Several people have asked me for my thoughts on the recent Cell study that indicated a fasting-mimicking diet might stimulate beta cell regeneration. I finally found some time to read the study, and, while I hate to be a nay-sayer, I’m not particularly optimistic. I present here a deep dive into the study. If you’re not interested in the details, skip to the “Conclusions” section at the end.
The Study Background
The researchers set out to test the effect of a “fasting mimicking diet” on beta cell regeneration.
- In both type 1 and type 2 diabetes (T1D and T2D hereafter), research has shown that beta cells behave oddly as the disease progresses and the cells begin to die, sometimes turning into other types of pancreatic cells, or starting to look like pluripotent cells that don’t produce insulin. Conversely, turning pluripotent cells into functioning beta cells is a big area of research that holds promise as a therapy for diabetes. So, the questions of how and why beta cells move across cell lines is an important one.
- Starving an animal induces a strong response across the body, and many tissues begin to waste as the body tries to limit energy use. Some researchers have shown that upon return to normal feeding, many of the damaged tissues are restored as the body dials up its regenerative capacity to compensate for the period of atrophy.
- Given points (1) and (2) above, the researchers in this study set out to determine the effect of fasting on beta cell regeneration. However, regular periods of fasting are not really practical as a treatment for humans, so the researchers defined a diet that mimicked the effect of fasting on the tissues of the body.
The fasting mimicking diet (FMD): a low-carbohydrate, low-protein, high-fat diet that results in similar changes in growth factors, glucose, and ketone bodies as seen with a water-only diet.
Cycles of FMD in a mouse model of type 2 diabetes
The researchers use a mouse model of T2D called db/db in which the gene for the leptin receptor is mutated. Leptin is a hormone produced by fat cells in the body, and when its receptor is mutated, mice do not properly regulate their appetites or hormones, resulting in obesity as well as increased leptin and insulin levels. Just as with human T2D, obesity in these mice leads to insulin resistance first, and, in later stages, beta cell failure. Despite these similarities, it is important to note that this is still just a model of diabetes, and not a perfect analogue. The symptoms of the human disease and the mouse model are similar, but the causes are very different. Notably, human T2D is not generally caused by leptin receptor mutations, or even a faulty leptin signaling pathway.
Nonetheless, in this study, the researchers let db/db mice develop diabetes (10 weeks of age), and then waited until the hyperglycemia had stabilized, indicating late-stage disease (12 weeks of age). At that point (14 days after hyperglycemia onset at 10 weeks), the mice were started on repeated cycles of FMD– four days of FMD followed by 7 – 10 days of normal feeding. These cycles were continued for several months. The cycles of FMD led to substantially less hyperglycemia in the treated mice by 60 days after the start of hyperglycemia as compared to db/db mice on normal diets.
How did this happen? The authors claim that the cause is primarily increased beta cell function rather than increased insulin sensitivity. They show that plasma insulin levels are higher in the FMD-treated mice two weeks after beginning treatment, and the mice had more insulin-producing beta cells about four weeks after treatment. Further, those beta cells contained proteins considered to be markers for proliferation, indicating that the mice were seeing the effects of beta cell regeneration, rather than just increased beta cell survival.
These are very cool results, but let’s pause here for a moment to be skeptical: first, let me remind you that this is a model of T2D that mimics the symptoms, but the underlying causes are quite different. Some research shows, for example, that beta cell dysfunction early on contributes to T2D, while in db/db mice, beta cell failure is a response to obesity and hyperglycemia, and the beta cells themselves are otherwise normal.
Second, it is important to note that the FMD-treated db/db mice weighed 25% less than the normally fed db/db mice. They were still about 20% heavier than normal, unmutated mice, but the weight loss here is important to consider. The reason that the db/db mice become hyperglycemic in the first place is that the leptin pathway is broken, leading to the mice eating too much, leading to obesity, which in turn leads to hyperglycemia and eventually beta cell failure. It’s possible, then, that the real cause of beta cell salvation here is lessened obesity, not the diet itself. What’s the difference if the diet leads to lessened obesity? Well, causality matters here; if studies like this lead to us prescribing fasting mimicking diets, but the real cause of change is weight loss, then the diet alone will not necessarily lead to any changes. And, while the mice in this study saw weight loss, that is not a guarantee of the diet.
Cycles of FMD in a mouse model of type 1 diabetes
Those concerns aside, the beta cell regeneration is an interesting effect, and the researchers next decide to look at what happens in a mouse model of type 1 diabetes. For T1D, they use mice treated with streptozotocin (STZ). STZ is a beta cell toxin, so mice treated with the chemical lose their beta cells, resulting in hyperglycemia. This is considered a T1D model because, unlike the db/db mouse, obesity is not involved, and the mice are not insulin resistant. In other words, STZ mice lose beta cells first, then become hyperglycemic, whereas it’s the other way around for db/db mice.
Why did the researchers use STZ-treated mice rather than the most common T1D model, the NOD mouse? As with db/db mice, STZ treatment results in symptoms that look like T1D, but causes are very different. In human T1D, an autoimmune reaction to proteins of the beta cells results in the destruction of beta cells by the immune system. The NOD mouse similarly develops an autoimmune response to beta cells, and is therefore the more common model of disease. For this study, though, NOD mice, and human T1D for that matter, pose a big problem– if FMD led to any beta cell regeneration, the immune system would likely just kill off the new beta cells, meaning the researchers wouldn’t be able to see any of the regeneration even if it did happen. So, the researchers used STZ mice, which remove the immune response from the equation, focusing on the symptoms of T1D rather than the causes.
Researchers began cycles of FMD five days after STZ treatment. In this case, the cycles were 4 days of FMD followed by 3 days of normal feeding, and the cycles only continued for about 25 days. (It’s not clear why this is different than with db/db mice; my guess, though, is that it’s practically easier for humans to work on weeklong cycles and they wanted results faster, but I don’t have any evidence for that guess.) The researchers saw glucose levels return to near normal values about three weeks after starting the FMD treatment, and, as with the db/db mice, the STZ mice showed higher levels of plasma insulin as well as increased numbers of beta cells in the pancreas. A greater percentage of the insulin-producing cells also contained protein markers for proliferation, indicating that the FMD treatment was leading to beta cell regeneration, just as in the db/db mice.
Notably, this study is useful as a test of my skeptical claim above– that weight loss rather than FMD resulted in beta cell regeneration. Here, the mice are not experiencing significant weight loss or weight gain, indicating that the effects seen are a result of cycling FMD. But, let’s put our skeptic hats back on: I would argue that is the primary value of this study, as the relevance to T1D is minimal. Because there is no immune reaction at play, it’s entirely unclear what effect, in any, FMD would have on a model of type 1 diabetes in which the beta cells are either under attack in early stages of the disease, or later when there are no functioning beta cells left.
And, a nitpick: the researchers claim that FMD cycles also reduced levels of immune signals associated with inflammation during beta cell damage. The evidence here is cherry-picked at best, and seems like a failure to correct for multiple hypothesis testing. This is presented as supporting evidence and not the key point, so I will let it go, but I suspect it’s an example of data massaging rather than a real effect.
Cycles of FMD in normal mice
The researchers next checked the effect of FMD cycles in normal, non-diabetic mice. Using normal mice allowed them to analyze the mice on tighter time scales, and also to isolate the effects of FMD from all the complicated symptoms of the disease models. Normal mice were fed on the fasting mimicking diet for four days, at which point samples were collected. Following the first four days, the mice ate normally, and samples were again collected after one day of normal feeding and after three days of normal feeding.
When looking at islet cells, researchers found that the four days of FMD led to “a trend of decrease” in the number and size of cells. Note that “a trend” here is journal-speak for “not statistically significant,” and though the box plots used in the paper show a decrease overall, the actual numbers shown in the supplemental figures are not very convincing. Also note that the researchers have switched from measuring beta cells to measuring islet cells, which includes both insulin-producing beta cells and glucagon producing alpha cells. This is because, as the supplemental figures show, the “trend” in beta cells is even less convincing.
Let’s take it at face value for a moment, and assume there really is a decrease in islet cells. Then, when researchers measure again after one and three days of normal feeding, it appears that the islet cells have returned, with more insulin- and glucagon- producing cells visible in the pancreas. The beta cells show increased presence of protein markers for cell proliferation, and there is an increased number of cells that appear to be in a transitional state, showing both alpha- and beta-cell markers. After three days of normal feeding, beta cells returned to pre-FMD levels.
The researchers next tried to establish what genes were contributing to this increased proliferation and transition to beta cells. They show significant changes in a handful of genes that have been previously shown to be relevant to lineage determination (that is, the process of going from stem-like ancestors of alpha and beta cells, called progenitor cells, to the fully-functional alpha and beta cells) in pancreatic islets.
But, but, but: I have some major statistical issues to pick with this part of the study. The genes they show are a set that happen to match the pattern they are looking for, but even then, some change insignificantly, and the others are assigned a p-value of less than 0.05 but more than 0.01. Knowing how science works, the researchers almost certainly checked a larger panel of genes than those shown, and presumably then picked out the ones that matched the expected pattern. I am almost guessing here, but it’s unlikely that the genes shown are really the only ones tried. And if that’s the case, the p-values should be adjusted for multiple hypothesis testing, which would render the differences shown insignificant.
In which case, I would argue that there is a possible explanation for what the researchers are seeing that doesn’t require beta cell regeneration at all: the period of deprivation during the FMD cycle leads to temporary shutdown of the hormone production programs in alpha and beta cells. As FMD is ended and normal feeding resumes, the increased availability of nutrients allows the cells to turn protein production back on, but the sudden commencement of activation signals throughout the cell also turns on some unintended genes that reflect the previous lineage of the cells. In metaphorical terms, the alpha and beta cells are like new college students from restrictive homes– they go a little bit wild. After a few days of normal feeding, the cells re-stabilize, and return to normal levels of hormone production, maybe even over-producing glucagon and insulin for a while.
This interpretation fits with the data shown so far; the proliferation and progenitor marker proteins don’t indicate real beta cell regeneration so much as quiet beta cells returning to full functionality. This would also explain why both alpha and beta cells seem to go quiet during FMD, but only beta cells subsequently express proliferation markers. The researchers claim only the beta cells regenerate, but that leaves the alpha cell disappearance and re-emergence unexplained. In the model I propose, both types of cells are shut off temporarily, and, when they wake back up, they enable a slightly different set of accidental genes.
How would one test this? You would want to show that the new beta cells did not exist prior to the period of FMD, and were split from progenitor cells only after normal feeding had begun. The authors do what’s called a lineage tracing experiment, but this only serves to show that the protein markers for progenitor cells appear after the FMD treatment. In my model, this would still be the case– the same beta cells go a little crazy and make progenitor markers, but that doesn’t mean they actually become progenitor cells that generate new beta cells. So, while I don’t have specific evidence for the alternate model, the current study has not ruled it out in my mind, which I don’t fully buy the claim of beta cell regeneration versus just beta cell suppression followed by over-excitement.
Cycles of FMD in human cells
But who cares about mice anyhow? There are an infinity of ways to cure diabetes in mice that go nowhere in humans. (My personal favorite: inject leptin straight into the brain.) Murine beta cells have been shown to be more capable of regeneration than human beta cells to begin with, and even a little bit of added insulin sensitivity in mouse models of diabetes can rescue the mice.
So the researchers next looked at human pancreatic cells. Specifically, they cultured pancreatic cells from non-diabetic and type 1 diabetic donors. First, they cultured the cells in a dish with serum derived from clinical trial patients that were on a five-day FMD. The serum itself had less glucose and lower levels of growth factors than serum from humans on a normal diet, which is consistent with the expected effects on glucose and growth factors in the blood of fasting patients.
When the pancreatic cells were treated with the two different types of serum, the researchers saw “a trend” towards increased progenitor markers in the pancreatic cells. As before, this “trend” means there were no statistically significant changes observed in the cells, indicating that the FMD serum was not having much of an effect, if any, on the treated cells.
The researchers then abandoned the real human serum, and instead treated the pancreatic cells with normal growth medium, which is typically 10% serum with high glucose, and a fasting-mimicking medium (STS), which had only 2% serum and low glucose. As with the real human serum, STS likely had lower concentrations of growth factors in addition to lower glucose levels. STS, as described, sounds equivalent to what is often called starvation media, which is frequently used in cell culturing experiments to set cells in a state of minimal activity before administering a treatment that researchers are interested in. The media is therefore not unusual, but researchers don’t usually study cells during the starvation itself.
In any case, the researchers see an increase in insulin production from the starved pancreatic cells as compared to the normal pancreatic cells. They also see increased presence of some of the same progenitor markers that were seen in the mouse cells. Note that the researchers do not show any indication of proliferation, but rather increased overall production of the progenitor markers, which would be in line with my theory above that the nutrient starvation results in some amount of perversion of normal regulatory pathways in the cell, but not beta cell regeneration.
The researchers conclude that “our study provides an example of a potent and coordinated dietary regulation of cell-fate determination with the potential to serve as a therapeutic intervention to treat diabetes and other degenerative diseases.” In other words, they claim they have shown that a fasting-mimicking diet can reprogram and regenerate beta cells, possibly to the point of being an effective therapy.
What do I think? As discussed above, I think their interpretation is over-optimistic. From where I’m sitting, they have shown that starvation can lead to genetic dysregulation at a cellular level, which is interesting, but more from a basic-research perspective than a medical one. In other words, I don’t buy the claims of beta cell regeneration, and am very skeptical that any real insulin-production could be found in humans from this sort of treatment.
And that’s all before we get to the standard caveats that it is very hard to translate treatments between mice and humans. The particular mouse models used here are only rough approximations of either type 1 or type 2 diabetes, and the human studies here are in cultured cells only.
But what if they’re right?
Let’s assume for a minute, though, that they correctly interpreted the results, and a fasting-mimicking diet leads to some amount of beta cell reprogramming and regeneration. Would I then believe that there was a viable therapy here?
Let’s consider the type 1 diabetes case first: it is not accidental that the researchers use a streptozotocin-induced model of type 1 diabetes rather than an immune-mediated model like the NOD mouse. In the NOD mouse, as in type 1 diabetic humans, even if the diet led to increased beta cell proliferation, those cells would quickly be killed off by the immune system, just as the original beta cells had.
What about newly-diagnosed type 1 diabetes? Isn’t it worth trying, just for the chance? Frankly, it’s not clear this would help rather than hurt. The increased insulin production and the progenitor proteins that are showing up where they don’t belong could lead to a more severe autoimmune reaction just as easily as a slower disease progression.
And for type 2 diabetes? Well, I actually suspect that this diet would help for many type 2 diabetics– but only because you are almost certain to lose weight on a fasting-mimicking diet, and losing weight will ease the burden of insulin resistance for most type 2 diabetics. But this would be true of any diet that makes you lose weight, and if you’re the kind of person who can commit to a fasting-mimicking diet for four days a week, then maybe you should consider a more moderate reduction in calories. That is more likely to be sustainable over the long term, and will probably be less exhausting and atrophying than a low-protein and low-carb diet. In other words, for type 2 diabetics, where the primary problem is insulin resistance, the effect of a fasting-mimicking diet is likely to be similar to that of any nutritionally questionable fad diet.
The Ohhhhhhhhh moment
I started to conclude: “In sum, if you really want to eat cod liver oil four days a week, go for it, but I will pass.” Cod liver oil is what I imagine a “low-protein and low-carbohydrate but high-fat” diet to contain. But the study had an actual diet that they fed to the humans– so I looked it up to determine what they actually ate. In the methods section of the paper, the researchers list: “Human diet: Fasting mimicking diet (FMD) — Propriety [sic] formulation belonging to L-Nutra”. The paper goes on to describe the diet: “The human version of the FMD is a propriety formulation belonging to L-Nutra. It is a plant-based diet designed to attain fasting-like effects on the serum levels of IGF-I, IGFBP1, glucose and ketone bodies while providing both macro- and micronutrients to minimize the burden of fasting and adverse effects (Brandhorst et al., 2015). Day 1 of the FMD supplies 4600 kJ [1099 calories] (11% protein, 46% fat, 43% carbohydrate), whereas days 2-5 provide 3000 kJ [717 calories] (9% protein, 44% fat, 47% carbohydrate) per day. The FMD comprises proprietary formulations of vegetable-based soups, energy bars, energy drinks, chip snacks, tea, and a supplement providing high levels of minerals, vitamins and essential fatty acids.” Well, that sounds… like an advertisement. I clicked through the provided link.
And, oh, look, it’s a 5-day meal box, all nicely packaged and marketed, complete with a modern-looking sans-serif font. Prolon, it says, “Promoting health and Longevity” [capitalization sic]. Well doesn’t that sound… homeopathic. A picture shows some of the foods contained in the box– pretty, white packages of olives, “minestrone blend,” a food bar, and a handful of teas. The site goes on to describe the product: “ProLon®, [sic] is designed to promote the body’s natural ability to protect, regenerate and rejuvenate. In clinical studies, ProLon® has been shown to reduce abdominal fat and maintain healthy levels of blood glucose, C-reactive protein (CRP), and insulin-like growth factor 1 (IGF-1). Unpublished clinical trials indicate that ProLon® may have other positive health benefits.”
And that’s when I realized what was going on here. This paper, published in a very reputable scientific journal, is selling a fad diet. Well. That’s… suspicious. I click through to the About pages of L-Nutra, trying to find the people involved, and, sure enough, the Chairman of the Board is Dr. Valter Longo, the corresponding author on the study. That seems like a significant conflict of interest.
I jump back to the paper, and look for the Disclosures section that often accompanies scientific papers. This one has none. But surely he disclosed the conflict of interest? Cell, like all major journals, (requires that authors disclose conflicts of interest). There it is, in the Acknowledgements, hidden away despite several mentions of the diet in question: “V.D.L. has equity interest in L-Nutra, a company that develops medical food. All shares will be donated to charitable organizations.”
All right, fine. So there’s a conflict of interest, and it’s not well-disclosed. That doesn’t by any means invalidate the study. But, it does make me suspicious, and even more skeptical than I was when I actually read the study.
So, I have a new conclusion sentence: In sum, proceed with caution. The science here has some holes, and someone stands profit off of the conclusions reached.