Traditional food diaries are flawed—now, scientists can analyze stool DNA to track what you eat with precision, unlocking new insights into diet and disease.
Study: Metagenomic estimation of dietary intake from human stool. Image Credit: Lightspring / Shutterstock
In a recent study published in the journal Nature Metabolism, researchers presented a method to quantify food-derived deoxyribonucleic acid (DNA) in human stool metagenomes using a computationally efficient, decoy-aware mapping strategy.
Nutrition and dietary intake are determinants of growth, development, health, and disease risk. Besides, diet shapes the composition of the gut microbiota, and in turn, the gut ecology influences the host's dietary effects. Dietary intake patterns can aggravate or alleviate various disease conditions throughout an individual’s lifespan.
Dietary and nutrient intake is usually assessed using self-reporting methods, such as food records and nutritional questionnaires, which require participant compliance and suffer from reporting biases. On the other hand, metagenomic shotgun sequencing (MGS) of fecal DNA could be leveraged to capture dietary intake data.
The Study and Findings
In the present study, researchers presented a metagenomic estimation of dietary intake (MEDI) to quantify food-derived DNA in human stool. Unlike traditional mapping approaches, MEDI incorporates a decoy-aware filtering process to minimize false-positive assignments from bacterial and human DNA, improving accuracy. They mapped food items to RefSeq genomes at the species or genus level, leading to a set of 459 foods mapped to 331 genome assemblies. Next, they searched the full National Center for Biotechnology Information (NCBI) nucleotide database to identify partial (genome) assemblies for foods without a complete reference assembly.
This approach identified 98 partial assemblies for 102 more foods. The resultant database consisted of 489 billion base pairs encompassing all major phyla of food components. Most genomic data were from the phyla Streptophyta and Chordata, which consist of most plant- and animal-based foods, respectively.
Because the size of the food genome database exceeded that of databases for classifying bacteria, viruses, and archaea genomes, the researchers developed MEDI, a computational method based on the Kraken 2 mapping scheme, optimized for processing large datasets. MEDI food quantification was based on relative read abundances without genome size corrections. The team tested MEDI on a ground-truth dataset with simulated reads.
They generated average abundance profiles of decoy organisms in fecal samples from 365 persons from the integrative human microbiome project (iHMP). Positive controls were generated by introducing 10% food reads from 10 random food items into each sample. Additionally, four background samples lacking food reads served as negative controls.
The team noted that MEDI quantified sequences derived from food in all samples. None of the reads in food-negative samples were categorized as food-derived. However, the researchers acknowledged that MEDI performs best for whole foods that retain more DNA through digestion, whereas highly processed foods, such as refined oils and added sugars, are often underrepresented in sequencing data. MEDI was highly sensitive and had over 80% power to detect a food item with an abundance as low as 10 reads per million. Next, MEDI was applied to metagenomic data from two studies (MBD and PATH).
In the MBD study, participants consumed a microbiome enhancer diet (MBD) or a Western diet (WD). MEDI estimates revealed significant differences in beta diversity between WD and MBD. There was about six-fold higher relative abundance of food reads in the MBD than in the WD. Additionally, MBD was also able to reveal specific enrichment of known MBD components, including quinoa, rye, strawberry, and spinach.
Further, participants in the PATH study received daily meals containing 90% of the same ingredients across study groups. The intervention group received daily meals with a large avocado daily, but the control group received meals without it. A differential abundance analysis by MEDI identified only avocados as the food item that differed in abundance.
Moreover, daily food diaries from the PATH study enabled the comparison of MEDI estimates of nutrient composition in fecal samples to overall intake data. The researchers noted agreement between intake data and MEDI estimates when fecal samples were obtained within 24 to 48 hours after food intake. Moreover, MEDI accurately estimated dietary intake for protein, energy, carbohydrate, cholesterol, and potassium. However, it struggled to quantify total fat and dietary fiber intake, likely due to DNA degradation in highly processed foods.
Next, the team applied MEDI to estimate the frequency of food-derived reads across different life stages. As such, MGS data were derived from a cohort of 60 infants aged 1–253 days. Besides, adult samples were obtained from 351 subjects from the iHMP. Expectedly, infant samples had a lower prevalence of food-derived reads; however, the prevalence of (infant) samples with food-derived reads steadily increased after the introduction of solid foods (around day 160).
By contrast, food-derived reads were detected in 98% of adult samples. While relative metagenomic abundances of human or bacterial reads remained stable, food reads exhibited high variability between individuals and across time, reinforcing the importance of sampling time and dietary diversity in data interpretation. Notably, MEDI-inferred diets were concordant with food frequency questionnaire (FFQ) data from adults and infants.
Health and Metabolic Syndrome Findings
Finally, the team assessed whether MEDI could capture dietary patterns linked to health and disease states. To this end, they applied MEDI to 274 healthy individuals and 259 subjects with varying manifestations of metabolic syndrome (MetS) in the METACARDIS study. MetS patients included 134 subjects receiving medication and 125 untreated subjects. Cocoa, wheat, oats, flax, hibiscus, and pork were the most common foods detected in this cohort.
MEDI-inferred metabolite and macronutrient intake was highly variable across individuals. A differential abundance analysis was performed to identify features associated with MetS relative to healthy subjects. Samples from MetS subjects had 69% more chicken and 121% more pork than those from healthy individuals.
By contrast, samples from healthy subjects had increased abundances of tomato, pineapple, and apple DNA. MetS was associated with a lower abundance of Streptophyta and a slightly higher abundance of Chordata. Further, elevated cholesterol and beta-lactose levels were identified in MetS subjects, aligning with previous research linking these components to metabolic dysfunction. MEDI-inferred diets from healthy people showed a higher abundance of sugars, ellagic acid, and myoinositol. However, the researchers noted that MEDI does not distinguish between naturally occurring sugars (e.g., from fruits) and added sugars, which are often absorbed earlier in digestion.
Conclusions and Limitations
The researchers developed a data-driven method to estimate nutritional and dietary intake from food DNA in human stool metagenomes. MEDI provides an alternative for measuring dietary intake that could be applied to existing, extensive human fecal MGS data lacking dietary information. Moreover, MEDI will be valuable to past, present, or future metagenomic studies for which dietary intake estimates would prove useful.
Despite its strengths, the study acknowledges several limitations: (1) MEDI underrepresents processed foods due to DNA degradation, (2) food DNA detection is highly variable across individuals and meals, (3) the food genome database is biased toward Western diets, limiting accuracy for global populations, and (4) certain low-abundance foods, such as shellfish, may fall below MEDI’s detection threshold.