Is Refined Carbohydrate Addictive?

[Note: in previous versions, I mixed up "LGI" and "HGI" terms in a couple of spots.  These are now corrected.  Thanks to readers for pointing them out.]

Recently, a new study was published that triggered an avalanche of media reports suggesting that refined carbohydrate may be addictive:

Refined Carbs May Trigger Food Addiction
Refined Carbs May Trigger Food Addictions
Can You be Addicted to Carbs?
etc.

This makes for attention-grabbing headlines, but in fact the study had virtually nothing to do with food addiction.  The study made no attempt to measure addictive behavior related to refined carbohydrate or any other food, nor did it aim to do so.

So what did the study actually find, why is it being extrapolated to food addiction, and is this a reasonable extrapolation?  Answering these questions dredges up a number of interesting scientific points, some of which undermine popular notions of what determines eating behavior.

Before we dive in to the study, I want to take a moment to discuss food addiction.  In common parlance, if we really like a food, we're motivated to get it, or we tend to eat more of it than we think we should, we often say we're "addicted" to it.  This is not quite accurate, but it does point toward an important truth.  Clinical food addiction is a serious phenomenon, akin to drug or gambling addiction, that only affects a few percent of the population.  Food is something we have to consume to survive, which makes it qualitatively different from drugs of abuse, although of course we don't have to consume Slurpees.

Addiction is a pathological over-activation of the reward system in response to a specific stimulus (e.g. crack cocaine), resulting in an enhanced level of motivation to obtain that stimulus, to a point where the behavior is damaging to a person's life and/or health.  Some have argued that obesity cannot be defined as a state of food addiction, though binge eating may qualify (1).  I think that's a reasonable perspective.

The more important point is that you don't have to be clinically addicted to a food to overconsume it.  Addiction is a pathological over-activation of the reward system, but there are many shades of gray between not caring about a food and being addicted to it.  Ordering chocolate cake and ice cream after a large meal doesn't mean you're addicted to chocolate cake and ice cream, it just means it's highly rewarding.  Your brain isn't broken; it's doing what it's supposed to do when it encounters a calorie-dense food rich in fat, sugar, and starch.  So while most people may not be literally addicted to food, they do experience an enhanced motivation to eat certain foods that works via the same brain systems that mediate addiction.

The Study

The paper was published by Dr. David Ludwig's group in the American Journal of Clinical Nutrition (2).  In a randomized crossover design, they gave 12 overweight or obese men a beverage that was either high glycemic index (HGI) or low glycemic index (LGI), meaning that the former caused a larger increase in blood glucose than the latter.  Then, they measured blood glucose and hunger over a 5-hour period.  At the 4-hour timepoint, they measured changes in brain blood flow (an indirect indicator of activity) in 25 different brain regions by fMRI.

As expected, people consuming the HGI beverage had a greater increase in blood glucose.  At 5 hours, the HGI group was back to baseline blood glucose, while the LGI group was still modestly elevated, with a small difference between groups.  Hunger ratings were higher in the HGI than LGI group.  This is consistent with some studies showing that lower glycemic carbohydrate foods tend to be more satiating at a single meal.  There are notable exceptions to this trend, such as potatoes, which are one of the most glycemic foods and also one of the most satiating (2B).

The key finding of the paper is that activity of the nucleus accumbens, a brain region central to reward processing and addiction, was higher in the HGI group than the LGI group at 4 hours, a timepoint at which there were differences in blood glucose levels and hunger between groups.

The authors suggest in the discussion that the drop in blood glucose in the HGI group after hitting peak glucose is responsible for the increased hunger, and activation of the NAc, which they interpret as an increase in the subjects' desire for food:

The decline in blood glucose (and other metabolic fuels) in the late postprandial period after a high-GI meal would not only constitute a powerful homeostatic hunger signal but also increase the hedonic value of food through striatal activation. This combination of physiologic events may foster food cravings with a special preference for high-GI carbohydrates, thereby propagating cycles of overeating

This interpretation goes far beyond the findings of the paper itself, and is based on a number of assumptions, which I'll examine below.

Discussion

One of the most pervasive ideas in the diet-health sphere, and to some extent even in the scientific literature, is that when you eat rapidly digesting carbohydrate, this causes an increase in blood glucose and insulin, followed by hypoglycemia (the "crash").  This low level of blood glucose then causes hunger and cravings.  It's a simple, logical idea that has virtually no scientific support.

Hypoglycemia definitely causes hunger and food cravings (3).  That isn't in question.  The problem is that the level of hypoglycemia that causes hunger (below 51-65 mg/dL depending on the study; 3, 3B) is rare in non-diabetics, no matter what kind of carbohydrate they eat.  For example,  Dr. Ludwig's study reported that participants had an average fasting glucose of 88 mg/dL.  This is a normal level of fasting glucose.  At the lowest level of the HGI group's post-meal "crash", they had a blood glucose of 85 mg/dL-- a trivial 3 mg/dL below fasting, and nowhere near hypoglycemic.  The LGI group had a blood glucose of 95 mg/dL, only 10 mg/dL higher than the HGI group.  Could a difference in blood glucose this small account for the measured differences in hunger?  Probably not, given that previous studies have struggled to find satiating effects of much higher levels of blood glucose in humans.

The unfortunate reality is that there is no compelling evidence that blood glucose fluctuations within the normal range cause differences in hunger and food intake.  When you get into the abnormal range of true clinical hypoglycemia and hyperglycemia, then you see big effects, but fluctuations within the range of what are observed when non-diabetic people eat food have not been convincingly linked to hunger, craving, or food intake, despite quite a bit of work in the area.  Here is a quote from a 2006 review paper on glucose-sensing neurons (4):

There is no question that severe [hypoglycemia] can stimulate feeding and that high concentrations of glucose placed in the brain can terminate feeding. However, it is uncertain whether any of these findings implicate glucose as a primary mediator of meal to meal intake under physiological conditions... In fact, no one has ever demonstrated that meal initiation or termination can be manipulated by altering brain glucose levels within the limits found during normal ingestion.

That doesn't necessarily mean this hypothesis is incorrect-- maybe future experiments will support it when we get better research tools-- but the hypothesis is often assumed to be correct in the media and even the scientific literature, despite a lack of clear evidence to support it.  This is putting the cart before the horse.

There is another hypothesis out there, proposing that it's not the absolute level of glucose that matters, but rapid changes.  Jenny Ruhl advocates for this idea in her book The Truth About Low Carb Diets.  Again, this is a logical hypothesis, and I would love for it to be correct because it's so simple, but it has not received much convincing support from science at this point.  To be fair, it's an exceedingly difficult question to answer properly, so it's possible that we just haven't found the right way to test the hypothesis yet.  Ruhl bases her hypothesis on observations she and other people have collected using their home blood glucose meters.

It is true that there are glucose-sensing neurons, and that they can impact hunger and food intake in response to glucose levels under some circumstances (e.g., true hypoglycemia).  These neurons often integrate a variety of signals of energy status (e.g. leptin, ghrelin, insulin, CCK, glucose, fatty acids) to determine hunger state, which is why they're sometimes called "metabolic sensing neurons" (4).  Glucose probably plays some role in hunger, but it may simply be too small to detect under normal circumstances because it's one of many signals these neurons integrate.  It's possible that the effect of glucose is heightened in certain metabolic contexts, but this remains to be tested.  If I had to speculate based my own personal gut feeling, I'd say that blood glucose (or some variable related to cellular glucose availability) probably does play some role in hunger and food intake, and that this effect is probably heightened in certain individuals.  However, that idea remains in the realm of logical speculation-- not fact.

This brings us to an additional point.  We see what we can measure, and we tend to base explanations on what we see.  The LGI and HGI meals caused a number of physiological effects, both measured and unmeasured.  For example, if blood glucose remained elevated in the LGI group at the four hour timepoint, that means glucose was presumably still present in the gut and being transported from the gut into the bloodstream, even though that wasn't directly measured.  Since we know that there are glucose receptors in the gut, and that these play a role in satiety, this presents a (in my opinion) more compelling explanation for the increased satiety they observed.  Longer residence time of glucose in the small intestine = prolonged satiety.

Now let's discuss the fMRI finding.  The nucleus accumbens (NAc) is intimately involved in reward processes, including food reward.  They showed that at 4 hours, the HGI group had a higher activity of the NAc than the LGI group.  This finding is interesting from an academic perspective, but it's difficult to interpret.  Normally, when you want to measure reward system activation, you present some sort of stimulus to the subject.  For example, as my colleague Dr. Ellen Schur does (5A), you present people images of tasty foods, average foods, and non-food items, and you see how brain activity differs between those conditions.  Then you know that the brain responses were specifically to food cues.  In contrast, in the current study, subjects were sitting in the fMRI machine getting scanned without any kind of stimulus.

So what does it mean for a person to be sitting in an fMRI machine and his NAc is lighting up spontaneously without any sort of cue?  Does it mean the brain has received a reward and is processing it; does it mean the brain is seeking an additional reward; or something else entirely?  Does it mean the subject wants to drive a fast car, eat a hamburger, or sleep with a beautiful woman?  No one knows!  The point is that we have no idea if this spontaneously increased NAc activity is causing an increased desire for food, favors addiction, etc.

If I had to offer an explanation for this result, I'd propose that prolonged small intestine exposure to glucose in the LGI group led to increased satiety, which led to decreased NAc activity relative to the HGI group.  We know that the satiety system suppresses food reward perception, which is why we aren't as interested in food when we're full.  This explanation is still speculative, but it rests on mechanisms that have been clearly supported by research in other contexts.

Practical Implications

Before moving on, I'll restate that this study has value in an academic sense-- it expands our understanding of the relationship between food intake, hunger, and brain activity.  But the practical implications, as suggested by news articles and the study's discussion section, are murky.

First of all, most meals don't differ in glycemic index as much as these two did (2.3-fold difference).  They were specifically designed to create a very large difference in glycemic index-- a difference that would be difficult to achieve in real life.

Second, as I've pointed out many times, the glycemic index is not a reliable indicator of the satiating effect of foods.  The white potato has one of the highest glycemic indices of any food, yet it's also highly satiating per calorie (5).  No one becomes addicted to plain potatoes, despite the high glycemic index, nor do people generally crave or overeat plain potatoes.  Now, put some sour cream, bacon, chives, and salt on that potato, and there's something people will crave and overeat.  Is that because of the glycemic index, which declined quite a bit after you put the toppings on?  Or something else about that food?

The glycemic index also doesn't correspond very well with how refined a food is.  For example, white bread and whole wheat bread have approximately the same glycemic index.  White pasta and whole wheat pasta do too.  So do white rice and brown rice.  If you think there's a difference in the health impact of these refined vs. unrefined foods, glycemic index doesn't capture it.

The biggest problem with the glycemic index literature (in my opinion) is one that does not apply to this particular study, which was relatively well controlled.  Many studies on the health impacts of the "glycemic index" compare diets that differ from one another in many ways.  If you feed one group a diet consisting of beans, nuts and fruit, and another group a diet of white bread, pastries and soda, you can call the diets "low GI" and "high GI", but that doesn't mean the effects you observed were attributable to differences in blood glucose.  There are so many other differences between the two diets that it's impossible to know what caused the effects you observed.  Yet these types of studies are frequently used to support the hypothesis that differences in blood glucose influence satiety, body weight, and health.

Another big problem is that the glycemic index literature overall contains little support for the idea that low glycemic index eating styles have a significant long-term effect on body weight and metabolic health.  If the authors' hypothesis is correct that high glycemic index foods "foster food cravings with a special preference for high-GI carbohydrates, thereby propagating cycles of overeating", then this should be observed in long-term controlled trials comparing LGI with HGI eating styles.  However, most studies over 10 weeks long have shown little or no effect on these outcomes in non-diabetic individuals (6), with the three longest trials (4, 6, and 18 months) showing no real advantage of LGI eating on body weight or glucose metabolism (7, 8, 9).  The large Diogenes study did find that LGI eating modestly slows weight regain following calorie restriction-induced weight loss, though the study wasn't really designed to isolate the glycemic index from other food properties like fiber content (10).  At least this shows that the advice to reduce glycemic index has a measurable impact in certain contexts, even if the impact is small.

Despite the unimpressive performance in controlled trials, the glycemic index is enthusiastically presented as an effective way to prevent and treat obesity and metabolic dysfunction in non-diabetics.  This is based largely on observational studies, which have sometimes found that HGI eating is associated with obesity and diabetes risk.  While I don't totally dismiss the findings of these studies, I think they're hard to draw firm conclusions from, given their inherent limitations and the divergent results of controlled trials.

Conclusion

Although it adds meaningfully to the literature on food consumption, hunger, and brain activity, this study did not support the conclusion that "refined carbs are addictive", nor did it show that differences in blood glucose within the normal range cause hunger or food craving, nor that a high glycemic index per se contributes to obesity.  Certain refined carbohydrate foods are indeed highly rewarding, and can lead to craving and overconsumption, but whether this relates specifically to their effects on blood glucose remains unclear.  I look forward to research clearly supporting the idea that changes in blood glucose within the normal range have a significant impact on hunger, satiety, food craving, obesity, and addiction.