During 2015–2017, we collected Campylobacter isolates representing human infections and food/animal sources. The collection period for isolates from humans (clinical isolates) and from food/animals overlapped by 18 months (October 2015–March 2017). We present WGS data for 1,509 C. jejuni isolates: 774 clinical and 735 food/animal.

The clinical isolates were supplied by 4 clinical microbiological laboratories (Aalborg University Hospital, Slagelse Hospital, Hvidovre University Hospital, and Odense University Hospital), which were performing clinical microbiological services for the regional hospitals and practitioners in 4 of the 5 geographic regions in Denmark. These 4 laboratories diagnosed ≈60% of the Campylobacter cases in Denmark. Collection of 904 isolates was timed to best capture seasonal variation. On the basis of data from previous years, each laboratory was to supply the number of isolates that could be expected to constitute a constant fraction of their positive samples per calendar month.

We performed WGS on all collected isolates. After removing 130 sequences because of poor sequence quality, we included in our analysis 774 C. jejuni isolates (representing 772 patients), representing ≈10% of the Campylobacter cases during the period.

Clinical isolates were collected during September 2015–April 2017 and represented 673 domestically acquired infections, 60 travel-associated infections, and 41 infections in persons of unknown travel status. We also included 14 isolates obtained from a fifth region (Aarhus University Hospital) during an outbreak investigation in June 2017. Of the domestically acquired isolates, 192 had been previously analyzed as part of another study to detect clusters among Campylobacter isolates in Denmark (October 2015–June 2016) (6).

The C. jejuni isolates from food/animals were collected during May 2015–March 2017 by the Danish Veterinary and Food Administration, where a total of 798 food/animal isolates were subjected to WGS. A total of 63 isolates were removed from the study because of poor sequence quality. The remaining 735 isolates represented animals from Denmark, hereafter referred to as domestic (27 pigs, 214 cattle, and 150 broilers [collected as cecal samples at slaughter]) or retail meat (172 domestic chicken, 111 imported chicken, 9 imported turkey, 4 domestic duck, and 22 imported duck). A limited number of isolates were available from 3 domestic retail vegetables, 2 seawater samples, and 21 pet dogs. The food/animal samples were primarily collected as part of official surveillance programs. Some animals, as well as turkey meat and duck meat, were collected specifically for the source attribution study (24). Isolates from pet dogs were obtained from healthy dogs and dogs that had diarrhea before sampling.

Genomic DNA (gDNA) from clinical isolates was purified by using a QIAGEN DNeasy Blood and Tissue kit (https://www.qiagen.com), and gDNA from food/animal isolates was purified by using the Invitrogen Easy-DNA gDNA purification kit (https://www.thermofisher.com). Sequencing was conducted by using Illumina technology (https://www.illumina.com) on either MiSeq or NextSeq sequencing machines and using the Nextera XT Library preparation protocol for paired-end reads of 150 bp or 250 bp. Raw data were deposited in the European Nucleotide Archive (accession no. PRJEB31119).

We processed WGS data by using a QC-pipeline (https://github.com/ssi-dk/SerumQC), in which isolate sequences were removed in case of contamination with >5% of another genus, as well as sequences representing C. jejuni isolates with genome sizes outside the range of 1.6–1.9 Mbp. We conducted further analyses by using BioNumerics 7.6 (Applied Maths, https://www.applied-maths.com), in which cgMLST assignment was based on the 1,343-loci cgMLST scheme from the Campylobacter Multi Locus Sequence Typing website (https://pubmlst.org/campylobacter) at the University of Oxford (25). We removed isolate sequences from the dataset if assemblies comprised >500 contigs, the cgMLST core-percentage was <95%, or >40 loci were present with multiple consensus sequences. When multiple isolates were available from the same patient, we kept only unique representative sequences. Clusters, or matches, were defined by the unweighted pair group method with arithmetic mean (UPGMA) clustering with cgMLST allele differences as a distance measure and a cluster threshold of 4. This value was chosen after evaluation of the phylogeny, with clustering of potentially related isolates and separation of nonrelated isolates (Appendix). We gave each cluster a unique name in which sequence type (ST) was followed by a number (#1, #2, etc.) to separate different clusters of the same ST.

We produced a rarefaction curve for the clinical isolates by using the rarefy function from the R vegan library (26). For each isolate, we used the specific cluster type as the input type for the rarefaction and assigned sporadic cases their own cluster type. We calculated data points for a subsample of n_subsample = 100 to 774, with a step of 100. From a random reordering of the same samples, at the same subsample steps as above, we selected the first n_subsample samples and counted the number of clusters of size >1 and >4.

Estimates of outbreak detection frequency were calculated by using the survival function S(cluster_threshold-1) under the binomial distribution B(N_cases,f_sampling). The complete code for the cost-effectiveness calculation is provided (Appendix).

Analysis of cgMLST results for all 1,509 isolates indicated that 732 of the isolates were part of 204 clusters consisting of >2 isolates. Of these, 63 clusters consisted exclusively of clinical isolates (n = 167), 66 consisted exclusively of food/animal isolates (n = 176), and 75 consisted of clinical (n = 233) and food/animal isolates (n = 156).

Separate analysis of the 774 clinical isolates showed that 366 (47.3%) formed 104 clusters consisting of >2 isolates. A considerably higher concentration of clusters occurred during the summer, when Campylobacter incidence peaks (27), whereas the number of sporadic infections was almost constant throughout the year (Figure 2).

Most (82) clusters were small clusters of 2–4 clinical isolates (total 190 isolates); 22 large clusters consisted of 5–17 clinical isolates (total 176 isolates). We detected a direct match to >1 isolates of food or animal origin for most (17/22; 77%) of the large patient clusters but only 24 (29%) of the 82 small patient clusters. Thus, 41/104 patient clusters could be matched to a potential source and represented 199/366 (54%) of the patients who were part of a potential outbreak.

The clinical isolates in this study represented ≈10% of the nationwide cases in the period. More extensive sampling would probably lead to detection of more clusters. Typing of all 774 clinical isolates indicated that 47% were part of the 104 identified clusters (Figure 3); had the sample size been reduced to, for example, 400 or 200 isolates, the percentage of isolates in clusters would have decreased to 39% or 25%, respectively.

To implement cost-effective surveillance of clusters of clinical isolates, it is essential to consistently find large clusters without investigating small insignificant clusters. With a sampling frequency of 10% and an investigation threshold of 4 clustered isolates, we expect to find 99.2% of large outbreaks (of 100 reported cases), and an investigation would be triggered within the first one third of the cases for 42% of the large outbreaks. A small outbreak of 10 reported cases would be observed only 1.3% of the time, and assuming a 1:100 ratio of large:small outbreaks, an average of 1.3 small outbreaks would be investigated per large outbreak.

A total of 75 clusters contained isolates of both clinical and food/animal origin (Table). Seventeen clusters contained >5 clinical isolates, but only 1 clinical isolate matched the source isolate(s) in 34 clusters.

Clinical isolates most often (192/233) matched isolates from domestic chicken meat, broilers, or both, corresponding to 25% of all clinical isolates in the study. In 10 clusters, 18 (2%) clinical isolates matched imported chicken meat, and 8 clusters included isolates from cattle and comprised a total of 16 (2%) clinical isolates. One small cluster matched 2 clinical isolates to an isolate from a dog, and a single cluster included isolates from imported chicken, imported turkey, and a domestic broiler. The chicken meat/broiler isolates in the same cluster were often linked to the same slaughterhouse, and some originated from the same farm. However, it was also common for cluster isolates to originate from different farms, but because complete data were not available, we could not make a systematic assessment.

Clusters often originated with the detection of a food/animal isolate, followed by the presence of several clinical isolates over a few months, as seen in the examples of epidemic curves of large genetic clusters (Figure 4). In some instances, the cluster type disappeared for months and later reappeared in new patients (Figure 4, panels A, C, E). With few exceptions, the clinical isolates present in clusters were detected across the 4 geographic regions.

Clusters occurred independently and generally reflected the seasonality of Campylobacter infections (Figure 2). Most clusters typically occurred over a few months. We determined the distribution of cluster isolates in the 22 large clusters over time (Figure 5). For most clusters, the isolates were distributed close in time around the median day (t = 0) of the specific cluster. Typically, most isolates were present within a 1-month period (+30 days) before and after this time point. However, it was not uncommon to detect a few cluster isolates outside this time range, and on a few occasions clusters lasted for several months (i.e., ST5#1 spanning 14 months and ST2254#1 spanning 17 months).

In the dataset, we found 2 examples for which distinct genetic clusters could be linked by epidemiologic data. The first example was an identified outbreak in which 14 isolates were sequenced as part of an outbreak investigation. These isolates split into 2 distinct clusters (ST21#9 and ST19#6), consisting of 8 and 6 clinical isolates, respectively. This multistrain outbreak involved ≈100 schoolchildren who had been served unpasteurized milk while visiting a dairy farm.

In the second example, a sample from a patient was positive for 2 strains belonging to distinct clusters (ST51#1 and ST21#2; Figure 4, panel F). The 2 clusters were observed for the same period of 2016, and each matched isolates from domestic chicken meat and broilers within a few weeks and contained clinical isolates from the 4 regions. The 2 isolates from broilers representing each of the clusters were obtained from the same slaughterhouse; the broilers originated from 2 geographically close farms.

Of the 66 clusters containing isolates of nonhuman origin, most (n = 59) included isolates from only 1 common source: domestic chicken meat and/or broilers (n = 37), domestic cattle (n = 11), imported chicken meat (n = 7), imported duck meat (n = 1), domestic pig (n = 1), imported turkey meat (n = 1), and domestic vegetables (n = 1). Isolates within each cluster often originated from the same farm, had been through the same slaughterhouse, or came from meat imported from the same country (data not shown). Cluster isolates were often isolated close in time, typically within a few days or months. Seven clusters contained isolates from >1 source: cattle and domestic chicken/broilers (3 clusters), cattle and dog (1 cluster), domestic chicken/broilers and dog(s) (2 clusters), and imported chicken and turkey meat and domestic chicken meat (1 cluster).