Pan-European Genomic Analysis

The researchers analyzed more than 10,000 Salmonella genomes collected from human, animal, and environmental sources across Denmark, England and Wales, France, Ireland, the Netherlands, Poland, Portugal, Spain, and Sweden. The total 3,548 isolates were collected and sequenced through routine surveillance activities conducted by national public health and veterinary institutes.

Using core-genome multilocus sequence typing (cgMLST) and a Random Forest classifier machine learning model, the team estimated the relative contributions of major sources for five key Salmonella serovars: S. Enteritidis, S. Typhimurium, S. Infantis, S. Newport, and S. Derby.

The model achieved moderate accuracy, although slightly lower than previous studies using WGS. Still, the findings align with previously established patterns: livestock remains the dominant source of human infection, while pets and wildlife play only a minor role.

Serovar-Specific Attribution Insights

The researchers’ analysis showed the most significant sources of human cases of salmonellosis to be:

• S. Typhimurium: Pigs were the leading source (63.8 percent), followed by dogs (7.1 percent), horses (6 percent), and broiler chickens (4.9 percent). Denmark and Portugal showed particularly strong pig associations (81 percent and 64.1 percent, respectively), while Dutch cases skewed toward broilers (66.7 percent).
• S. Enteritidis: Two-thirds (68.8 percent) of cases traced back to layer hens, with broilers accounting for another 19.7 percent. Many isolates across Europe originated from Poland and Spain.
• S. Infantis: Broilers were the most significant source (56.1 percent), followed by pigs (14.4 percent), dogs (9.8 percent), layers (7.6 percent), cattle (6.8 percent), and turkeys (4.5 percent). Most human clinical isolates were attributed to sources in Poland (37.9 percent) and the Netherlands (31.1 percent). Interestingly, 64.3 percent of Portuguese isolates originated in Poland, and none of them in Portugal itself. 
• S. Newport: Contrasting with historical trends linking S. Newport to poultry, the present data pointed to reptiles (28.1 percent) and cattle (25.9 percent) as the primary sources; this finding was likely influenced by reptile isolates being a more homogenous population. All Portuguese sequences were attributed to Poland, as well as half of Dutch sequences.
• S. Derby: Pigs were the greatest contributor (62.9 percent), followed by turkeys (22.7 percent) and cattle (6.1 percent). Sequences of human isolates were attributed to sources in Poland (43.9 percent), England and Wales (37.9 percent), and Spain (18.2 percent). 

Cross-Border Connections

Country-level attribution revealed that only one-third (33.1 percent) of human salmonellosis cases were linked to domestic sources. Many isolates were attributed to countries other than where the associated salmonellosis case was reported, underscoring the interconnectedness of European food supply chains. Poland, for example, was a frequent source for cases in other countries (32.2 percent), reflective of the country’s role as the origin of a prolonged, multi-country S. Enteritidis outbreak. Additionally, approximately 20 percent of S. Typhimurium human isolates in England and Wales, Portugal, and Denmark were attributed to Poland.

At the same time, a significant percentage of non-clinical sequences in the dataset were provided by Poland (39.5 percent); Poland, England, and Wales were the only countries to submit non-clinical sequences for each serovar. Therefore, an overrepresentation in the attributions is not unexpected. However, attributions were not proportional to the number of sequences for all other countries and/or sources for each serovar, and the model performed well for country attributions.

Implications for Controlling Salmonella in Europe

The study reinforces what is already understood by risk managers: poultry and pigs remain the primary reservoirs for human salmonellosis. While national action plans targeting poultry have driven down S. Enteritidis infections, similar efforts for pigs and pork are still lacking in most EU countries.

The findings also highlight the need for internationally harmonized control measures and improved genomic surveillance to address cross-border transmission.

Usefulness of Machine Learning for Foodborne Pathogen Surveillance

Machine learning offers a promising complement to traditional typing methods, particularly for real-time surveillance. However, challenges remain, especially in predicting minor sources and accounting for non-clonal populations like pets and wildlife.

Future work should focus on expanding datasets, refining genomic markers for host adaptation, and integrating attribution models into One Health strategies.