Most nutrition research converges on the same conclusion: the calorie deficit decides the outcome, and the diet composition is just packaging. A trial from Washington University School of Medicine, published August 27 in *Cell Metabolism*, refuses to cooperate—three groups lost the same amount of weight, yet the liver only rewarded one of them. The ketogenic group's intrahepatic fat fell by an average of 67%, while the Mediterranean and plant-based low-fat groups both dropped 45%.
Trial Design
The elegance is entirely in the design:
- Population: metabolically unhealthy adults with obesity (prediabetes + fatty liver)
- Three arms, randomized:
- Ketogenic: 4% carbs / 73% fat
- Mediterranean: 50% carbs / 35% fat
- Plant-based low-fat: 70% carbs / 15% fat
- 100% catered meals for five months, with individually rationed calories
- WashU Medicine press release
- ScienceDaily reprint
- EurekAlert
- Paper: *Cell Metabolism*, DOI 10.1016/j.cmet.2026.07.020
That last point is the crux: calories were individually allocated to deliberately equalize weight loss across all three arms—about 10% each. With the scale equalized, differences in liver outcomes can only come from *what* was eaten, decoupled from *eating less*. Most diet trials cannot do this, because weight loss is never synchronized and hepatic fat changes are confounded by weight-loss contributions.
Results
| Diet | Intrahepatic fat reduction | |---|---| | Ketogenic | 67% | | Mediterranean | 45% | | Plant-based low-fat | 45% |
The paper offers a fairly complete mechanistic chain:
1. Carbohydrates cut to 4% 2. 24-hour insulin down 74% (vs -44% Mediterranean, -27% low-fat); glucagon rose the most 3. Greatest suppression of hepatic de novo lipogenesis (DNL—*de novo lipogenesis*, literally "making fat from scratch"; at 4% carbs this assembly line essentially shuts down) 4. Liver fat down 67%
Additional findings: hepatic insulin sensitivity improved 2-3 fold (p<0.001), and HbA1c fell the most in the keto arm. Interestingly, on the negative side: LDL, apoB, and 24-hour triglycerides showed no differences across groups—the cardiovascular storyline gets no award from this trial.
The first author is Max C. Petersen (assistant professor); the corresponding author is Samuel Klein (Danforth Distinguished Professor, a veteran of the field).
Caveats, ranked by importance
1. Population: metabolically unhealthy adults with obesity (obesity + prediabetes + fatty liver)—not healthy people. This is both a limitation and a selling point, since it matches the fatty-liver demographic, but results do not extrapolate to lean or healthy individuals. 2. Small sample, conflicting press releases: the WashU release says 55 randomized; the Cell Press release says 42. Likely 55 enrolled, 42 completed—neither states this explicitly. Unverified; the only number this post could not fully confirm. 3. Five months, no long-term outcomes. 4. Imaging method for liver fat not specified in public releases (presumably MRI-based). Unknown; awaiting the full text. 5. Klein has consulting relationships with multiple pharma companies (AbbVie, 89Bio, Boehringer Ingelheim, Merck, Viking)—disclosed per convention, but readers should know.
The authors themselves acknowledge open questions: the underlying "basic mechanism" is not fully understood (is GLP-1 involved?); whether carbohydrate restriction must induce "ketosis" to work, and whether the benefit must occur during active weight loss, both remain unanswered.
Why it matters
Fatty liver is strikingly common—Petersen said in an interview that about 75% of adults with obesity have it (as quoted; not independently verified). In the GLP-1 era, the default assumption is "weight loss is the whole therapeutic effect." This trial says: wait—identical scale numbers, different organ outcomes. For people with fatty liver, the carbohydrate side of the plate may be the real lever.
The verdict here: wait for replication before changing one's own diet—nutrition science has been burned by single-arm studies too many times. Still, catered meals, randomization, three arms, and full mechanistic markers—this is the ceiling of evidence quality for a dinner-table question. The remaining question is whether that ceiling is high enough.
References: