Source attribution of Campylobacter jejuni shows variable importance of chicken and ruminants reservoirs in non-invasive and invasive French clinical isolates

Abstract

Campylobacter jejuni is the most common cause of bacterial gastroenteritis worldwide. Mainly isolated from stool samples, C. jejuni can also become invasive. C. jejuni belongs to the commensal microbiota of a number of hosts, and infection by this bacterium can sometimes be traced back to exposure to a specific source. Here we genome sequenced 200 clinical isolates (2010-2016) and analyzed them with 701 isolate genomes from human infection, chicken, ruminants and the environment to examine the relative contribution of different reservoirs to non-invasive and invasive infection in France. Host-segregating genetic markers that can discriminate C. jejuni source were used with STRUCTURE software to probabilistically attribute the source of clinical strains. A self-attribution correction step, based upon the accuracy of source apportionment within each potential reservoir, improved attribution accuracy of clinical strains and suggested an important role for ruminant reservoirs in non-invasive infection and a potentially increased contribution of chicken as a source of invasive isolates. Structured sampling of Campylobacter in the clinic and from potential reservoirs provided evidence for variation in the contribution of different infection sources over time and an important role for non-poultry reservoirs in France. This provides a basis for ongoing genomic epidemiology surveillance and targeted interventions.

Figure 1

Maximum-likelihood tree based on 1422 concatenated core genes from the 899 strains of C. jejuni. Filled circles represent French strains, white circles represent strains from the rest of the world. Clonal complexes obtained from MLST that contain more than 10 strains are labelled in small font. Clonal complexes containing more than 20 strains are labelled in larger font.

Figure 2

Self-attribution of isolates from chicken, cattle and environment sources. Attribution tests were performed using the STRUCTURE software with 10 replicates. For each sub-dataset of 20 isolates coming from a known source, the proportion of isolates attributed to each reservoir is represented. (A) Original self-attribution, uncorrected STRUCTURE results. (B) Corrected self-attribution, after correction step based on a system of equations that balances the bias observed in the self-attribution internal results of each attribution test.

Figure 3

Attribution of French clinical non-invasive (n = 104) and invasive (n = 105) isolates, collected between 2014 and 2016, to isolates from chicken, ruminants and environment sources (352, 136 and 95 isolates respectively). Attribution tests were performed using the STRUCTURE software with 10 replicates. The corrected proportion of isolates attributed to each reservoir is represented after the model correction step based on the observed self-attribution results for each attribution test.

Figure 4

Attribution of French clinical isolates collected between 2009 and 2016 over time based on a collection. Attribution tests were performed using the STRUCTURE software with 10 replicates. A corrected proportion of isolates attributed to each reservoir (after model correction) is represented according to the year of isolation. (A) Attribution of French non-invasive clinical isolates collected in 2009 (n = 39), 2015 (n = 78) and 2016 (n = 26). (B) Attribution of French invasive clinical isolates collected in 2011 (n = 17), 2012 (n = 18), 2013 (n = 33), 2014 (n = 35), 2015 (n = 33) and 2016 (n = 37).