Using software to compare genetic information in bacterial isolates from animals and people, researchers have predicted that less than 10% of Escherichia coli
0157:H7 strains are likely to have the potential to cause human disease.
According to Nadejda Lupolova, from the University of Edinburgh, Scotland, and colleagues, “machine-learning approaches have tremendous potential to interrogate complex genome information for which specific attributes of the organism, such as disease or isolation host, are known.” The researchers published the results of their study
in Proceedings of the National Academy of Sciences.
0157:H7 is a pathogenic strain of enterohemorrhagic E. coli
(EHEC). Although most E. coli
strains live in the gastrointestinal tracts of people and animals without causing disease, infection with E. coli
0157 is associated with serious illness in people. E. coli
0157 was first identified as a cause of disease in the United States in 1982, during an investigation into an outbreak of hemorrhagic colitis.
Since then, EHEC infections have emerged as a serious public health concern because these organisms produce Shiga toxin (Stx). E. coli
strains that encode Stx subtype 2a and a type 3 secretion system are often associated with the most severe human infections, which can lead to bloody diarrhea and kidney damage.
Ruminants—in particular, cattle—have been identified as the predominant reservoir of Shiga toxigenic E. coli
(STEC). These animals are infected asymptomatically and shed the organism in their feces, making it difficult to determine which animals are carrying strains that are harmful to people.
Therefore, the researchers wanted to test whether machine-learning approaches such as support vector machine (SVM) could be used to identify E. coli
strains in cattle that might represent a threat to human health and would therefore allow more targeted interventions in cattle.
Although machine-learning approaches have been routinely used to investigate complex data in numerous areas of science, it has not previously been used to analyze bacterial genomic data in order to use genotype to predict phenotype.
In this study, the researchers applied machine learning to predict the zoonotic potential of bacterial isolates from the United Kingdom and the United States. They used software to compare the DNA sequences of 185 E. coli
0157 strains that were isolated from 91 people and 94 cows. The approach was tested across isolates from the United Kingdom and United States and verified with food and cattle isolates from outbreak investigations.
According to the researchers, only a small subset—less than 10%—of cattle strains is likely to cause human disease. “[O]ne of the cattle isolates (apart from outbreak trace-back isolates) achieved very high human association probabilities (>0.9), potentially indicating that those posing a serious zoonotic threat are very rare,” the authors write.
This finding has important implications for public health management of this disease, they say, because researchers can now potentially identify such harmful strains in cattle. As a consequence, experts could use targeted control strategies, including vaccination or eradication, in cattle carrying strains of high zoonotic potential, in order to better protect human health.
“Machine-learning approaches should be applied broadly to further our understanding of pathogen biology,” the authors conclude.
Dr. Parry graduated from the University of Liverpool, England in 1997 and is a board-certified veterinary pathologist. After 13 years working in academia, she founded Midwest Veterinary Pathology, LLC where she now works as a private consultant. She is passionate about veterinary education and serves on the Indiana Veterinary Medical Association’s Continuing Education Committee. She regularly writes continuing education articles for veterinary organizations and journals, and has also served on the American College of Veterinary Pathologists’ Examination Committee and Education Committee.
To stay informed on the latest in infectious disease news and developments, please sign up for our weekly newsletter.