Researchers find genetic clues to infant formula pathogen’s global persistence

Researchers from the University of Maryland’s Department of Nutrition and Food Science are shedding new light on how a dangerous food-borne pathogen—Cronobacter sakazakii—may have adapted to thrive in dried and powdered foods across the global supply chain.

Their study, published in the International Journal of Food Microbiology, could transform how we monitor, control, and prevent contamination in critical food products like powdered infant formula.

C. sakazakii has made international headlines in recent years following recalls of powdered infant formula and has been linked to life-threatening infections in premature infants, the elderly, and other vulnerable populations. Although infections are rare, the consequences can be devastating—ranging from meningitis to long-term developmental issues.

To better understand the pathogen’s persistence and transmission, the researchers conducted the first-ever genomic meta-analysis of C. sakazakii bacteria strains from all over the world. Using an AI language model to standardize the data and machine learning to identify potentially significant genes, the team demonstrated the potential for AI to unlock information held in massive amounts of inconsistent and otherwise challenging data.

“We’re seeing how certain accessory genes—those not essential to survival but beneficial under specific environmental conditions—could confer advantages that help Cronobacter sakazakii persist in food systems and possibly even resist sanitation protocols,” said Assistant Professor Ryan Blaustein, the senior author of the study.

Individuals of a species carry a core set of genes that are shared across the species. But different strains or variants from different regions contain additional accessory genes unique to that strain. Blaustein and his colleagues analyzed 748 whole genome sequences collected from food, clinical, and environmental sources across North America, Europe, and Asia to identify the most complete set of C. sakazakii genes—also known as a pangenome—to date.

One of the major innovations of the study was the integration of artificial intelligence, including a Large Language Model (similar to Chat GPT technology) that standardized inconsistent metadata about the origins and sources of each sample, making large-scale comparison possible.

“There’s so much data available, but that data is not always standardized,” Blaustein said. “It’s not just the assembled DNA sequences, but the descriptor metadata. Everyone enters things differently, from the date and time to things like ‘powdered infant formula’ using a capital ‘P’ or lower case ‘p’ or just ‘powdered formula’ or even ‘PFI.’ We used the language model to recategorize everything that was already in the public database and assign it with a very high accuracy. That hadn’t been done in this setting before.”

Once they had standardized the data, the team used machine learning models—including random forest classifiers—to identify core genes and paint a clearer picture of how the accessory genes varied among samples from different locations, environments and conditions. This helped them identify genetic signatures associated with where and how the sample was taken.

They found that samples from powdered foods (including infant formula and powdered milk) relative to other types had larger genomes, and a higher frequency of genes involved in DNA recombination, repair, and desiccation resistance, all of which could contribute to the bacterium’s survival in dry conditions. In addition, there was a greater prevalence of genes associated with higher virulence in strains that were likely to persist in the food chain and cause illness.

The team also found correlations between geographic regions and genes associated with the formation of biofilms and resistance to heavy metals like copper, which shows up in some food systems as a component of pesticides or as an essential nutrient, but can also act as an antimicrobial at high levels.

The presence of so many accessory genes with potentially adaptive traits may be what enables C sakazakii to persist across a variety of ecological niches, including hospitals, food facilities, and dried food products. Understanding which genes support C. sakazakii’s survival in a variety of environments could help target sanitation measures and guide the development of safer processing protocols and technologies. The findings also raise important considerations for the food industry, especially manufacturers of powdered foods.

Importantly, this study provides a pathway to identify genes with key traits of interest for a variety of pathogens. The integration of AI models to clean, standardize, and interpret genomic and epidemiological data could help create faster, more accurate molecular surveillance systems for emerging pathogens.

This research underscores the need for international cooperation in understanding how food-borne pathogens evolve and move through the food system. With food products routinely crossing borders and oceans, tracking genetic markers of virulence and resistance has never been more critical.