Tri-protein signature indicates prediabetes

This week, scientists at the University of Cambridge have found some markers in the bloodstream that can flag up individuals on the verge of developing diabetes. Intervening early in cases like this, can help patients to take action sooner to avoid the consequences, or even not develop diabetes at all. But existing, traditional techniques for doing this can be less reliable and they’re inefficient. The new approach can help to change that. With us is Julia Carrasco Zanini, who did the work.

Julia - There are a subset of people with pre-diabetes that can only be identified through a very inconvenient and time consuming test. This is called the oral glucose tolerance test, which essentially involves giving people a sugar drink and then, two hours afterwards, taking a blood sample and measuring their blood glucose levels. Now the new test that we are proposing would only require a one off blood sample to measure three proteins that would enable us to predict the outcome of this oral glucose tolerance test.

Chris - So these are chemicals that are washing around in these people's bloodstreams, and you're saying if we measure those, they can predict what would happen if we were to feed those people a whole heap of sugar and tell us if they were at risk of diabetes?

Julia - Yes. Those three proteins would then enable us to classify individuals that would be predicted to be at high risk of having this specific type of prediabetes.

Chris - How did you find the three chemicals? Because there are obviously millions of different molecules in the body. How did you hone in on those three?

Julia - So we have a really large study with over 11,000 people in which we performed this oral glucose tolerance test, but in whom we'd also taken a prior blood sample where we measured around 5,000 proteins circulating in their blood. So then the question really was, out of these 5,000 proteins, can we identify those markers that are very informative to do this prediction? Every single time that we want to make the prediction, we don't want to measure again 5,000 proteins. So then for this we developed a machine learning framework, which essentially enabled us to extract those three proteins that were considered the most predictive of the outcome of this oral glucose tolerance test.

Chris - I suppose you're sorting through a biochemical haystack here, aren't you? You're saying we've got millions of molecules we could consider. You honed in on 5,000 to start with. And you're just saying "what, of those, when the person has this situation where they're at risk of developing diabetes, which ones consistently are changed either up or down relative to a normal person." And those are the ones you then have found that are the most predictive?

Julia - Yes.

Chris - How reliable is it?

Julia - If we were able to implement it along with diabetes screening strategies such as the NHS diabetes prevention programme, we could actually reduce the number of oral glucose tolerance tests that would need to be performed to identify one of these people with pre-diabetes by half.

Chris - So it's quite a big difference. How long before a person develops diabetes would they go positive on your test?

Julia - So this is one of the things that would require further follow up studies and trials that are large enough specifically to assess the effectiveness of these three proteins.

Chris - There are lots of databases available now, aren't there? The UK Biobank is one where, over the last 20 years, people have given blood samples, had very close follow up, and we've got good histories on what happened to those people. Presumably you could go to big data sets like that and now begin to comb through those looking at the same molecules and test it that way.

Julia - Absolutely. I think it's really important to consider absolute risk in people that have high levels of these three proteins, yes.

Chris - So what sort of a difference could this make then if this did become policy and they did decide to turn this into a test? We can look at the practicalities of that in a moment, but what sort of difference could you make if this were implemented instead of the current practice?

Julia - It would enable us to identify, at the early stages, the diseased people who are at very high risk. And then of course this would be very beneficial because it is known that behavioural and lifestyle interventions are really effective in delaying and preventing the onset of type two diabetes.

Chris - You could give people preventative advice? So if I did this test and it found that I was at higher risk of developing diabetes, you are saying that would give me advanced notice and I could make changes that would reduce the chance of that happening?

Julia - Yes, absolutely. Because these people remain undetected for so long, they don't really know that they are at high risk and that they should make these behavioural modifications.

Chris - In those people, we're slamming the door after the horse has bolted. Whereas with what you are doing, you give people some warning beforehand so there's still time to intervene meaningfully.

Julia - Yes, absolutely.

Chris - Is it easy to detect these proteins? Because at the moment you've got a powerful laboratory to do this. If this were to be turned into a test, how practical is that?

Julia - So this is one consideration that has to be taken into account before thinking this could be implemented in clinical practice because how we measure these proteins in blood actually really matters. So in our study for example, we compared the measurements for these three proteins with two different technologies and we found good agreement between them. However, if we were to roll out this more widely, it is very likely that we would have to develop a single robust universal test that could be rolled out in a wide range of clinical settings.

Chris - Does it miss any cases? Because that's the other important thing, isn't it? It's very good to have a test that finds people really fast and gives us loads of lead time, but if it misses critical cases, that's dangerous.

Julia - I think this is one of the things that would really require a larger trial to be able to specifically say how effective are these three proteins and how many people they miss and how many people they identify. And ultimately the cost effectiveness of a strategy such as this for healthcare systems.