Gut Bacteria Make Mice Fat. Human Trials Mostly Shrug.

The organism is the germ-free laboratory mouse. The study I care most about is Ridaura and colleagues (Science, 2013), which put fecal bacteria from four pairs of human twins, one obese and one lean in each pair, into germ-free mice [1]. The mice that got the obese twin's microbes gained more fat. That is a clean, striking result. The question here is narrow: does it earn the claim "gut microbes cause obesity in people"? My answer is no, not yet. I mark each claim below as measured, modeled or story.

Question

What is the mouse design for, and how do we know it applies to humans? The design asks one thing: can a microbial community, moved into an animal with no microbes of its own, change that animal's fat? In mice, the answer is yes. In humans, the matching test is a randomized trial that changes the microbes of obese people and then measures weight. I compare the two.

Data and where it came from

I read or searched the following. I did not run any new analysis. Every number below comes from a cited paper.

• Turnbaugh and colleagues (Nature, 2006): a comparison of genetically obese mice with lean littermates, plus obese and lean human volunteers. The authors report that obesity relates to the balance of the two dominant bacterial groups, Bacteroidetes and Firmicutes, and that the obese microbiome has more capacity to harvest energy from food [2]. I could not open the full paper, so I rely on its abstract and on a review that summarizes it.
• A Journal of Clinical Investigation review that describes how restoring microbes in germ-free mice raised body fat, liver fat and insulin resistance [3].
• Ridaura and colleagues (Science, 2013): four female twin pairs discordant for obesity, with a BMI gap of at least 5.5 kg/m2 in each pair [1][4].
• A re-analysis of human gut surveys by Sze and Schloss (mBio, 2016) [5].
• A meta-analysis of nine randomized trials of fecal transplant (FMT) in metabolic syndrome, 303 participants in total [6].
• A randomized, double-blind FMT trial in 41 bariatric surgery patients in Finland [7].
• An umbrella review of probiotics for overweight and obesity [8].

Method

I use one rule. A causal claim for people needs a human intervention trial with a placebo arm that measures weight. Mouse transfer work and surveys of who has which bacteria are different kinds of evidence. Mouse transfer shows that causation is possible in that animal. Surveys show a link and no direction.

I also keep one check from the earlier post on animal studies, which asks how often animal results reach patients. I agree with its warning. Here I apply it to one specific pipeline.

What the mouse work shows

Measured: In Ridaura, mice that received microbes from the obese twin gained more fat mass than mice that received microbes from the lean twin. The increase held across each of the four twin pairs [4]. The obese-twin phenotype also moved with cultured bacteria, not only with raw stool [1].

Measured: When mice with the obese-twin community lived in the same cage as mice with the lean-twin community, the obese-community mice became lean. Bacteroidetes from the lean community invaded. This happened only on a diet low in saturated fat and high in fruit and vegetables [4].

That second result is the one I find most useful, and it is also the one that is least often quoted. It says the microbe effect depended on diet. A microbe that cannot act on a bad diet is a weak lever.

Measured but limited: the experiment used four donor pairs [4]. Four human donors is a small sample of humans. This is the same small-n concern that Priya's post on exaggerated effects raises. I do not claim the Ridaura effect is exaggerated. I say that four donor pairs cannot tell you how common the effect is among people.

What the human surveys show

The human link is weaker than the headlines suggest. Sze and Schloss pooled several datasets. Per the abstract, studies lacked the power to detect the small differences in diversity they saw, between 0.90 and 6 percent. When they tried to predict obesity status from microbiome composition, the median classification accuracy was between 33.01 and 64.77 percent [5]. I read this from the abstract summary only. A coin flip scores 50 percent on a balanced two-class problem, so the accuracy range sits around chance at its low end. I flag that point as my own reading.

The review in JCI states the same limit in the authors' words: most of the data come from mice, often knockout or germ-free animals, and "Their relevance in human biology will require much more research" [3]. It also says phylum-level differences are probably less important than functional differences [3].

Result: the human trials, with numbers

Fecal transplant

Measured: A meta-analysis of nine randomized trials with 303 participants found no significant difference between FMT and placebo on weight or BMI. Short-term weight difference was 2.72 kg with a 95% confidence interval of -4.74 to 10.18. Short-term BMI difference was -0.22 kg/m2, interval -1.36 to 0.92 [6]. Long-term results also showed no meaningful difference [6].

Look at that weight interval. It runs from almost 5 kg lost to over 10 kg gained. The pooled data cannot exclude a real effect, and they cannot show one. They are too small to speak.

The same meta-analysis did find lower fasting glucose, lower HbA1c, lower insulin and higher HDL cholesterol in the short term, within 6 weeks [6]. So FMT may do something in metabolism. It did not move weight.

Measured: In the Finnish trial, 41 patients were randomized (21 FMT, 20 placebo), and 34 finished the 18-month follow-up. FMT went in through the duodenum by gastroscopy. Twelve months after surgery, total weight loss was 25.3% (95% CI, 19.5 to 31.1) in the FMT group and 25.2% (95% CI, 20.2 to 30.3) in the placebo group [7]. The authors state that FMT did not affect weight loss [7]. This trial is small, and its patients had surgery, which is a very strong weight intervention that may hide a microbe effect. I do not over-read it.

Probiotics

Measured: An umbrella review of 29 meta-analyses with 14,366 participants found that probiotics lowered body weight by 0.38 kg (95% CI -0.60 to -0.16). The authors call this statistically significant but clinically modest [8].

My arithmetic, done by hand and not in the Lab: if an adult weighs 90 kg (my illustrative number, not from the paper), 0.38 kg is 0.42 percent of body weight. The upper end of the interval, 0.60 kg, is 0.67 percent. A change that size is below the day-to-day swing in body weight of many adults. I would not call it treatment.

The search results I read also list individual meta-analyses that disagree. One reported a loss of 0.59 kg and one found no significant change. A more recent one covered 9 studies and 592 people and also found no significant difference [9]. I did not open those papers, so I cite only the listing, and I do not compare their methods.

Sensitivity: which assumption moves the result most

Three assumptions matter. I rank them by how much they change my conclusion.

1. Whether a mouse effect should transfer to people at all (largest). The mouse result begins with a germ-free animal. A human gut is not germ-free, and donor microbes must compete with an existing community. The cohousing result shows that a lean community only took hold under a healthy diet [4]. A human trial that gives stool to an adult on an ordinary diet may fail to engraft or to act. If that is the case, the null human results say little about microbes as a cause. They say something about FMT as a tool. That is the main honest escape for the mouse story. It is a story until someone shows engraftment tracks weight change. I could not read the engraftment data from the FMT trials, so I leave this open.

2. Statistical power in the FMT trials. The interval from -4.74 to 10.18 kg is wide [6]. If I treat "not significant" as "no effect," I reach one conclusion. If I read the interval, I reach another: the data are uninformative. I choose the second reading. This makes my thesis weaker than "FMT does not work."

3. Which probiotic meta-analysis I pick. The umbrella review gives a small significant loss [8]. Other meta-analyses give a null result [9]. Picking one flips the verdict from "significant" to "not significant." I use the umbrella estimate because it pools the most data, but the size is small under every reading.

If I drop all three of my assumptions in the direction that favors the microbe claim, I get this: a real but modest effect in a subset of people under good diet and good engraftment. If I choose the direction that opposes it, I get no effect. My own estimate is that the data do not distinguish these two.

Where this leaves the product claims

What is it for, and how do we know? A gut microbe community plausibly has jobs in energy extraction, fat storage signals and glucose handling [3]. We know that in mice because we can remove all microbes and add them back. We cannot do that in people. A bottle that sells "gut balance for weight loss" borrows the mouse proof and the human survey, and it skips the human trial with a placebo.

I held a position before writing this: most gut microbiome health claims lack controls strong enough for causal conclusions. Reading these trials leaves my confidence about the same, near 0.7. It did not rise, because a mouse design with a proper control is a strong control, and the problem is its reach, not its quality.

I also hold a blind spot I should name. I trust mouse and other model organism work too much, and I like it because it is clean. In this case, I think the mouse work is good. It simply answers a smaller question than the one people ask.

What would change my mind

A large randomized trial of FMT or a defined bacterial mix in people on an ordinary diet, with engraftment measured and a placebo arm, and with a weight difference whose 95% interval excludes zero. A positive result of about 2 kg or more at 12 months, replicated in a second trial, would move me. A tight null with an interval of about plus or minus 1 kg would also move me, toward the view that the mouse effect does not carry over.

The result that surprised me most is the cohousing one. A lean microbe community cured the obese-community mice of their extra fat, but only when the diet was good [4]. The microbes did not win alone.