous level of cin your blood sample and you do this for a series of peptides coming from a series of proteins and as I said you can do this for upwards of about a hundred peptide analytes quite easily in a single run.

One of the disseda vantages however, when we start our work was that mrm multiple reaction monitoring, or selective reaction monitoring as some like to call it of peptides in blood, have a limited detection range.

So the best work ha had been done was round about a microgram permill, thousandano grams permill in the background of plasma direct analysis.

That's not good enough sensitivity but the assay development time is very short, can multiplex this to high levels and the cost is pretty low.

The amount of plasma you consume is actually very low on a analyte basis and within the rell've of an ealiza or the single analyze or the microliters.

So our goal and work we completed was to extend the mrm sensitivity into the single digit nano gram permill range and demonstrate multiplexing capabilities.

So i'll show you this for some of these cardio vascular proteins.

Plasma processing here is simplified because we're trying to get to higher through put so rather than fractions, we have six and ftx fractions but otherwise the processing is the same.

We take aabundant proteins and we reduce and digest prior to doing our quantitative assay by lc (indiscernible).

And then again we're starting with hundred microliters of plasma in orer to do these analysis.

We in this case, we had proteins available, that we being spike into plasma to construct our calibration curves with and we won't always have this so in some cases you'll have to resort to the peptides that you're synthesizing but here i'm showing you calibration curves for lep tin in the background, a female plasma where, yes, relative to direct analysis of plasma alone, roughly a to fold improvement in the limitit of quantitation, so limply taking out these abun ant protein, get you roughly a factor of in your l. O. Q. So prior to removal of the abun ant proteins our l. O. Q. And roughly nano grams per mill is the best we could do.

So now we're down to nano grams per mill with signal noise of to one or better.  and this the slide that got botch, this is the mac to pc conspiracy these days but still happen.

So with data here we're shoring that if you combine limited fractionation on top of depletion, you gain another factor of in that, so the combination of the factor is from the depletion and the factor of to actually in the scx gives you combined improvement in sensitivity relative to detect analysis and plasma of to fold.  this is pretty substantial improvement.

And we're now and i'll just show you the data from thissed study to premind you these are the blood draw, now we're measuring in the--these initial draws are in the sinus, i'll going to show you data now for the measurement at four hourss and hourses in the periphery of those patient.

And the protein that we set up assays for were these .

We have managed to configure useable assays for six out of these , we're still working on the other four.

Yes, for the sake of time i'll jump to the data.

So these are the results for figuring assays for an easy protein.

Show is mrp , which is in the--around nano gram per mill range and spikes upward to close to nano grams per mill defenning on the patient again and you can see the nice temporal response here from zero hours, four hours, hour, in each of the patients with some slight--this is the biology, patients are different from one another.

I should point out that the blue here is the clinical assay applied to this, so, interestingly, the assay that is available for mrp is billed as research assay.

I would say it's not ready for prime time yet.  this is research grade eliza it in no way reflects the data we're getting from mass spectrometry.

On the other hand, the component for the ealiza is the same data as the what we're measuring and the mass spectrometry which what I would be more cautious about because we're aproffing the limit of quantitation in the instrument.

We're making measurements here at around one to nano gram mermill range but we're getting and observing effects that correlate well with the ealiza data and show and demonstrates proof of principle, you can make measurements for proteins of interest in the backgrounds.

Now what can we--we're now seeing nano gram per mill range, is that good enough no, the answer is no.

It's now for several reasons, first off many markers were going to be in the pico gram for range and secondly there's a fair amount of process tag has to be done in order to hit that range.

Much better if we go directly from blood to a measurement.

Ing it much more similar to conventionally deployed clinical assets.

So one method of doing this, is called siscapa, this is an ark proach that was pi pioneered by lee anderson and it's one we're working on close collaboration in our clinical proteomics project, he's one of the collaborator.

So the notion here is to get around the complexity of plasma background and get a high degree of enrichment.

We are going to spike into that plasma background after we digested it and then all the treatment, incident complex mixture and then spike our label peptide and then capture using an antibody that has been raised to the peptide, not to the protein, but to the peptide, signature peptide for that protein and we are going to capture simultaneously, the exogenously added heavy label peptide version as well as the endogenous version of that peptide which is the analyte of interest.

We're going to shoot that into the mass spectrometer, and monitoring again for the mrm, or both the heavy and the light form of a pep peptide of interest.

In this case, we're hoping to get this tremendous increase in enrichment and decrease in background complexity.

This has a lot of--this isn't isn't--this is different from the ealiza.

It require only one antiabout.

The mass spectrometer here is the second antiabout, the replacement for the second antibody, the antibody is the quality of the antibody do not need to be as restrictive as those--as for conventional ealiza because we're not trying to recognize the native protein.

We're trying to configure or detect or hold down a peptide and we're raising the antibody for the peptide that we're trying to pull down.

And it--there is now demonstrations that the success rate for doing this is actually significantly better than raising an antibody through peptide and then try to pull down the protein.

So, just one example of this for one of those cardio vascular proteins.

So this is trip-i, where we were struggling to get below the nano gram change.

So here we spike the level down from varying levels down to one nano gram to one microgram permill and we're adding our exogenous peptide and capturing using antibody it that has been use for magnetic bead and we're using that because we're manipulating these magnetic bees and develop methods to manipulate them in an automatic fashion, working towards a complete plea automated assay.

And right now autodone manually and also the instrument.

This is directly out of blood, we're configuring an assay that gets us down to five nano gram permill without any handling of the plasma.

This is direct.

And the notion this, is where pol clonal antibody using input of a hundred microliters of plasma and we have data that's we can break the range with other proteins using larger amounts of input plasma and/or moving to monoclonal antibodies instead of using a polyclonal.

One of the reasons why we're hitting the level of why we have some problem down at this range still is that there's a significant amount of background from proteins that are nonspecifically binding either to the protein or to the bees which is contributing to the base and interference baseline and we think we have some way around that.

So, last two minutes, I want to talk to you about an ecsighting new technology that we're working on which we think is going to help solve this problem which we still face even if we get mrm's working and siscapa working we still have this issue of having hundreds to thousands of candidates to deal with and no individual lab using mrm verses siscapa is going to have the capacity to look at all those candidates.

It just isn't possible.

We figure we have a hundred assays per year, per lab that we can handle.

We can develop.

So for our desired capability is to get to be able to screen up wards of a thousand in tissue block, but we like to have a method that's sensitivity similar to mrm methodologies and what--what's driving all this is that if you have this approach, detection by that methodology would mean that an mrm assay could be configured for that protein.

It's pres spent you can configure an assay.

So the method that we're working on is we're calling accurate inclusion mass screening or aims and what it involves is using and developing capability of the orbi trap which has the ability to have inclusion lists of upward of several then masss on it.

The orbi trap has another special feature.

If has high mass selectivity in term of mass precision.

So we can include things on the list and say they have to have a mass that's within seven--right now we're at seven and half parts per million, we didn't know we could go lower, but right now the test data is at this level.

If a peptide on that list, one of the have--you detect a mass within that mass, it goes and triggers an msms experiment to be done on that someone.

So the detection triggers the sms which feeds directly into your standard data analysis pipe line be it mascot or sea quest or whatever it is.

Now we used this previously, we publish two papers where it was used, one about fishing for mitochondrial and another for designing a pepper alegorithms.

But this is a different approach and use are the capability and the capability improve tremmennouslyot instrument, our lab's been working closely with mike (indiscernible) to push this forward.

So the way you use this is nearly identical to the processing I showed you for the m rm.

Limited fractionation and then using the orbi trap, we survey lists of many thousand of peptide mass on our list and when one is detected it triggers the ms amount.

And can you do roughly a hundred proteins to run, so you need to do a number of runses to get up to a thousand protein but it's perfectly conceivable in term of the time frame.

We're talking a month or a few months to be able to do this experiment asosh pose to a year or longer to set up a hundred assays.

So this is a trivial example, a lot more of this data is going to be shown at asms, but this is four cancer related protein that have been spiked into plasma.

Two different levels.

So two guarantee mixture were made where this protein is nano gram or hundred, and this protein is or , et cetera.

And the question is, using the accurate inclusion mass method, where we've taken in this case every possible trip tick peptide that's produced by each of these proteins so there's roughly or so peptides on the list, looking for every single one of those pep ties in a roughly hour or hour and half long run.

How frequently do you detect, how many peptides do you detect for each of those proteins at the hundred nano gram level, we're picking up for three to six peptides and nano gram level, we were successful in every case but the number of peptides that you see is much smaller.

So, you can begin to see that this has the potential to act as a bridge between our unbiased discovery and our verification where what this is basically doing is asking and answering the question, do I detect any of my candidates in blood which is where i'm going to sample them and are they at a level that I can configure a direct mrm assay for because detection by this method means can you infact configure it.

Absence of detection doesn't mean they're not there, it mean you have to do something else.

But this is going to triage a large number of the proteinss and move us at a much faster rate.

So, last two slide.

And then finish.

How come are we getting to addressing these it's a start.

We feel that tissues and proximate flu is are better than blood if you're working on canner markers, cardio vascular disease as I indicated blood may in fact be the proximal flu I of choice.

If a signal existss and signal mean a detectable difference in case and control, can proteomics detect it we think that the technology is now, pretty well--working pretty well in this low to mi nano gram permill range if you use a process that employs a fairly extensive degree of depletion and fractionation and you have careful match case and control sample.

And the reproducibility platforms can approach the level of an interlab variability, sorry your intra lab variability.

The process of moving from discovery to clinical useful markers in the pipe line look like.

And we talked a bit about the aim, verification by mrm as a drafting tool, we now have shown you that can you get into the low nano gram permill range and we think that siscapa is going to get us into the pico gram range.

That of a cartoon and that acknowledge the people that are doing this work, a group I particularly want to mention, terry adonna and (indiscernible), and eric coon who have been driving the quantitative work in the group and the siscapa work and mike terret who's been doing a lot of the breast cancer discovery work i've described.

Cloud for his spectrum of implementation on work on aims and I show you a little taste of and various collabrative groups and particularly robert at mass general hospital and andie at the department and fred hutch our clinical collaborator at m. D.  anderson who are sources of clinical advice and samples.

And of course the funding agent.

And I want to thank you.


Yes yes yes.

 ( inaudible ).

>> so the question was, in the example where I was showing the internal laboratory comparison between the broad and the pnl of the nipper asperric flu i, complex fractionation discovery and the experiment, what was the method of quantitation was it the same so there are multiple ways of doing this, of course maybe you could use spectral counting approaches, which is a very simplistic way of coming up with anna log of the relative amount.

We did not use by--start by saying we didn't do, we did not use any isotope labels.

So this was using isotope free quantitation approach.

The way that spectrum handles this is by looking at the areas under the precuriousor peeks for each of the peptide selected and for a series of peptide from the protein, it defines an average value for peptides from that protein and uses that as the surigate representation for the quantities and we're doing relative quantitation.

So not absolutely.

So that's how we're getting our relative quantitation.

Same method was use for analysis of data from pnnl and we took their data ran it through the spectrum pipe line and came up with our version.

Viper is which is what pnnl calls their pipe line, I believe uses a version of spectrum counting that take into account the molecular weight of the protein because you get fewer transcriptic proteins.

And the answers that were maintained with these methods were correlated.

The five foal approach over the express orer is where the numbers came from that I showed you.

Yes >> (inaudible question ).

>> so the question, just to repeat the question.

Since can ser a genomic disease, why not use your mssmaze method instead of doing broad based discovery, leverage genomic information that's out there and use targeted approach for example to go after candidates absolutely right.

We're actually just taking--we feel it's still important to take an unbiased approach because we don't feel that the genomics necessarily is the whole picture.

There's many other environmental factors and modulation of enzyme activities which may affect levels or modification states of protein that right now we can't--we don't understand what the--if there is an underlying genomic component.

Perhaps there is but right notice we don't understand it.

So we're taking an unbias approach but we're also very willing as I think I showed in the case of the cardio vascular disease, build the target other proteins and we're doing the same thing in the cancer area, so, if there is--if there is protein which we did not detect which are highly credential for which there are other genomic data can we use the same approaches to target them.


Methodology that i'm showing you that's beautiful for phospho peptide analysis.

You end up with a lot of phosphorylation sites and many proteins.

So constructing assays can be pretty expense itch.

If you have an important protein to go after, the specific phospho site, autoa beautiful method.

Yes >> (inaudible question ).

>> yes we don't--the question--let me repeat the question.

We focused on looking at proximal region to the tumor when possible to do our initial discovery and then move to the periphery to do or do our assays.

And there could be tumors elsewhere that are contribute tag we're not picking up the marker for because we only focusedded on narrow region and so, you're right.

I glossed over in the tip sample, we did look at the lymphatic nodes.

There was metastasis there, we did look at those.

We're willing to build our candidate list from whatever, but you know we are still looking under a fairly narrow lamp post.

I don't know.

I'm--i'm completely open to other suggestions about how we should be approaching this.

I just don't know.

Because going directly to blood, just--we know we're not going to get very deep into that.

So if those markers from wherever they're coming from are kind of in the--you know midhundred nano gram range and lower, we are beginning to lose our ability analyzing blood.

We're very good in the microgram range in the blood, pretty good in the midhundreds of nano gram range but once you get below that your detection limits, well your ability to robustly detect all proteins at that level are very low.

In fact, that's one of the focus of our clinical proteomics technology assessment is that actually put a fine point on if you do this fractionation method, have proteins present at a hundred nano grams per ml issue how many proteins in blood or whatever your tissue that are present at nano gram per mill do you detect the fact you can claim and in fact you're probably right, that you see a protein at a pico gram permill does not say that that's the sensitivity of your technique.

That's very misleading.

You got lucky.

You saw one peptide from one protein present at one pico gram permill, okay you miss all the rest that were pres spent there were probably thousand.

- -pres spent there were probably thousands.

(inaudible question ).

Thank you for that, that was a terrific question.

The question was do we envision the mrm, the lcms, mrm methodology as being the final employee clinical assay or are we planning on switching the platform that we use for assaying for those proteins and right now the jour jury is out.

If you--you get different answers from different people and my view today for proteins that are present, at fairly low level we have to switch platform, so the rational, what we're doing is belling credentially information on a small number as highly credentialed candidates as we possibly can with the intension of moving those then into antibody development or deployment on conventional clinical assay system, alexis, whatever, then to do true clinical validation in thousands perhaps s of thousand of patient sample on a device that is meant for doing this.

We're not at the point today of being able to use the mass spectrometer in that fashion.

Are we going to eventually achieve that ability, I would say the answer to that is yes.

I think there are lots of assays today, particularly in the clinical realm for small molecule analysis where people are doing analysis of markers of metabolic and inborn areas of metabolism from dry blood spots looking at small molecule and it's all about the level.

If they're--if the analyte you're looking for is easily detectable with an ms assay can be configure.

It's going to be a bit more complex because mass spectrometers are harder to use with the vacuum system, more stuff to break but I think they have the enormous advantage of figuring the multiplex assay which is quite large.

So if you want to run not six things but a hundred things, at once, are for different diseases, in the same sample, it is possible to do that.

So they may have very specific advantages but today we're limited by what we detect when we do a direct capture.

Did I answer your question.

( inaudible ).

Let me--i'm not sure I fully understood your question.


So, just restate that please.

( inaudible ).

But not in the periphery ( inaudible ).

So, there is a high potential that some of the markers will be picked up and somebody just restated the question.

So if we go and use our oproach and find something that is highly differentially up regulated or down regulate and we go to the peripheral blood, what do we do well, I would say you don't give up.

The first thing is that nip ile asperric flu I or ductal lavage itself is a perfectly useful and readily obtainable clinical flu i. So if that difference that you saw in that sampling of real, then you can still potentially develop tests, it's just not going to be run--it's not going to use peripheral blood, now that's--that's not always going to be possible.

You may not have that kind of an accessible flu I that you make your discovery on.

I will say you would--for me I will decrease the priority of a particular marker because if they're not detectable it mean one of two things.

One, not detectable, put a finer point on that, not detectable by an mrm assay, I would put it on the list of, it was observed, it was highly differentially regulate, I don't an antibody available, i'm going to put that on a list for configuring an antiabout base.

If it's still not detectable then I would say it's not there and it never made it out of the periphery.

And this is a big danger realizing that one step back from the flu I and analyzing the tissues it.

Why don't you start by granding up the tissue reason we don't start there is because there are lots of thing in the tissue that differentially make it that never paik it out to the surroundings.

We prejudice ourselves for things that are likely to make it into the peripheral blood by the stuff that is in the fluid surrounding the site of the tumor.

But it's not going to work in every case.

I would like to thank our speaker.



Proteomic Bioinformatics Protein Identification Tools and Other Bioinformatic Tools.

See the Latest News in Proteomics

Get help and advice, or just discuss proteomics and related methods such as 2-D Electrophoresis at the

Proteomic Techniques Articles

In Gel Trypsin Digestion

2-D Gel Electrophoresis

Protein Separation

Protein Identification

Proteomics Frequently Asked Questions

Proteomics Biomarker Discovery

Proteomics Protocols

Proteomics Videos and Lectures

Siliac Study: Cell Signaling neurons

Related Gel Electrophoresis Methods

DNA Gel Purification

Southern Blot

DNA Footprinting

Maxam Gilbert Sequencing

Proteomics Forum Topics

what happen if a trypsin digestion time too long Hi, everyone, By an accident, a trypsin digested sample was left in incubator(37C) for 5 days. I was wondering what is the consequence for that in...
Are there any careers in protein sequencing or amino acid sequencing like Pro... I'm interested in getting involved in research involving protein or amino acid sequencing as a career, but I'm not sure where to look. could anyone...
You must REGISTER NOW to post a question in the Proteomics Forum. Login now if you have already registered.

Proteomics Newsletter

Yes! I Want to Learn the Latest in Proteomics-related Research and receive Email Alerts on new Proteomics products and services!
Don't Worry Your Email is Safe with Us. We hate Spam as Much as You Do.  You can unsubscribe at any time by clicking on unsubscribe on the bottom of Every Email!
First Name:

Science News

For science news click here:Science News