A feeding trial that took the kitchen away
Diet studies fail for a reason that has almost nothing to do with biochemistry. People do not eat what they say they eat. They snack. They under-report. They drift back toward the foods they already like. A paper in Cell Metabolism tried to close that gap the expensive way: it ran a randomized trial that gave all meals to 42 adults with obesity, prediabetes, and fatty liver, and it forced similar weight loss, about 10 percent, on three very different plans. The menus were keto, Mediterranean, or a low-fat plant-forward diet. Adherence topped 95 percent.
That last number is the study’s hidden engine. In free-living trials, a ketogenic prescription and a Mediterranean prescription often collapse into the same blur of “I tried.” Here the blur was removed. Participants did not have to interpret a handout. They were fed. The scale was not left to chance either. Weight loss was matched across arms, which is the only honest way to ask whether how someone loses weight matters as much as whether they lose it.
The cluster of conditions was not accidental. Obesity, prediabetes, and fatty liver travel together. Fat in the liver is not a cosmetic finding on an ultrasound. It is a metabolic organ running a surplus, dumping glucose, resisting insulin, and helping to push a person from “at risk” toward type 2 diabetes. If a diet can empty that organ while the rest of the body is losing the same amount of weight, the diet is doing something the scale cannot see.
Three maps of the same calorie cut
A ketogenic diet, in clinical use, is a severe carbohydrate restriction that forces the body to run on fat and on ketone bodies made from fat. Protein is moderate. Fat is high. Bread, rice, fruit juice, and the usual scaffolding of a Western plate fall away. The Mediterranean pattern is the opposite kind of fame: olive oil, vegetables, legumes, fish, some grain, some fruit, a cuisine that public-health agencies have spent decades recommending as a heart-healthy default. A low-fat plant-forward diet pushes the macronutrient mix the other way still, emphasizing plants and keeping fat down.
Those three patterns are not three versions of “eat less.” They are three different signals to the liver. Carbohydrate restriction lowers the raw material for de novo lipogenesis, the pathway that turns sugar into fat inside hepatocytes. A Mediterranean plate still delivers carbohydrate, but it delivers it with fiber, unsaturated fat, and a food matrix that has a long epidemiological reputation. A low-fat plant-forward plate delivers still more carbohydrate and less fat, on the theory that the fat you eat is the fat you store.
The trial’s design was a test of those theories under a constraint that diet wars usually ignore. Everyone lost about 10 percent of their weight. If the only thing that mattered were energy balance, the liver should have improved in lockstep. Muscle should have improved in lockstep. Blood insulin should have moved together. The paper’s claim is that they did not.
Muscle agreed. The liver did not.
Muscle insulin sensitivity rose about 50 percent on every plan. That is a large, uniform, and clinically meaningful change. Skeletal muscle is the body’s main sink for glucose after a meal. When muscle starts listening to insulin again, blood sugar has somewhere to go. Weight loss of this magnitude is already known to produce that effect. The fact that keto, Mediterranean, and plant-forward eating all delivered roughly the same muscle result is a reminder that the scale is not a fraud. Losing about a tenth of body weight, with meals provided and adherence above 95 percent, is a potent insulin-sensitizing intervention no matter which cultural flag is flying over the plate.
The liver told a different story. Keto pulled ahead on the liver. Liver fat fell 67 percent on keto versus about 45 percent on the other two. That is not a rounding error. It is a gap wide enough to matter for a person whose diagnosis, at enrollment, included a fatty liver. The same hierarchy appeared in how the organ handled insulin. Hepatic insulin sensitivity improved two to three times more on keto than on the other arms.
Hepatic insulin sensitivity is a technical phrase with a practical meaning. When the liver is insulin-resistant, it keeps making glucose in the morning and after meals, as if the body were starving, even when it is not. That extra glucose output is one reason prediabetes exists. A diet that clears fat from hepatocytes and restores the liver’s willingness to quiet down under insulin is doing work that a matching loss of pounds, by itself, did not fully do on the other two menus.
The clinical downstream was blunt. Half the keto group no longer met prediabetes criteria after four to five months, versus 29 percent on Mediterranean and 7 percent on the plant-forward diet. Those percentages are not interchangeable talking points. They describe, in a small sample, how many people crossed a diagnostic line that insurers, guidelines, and patients all treat as real. Mediterranean eating, the pattern most likely to be praised in a primary-care visit, still beat the plant-forward arm. Keto beat both.
Four to five months is not a lifetime. It is long enough for liver fat, insulin, and a prediabetes label to move, and short enough that no one should confuse the result with a twenty-year outcome study. The paper itself sits inside that tension.
Insulin down, glucagon up
Weight-matched diets can still send opposite hormonal messages. Keto also drove the biggest drop in insulin and the biggest rise in glucagon, a hormone that helps the liver burn fat. That pairing is the metabolic signature of carbohydrate restriction. Insulin is the storage hormone. Glucagon is, among other jobs, the hormone that tells the liver to stop treating fat as a warehouse and start treating it as fuel.
In a high-carbohydrate pattern, even a “healthy” one, insulin stays in the conversation after meals. In a ketogenic pattern, insulin falls because the meals no longer demand it at the same scale. Glucagon, relatively unopposed, can then do what textbooks say it does: mobilize hepatic fat. The Cell Metabolism trial did not need a new hormone to explain its liver results. It needed a design that let that old hormonal contrast show up while body weight was pinned.
The muscle data keep the story from becoming a cartoon. If keto were simply “better at everything,” muscle insulin sensitivity would have pulled away too. It did not. It rose about 50 percent on every plan. The divergence is organ-specific. Muscle, in this sample, cared about the weight. The liver cared about the carbohydrate.
That distinction is why the study is being read as more than another round in the diet-tribe argument. Fatty liver disease is not a muscle disease. Prediabetes is not only a muscle disease. For adults who already have the cluster of obesity, prediabetes, and fatty liver, the organ that is most overloaded is the one that responded most to carb restriction.
The fat on the plate did not wreck the usual lipid alarms
The standard objection to keto is not that it cannot empty a liver. It is that a high fat load will shove LDL cholesterol and apoB in the wrong direction, trading a softer liver for a harder artery. LDL and apoB did not worsen despite the high fat load.
That sentence is not a claim that keto is cardioprotective. It is a claim about what happened in this trial, over this interval, in these 42 adults. ApoB counts the particles that actually carry cholesterol into artery walls. LDL cholesterol is the more familiar number on a lab printout. Either rising would have given critics an easy, quantitative rebuttal: the liver got better and the lipid panel got worse. That rebuttal did not materialize here.
Absence of worsening is not the same as improvement, and a few months is not a cardiovascular outcomes trial. Readers who care about heart disease should hold those caveats in the same hand as the liver-fat percentages. The paper does. The high fat load was real. The expected lipid penalty did not appear in the window the investigators measured.
Context still matters. Ketogenic diets in the wild are not all salmon and olive oil. They can be cheese, processed meat, and butter. They can also be avocado, nuts, and oily fish. This was a feeding study. The fat quality was whatever the kitchen plated. The result says that a high-fat ketogenic menu, as delivered here, did not worsen LDL and apoB while it was emptying the liver. It does not say that every grocery-store “keto” snack bar will do the same.
Small, short, and not a coronation
The trial was small and short. Forty-two people is enough to see a large, consistent organ-level effect when meals are controlled and adherence is above 95 percent. It is not enough to settle debates about bone, kidney stones, athletic performance, micronutrients, LDL hyper-responders, or what happens when the research kitchen closes and the participant has to cook. Four to five months is enough to reverse a prediabetes label in half of one arm. It is not a surrogate for a decade of eating.
It does not crown keto as the healthiest diet. That line is not politeness. Healthiest, as a word, smuggles in blood pressure over thirty years, cancer, dementia, the microbiome, the pleasure of food, the ability to eat at someone else’s table, and the simple fact that a diet people cannot follow is a diet that does not exist. Mediterranean eating remains the pattern with the deepest bench of long-term observational and trial evidence for cardiovascular events. A low-fat plant-forward pattern remains the pattern many guidelines still reach for when they want fiber high and saturated fat low. This paper does not erase that literature. It adds a narrower, sharper finding on top of it.
What it adds is specific. It says that for this metabolic cluster, cutting carbs did more than the scale alone. The scale did a lot. Every arm lost about 10 percent of body weight. Every arm improved muscle insulin sensitivity by about half. Two arms dropped liver fat by about 45 percent. Keto dropped it by 67 percent, improved hepatic insulin sensitivity two to three times more, produced the biggest drop in insulin and the biggest rise in glucagon, and moved half its participants out of prediabetes, against 29 percent and 7 percent. Those are not vibes. They are the measurements.
What “this metabolic cluster” actually is
The phrase is doing real work. Adults with obesity alone are not the same population as adults with obesity plus prediabetes plus a fatty liver. The third item is the tell. Once fat has accumulated in the liver, the organ that manufactures glucose and exports lipids is already in trouble. A diet that lowers insulin and raises glucagon is a diet aimed at that organ’s job description.
Prediabetes is a threshold, not a personality. People cross it and recross it. In this trial, crossing back below the line was most common on keto, less common on Mediterranean food, and uncommon on the low-fat plant-forward menu, even though weight loss was matched. That gradient is the practical takeaway for a clinician who is staring at a patient with a high liver-fat fraction and a rising A1C, and who has been telling that patient, correctly, that any diet they can stick to will help if the weight comes off. Stick-to-it-iveness was removed as a variable. Composition was not.
The Mediterranean result should not be discarded in the keto headline. Nearly a third of that group left prediabetes as well, and liver fat fell by about 45 percent. That is a successful metabolic intervention. It is just not as successful, in this tightly controlled comparison, as carbohydrate restriction. The plant-forward arm’s 7 percent figure is the one that should unsettle people who assume that “low fat” and “plant-forward” are automatically the right levers for a fatty liver. In this sample, with meals provided, they were the weakest levers for the diagnosis that sent people into the study.
What a reader should not do with a 42-person paper
Do not treat a feeding trial as a grocery list. Do not assume that a do-it-yourself ketogenic month, without the provided meals or the matched weight loss or the medical monitoring, will reproduce a 67 percent drop in liver fat. Do not ignore that LDL and apoB did not worsen here and also do not assume they never will in someone else. Do not skip the Mediterranean arm because it lost the headline. Do not skip the caveats because the percentages are large.
Do notice the design. Randomization. All meals provided. Weight loss forced to be similar. Adherence at 95 percent and above. An endpoint stack that includes liver fat, muscle insulin sensitivity, hepatic insulin sensitivity, insulin, glucagon, lipids, and the prediabetes label. Publication in Cell Metabolism, a journal that does not usually platform anecdote. The limitations are listed in the same breath as the results because they belong there: small, short, not a crown.
The diet debate has spent years arguing about which pattern is virtuous. This paper is arguing about which pattern, when calories and pounds are pinned, does more for a stuffed liver and a prediabetic glucose line. For 42 adults fed every meal for four to five months, cutting carbohydrate did more than the scale alone. That is a narrower claim than the internet will make of it. It is also, for the metabolic cluster the investigators enrolled, the claim the data support.
The scale was never the whole story
People who have been told to lose 10 percent of their weight are not being told a lie. In this trial, that loss arrived with a 50 percent improvement in muscle insulin sensitivity no matter which menu produced it. The lie, if there is one, is the older idea that all 10-percent losses are metabolically interchangeable once the clothes fit better.
They are not interchangeable in the liver. They were not interchangeable for glucagon. They were not interchangeable for how many people still met prediabetes criteria at the end of a short, tightly run experiment. Keto, in this randomized, fully fed comparison, beat a Mediterranean diet and a low-fat plant-forward diet on the organ that was fattest and on the diagnosis that sits one step before diabetes.
Hold the result with both hands. One hand is the measurement: 67 percent versus about 45 percent, two to three times the hepatic insulin-sensitivity gain, half versus 29 percent versus 7 percent, insulin down, glucagon up, LDL and apoB not worse. The other hand is the frame: 42 people, four to five months, meals provided, not a verdict on the healthiest way for a human being to eat for the rest of a life.
For adults who already carry obesity, prediabetes, and a fatty liver at the same time, the paper’s point is simpler than the culture around keto. The kitchen can change the liver even when the scale is told to move the same amount. In this cluster, the kitchen that cut the carbs moved it more.