Human Campylobacteriosis in Luxembourg, 2010–2013: A Case-Control Study Combined with Multilocus Sequence Typing for Source Attribution and Risk Factor Analysis

Campylobacteriosis has increased markedly in Luxembourg during recent years. We sought to determine which Campylobacter genotypes infect humans, where they may originate from, and how they may infect humans. Multilocus sequence typing was performed on 1153 Campylobacter jejuni and 136 C. coli human strains to be attributed to three putative animal reservoirs (poultry, ruminants, pigs) and to environmental water using the asymmetric island model. A nationwide case-control study (2010–2013) for domestic campylobacteriosis was also conducted, including 367 C. jejuni and 48 C. coli cases, and 624 controls. Risk factors were investigated by Campylobacter species, and for strains attributed to different sources using a combined case-control and source attribution analysis. 282 sequence types (STs) were identified: ST-21, ST-48, ST-572, ST-50 and ST-257 were prevailing. Most cases were attributed to poultry (61.2%) and ruminants (33.3%). Consuming chicken outside the home was the dominant risk factor for both Campylobacter species. Newly identified risk factors included contact with garden soil for either species, and consuming beef specifically for C. coli. Poultry-associated campylobacteriosis was linked to poultry consumption in wintertime, and ruminant-associated campylobacteriosis to tap-water provider type. Besides confirming chicken as campylobacteriosis primary source, additional evidence was found for other reservoirs and transmission routes.


Methods
Human data. A nationwide case-control study of risk factors for human campylobacteriosis was conducted in the Grand Duchy of Luxembourg. In total, 1153 human cases of C. jejuni infection (89.4%) and 136 of C. coli (10.6%) were identified by the National Health Laboratory of Luxembourg (LNS) during December 2010-May 2013. The LNS serves as the national reference laboratory for enteropathogens and receives all Campylobacter isolates from private and hospital laboratories in Luxembourg. Campylobacter isolates were cultured and characterized at the species level using molecular methods 4 . MLST was performed according to the protocol for amplification and sequencing of the seven housekeeping genes of Dingle et al. 26 , with slight modifications 4 . Automated data analysis and library matching were performed with SeqScape ® v2.5 (ABI, Life Technologies, Belgium). For Scientific RepoRts | 6:20939 | DOI: 10.1038/srep20939 Source data. A dataset was compiled including 498 C. jejuni and 340 C. coli isolates from food-producing animals (n = 509) and from environmental water (n = 329) typed using MLST as per the human isolates. Animal isolates were obtained from poultry (n = 338), ruminants (n = 109), and swine (n = 62). Most of these isolates were collected in Luxembourg (n = 799), and only few swine isolates were collected in the geographically close countries the Netherlands (n = 17), France (n = 16), and Belgium (n = 6). Animal isolates originated from either faeces or meat and were collected during 2003-2013 as part of different research and surveillance activities on farm, slaughterhouse, and retail ( Table 1). The environmental water isolates were sourced from multiple sites in Luxembourg, including rivers, ponds, recreational waters, and wastewater treatment plant outlets. Environmental water was treated as a source in the attribution analysis (see below). Although environmental water cannot be considered as an amplifying host for campylobacters, it represents a 'sink' collecting strains from a variety of different hosts, including those commonly found in animals and humans 14,16,17 . Therefore, similar to previous studies 5,14,16,17 , environmental water was considered here as a proxy for other unidentified (animal) reservoirs, including wildlife. Part of the source isolates employed here have already been used in previous studies conducted in Luxembourg 4,29 and the Netherlands 5,9,17 . Source data are available in the Campylobacter PubMLST website.
Source attribution analysis. Attribution of the 1289 MLST-typed human campylobacteriosis cases to poultry, ruminants, swine, and environmental water was performed using the asymmetric island model (AIM), which is presented in detail elsewhere 15 . Briefly, the AIM uses a Bayesian approach for attributing human Campylobacter strains to different sources based on their MLST profiles, accounting for genetic relatedness amongst STs. The AIM represents the Campylobacter population as separate islands, one per each source. The population is homogeneously mixing within each island and strains can migrate between islands, determining different levels of genetic differentiation. The AIM assumes that the observed genotypes arise from mutation (where an allele at a locus is novel), recombination (where an allele at a locus has been observed before in another allelic profile), and migration (where an allelic profile has been observed elsewhere). After estimating the mutation and recombination rates within each source, and the migration rates between sources and from each source to humans, the AIM uses these rates to estimate the relative contribution of each source to human cases. Attributions rely on the calculation of sampling probabilities, i.e. the likelihood that a given human genotype is sampled from a given source. One of the major advantages of the AIM is that it can account for novel (combination of) alleles in strains from humans unobserved in source populations 15 . For every human ST, the AIM estimated a relative posterior probability (Pr) to come from poultry (Pr P ), ruminants (Pr R ), swine (Pr S ), and environmental water (Pr W ).
Risk factor analysis. We conducted a risk factor analysis for human campylobacteriosis as previously performed [5][6][7]9 , including the 415 domestic cases and the 624 non-travelling controls. C. jejuni and C. coli cases were at first analysed together to identify risk factors for campylobacteriosis as a whole, and then separately to identify species-specific risk factors, as compared to the controls. To identify factors uniquely associated with either Campylobacter species, a case-case comparison was also performed using the C. jejuni cases as 'controls' 7,30 .
For preliminary statistical testing, 48 putative risk factors (Supplementary Information S2) were tested for association with the outcome using unconditional logistic regression, with the variables age group, gender, urbanization degree, and season always included as covariates to control for confounding. Variables showing a p ≤ 0.10 were selected for inclusion in a multivariable logistic regression model built in backward stepwise fashion. Variables were dropped one by one if they showed a p ≥ 0.05 and their exclusion from the model did not influence the association of the other covariates. Biologically plausible interactions were also tested, and the model was expanded to include significant interaction terms. Selection between collinear variables was made based on improved model fit as revealed by the Akaike information criterion (AIC).
Subsequently, risk factors for human campylobacteriosis caused by Campylobacter STs putatively originating from poultry, ruminants, swine, and environmental water were studied by building logistic regression models including separate subsets of cases (either Campylobacter species combined) attributable to each of the sources based on their estimated Pr values. The assignment of cases to sources was performed like in previous studies 5,31,32 . A cut-off point of the Pr distribution for each source was determined to provide a balance between the number of cases assignable to each source and the confidence as to their correct assignment derived by the highest possible Pr value. The final cut-off points represented the best compromise between the increasing Pr for a given source (i.e. increase in source specificity) and the decreasing number of cases included in the models (i.e. decrease in statistical power and failure of the model to converge) 5 . A separate model for each group of source-assigned STs was built to identify risk factors for Campylobacter infection. For cases of probable poultry and ruminant origin, a Pr cut-off point of 0.80 was determined, resulting in the selection of 161 cases with a mean Pr P of 0.93 (range 0.83-0.99) -hereafter referred to as poultry-assigned cases -and of 68 cases with a mean Pr R of  Table 1. MLST typed Campylobacter isolates used in the source attribution analysis. 1 Includes broiler chickens (n = 254), turkeys (n = 49), meat ducks (n = 3), quails (n = 4), Guinea fowl (n = 2), and other unidentified domestic poultry (n = 26). 2 Includes cattle (n = 101), sheep (n = 4), and goats (n = 4).
0.84 (range 0.81-0.94), i.e. ruminant-assigned cases. For cases of probable swine origin, the construction of any regression model was not possible since there were only three cases with Pr S ≥ 0.25. Moving the cut-off point to a Pr S < 0.25 would have resulted in the inclusion of many cases nearly equally, or even mainly, attributed to the other sources, making the risk factor analysis for swine-assigned campylobacteriosis unclear and uninformative. Therefore, the swine source was not further considered here. For campylobacteriosis of probable environmental water origin, only 21 cases had a Pr W ≥ 0.25; analyses were thus performed using these 21 cases, having a mean Pr W of 0.38 (range 0.28-0.94). Given the limited number of environmental water-assigned cases, the above a priori confounders were included in the model only if they influenced the variable of interest by ≥ 10%. The complete list of STs and respective cases assigned to each of the sources included in the case-control study is appended as Supplementary Information S3. Logistic regression models for poultry-and ruminant-assigned campylobacteriosis were built the same way as described above, using the 624 controls as comparison group. Finally, to identify factors uniquely associated with campylobacteriosis from poultry or ruminants, a case-case analysis was performed comparing exposures of poultry-and ruminant-assigned cases with one another. For simplicity, while all available risk factors were tested for all the outcome variables considered here, only the results of the final multivariable regression models based on the aforementioned stepwise procedure were then presented. All final models showed an overall statistical significance (likelihood ratio χ 2 test, p < 0.05) and goodness-of-fit (Hosmer-Lemeshow test, p > 0.05). Statistical analyses were performed using STATA 13 (StataCorp LP, College Station, USA). Ethics statement. This study was approved by the National Research Ethics Committee (CNER) of Luxembourg (www.cner.lu) and was conducted in accordance with the principles of the Declaration of Helsinki. All methods were carried out in accordance with approved guidelines and regulations. Informed consent was obtained from all subjects. No subject-identifiable data were generated.
Campylobacter genotypes in humans. The   respectively) and peaked in summer (48.3% and 46.9%, respectively). No significant effects of age, gender, and urbanization were found for the other STs and CCs.
Risk factors for human campylobacteriosis. The significant risk factors for human campylobacteriosis as a whole (C. jejuni and C. coli combined) are reported in Table 2. Consuming chicken outside the home (OR 2.10), and both at home and outside (OR 4.77), but not at home only, were associated with campylobacteriosis. Consuming poultry other than chicken outside the home was also associated with campylobacteriosis (OR 1.98). Campylobacteriosis was also more likely to occur in summer than winter (OR 2.00), and in those who had had contact with garden soil (OR 2.94).
Consuming organic fruit, contact with fresh produce, performing outdoor sport/recreational activities, and contact with dogs outside the home were protective factors (OR 0.64, 0.37, 0.54, and 0.51 respectively) ( Table 2).
Risk factors for human C. jejuni and C. coli infections. The significant risk factors for C. jejuni infection ( Table 2) resembled those for campylobacteriosis in general. These were: consuming chicken outside the home (OR 2.05) and both at home and outside (OR 4.52), consuming poultry other than chicken outside the home (OR 2.01), contact with garden soil (3.03), and summertime (OR 1.92). Protective factors for C. jejuni infection were consuming organic fruit (OR 0.64), contact with fresh produce (0.38), performing outdoor sport/ recreational activities (0.52), and contact with dogs outside the home (0.54).
For C. coli infection (Table 2), risk factors were consuming chicken outside the home (OR 3.02) and both at home and outside (OR 6.33), consuming beef at home (OR 2.89) and both at home and outside (OR 3.54), and consuming hamburger outside the home (OR 2.74). Female gender was associated with C. coli infection (OR 2.36), and so were the age groups 25-44 years (OR 7.64), 45-64 years (OR 6.27), and ≥ 65 years (OR 8.03) vs. the ≤ 4-year-olds. The only protective factor for C. coli infection was contact with fresh produce (OR 0.20). In the case-case comparison (Table 2), the only factor associated with C. coli infection besides female gender (OR 2.05) was consuming beef both at home and outside (OR 5.04).  (Table 3).

Risk factors for ruminant-assigned human campylobacteriosis. Contact with garden soil and hav-
ing both a local and a regional tap water provider in the household (as compared to having a local provider only) were the only risk factors for ruminant-assigned campylobacteriosis (OR 2.98 and 3.80, respectively) (  outdoor sport/recreational activities (OR 0.49). The risk for ruminant-assigned campylobacteriosis was higher in summer than winter (OR 2.94).

Risk factors for environmental water-assigned human campylobacteriosis.
The only factor associated with campylobacteriosis of probable environmental water origin was contact with garden soil (OR 3.27) ( Table 3).

Discussion
This study was performed to determine the Campylobacter genotypes causing human disease, their likely sources, and the associated risk exposures in Luxembourg, a country with a high campylobacteriosis incidence. A large number of cases (and controls), as well as non-human Campylobacter isolates, was included in this study, allowing for a multitude of analyses to infer the sources of human cases and to study source-specific risk factors in addition to those by Campylobacter species. Source attribution analysis confirmed that poultry is the main reservoir for both C. jejuni and C. coli, though a relatively high contribution from cattle to C. jejuni and from swine to C. coli was found, supporting previous evidence that C. jejuni is more prevalent than C. coli in cattle and that the inverse holds for swine 5,33,34 .
With 282 STs identified amongst 1289 human cases, our results indicate a considerable diversity of genotypes in Luxembourg's Campylobacter population. Rare genotypes were also relatively frequent, as STs occurring ≤ 5 times accounted for 27% of all cases. The top five STs and CCs identified here have been reported worldwide 5,16,19,26,[35][36][37][38][39] and rank amongst the top genotypes of other EU countries 5,19,37,39 , suggesting that these genotypes are endemic also in Luxembourg. Although ST-48 and its CC seemed to occur more often in summer, the other predominant genotypes were evenly distributed over demographic groups, urbanization degrees and seasons.
Poultry was confirmed to be the primary source of campylobacteriosis in Luxembourg, accounting for 61% of cases, followed by ruminants (33%). This ranking is in line with other studies from developed countries using the AIM 5,15,16,18,20,33 . Interestingly, our fraction of cases attributed to ruminants was somewhat high, but comparable to previous research from Switzerland 33 and Scotland 20 where 36% and 38% of C. jejuni cases were attributable to ruminants, respectively. Altogether, these findings support the growing body of evidence (reviewed by Stanley and Jones 40 ) indicating that the significance of non-poultry sources should not be underestimated. Campylobacter colonization of ruminants does not only relate to the potential contamination of meat and dairy products, but also to contamination of the environment, including water, from disposal of slaughterhouse effluents and farm slurry/manure. Contaminated irrigation water can transfer Campylobacter to growing leafy vegetables and herbs 41 . Ruminants are often implicated in cases of drinking water contamination with Campylobacter from agricultural run-off 42 . Our findings that drinking tap water was associated with increased risk for infection with ruminant-assigned vs. poultry-assigned STs and that the risk of infection with ruminant-assigned STs was higher in households with both a regional and a local water supplier, warrant further attention on this exposure.
While source attribution analysis quantified the relative contributions of the different sources to human cases, the risk factor analysis identified factors associated with increased or decreased risk for infection. Risk factors could be analysed in more detail for groups of STs with high probability to originate from a specific source, allowing for the identification of possible pathways by which these STs may have reached humans from their attributed sources. We found consuming chicken and other poultry to be a risk factor only when these were consumed outside the household, and for chicken the effect size was even higher when this was consumed both at home and outside, suggesting an effect of (multiple) exposures to chicken-assigned campylobacters besides domestic food handling and consumption. We hypothesized that this risk may be due to increased chance of encountering (higher doses of) Campylobacter strains different from those to which people are (usually) exposed in their kitchen, as repeated exposure to Campylobacter strains may lead to sufficient immunity to provide protection against (severe) clinical illness 24 . Whereas these differences were not so pronounced when splitting cases according to their attributed reservoirs, we found significant seasonal effects in the risk for poultry-assigned STs. The higher risk derived from consumption of chicken in winter as compared to summer was quite unexpected, as we hypothesized that this risk would be higher in summer since Campylobacter incidence in broilers follows the temperature pattern 43 . The increased risk posed by consuming chicken or other poultry during wintertime may therefore be due to whom (regular/irregular consumers) and to the manner this poultry is consumed in winter. It has recently been suggested that meat fondues, such as 'fondue chinoise' , a popular dish that is traditionally consumed during the festive season in countries like Switzerland and Germany, are a major driver of the winter epidemic peak of campylobacteriosis 44 . At these meals, the individual handling/preparation of raw poultry meat directly at the point of consumption poses a high risk for Campylobacter transmission and cross-contamination. Since meat fondues are popular in Luxembourg, the seasonal effect observed here might well be, to some extent, the result of this culinary tradition.
Surprisingly, we found that consuming beef (both at home and outside) was a distinct risk factor for C. coli infection. The risk for C. coli infection posed by consumption of beef products, particularly offal and tripe, has been reported before 7 . There is, however, little evidence that consumption of red meat poses a risk for campylobacteriosis in general 5 , as it is rarely contaminated with Campylobacter, and where contamination exists, it is at low doses 45 . Moreover, we found that female gender and older age groups had an increased risk for C. coli infection, possibly reflecting some hitherto unknown exposures linked to typically female-and adulthood-oriented tasks and eating habits.
Contact with garden soil stood out as a novel and consistent risk factor for Campylobacter infection, irrespective of the attributed reservoirs. Garden soil is often obtained by aerobic composting in which the organic matter in manure and other substrates is transformed and stabilized into humus-like products. When performed properly, this process eliminates coliform bacteria 46 , parasite (oo)cysts 47 , and some viruses 48  does not survive well in solid manure 48,49 , and compost is a very inhospitable environment, recent research has demonstrated that Campylobacter in (cattle) faeces can persist in compost for prolonged periods, calling into question the common belief that Campylobacter has poor fitness outside the host 50 . Further research into this newly identified potential at-risk exposure is therefore warranted.
Consumption of organic fruit, contact with (and presumably consumption of) fresh produce, and performing outdoor sport/recreational activities were protective factors. These are not unusual findings 5,7,21,22 , as it is generally believed that a diet rich in fruit and vegetables may have genuinely beneficial effects on health, and performing outdoor sport/recreational activities may be a proxy of healthy living. However, selection bias might also have occurred, as our controls might have been particularly motivated people with a generally healthier lifestyle. An advantage of having performed the case-case analysis was that selection (and recall) bias were minimised. Another protective factor was contact with dogs outside the household. This has been found in other studies 5,6 , and a possible explanation is that contact with dogs other than their own encourages people to undertake protective actions such as hand washing 5,6 .
A limitation of this study was the residual contribution to Pr values from sources other than those to which human cases were assigned. To address this, regression models were restricted to subsets of cases with the highest possible Pr for each source. However, this residual contribution, although minimized, could have masked or diluted some associations, or led to some additional associations, in the risk factor analysis. The latter option could be the case of consuming grilled sausages as a protective factor for ruminant-assigned campylobacteriosis, though sausages are more commonly made with pork than ruminant meats. We did not have information on underlying conditions and medication use, such as proton-pump inhibitors, that may have had an effect on campylobacteriosis, being therefore a major weakness of this study. For practical reasons, we could not test the controls for asymptomatic Campylobacter infection, which might lead to some misclassification. However, campylobacteriosis cases were identified by passive surveillance of gastroenteritis; thus, they represented the most severe infections occurring in the population. It follows, therefore, that our results mainly refer to severe campylobacteriosis. The human and source isolates were not perfectly contemporaneous, but this is unlikely to be a cause for concern. Indeed, Smid et al. 17 showed that while the MLST frequencies of two sources become increasingly dissimilar over time, this is far less problematic than the potential bias introduced by using data from geographically distant regions. We therefore used all the locally sourced (albeit not always fully contemporaneous) data at our disposal to minimize bias and reach satisfactory statistical power 17 . Moreover, the source isolates were obtained during or before (but not after) the human isolates; thus, given the unidirectional transmission (from sources to humans) on which the attribution implicitly relies, attributing human cases based on what goes or went on in the sources (i.e. the current and previous STs circulating in sources), rather than the other way around, would theoretically make the consequences of any temporal bias less severe than when attributing cases forward in time. While our sampling reflected the actual animal production in Luxembourg, we supplemented the local pig isolates with some isolates from Luxembourg's neighbouring countries, with which Luxembourg has intensive trading relationship of foodstuffs, including pork products. Therefore, the non-local pig strains were very representative of what could be found in Luxembourg's food market. Accordingly, bias due to the use of non-local data can be minimized by using data from geographically close areas 17 . Finally, for logistic reasons, cases and controls were interviewed in a different way, i.e. paper questionnaire for cases and web-assisted interview for controls. This might have influenced the participation (as this usually tends to be lower in web-based questionnaires, although we compensated for that by offering a financial incentive to controls) as well as the recall, which was probably more accurate for cases since they had more time and opportunities to remember and possibly correcting the answers, potentially leading to the exaggeration of some of the effects seen here.
In conclusion, MLST allowed us to investigate Campylobacter diversity in Luxembourg, though the predominant genotypes in humans were also typical of other EU countries. Regarding where these strains came from, we provided evidence that interventions against Campylobacter should not only focus on poultry, but also on ruminants. This has significant implications since there has been some reluctance to accept that also non-poultry Campylobacter strains pose a substantial public health burden 40 . Several risk factors were assessed to ascertain how Campylobacter might infect humans, and some of the identified risk factors were novel (e.g. consuming beef, contact with garden soil, and tap water provider type) or presented in a somewhat more nuanced way (e.g. consuming poultry within or outside the household) than has been shown previously. Combining epidemiological and genotype-based source attribution data was helpful in enhancing risk factor characterization. This approach helped supporting and generating hypotheses, such as that of the potential role of meat fondues for poultry-associated campylobacteriosis during wintertime, as well as the increased risk for ruminant-associated campylobacteriosis linked to drinking tap water and having both a local and regional water provider. From a larger perspective, this study provided insights in the public health importance of different Campylobacter genotypes, sources and risk factors in a high-incidence setting, enhancing our understanding of the underlying transmission pathways.