Probiotics are emerging tools used by neonatal intensive care units to promote healthy outcomes and prevent intestinal diseases such as necrotizing enterocolitis. Approximately one in ten of the youngest preterm infants in the U.S. are treated with probiotics, and studies show that this therapy can reduce all causes of mortality.
Probiotic treatment often includes the administration of bacterial strains that belong to the Bifidobacterium genus. Bifidobacterium strains are especially abundant in the guts of children — particularly children who are breastfed — and are considered beneficial to human health.
"Bifidobacteria confer multiple positive properties, beginning with inhibiting the growth of pathogenic bacteria by outcompeting them for space and nutrients," said Aleksandr Arzamasov, PhD, a postdoctoral associate at Sanford Burnham Prebys. "They also help with the development of an infant's immune system."
Scientists have sought to use probiotics to deliver the benefits of Bifidobacterium to infants suffering from malnutrition. Studies showed that probiotic treatment in Bangladeshi infants suffering from severe acute malnutrition promoted weight gain, but the beneficial bacteria did not carve out a permanent home in the infants' microbiomes as expected from testing in the U.S.
"We wondered if the strain was less effective because it was not adapted to the local diets of Bangladeshi children," said Andrei Osterman, PhD , a professor in the Center for Data Sciences at Sanford Burnham Prebys and vice dean and associate dean of Curriculum in the Graduate School of Biomedical Sciences . "And we thought we may be able to predict which strains will thrive in different conditions, allowing us to match probiotics to children based on where they live and what they eat."
Osterman, Arzamasov and colleagues at Washington University School of Medicine, Sabanci University and the University of California San Diego published findings July 16, 2025, in Nature Microbiology demonstrating the ability to predict the nutritional adaptations of Bifidobacterium strains by analyzing the distribution of hundreds of metabolic genes in thousands of Bifidobacterium genomes.
The scientists first had to define the metabolic genes enabling Bifidobacterium to break down specific carbohydrates to create energy.
"When we eat food, many of the dietary carbohydrates are not digested by our bodies, especially the more complex fibers," said Arzamasov, the lead author of the study. "Instead, they go straight to the large intestine where they can be metabolized by gut bacteria."
After manually analyzing 263 Bifidobacterium genomes and integrating data from hundreds of previously published studies, the research team reconstructed 68 metabolic pathways that determined if a bacterium could digest a specific carbohydrate. The group then extended those findings by training an artificial intelligence-based model to analyze more than 2,800 additional genomes to predict the encoded capability to chow down on each of the 68 identified glycans.
The scientists followed up by putting their predictions to the test on 30 geographically diverse Bifidobacterium strains by observing their ability to grow when feasting (or not) on 43 carbohydrates corresponding to the predicted carbohydrate utilization pathways. When comparing the predicted growth to actual growth in these validation experiments, the accuracy rate of predictions was more than 94 percent.
The group uncovered differences in carbohydrate utilization based on geographic location, diet and lifestyle. For example, they discovered distinguishing features of Bifidobacterium strains isolated from fecal samples of Bangladeshi children. These strains had a unique capacity to digest both carbohydrates from human milk and plant fibers, which may indicate that these strains had adapted to changes in nutrients as an infant weans from milk to other foods.
"We found that these Bangladeshi isolates have unique gene clusters and unique metabolic phenotypes not found in any other genomes of strains isolated from other parts of the world," said Arzamasov. "This reinforces the importance of studying the gut microbiomes in understudied populations around the world in a culturally sensitive way, as they have unique biological diversity which is currently underappreciated."
By showing how carbohydrate metabolism strategies vary across and within Bifidobacterium species and are shaped by ecological factors — including host age, diet and lifestyle — Osterman, Arzamasov and their colleagues have provided a critical resource for future research and the development of personalized probiotics.
"With this encyclopedia of sugar utilization pathways in hundreds of strains with sequenced genomes, you can now confidently predict what are the nutrients that support their growth and what are the nutrients that do not support their growth," said Osterman, the senior and co-corresponding author of the study. "In addition to a compendium of hundreds of already known bacterial isolates, we built a tool that can be used to provide the same type of predictions for thousands and thousands more."
"You can use this knowledge to select the strains as probiotic candidates for a given situation," said Arzamasov. "And you can define very precisely what nutrients would support these probiotic strains to guide the rational development of supplementary foods to make them even more effective."
Jeffrey Gordon, MD, the Dr. Robert J. Glaser Distinguished University Professor and a professor of Pathology and Immunology at the Washington University School of Medicine in St. Louis, is co-corresponding author of the study.
Additional authors include:
- Dmitry A. Rodionov, James E. Kent, Semen A. Leyn and Marinela L. Elane from Sanford Burnham Prebys
- Matthew C. Hibberd, Janaki L. Guruge and Michael J. Barratt from Washington University School of Medicine
- Marat D. Kazanov from Sabanci University
- Kristija Sejane, Annalee Furst and Lars Bode from the University of California San Diego
The study was supported by the National Institutes of Health and the Bill & Melinda Gates Foundation.
The study's DOI is 10.1038/s41564-025-02056-x .
