By Tony Fong
Name: Colin Creaser
Position: Professor, analytical chemistry and head of the Centre for Analytical Science, Loughborough University, July 2007 to present
Background: Professor and associate dean for research and graduate studies, Nottingham Trent University, May 1993 to June 2007; lecturer in analytical chemistry, University of East Anglia, September 1984 to May 1993; senior scientific officer, UKAEA Harwell Laboratory, June 1978 to August 1984
As sports doping has become increasingly sophisticated and new designer drugs escape detection by conventional screening and testing methods, the field is exploring new technologies to combat the use of illegal performance-enhancing drugs.
In two recently published papers, researchers in the UK describe a method combining mass spectrometry and bioinformatics to discover, and then validate, a biomarker that they said can be used as part of a panel of biomarkers for detecting growth-hormone use by athletes.
In the studies, the researchers used both MALDI-MS and LC-MS as well as artificial neural networks and identified six protein ions and six peptide ions capable of differentiating serum samples of people who were administered human growth hormone from samples of control subjects. The peptide ion with the "highest significance," was associated with leucine-rich alpha2 glycoprotein, they wrote.
They then tested LRG as a candidate biomarker for growth hormone use, combining it with insulin-like growth factor 1, a known biomarker for recombinant human growth hormone abuse. "Combining IGF-1 and LRG data improved the separation of treated and placebo states compared with IGF-1 alone, further strengthening the hypothesis that LRG is a biomarker of rhGH administration," the researchers said.
ProteoMonitor spoke with Colin Creaser, a professor of analytical chemistry at Loughborough University in the UK, and the corresponding author on the Proteomics: Clinical Applications study, this week. Below is an edited transcript of the conversation.
Can you describe the issues around these new designer drugs and the need for new methods detecting them?
The research that we've been doing on the growth hormone abuse really comes out of current difficulties in detecting growth-hormone abuse in athletes. First, the GH has a very short half-life so that levels drop to basal levels within about 24 hours of administration. Secondly, it's very pulsatile and changes concentration very dramatically, for example, up to 70-fold following exercise.
So the detection of GH directly is problematic because it would be difficult to distinguish exogenous abuse from endogenous changes. The interest has been to look at downstream biomarkers of growth-hormone administration. A number of those have been identified. One in particular is IGF-1 protein, which is a downstream one, and that does act as a good marker for growth-hormone administration and is not subject to, for example, changes as a result of exercise.
But on its own, unfortunately, it's still not clear enough [of] a biomarker to identify growth-hormone administration [because] some individuals have high IGF levels.
What we were trying to do with this work was to see if we could identify additional biomarkers, which would help add confidence to the downstream detection of growth-hormone administration.
[ pagebreak ]
The work that we did here is an extension of work, which originally started looking at cancer patients, designed to develop an integrated mass spectrometry/bioinformatics strategy for proteomic biomarker detection, discovery, and validation.
In this work we were interested in growth-hormone administration. We've also been interested in gene doping in sports. … What we've actually been doing in our research for a number of years is to try and develop an effective yet generic methodology for biomarker detection, which could be applied to a variety of different applications.
It could be applied to the diagnosis of biomarkers of cancer, for example. We published on discriminating metastatic melanoma patients from healthy controls using an earlier version of [this] methodology, which is obviously undergoing developing work.
Briefly describe the method you developed.
The method basically starts by taking serum samples [but] it can be applied to other fluids, as well. We carry out a mass spectrometric profiling of those, and that can be done using either matrix-assisted laser desorption/ionization MALDI or electrospray [ionization] mass spectrometry.
In our early work, we were MALDI based, and later on we've been moving toward liquid chromatography electrospray mass spectrometry.
We were looking toward developing rapid assays. MALDI is potentially a very rapid technique, whereas in the early part of the work … we actually had an LC-MS run, which took over an hour. So, for high-throughput screening, it was really not suitable for the time.
Subsequently … we've changed that one-hour LC-MS run down to a five-minute UPLC run, and so now that we can do fast LC-MS analyses, the timescale is fast enough to be useful in high-throughput diagnosis and diagnostics. We're moving more toward the LC-ESI route, but we're still doing MALDI work, as well.
So, we obtain a mass-spectrometry profile, and then we apply bioinformatics to [the data], and the approach that we've been using in collaboration with my colleague Graham Ball at Nottingham Trent University is to use artificial neural networks. We've used that as opposed to more traditional techniques such as principal components analysis because the ANN approach is very good at looking at very large numbers of data points, which you get from mass spectrometry data, and can handle things such as nonlinear changes in concentrations.
We found this technique to be a very powerful way of looking through very large datasets to try to look for mass spectrometry ions [that] are either up-regulated or down-regulated as a result of some sort of change, such as growth-hormone administration.
When we do that, what we do is effectively reduce the number of ions down to a relatively small number. In the case of this GH work … we had six ions, which between them provided a good diagnostic test for the presence of growth-hormone administration. So what we have is what we call a bioinformatics-directed biomarker discovery route that [resulted in] a very small number of ions. And we then look at those in more detail to try to obtain sequence information.
In this respect, we're different from the bottom-up proteomics approach, where you're trying to make comparisons of very large number of proteins. Here, we're trying to use our bioinformatics to reduce the number of proteins to a small number of targeted ones [that] we try to identify by peptide sequencing.
There are other mass spec-based or proteomics-based approaches in the area of dope-testing. What is unique about your approach?
The key difference, as far as we're concerned, is the use of artificial neural networks to help us pick out the ions of interest, rather than a large screening approach where we try to identify as many proteins as we can in control and some other state samples.
[ pagebreak ]
Can you describe this artificial neural network?
Broadly speaking it works a bit like a brain. … What you basically do is put data into the system and you ask it a question, effectively a 'Yes' or 'No' question: Is somebody hormone treated or not hormone treated?
The first data set is a training set, so you present data, which are in growth-hormone or non-growth-hormone- treated states, to the ANN and you ask it to look for differences between them. It starts by looking for a difference, and if it finds a difference, it strengthens the links that show up and emphasizes that particular difference. And then it goes on to look for other differences in other samples.
What it's basically doing is learning and getting better as it's being trained.
This is different from a PCA approach … where you put all the data in and you get one result from that. There's no learning process [as] you get with an ANN, and therefore the models as they run get better and better at predictions.
Is your method looking for markers that would be a definitive proof that someone has doped, or is it looking for a biological change that only may suggest that someone is guilty of doping, but could also be the result of something else?
Essentially, our approach is non-hypothesis-based. We're making no prior assumptions about what might be changing or the significance of the change in a protein from our samples. And that's a very important element of what we do, making no prior assumptions, whatsoever.
We're just asking the ANN, effectively, to tell us what changes have occurred, and then to identify the proteins that come from it. That gives us, first, mass spectrometric ions, which in themselves could be part of a diagnostic test, but I think the acceptance of something that just has ions rather than identities by the community is problematic.
Obviously, we [have to] proceed on, first, to enzymes and then to find the sequence of a peptide generating those ions, and then obviously to the protein identification. And that constitutes, if you like, the discovery phase of our work. And then we need to do a validation phase.
Our two papers … the first one in Proteomics: Clinkical Applications is essentially the discovery phase, and the second one in Rapid Communications in Mass Spectrometry is the validation phase. And taken together, those papers represent the full cycle from the initial samples through discovery and into validation.
This is the first time we've demonstrated the complete loop.
What this second paper has done is actually try to do some high-quality quantitative measurements, by developing a high-throughput method using ultra-performance liquid chromatography, which allows us to reduce the time.
But also using the technical isotope dilution mass spectrometry in order to get accurate quantitative measurements… my background is in small molecule work and in small molecule work, for example, in detection of abuse of steroids, the gold standard method is to use the isotopic dilution mass spectrometry approach, whereas an immunoassay will be viewed as a screening method and not a definitive one.
We've tried to use the same small molecule approach that's used regularly [and] routinely in doping laboratories. And we extend that to these protein biomarkers.
How confident are you in your findings?
I think the point, which we wanted to make from the work that we've done, is that you can identify a biomarker such as IGF-1, or in our case the leucine-rich alpha2 glycoprotein. But on their own, these biomarkers may not give sufficient confidence of the evidence of doping.
Our essential thesis is that actually you want to be able to combine together more than one biomarker, perhaps panels of biomarkers that between them will give enhanced confidence in the detection of GH administration.
[ pagebreak ]
What we've shown in our papers is that if you take the concentrations of IGF-1 for a placebo group as opposed to a GH-administration [group] you can see that the IGF-1 rises, but, unfortunately, it's not definitive because some individuals have high IGF-1 levels.
If you take the IGF-1 and LRG data together, then you improve confidence in the identification of GH administration. In every case, you get improved confidence by taking, in this case, a panel of two biomarkers together. Currently, our view is if you take several biomarkers together, and measure them at the same time, then the confidence would increase further.
Going back to my earlier question about whether you're detecting actual doping or biological changes that may suggest doping, it sounds like you're saying that even with a multi-panel test that you would never be able to get 100 percent confidence that doping occurred because ultimately, you're still detecting only a biological change, not the presence of a doping agent.
I don't think that we would necessarily never reach 100 percent, but the point is that the more biomarkers that you look [at] together, the more confident [you] can be in the results of the data. Two markers are better than one, three is probably going to be better than two, four better than three, and so on.
Our current work is directed toward looking at that bigger panel of biomarkers, having established this principle.
Is this panel composed of these six peptide ions you identified in the Proteomics: Clinical Applications paper?
No, the panel of biomarkers are proteins or peptides derived from proteins that have been implicated in growth-hormone administration, so LRG and IGF would be amongst them but there are other markers that have been used and been implicated in growth-hormone administration.
The idea is to be looking at a selection of proteins by monitoring and quantifying their tryptic peptides.
Do any of the other five peptide ions have promise as a candidate for a doping marker?
Our belief, and the reason we're continuing this work, is that taken together, they would be. So if you add in procollagen 3 N-terminal peptide — which is one of the ones that we've added to the list because it's been implicated in growth-hormone administration — that it would enhance our confidence and added to the data that we get from IGF-1 and added that to the [data] from the LRG protein.
Is the goal to translate this into an immunoassay?
No. … From the community that I come from and the data that's used in the detection of small molecule drugs, such as steroids, the gold standard is viewed to be the LC-MS approach using isotopic dilution, and the screening is the immunoassay.
So you can screen with an immunoassay and then move onto your gold-standard confirmation. The problem is if you're looking for not one protein, but, say, six proteins, then you might have to use six separate immunoassays, which is relatively time-consuming, whereas you can actually look for six proteins simultaneously in the same five-minute UPLC run by targeting these specific proteins.
How it might work out in the end, we don't know, but we're really working that out by using what is a fairly established LC-MS approach, it might be possible to screen for a panel of proteins very rapidly as an alternative to using a number of immunoassays.
We may be prejudiced because we come from the mass spectrometry community and because it's so widely used for small molecule drugs of abuse, it seems to us to be grounds for believing that perhaps in the future fast LC-MS assays using isotopic dilution mass spectrometry may be the way forward.