Bad Science: Does Metformin Inhibit DPP-4?

A study in the news a few weeks ago purports to show that Metformin inhibits DPP-4, the protease that cleaves GLP-1. If this were true, it would be a major concern to people with diabetes, because DPP-4 is a known tumor suppressor and inhibiting it increases the likelihood that cancerous cells will proliferate without being destroyed by the immune system. Since the study also found that this supposed effect was suppressed when metformin was taken along with a meal, you might think I was going to suggest you protect yourself by not taking metformin in the fasting state.

But that would be only if I hadn't reviewed the study, which breaks new ground for published awfulness. In fact, I've seen better designed research presented at Junior High School science fairs.

Here's the abstract of the study.

Investigation of the effect of oral metformin on dipeptidylpeptidase-4 (DPP-4) activity in Type 2 diabetes. Cuthbertson, et al. Diabetic Medicine Volume 26, Number 6, June 2009 , pp. 649-654(6)

I don't have access to the full text, and neither does Diabetes in Control, which reported on this study, citing only the information in the abstract. But what a train wreck of an abstract it is!

The entire methods section is this:

Ten subjects with Type 2 diabetes (6 male/4 female, age 65.8 ± 2.6 years, body mass index 30.0 ± 1.2 kg/m2, glycated haemoglobin (HbA1c) 6.3 ± 0.2%, mean ± sem) received metformin 1 g orally or placebo together with a standard mixed meal (SMM) in a random crossover design. Six subjects re-attended fasting and received metformin 1 g without a SMM. [Emphasis mine]

They start with a whopping ten subjects and then compare them to a subgroup of six subjects taken from the original group of ten. Note that the metformin was taken with the meal first and then fasting second. There is no indication if these subjects were on metformin before the study or if these were the the first doses of metformin these subjects had taken. This is significant as metformin's impact increases with each dose a naive user takes and it takes about a week to start showing many of its physiological effects.

Note also that there is no explanation of what was measured in these subjects, when it was measured, or, for that matter, why.

The "Results" section is not much more informative. The researchers report that:

DPP-4 activity was not suppressed by metformin compared with placebo [area under curve (AUC)0-4 h 1574 ± 4 vs. 1581 ± 8 μmol/ml/min, respectively]. Plasma glucose, insulin and active GLP-1 were not different. However, DPP-4 activity was suppressed with metformin following fasting compared with a SMM (n = 6) (AUC0-4 h 1578 ± 4 vs. 1494 ± 9 μmol/min, P < 0.02). "

Ahem, AUC0-4 h of what? "DPP-4 Activity" is not something you measure with a blood draw. DPP-4 concentration would be, but not activity. And since "DPP-4 activity" is defined nowhere in this abstract we are left scratching our heads as to what exactly they measured. And what "placebo" are they talking about? I don't see a placebo described anywhere in the methods section.

As far as I can figure from this extremely poorly written abstract the finding the researchers and peer reviewers considered worthy of publication was a 5% difference in an average of some undefined "Area Under the Curve" between the larger group--which ate a meal with their metformin and a subgroup which took metformin fasting.

Did they assess DPP-4 "activity" by working statistical magic on some mix of measured GLP-1 levels, insulin levels and glucose levels which are what they mention measuring? If so, you would want to ask yourself why this would be a meaningful measure of "DPP-4 activity?" There may well be other factors that affect GLP-1 levels, insulin, and glucose that have nothing to do with DPP-4, and indeed there are.

They also note that "Metformin serum levels were significantly lower (P < 0.001) after SMM than fasting (AUC0-4 h 350 ± 66 vs. 457 ± 55 mg/ml/min)," though what this is supposed to mean in relation to their other reported findings is completely unexplained.

But it really doesn't matter what these researchers report, because, folks, when you are comparing an average of a measurement taken in a group of 10 people to an average of a measurement taken in 6 people who form a subset of that original 10 people, you can run all the statistical software you want, and come out with numbers to ten places of significance, but your result is still junk.

Your sample size is too small to yield averages that eliminate the impact of individual variations.

To understand this better consider this hypothetical example: Let's say you are studying whether eating cheese sandwiches for lunch makes people wealthy. You start with ten people, five of whom are corporate executives and five of whom are janitors.

You feed them something other than a cheese sandwich for one lunch and average their income over the next month. The next month you call back six of the original ten subjects, randomly chosen, and end up with a subgroup where five are corporate executives and one is a janitor. You feed them cheese sandwiches for lunch, and by golly, a month later you see a dramatic difference in average income between the two groups. Have you proven that eating cheese sandwiches for lunch raises income? Of course not.

Your "result" here is caused entirely by differences between the larger and smaller group which become more significant because the group was so tiny to start with and you've reduced it to a size where two or three people's individual variations can have a huge impact on your average. In the metformin study the "finding" is a difference of 5% between the two groups, a difference in an average no less, which makes it statistically significant only if you ignore everything but the results displayed by your statistical software which is too polite to tell you that only an imbecile would run statistics in this kind of situation.

In fact, the result in the smaller, fasting subgroup which is what the researchers considered worth a journal publication may differ because that group may have selected, from the original group, more people with high natural GLP-1--or lower DPP-4--production. The second subgroup may have had different average insulin producing capability. There might be differences in the degree to which the subjects digest metformin. There may be differences in their body size which affects all these parameters. There may be huge differences in how responsive their livers are to metformin. With the study design here, we will never know.

But the main point is that with a tiny study population like this one, if you compare any average measurement made in the total group to a measurement taken in a significantly smaller subgroup, any result is junk.

And that doesn't even get into the question of what did this study do to determine that inhibition of DPP-4 was what produced its reported result, if that result were, miraculously, not a statistical artifact.

So this study tells us nothing about whether metformin inhibits DPP-4 and whether this effect is wiped out by administering metformin with food.

In case you wonder if there is any research on this subject that is more useful, here's an old study that looked at this question in a more scientific way:


Metformin effects on dipeptidylpeptidase IV degradation of glucagon-like peptide-1. Hinke SA et al. Biochem Biophys Res Commun. 2002 Mar 15;291(5):1302-8.

Note that this old study cites other studies that have found that Metformin increases GLP-1, so it is quite possible it does. But there is no reason to believe that GLP-1 is only increased by inhibiting DPP-4. It is quite possible you could increase GLP-1 by doing something to stimulate the secretion of GLP-1 for a longer time period after eating. And it's worth noting that this older second study could not find any evidence that, based on sophisticated in vitro examination, metformin had the biochemical ability to inhibit DPP-4.