I'd like to begin by emphasizing that carbohydrate restriction has helped many people lose body fat and improve their metabolic health. Although it doesn't work for everyone, there is no doubt that carbohydrate restriction causes fat loss in many, perhaps even most obese people. For a subset of people, the results can be very impressive. I consider that to be a fact at this point, but that's not what I'll be discussing here.
What I want to discuss is a hypothesis. It's the idea, championed by Gary Taubes, that carbohydrate (particularly refined carbohydrate) is the primary cause of common obesity due to its ability to elevate insulin, thereby causing increased fat storage in fat cells. To demonstrate that I'm representing this hypothesis accurately, here is a quote from his book Good Calories, Bad Calories:
This alternative hypothesis of obesity constitutes three distinct propositions. First, as I've said, is the basic proposition that obesity is caused by a regulatory defect in fat metabolism, and so a defect in the distribution of energy rather than an imbalance of energy intake and expenditure. The second is that insulin plays a primary role in this fattening process, and the compensatory behaviors of hunger and lethargy. The third is that carbohydrates, and particularly refined carbohydrates-- and perhaps the fructose content as well, and thus perhaps the amount of sugars consumed-- are the prime suspects in the chronic elevation of insulin; hence, they are the ultimate cause of common obesity.There are three parts to this idea. I'll discuss them each separately.
Part I: A Defect of Fat Metabolism?
The first part of this hypothesis states that energy balance is not the ultimate cause of fat gain, it's the proximal cause. That is, Taubes is not disagreeing with the first law of thermodynamics: he understands that fat accumulation depends on how much energy is entering the body vs. leaving it. However, he feels that the entire industrialized world didn't just wake up one morning and decide to eat more calories, therefore something must be driving the increased calorie consumption.
He cited the research of Drs. Jules Hirsch and Rudy Leibel, various underfeeding and overfeeding studies, lipectomy studies, and evidence from genetically obese rodents, to demonstrate that body fatness is biologically regulated rather than being the passive result of voluntary food intake and exercise behaviors. He then advances the idea that it's an alteration in this body fat regulatory system that is behind obesity. This may sound familiar because I've written about it several times. So far, so good.
This is where he should have mentioned leptin signaling, and the circuits in the brain that regulate body fat mass, which would have taken the book in a more compelling direction. According to literally thousands of publications spanning nearly two centuries, the brain is the only organ that is known to regulate body fat mass in humans and other animals-- neither fat tissue itself, nor the insulin-secreting pancreas have the ability to regulate body fat mass as far as we currently know. Leptin is the system that Drs. Jules Hirsch and Rudy Leibel have shown in carefully controlled human studies is responsible for the metabolic defect Taubes alluded to (1). It's also the system that is mutated in the genetically obese rodents he discusses (2, 3). Yet it receives no mention in the book. This is a fork in the road, where Taubes discards a solid hypothesis in favor of a shaky one.
Part II: The Role of Insulin in Body Fatness
Insulin has many functions throughout the body. The primary role of insulin is to manage circulating concentrations of nutrients (principally glucose, amino acids, and fatty acids, the body's three main fuels), keeping them within an optimal range, and coordinating the shift between metabolic fuels that is required when a person consumes more of one or the other. Any time insulin suppresses fat burning, it increases carbohydrate and/or protein burning by an equivalent amount. That is what insulin does.
Insulin has a number of actions on fat and lean tissues that favor fat storage and suppress fat burning, and this is the crux of Taubes's basic argument in support of the idea that insulin causes fat accumulation. Some of these actions have been recognized for many decades. Taubes's idea is so simple, you might think someone had already thought of it. In fact, the idea has been around for a long time, but it has very little traction among obesity researchers today because it doesn't fit with a variety of basic observations, as I will explain.
The reason insulin suppresses fat burning is because it's a signal of glucose abundance. It's telling tissues to stop burning fat because carbohydrate is the available fuel. If you eat a meal of 500 calories of carbohydrate, you will burn that carbohydrate under the direction of insulin, which will also make sure body fat mostly stays inside your fat cells during the process. If you eat a meal of 500 calories of fat, you will burn fat instead of carbohydrate, but since you just ate fat, you aren't dipping into your body fat stores any more than you were when you ate carbohydrate. So even though insulin temporarily suppresses fat burning and the release of fat from fat cells when you eat carbohydrate, at the end of the day if you ate the same number of calories you end up with the same amount of fat in your fat cells either way. You now know more about insulin than many popular diet gurus.
As we are all on the same page (I hope) that the first law of thermodynamics applies to humans, for insulin to cause fat gain, it must either increase energy intake, decrease energy expenditure, or both. Let's see if that's true.
Let's look at the effect of insulin on food intake. To keep it as realistic as possible, let's compare satiety and subsequent food intake among foods that raise insulin to varying degrees. If calories and protein are kept the same, high-carbohydrate meals cause equal or greater satiety than high-fat meals, and equal or less subsequent food intake, despite a much larger insulin response (4, 5, 6, 7). Due to the insulin-stimulating effect of protein, low-carbohydrate high-protein meals can sometimes stimulate insulin to an equal or greater degree than high-carbohydrate meals, yet even in these cases higher insulin release is associated with increased satiety (8). Experiments in which investigators feed volunteers protein foods that stimulate insulin to different degrees show that the amount of satiety is positively correlated with the degree of insulin release (9), which is not consistent with the idea that insulin stimulates food intake. In the long term, low-carbohydrate diets suppress appetite in many overweight/obese people, however this is unlikely to be related to insulin.
If elevated insulin leads to increased fat storage and increased food intake, then experimentally elevating insulin in animals should replicate this (since insulin acts on fat cells in the same manner in humans and non-human mammals). However, this is not observed. Insulin injections at a dose that does not cause frank hypoglycemia do not increase food intake, and in some cases they even reduce it (48). Chronically increasing circulating insulin without causing hypoglycemia reduces food intake and body weight in non-diabetic animals, without causing illness, contrary to what this idea would predict (49, 50). If anything, insulin constrains food intake and body fatness, and research indicates that this action occurs via the brain. Insulin infused into the brains of baboons causes a suppression of appetite and fat loss, which is consistent with the fact that insulin and leptin have overlapping functions in the brain (10, 11). Knocking out insulin receptors in the brain leads to increased fat mass in rodents, suggesting that its normal function involves constraining fat mass (12). Insulin is also co-secreted with amylin, which suppresses food intake and body weight (13). This is why insulin is viewed by some obesity researchers as an anti-obesity hormone.
Now let's look at energy expenditure. If insulin is increasing fat accumulation due to a decrease in energy expenditure (presumably because elevated insulin is locking fat away inside fat cells), then people with higher fasting insulin should have lower resting energy expenditure. Lucky for us, that hypothesis has been tested. At least two studies have shown that higher fasting insulin is associated with a higher resting energy expenditure, independent of body fatness, not a lower expenditure (14, 15). If anything, this is the opposite of what the hypothesis would predict. How about post-meal insulin spikes due to eating carbohydrate? A number of studies have consistently shown that under isocaloric controlled conditions, substantially different carbohydrate:fat ratios do not influence energy expenditure in any measurable way, even over long periods of time (16, 17).
Therefore, if insulin doesn't increase energy intake (if anything, the combination of insulin and amylin that the pancreas releases in response to carbohydrate decreases it), and doesn't decrease energy expenditure, then how exactly is it supposed to cause energy accumulation in the body as fat? There is no energy fairy. Obese people are obese despite having higher fasting insulin, not because of it. The fact is, insulin spikes after meals temporarily decrease fat release from fat cells, but if you look at total 24 hour energy balance, insulin spikes, in conjunction with all the other hormones that are released in response to food ingestion, do not cause fat accumulation. This is exactly how you would expect the system to work if it were designed to constructively handle a wide variety of macronutrient ratios, which it is. Just as cholesterol did not evolve to give us heart attacks, insulin did not evolve to make us fat.
Now let's address the common sense arguments that are used to support the insulin hypothesis of obesity. These include:
- Type I diabetics, who don't produce enough insulin, lose fat.
- Animals lacking insulin receptors on fat cells are resistant to fat gain.
- Insulin therapy often causes fat gain in diabetics.
- Repeated insulin injection into the same site causes fat accumulation at that site (lipoma).
- Two drugs that suppress insulin secretion, diazoxide and octreotide, sometimes cause weight loss in controlled trials.
Arguments 1-5 listed above are cases where insulin levels and/or insulin sensitivity are changing independently of one another, either through a pathological process (islet autoimmunity), genetic manipulation (fat cell insulin receptor knockout), or through drugs. This is why they're irrelevant to common obesity, where insulin levels and insulin resistance rise in parallel, such that total insulin action is either maintained or diminished. If we want to do an experiment that's actually relevant to the question, we can use animal models that are genetically manipulated to maintain insulin sensitivity in response to fattening diets, which as expected eliminates the increase in insulin that is typically observed on these diets. These experiments show that fat mass accumulation does not consistently differ between animals that experience an increase in insulin, and those that don't-- they all get fat at approximately the same rate (17a, 17b, 17c).
In addition to what I just explained, both diazoxide and octreotide (argument #5) are extremely nonspecific drugs that have actions in the hypothalamus (brain) that would be expected to influence fat mass, so we actually have no idea if they act by reducing circulating insulin levels or through some other mechanism.
The idea of fat gain in insulin-treated diabetics (argument #3) is not as airtight as it might at first seem. On average, diabetics do gain fat when they initiate insulin therapy using short-acting insulins. This is partially because insulin keeps them from peeing out glucose (glycosuria) to the tune of a couple hundred calories a day. It's also because there isn't enough insulin around to restrain the release of fat from fat cells (lipolysis), which is one of insulin's jobs, as described above. When you correct this insulin deficiency (absolute or relative), obviously a diabetic person will typically gain weight. In addition, short-acting insulins are hard to control, and often create episodes where glucose drops too low (hypoglycemia), which is a potent trigger for food intake and fat gain.
So what happens when you administer insulin to less severe diabetics that don't have much glycosuria, and you use a type of insulin that is more stable in the bloodstream and so causes fewer hypoglycemic episodes? This was recently addressed by the massive ORIGIN trial (17d). Investigators randomized 12,537 diabetic or pre-diabetic people to insulin therapy or treatment as usual, and followed them for 6 years. The insulin group received insulin glargine, a form of long-acting "basal" insulin that elevates baseline insulin throughout the day and night. In this study, insulin treatment brought fasting glucose from 125 to 93 mg/dL on average, so it was clearly a high enough dosage to have meaningful biological effects. After 6 years of divergent insulin levels, the difference in body weight was only 4.6 lbs (2.1 kg), which is at least partially explained by the fact that the insulin group had more hypoglycemic episodes, and took less metformin (a diabetes drug that causes fat loss). A previous study found that three different kinds of long-acting insulin actually caused a slight weight loss over three months (17e). This is rather difficult to reconcile with the idea that elevated fasting insulin is as fattening as claimed.
In obesity, fat tissue is insulin resistant. The fat tissue of obese people doesn't suppress fatty acid release in response to experimentally elevated insulin or mixed meals as effectively as the fat tissue of a lean individual (18, 19). In fact, obese people release an equal or larger amount of fatty acids from their fat tissue than lean people under basal conditions as well (20, 21). If this is true, then why do they remain obese? It's simple: the long-term rate of fat entering the fat cells is equal to the rate exiting, or higher. There is no defect in the ability of fat cells to release fat in obesity, the problem is that the fat that is released is not being oxidized (burned) at a rate that exceeds what is coming in from the diet, therefore it all ends up back in the fat tissue.
While we're on the subject, let's address the idea of "internal starvation". Taubes suggests that people overeat because they can't access their fat stores due to elevated insulin. However, obese people have normal or elevated levels of circulating fat (22, 23), so how is that possible? The internal starvation model was interesting, if speculative, at the time it was proposed, however the evidence for it has simply failed to materialize. If anything, obesity is a condition of "internal excess".
Let's also address the claim that obese people don't necessarily eat more than lean people. Food records are notoriously inaccurate, however there is at least one way to measure total energy intake in a precise and unbiased manner. It is called the "doubly labeled water method" (DLW). DLW studies have shown that after controlling for confounding factors (gender, age, physical activity), obese people almost invariably expend more, and consume more calories than lean people (24, 25). Weight stable obese people have a higher energy flux out of fat cells, and a higher metabolic rate, but it is not enough to overcome the higher calorie intake that is also observed (26, 27). That has been repeatedly confirmed and it is simply a fact at this point.
If elevated insulin leads to fat gain, then this should be scientifically observable. All we have to do is look for people with different levels of circulating insulin (controlling for baseline fat mass), and see if it predicts fat gain over time. Fortunately for us, this has been studied many times. In most studies, insulin levels are unrelated to future fat gain, or people with higher fasting insulin at baseline actually gain less fat over time that people with lower fasting insulin (27a). In the most recent study, higher insulin (and insulin resistance) at baseline was associated with less fat gain over time, but this relationship was eliminated by adjusting for baseline fat mass, leaving no relationship between insulin and fat gain after adjustment (27b). Again, I don't see how this can be reconciled with the idea that elevated fasting insulin is the cause of common obesity.
Therefore, the insulin hypothesis is not consistent with basic thermodynamics, it's not consistent with research on the biological functions of insulin, and it's not consistent with observational studies. Obese people do not have a defect in the ability to release fat from fat cells and burn it, to the contrary. They release more fat from fat cells than lean people, and burn more of it. However, this is compensated for by a higher energy intake, and a higher rate of fat incorporation into fat cells that counterbalances the increased expenditure. This shows that insulin does not cause obesity by acting directly on fat cells to cause fat storage. To understand obesity, we have to understand what causes increased food intake, and that factor is not insulin.
Part IIB: Insights From Human Genetics
Genetic studies can give us important clues to the biological processes underlying common diseases. For example, common genetic variants associated with type 2 diabetes risk tend to be in genes that regulate the insulin-secreting pancreas (38). This tells us, as one would expect, that pancreatic function is important in diabetes. What does genetics tell us about the mechanisms of obesity?
There are a handful of rare single-gene mutations in humans that lead to severe obesity. Every single one that has been discovered to date that does not also result in deformity (nondysmorphic monogenic obesity) is in the leptin signaling pathway (39), and even those that do result in deformity all influence how the brain regulates body fatness, suggesting that body fatness is normally regulated by the brain, not by fat tissue. From a 2009 review paper (40):
There are now at least 20 single gene disorders that clearly result in an autosomal form of human obesity. Notably, so far all these disorders affect the central [i.e., brain] sensing and control of energy balance.Genome-wide association studies (GWAS) give us a different perspective-- they look for common genetic variants that associate with higher or lower body mass index (BMI) in the general population. These are not mutations that make genes non-functional, they are simply common differences between genes that in some cases subtly influence their activity. Of the numerous common gene variants that have been found to associate with BMI variability, and whose function is known, the large majority are expressed in the brain, particularly the hypothalamus, and some are in the leptin signaling pathway (41, 42). That's why these papers often make statements like this (43):
...when we look at the information gleaned from the past 15 years of molecular genetic activity we cannot avoid concluding that, as much as type 2 diabetes is clearly a disease in which pancreatic beta-cell dysfunction is a critical element, obesity is a condition in which inherent genetic predisposition is dominated by the brain.And this (44):
Many of our associated loci highlight genes that are highly expressed in the brain (and several particularly so in the hypothalamus), consistent with an important role for CNS [central nervous system] processes in weight regulation.If insulin action on fat cells is a dominant factor in obesity, why don't genes linked to insulin signaling show up at the top of the list in these studies? There are enough proteins that regulate insulin secretion in the pancreas and insulin signaling in fat cells that one would expect genetic variability in these genes to turn up frequently if they were important regulators of fat mass, but instead the list is dominated by genes that relate to the brain, and leptin signaling in particular. This is consistent with a huge body of literature implicating the brain in body fat mass regulation and the development of obesity.
Part III: Carbohydrate, Particularly Refined Carbohydrate and Sugar, Cause Fat Accumulation by Increasing Insulin?
I've already demonstrated that it makes no sense to invoke insulin as a mechanism between carbohydrate consumption and body fatness. Another problem with the hypothesis is a thing called the insulinogenic index (II). The II is simply a measure of how much eating a food increases insulin, per unit calorie (28). It turns out, it doesn't correspond with the carbohydrate content of a food very well. In particular, protein-rich foods such as beef can increase insulin secretion as much as certain starch foods such as pasta, or more. High-protein diets, as many of you know, aid with weight loss. Some have suggested that this is because of glucagon release by the pancreas in response to protein. That may well play a role, but if we are going to invoke glucagon, then aren't we acknowledging that other signals besides insulin play an important role in this process? That's the larger point I'm trying to make here-- you can't just look at insulin, you have to consider amylin, glucagon, GLP-1, ghrelin, leptin, stomach distension, and all of the other short- and long-term signals that are activated in response to nutrient ingestion and changes in body fat mass. These collectively regulate food intake and long-term body fatness via the brain.
The other problem is that refined and unrefined carbohydrates often have a similar II. Pasta made from white and whole-grain wheat have the same II, and the same goes for white and whole-grain bread (29). Doughnuts and cookies are on par with whole grain bread. So post-meal insulin is not a compelling explanation for the potentially different effects of protein, unrefined carbohydrate, refined carbohydrate and sugar on body fatness.
I think it's likely that refined carbohydrate and sugar can contribute to obesity, but by what mechanism? Insulin is not a compelling explanation.
But let's forget about insulin for a minute. Without worrying about the mechanism, let's simply consider the hypothesis that carbohydrate consumption per se causes body fat accumulation. At this point, I know some people will be insisting that Taubes is talking specifically about refined carbohydrate, not carbohydrate in general. Taubes does repeatedly suggest in GCBC that all carbohydrate is fattening, although refined carbohydrate is more fattening. Otherwise, why would he write "...carbohydrates, and particularly refined carbohydrates... are the ultimate cause of common obesity", rather than simply stating "...refined carbohydrates... are the ultimate cause of common obesity"? This wording, used throughout CGBC, implies that all carbohydrate is fattening to some degree. There is also the example in GCBC of the Massas tribe fattening on unrefined sorghum, described below. If Taubes doesn't think unrefined carbohydrate is fattening, then why does he recommend a low-carbohydrate diet rather that suggesting that people replace refined carbohydrate with unrefined carbohydrate?
To address this hypothesis, first let's find some cultures that have a very high carbohydrate intake and see how fat they are. Let's start with a culture that eats more carbohydrate than any other I know: the New Guinea highland tribe at Tukisenta that was studied extensively in the 1960s and 70s. They ate 94 percent of their calories as carbohydrate, mostly from sweet potatoes, for a total calorie intake of 2,300 kcal/day in men and 1,770 kcal/day in women. Investigators found them to be fit, lean and muscular, with no sign of protein deficiency (Trowell and Burkitt. Western Diseases. 1981).
West Nile district, Uganda, 1940s. The diet consisted of millet, cassava, corn, lentils, peanuts, bananas and vegetables (Trowell and Burkitt. Western Diseases. 1981). Despite food abundance, "in the 1940s it was quite unusual to see a stout man or woman." "In recent years, however, a fair number of upper-class middle-aged West Nile women have begun to look rather stout, and some men have become very obese, especially those who hold lucrative posts and can purchase whatever food they like." This corresponded with an increase in "sugar, cooking oils, milk, fish and meat" and a corresponding decrease in "home-grown starchy staple foods." This same scenario has happened to hundreds, if not thousands of African communities whose traditional diets are very high in carbohydrate.
Northern Cameroon, 1980s. The Massas tribe (also spelled Massa) is known for its overfeeding ritual called Guru Walla, which Taubes describes in GCBC:
The Massa tribe of northern Cameroon fattens their males using both milk and a porridge made from sorghum, a corn-like grain that provides sweet syrup from the stalk. One man gained seventy-five pounds on a ceremonial binge. The average weight gain tends to be fifteen to twenty pounds using milk and porridge. The Massa are cattle herders and their staple diet is primarily milk. This fattening comes about by the addition of carbohydrates (sorghum) almost exclusively.Taubes states here that the typical diet is "primarily milk", therefore by inference, low in carbohydrate. Let's follow his reference and see what it says. It leads to a freely accessible paper by Drs. Igor de Garine and Georgius J.A. Koppert titled "Guru Fattening Sessions Among the Massa" (30). The Massas indeed herd cattle, but "their main use is not as food." The typical diet (not during overfeeding) is described as containing 516 grams of carbohydrate per day, and only 32 grams of fat (Table VIII). The typical diet is 81% carbohydrate, and primarily based on sorghum, according to his reference. This account is consistent with other freely accessible references in respected peer-reviewed journals (31). These people are lean on their typical high-carbohydrate fare until they deliberately overconsume a mixture of sorghum and milk.
Most of Asia, 20th century. Many Asian countries, including China, Japan, Taiwan and India, have a traditional diet that is very high in carbohydrate. In many cases, the dominant carbohydrate was white rice, a refined carbohydrate. Yet traditional Japanese, Chinese and Southern Indians eating mostly white rice were renowned for their leanness. Any plausible hypothesis of obesity needs to account for these observations.
Kitava, 1990s. Dr. Staffan Lindeberg showed that the Kitavan diet is 69% carbohydrate, mostly from taro, breadfruit, sweet potatoes and cassava (32). Thus, their diet would have had a high glycemic load and high II. They also obtain 50 g/day of carbohydrate from fruit, most of which would presumably been sugar (unrefined). Yet there was no obesity on the island, and only a few individuals that were slightly overweight (33). Fasting serum insulin was low, consistent with other high-carbohydrate cultures. Dietary carbohydrate does not cause insulin resistance.
Pima, 20th century. The Pima of New Mexico currently have one of the highest obesity rates in the world, on par with Nauru. It is rather ironic that Taubes uses them as an example in GCBC, when they are at odds with his hypothesis. The Pima were first contacted in 1539 by the Spanish, who apparently found them to be lean and healthy. At the time, they were eating a high-carbohydrate, low-fat diet based on corn, beans, starchy squash, and a modest amount of gathered animal and plant foods from the forest and rivers in the area. In 1869, the Gila river went dry for the first time, and 1886 was the last year water flowed onto their land, due to upstream river diversion by settlers. They suffered famine, and were rescued by government rations consisting of white flour, sugar, lard, canned meats, salt and other canned and processed goods. They subsequently became obese and have remained that way ever since. Their diet consisted mostly of bread cooked in lard, sweetened beverages and canned goods, and they also received salt. More recently, their diet has modernized but still relies heavily on processed food (34, 35).
Finally, let's take a look at my country, the United States of America. Total calorie intake has increased since the 1970s, and the excess calories came mostly from carbohydrate (primarily refined), and also from fat and protein to a lesser extent. But what happens if we go back further, to the turn of the 20th century? Here's our per capita macronutrient consumption in calories per day from 1909 to 2006, according to USDA data*:
If we take the long view, the only thing that has consistently increased is fat, not carbohydrate. The prevalence of obesity was very low at the turn of the century (36), yet our diet was 57% carbohydrate by calories, much of which came from white flour. These USDA figures account for food produced and consumed on farms and in home gardens, in addition to what passed through commercial sales (37). Why would carbohydrate promote obesity today when it didn't 100 years ago, and it continues not to in numerous high-carbohydrate cultures around the world?
I hope you can see by now that the carbohydrate hypothesis of obesity is not only incorrect on a number of levels, but it may even be backward. The reason why obesity and metabolism researchers don't typically subscribe to this idea is that it is contradicted by a large body of evidence from multiple fields. I understand that people like ideas that "challenge conventional wisdom", but the fact is that obesity is a complex state and it will not be shoehorned into simplistic hypotheses.
Carbohydrate consumption per se is not behind the obesity epidemic. However, once overweight or obesity is established, carbohydrate restriction can aid fat loss in some people. The mechanism by which this occurs is not totally clear, but there is no evidence that insulin plays a causal role in this process. Carbohydrate restriction spontaneously reduces calorie intake (as does fat restriction to a lesser extent), suggesting the possibility that it alters body fat homeostasis, but there is no compelling evidence that this happens due to a hormonal influence on fat tissue itself. The brain is the primary homeostatic regulator of fat mass, just as it homeostatically regulates blood pressure, breathing rate, and body temperature. This has been suspected since the early brain lesion studies of the 1940s (47) and even before, and the discovery of leptin in 1994 cemented leptin's role as the main player in body fat homeostasis. In some cases, the setpoint around which the body defends these variables can be changed (e.g., hypertension, fever, and obesity). Research is ongoing to understand how this process works.
* I've adjusted these data for loss, using the standard USDA adjustment of 28.8 percent, to get a more accurate picture of actual consumption rather than sales. I've also adjusted for an artifact in the fat data in 2000, where there was a big spike due to a change in the assessment method.