Aug 22, 2018, 10:30 AM ETInterview

David Zeevi on Personalized Nutrition Based on Your Gut Microbiome

The research focuses on "personalized nutrition" to address the global epidemic of metabolic disease, driven by the finding that individuals exhibit highly variable post-meal glycemic responses to identical meals.
Prevalence of obesity in the U.S. has risen from 1 in 10 adults in the 1980s to 4 in 10 adults currently, with Type 2 diabetes affecting approximately 1 in 10 adults.
Annual direct costs for diabetes and related complications were estimated at $250 billion in the U.S. (2012).
Researchers identified a dietary shift over the last 30–50 years as a primary driver: fat intake decreased from 20% to 15% of calories, while sugar intake increased drastically (from 1700s levels to current daily consumption).
Blood glucose response was selected as the primary metric for the study because it offers high-resolution, rapid feedback (measurable every 5 minutes) compared to the slow and noisy nature of weight measurement.
The study posits that high glucose spikes trigger insulin secretion, signaling the body to store energy as fat and contributing to long-term metabolic diseases.

The primary cohort consisted of approximately 800 participants recruited for a standardized, one-week study.
Participants were given a continuous glucose monitor (CGM) to record interstitial fluid glucose levels every five minutes.
A standardized breakfast containing 50 grams of available carbohydrates (bread, bread and butter, glucose, or fructose) was administered to all participants after an overnight fast, with no exercise allowed for two hours post-meal.
Data collection included blood samples, questionnaires on medical history and food frequency, stool samples for microbiome analysis, and detailed food logs using a custom app with provided scales.
A secondary cohort of 100 participants was used specifically to test the generalizability of the prediction algorithm.
A follow-up intervention study involved 26 pre-diabetic participants who underwent a "good week" (personalized diet to lower glucose) and a "bad week" (non-personalized diet) in a double-blinded, isocaloric design.

Intra-individual consistency was high: a person's response to the same meal on different days correlated with an r-value of 0.7 to 0.77.
Inter-individual variability was massive: the same food (e.g., white bread) elicited the full spectrum of glucose responses across the population, from flat responses to massive spikes.
Food categorization by "good" or "bad" is inaccurate; foods like bread or ice cream elicited opposite responses in different individuals despite identical carbohydrate content.
The predictive algorithm achieved a correlation of r=0.68 between predicted and actual glucose responses, approaching the theoretical upper bound of intra-individual stability (r=0.7).
The algorithm was trained on 40,000+ meals and utilized 137 features, including microbiome composition (species and genes), nutrient profiles, meal timing, sleep data, and blood parameters.
The correlation of r=0.38 found by traditional carbohydrate counting was significantly inferior to the new algorithm's performance.
In the 100-person validation cohort, the algorithm achieved a prediction accuracy of r=0.7.
The algorithm's effectiveness is specific to the population it is trained on; it was validated on the Israeli public, who consume a Western diet with a higher vegetable content.

The gut microbiome is defined as an ecosystem of bacteria, archaea, fungi, viruses, and worms, containing roughly 3 million genes (150x the human genome) and weighing as much as the human brain.
The microbiome is distinct from genetics (fixed) and lifestyle (modifiable), making it a key variable for personalized nutrition; unlike the human genome, the microbiome composition can be altered.
Specific microbial pathways can influence health outcomes; a study by Hazen's group showed that specific microbes convert carnitine (from red meat) into TMAO, which accelerates atherosclerosis.
Jeff Gordon's research demonstrated that transplanting the microbiome from obese twins into germ-free mice caused the mice to become obese, while lean twin microbiomes resulted in lean mice, even with identical diets.
High diversity in the microbiome is generally associated with better host health.
In the study's intervention, microbiome composition changed consistently over "good" and "bad" weeks, with beneficial bacteria increasing and deleterious bacteria decreasing.
A specific analysis of 5,000–6,000 genomic regions found that the presence of a specific 1% genomic region in a person's microbes correlated with those individuals being approximately 15 pounds thinner.
This specific region appears to enable microbes to convert sugar/sugar alcohols into butyrate, a compound that reduces inflammation and improves the host's glucose metabolism.

In the 26-person intervention, personalized diets successfully normalized blood glucose peaks for most participants; some saw glucose responses decrease by a factor of two or three.
The intervention confirmed that personalized diets could reduce glucose spikes without requiring drastic caloric restriction or eliminating specific food groups universally.
The researcher personally ceased fearing dietary fats after reviewing the historical context of the 1960s McGovern Committee and the 1950s Keys study, which he argues omitted data from high-fat, low-cardiovascular-disease countries like France.
Current research is expanding to the ocean microbiome, investigating how marine bacteria sequester CO2 and metabolize plastics or oil spills to aid in treating ocean acidification and pollution.
A long-term follow-up study involving 200–300 participants is underway to track microbiome and metabolic changes over six months to a year.
Commercial application is being pursued through "Day Two," a company offering personalized predictions based on questionnaire and microbiome data.
The researcher notes that fecal transplants are currently effective primarily for Clostridioides difficile (C. diff) infections but lacks sufficient evidence for broader dietary modulation in healthy individuals.
Moving forward, the research aims to move from broad microbial counting to identifying specific genomic regions and pathways that influence host metabolism.