Two Distinct Patterns of Clostridium difficile Diversity Across Europe Indicating Contrasting Routes of Spread

Whole-genome sequencing of Clostridium difficile across Europe demonstrates 2 distinct transmission patterns. Healthcare-associated ribotypes showed genetic clustering by country and region, and widespread fluoroquinolone resistance. However, multiple common ribotypes showed no within-country clustering, consistent with Europe-wide dissemination via nonhealthcare routes/sources.

isolates from pigs, healthy farmers, and CDI case patients in the Netherlands [27]. Ribotype 078 infections are more likely to be community acquired and affect younger patients [28]. In the United States in 2011, ribotype 078/NAP7, ribotype 002/NAP6, and to a lesser extent ribotype 020/014/NAP4, occurred predominantly in community-associated cases, based on surveillance definitions [26]. Community-associated ribotype strains may be widely distributed in the environment; for example, ribotypes 014, 015, and 078 were all identified at Swiss wastewater treatment plants [29].
A point-prevalence survey (EUCLID) has demonstrated marked Europe-wide variation in both CDI and testing rates [30] and prevalent ribotypes [31]. We determined the extent of Europe-wide diversity within the 10 most prevalent ribotypes from that study to investigate whether different modes of transmission could plausibly occur across these ribotypes. We also investigated the relationship between geographic distribution and antimicrobial susceptibilities to explore potential drivers of healthcare-and community-associated infections.

Study Design and Samples
Samples were obtained during a European, multicentre, prospective, point-prevalence study of hospital inpatients with diarrhea (EUCLID) [30]. The study included 20 European countries, each with a national coordinating laboratory. National coordinators selected the 482 participating hospitals to cover all major geographic regions, at a rate of 1 per 1 million population. Ethical approval was granted in the Netherlands, Sweden, and Slovenia; the remaining countries did not require such approval because the study was classed as surveillance.
All inpatient diarrheal samples submitted to the microbiology laboratory of the participating hospitals on 2 sampling days (1 in winter [December 2012 through January 2013], and 1 in summer [July-August 2013]) were eligible, irrespective of the original test(s) requested. Samples were included from all patients >2 years old currently occupying a hospital bed, including patients in the emergency department but excluding outpatient clinic attendees. Only 1 sample per patient on each sampling day was included. All samples were tested for CDI at the national coordinating laboratory using the same standardized method (C. DIFF QUIK CHEK COMPLETE; Techlab); those that were glutamate dehydrogenase or toxin A/B enzyme immunoassay positive were also cultured and isolates sent to the Leeds General Infirmary microbiology laboratory for polymerase chain reaction ribotyping [32]. A total of 7297 inpatient diarrheal samples were tested at 482 European hospitals [30], resulting in 1211 isolates sent for ribotyping, with distribution and characteristics (eg, age) as described elsewhere [31].

Sequencing
Isolates from the 10 most prevalent toxigenic ribotypes underwent whole-genome sequencing. DNA was extracted from subculture of a single colony and sequenced using the Illumina HiSeq 2500 system. Sequence data were processed as described elsewhere [6,33], mapping reads to the C. difficile 630 reference genome (AM180355.1), except for isolates from ribotypes 027 and 078, whose reads were mapped to CD196 (NC_013315.1) and M120 (FN665653.1), respectively. Sequences were compared using single-nucleotide polymorphisms (SNPs), obtaining differences between sequences from maximum likelihood phylogenies corrected for recombination [34]. BEAST (version 2.4.3) [35] and BASTA (version 2.3.0) [36] software were used to reconstruct time-scaled phylogenies and to infer the geographic location of each ancestral lineage.
A strength of this approach is that it is possible to reconstruct past geographic locations and patterns of geographic spread for each lineage, despite sequencing isolates at only 2 time points. Antimicrobial resistance determinants for fluoroquinolones, tetracycline, clindamycin/macrolides, aminoglycosides, and fidaxomicin were identified from mapped data and de novo assemblies.  Table 3 for study sample metadata).

Country-based Clustering
For each ribotype, the median number of SNPs between isolates within a country (SNP within ), between sequences across different countries (SNP across ), and overall were calculated. Country-based sequence clustering was assessed using the SNP within /SNP across ratio; values closer to 0 indicate greater country-based clustering. To determine whether observed SNP within / SNP across ratios were compatible with the null hypothesis of no country-based clustering, for each ribotype, the country labels of each tip of the phylogenetic tree were permuted 1000 times. For each permutation, the ratio was calculated. Country-based clustering was supported if the observed ratio was less than the 2.5% quantile of the distribution of permuted ratios. Because both the overall diversity within each ribotype and the number of samples varied, SNP within /SNP across ratios are not directly comparable between ribotypes. A similar approach was used to determine, by ribotype, whether there was evidence of within-hospital clustering in sequences obtained from the same country.

Ribotype Diversity Within and Between Countries
Within ribotypes, the overall median number of SNPs between isolates ranged from 9 to 298 (Table 1). Using time-scaled phylogenies, the most recent common ancestor for most ribotypes was estimated to date from the 18th to the mid-20th century, with the exception of ribotypes 176 and 356, which emerged more recently, Five ribotypes had evidence of clear within-country clustering: ribotype 356, found only in Italy; ribotype 018, predominantly isolated in Italy; ribotype 176, with distinct Czech and German clades; ribotype 001/072 including distinct German, Slovakian, and Spanish clades; and ribotype 027, with multiple predominantly country-specific clades, including in Hungary, Italy, Germany, Romania, and Poland ( Figure 2 and Supplementary Figures 1-3). For these ribotypes, the SNP within / SNP across ratio was significantly lower than expected by chance (Table 1); for example, for ribotype 027 the ratio was 0.60, versus an expected value of .93-1.07 without any country-based clustering (95% confidence interval from permutation test).
By contrast, there was no evidence of within-country clustering for ribotypes 078, 015, 002, 014, and 020 ( Figure 3 and Supplementary Figures 4-7), consistent with recent Europewide distribution. For these ribotypes, the SNP within /SNP across ratio was approximately 1 and fell within the 95% confidence interval expected by chance, assuming no country-based clustering (Table 1).
For most ribotypes the median number of SNPs between sequences from the same hospital was less than that between sequences within countries-for example, 8 (interquartile range [IQR], 3-26) versus 49 (17-71) SNPs for ribotype 027 (Table 3), compatible with onward transmission within institutions. For ribotypes showing within-country clustering, based on permutation tests, there was evidence of smaller-scale within-institution clustering for ribotypes 027, 001/072, and 018. There was limited local and national diversity within ribotypes 176 and 356. Among ribotypes compatible with a recent pan-European distribution, ribotypes 014, 002, and 015 had lower median numbers of within-hospital than within-country SNPs, but ribotypes 020 and 078 did not. The median numbers of SNPs within countries between pairs of sequences not from the same hospital were very similar to those within countries overall, demonstrating that, where observed, country-based clustering is not just driven by hospital-based clustering. Distribution of isolates from sequenced ribotypes by country. Study hospitals were recruited at a rate of 1 per million head of population; hence, differences in total isolates obtained between countries represent differences in within-country testing rates and incidence of Clostridium difficile infection (see Table 2). To calculate the number of miles between the most closely related isolates, within each ribotype, a minimum spanning tree was constructed using SNP distances as edge weights and isolates as nodes. The properties of the edges (SNPs and miles) of the minimum-spanning tree were analyzed, such that each isolate but 1 contributes 1 data point (see Figure 5). b CIs were based on permutation tests (see Methods).

Country-based Clustering and Antimicrobial Resistance
There was a strong association between within-country clustering and the presence of fluoroquinolone resistance determinants ( Figure 4). The proportion of isolates with fluoroquinolone resistance was significantly higher in the 5 ribotypes showing within-country clustering than in the remaining 5 ribotypes (rank sum P = .009). There was no association between clustering and resistance determinants for 3 antimicrobials, tetracyclines (P = .75), aminogylcosides (P = .41), and clindamycin/ macrolides (P = . 22). No isolates had previously described fidaxomicin resistance determinants. All assessable isolates from ribotypes 027, 176, and 356 had fluoroquinolone resistance mutations. Only a subset of ribotype 018 and 001/072 isolates had resistance mutations allowing the date of the most recent common ancestor of each resistant clade to be estimated. In ribotype 018, fluoroquinolone resistance was estimated to have emerged in 2001 (95% HPD, 1993-2008), followed by a clonal expansion restricted to Italy, from which ribotype 356 descended (Supplementary Figure 1). In ribotype 001/072 fluoroquinolone resistance was strongly associated with the presence of ermB, conferring resistance to clindamycin and macrolides. Country  Figure 2).

Geographic Distance and Genetic Clustering
Ribotypes that clustered by country also showed evidence of tighter geographic clustering by region ( Figure 5 and Table 1). The median geographic distance between each isolate and the next most closely genetically related isolate (using minimum spanning trees to avoid double counting; see legend to Figure

DISCUSSION
Here we show 2 distinct patterns of C. difficile ribotype spread. Ribotypes previously associated with healthcare-based transmission, including ribotypes 027 and 001/072, demonstrated evidence of country, regional, and within-hospital clustering. This geographic structure was preserved over several years of C. difficile evolution in each ribotype. In many instances the most plausible source of all cases of a particular ribotype, or subclade of a ribotype, in a country could be explained by a single imported case, suggesting that intercountry transmission of C. difficile with these ribotypes is unusual.
By contrast, other ribotypes (002, 014, 015, 020, and 078) showed no evidence of country-based clustering, and in many cases the most closely genetically related strains were separated by several hundred miles. This suggests that these ribotypes are widely disseminated across Europe, and sustained local transmission is less common than with the healthcare-associated ribotypes described above. It is likely that there is still some secondary transmission of these ribotypes within hospitals. Although by sampling on only 2 days 6 months apart this study did not fully capture the extent of local transmission, we did find evidence of within-hospital clustering in ribotypes 014, 002, and 015.
However, the overall pattern observed is consistent with a dominant route of spread other than healthcare. The association of ribotypes 078 [27] and 014 [37] with pig farming, and previous findings of C. difficile in food, raises the possibility that the food chain may be an important vector for these ribotypes. The most recent ancestor of pairs of closely related samples from these ribotypes was often 5-10 years earlier. Therefore, the spread of these strains across Europe may have occurred several years previously with persistence in local reservoirs; alternatively, more recent spread may have occurred from a reservoir supporting a more diverse C. difficile population than that found in a single human host. We also found that country-based clustering of C. difficile, and therefore probably healthcare-associated transmission, was strongly associated with fluoroquinolone resistance, not just in ribotype 027 but also in ribotypes 001/072, 018/356, and 176. In ribotype 001/072, fluoroquinolone resistance was found together with ermB conferring resistance to clindamycin and macrolides, possibly reflecting fluoroquinolone resistance acquisition by an already successfully healthcare-adapted clindamycin-resistant clone or vice versa; otherwise, there was no evidence of associations between country-based clustering and resistance to other antimicrobials, including tetracycline and aminoglycosides. We found no evidence of fidaxomicin resistance. Clostridium difficile is widely resistant to cephalosporins; therefore, although cephalosporins can be a risk factor for some infections, their use is unlikely to differentially affect the transmission of some ribotypes.
Supporting the importance of fluoroquinolone use as a driver of healthcare-associated CDI, Dingle et al [14] demonstrated that declining CDI incidence in England this millennium can be accounted for by falls in fluoroquinolone-resistant   C. difficile, while levels of fluoroquinolone-susceptible C. difficile have remained largely unchanged. Clonal expansions, representing rapid successful transmission, followed the acquisition of fluoroquinolone resistance in 2 ribotype 027 lineages [4]. Here we provide additional examples of clonal expansions following acquisition of fluoroquinolone resistance in ribotypes 018 and 001/072 dating from the 1990s onward, consistent with the launch of fluoroquinolones in Europe in 1987.
In contrast to previous findings of He et al [4], who reported frequent and extensive long-range geographic transmission within the United Kingdom of ribotype 027, transmission events between European countries were relatively rare. They typically involved geographically adjacent countries, although we observed probable spread from the United Kingdom to Poland as well, possibly reflecting the population flows between these countries. We were able to observe marked European country-based clustering of some ribotypes because a sufficiently large number of isolates was obtained from each country in our study.
A strength of our study is its unbiased sampling of C. difficile across Europe, with samples obtained on a single day in summer and winter, and the number of hospitals from each country based on its population. However, because only 2 single time points were captured, analyses will tend to underestimate the extent of local secondary transmission for all ribotypes, because this results in genetically related cases that typically occur close in time but are not often sampled on the same day.
Other study limitations include the dependence on the rate of submission of diarrheal samples to hospital laboratories. Despite efforts to account for some underreporting by testing all samples, irrespective of the original test requested, patient testing rates varied widely between countries (Table 2). This may have affected the relative prevalence of different ribotypes, but our conclusions regarding within-country clustering should be robust to this. In addition, the reconstruction of the ancestral country for each lineage is only an approximation, given that, owing to limited sample size, we had to assume the same effective bacterial population size in each country, and the same rates of exchange of strains between all countries. The study is also limited by only considering transmission between countries as discrete entities. The study was insufficiently powered to explore regional transmission, including the impact of referral networks, hospital characteristics, and differing patient catchments. These could be investigated in larger national studies, as has been done for other healthcare-associated pathogens, for example, methicillin-resistant Staphylococcus aureus [38].
Our findings have important implications for C. difficile surveillance and control. Its transmission may be more heterogeneous than is usually assumed, with different interventions required for predominantly healthcare-associated and community-associated lineages. Surveillance programs may need to consider routine use of typing to allow the incidence of commonly healthcare-associated strains to be followed independently of community-associated strains. Although this is theoretically possible with surveillance definitions based on the timing of the last healthcare exposure, recent healthcare exposure may merely be a marker of exposure to a precipitant of symptoms (eg, antibiotics), rather than the location of C. difficile acquisition.
Measures to control healthcare-associated C. difficile are well established. However, it is less clear how to control the spread of ribotypes that are widely distributed across Europe. Further studies are required to better understand potential vectors for transmission, for example, contaminated food or use of animal-based fertilizers. In principal, better antimicrobial stewardship should reduce the risk of all types of CDI, by reducing patients' exposure to provocative antibiotics. However, because C. difficile antimicrobial resistance is associated with country-based clustering, stewardship may have a greater effect on healthcare-associated strains. If the nonclustered ribotypes are found very widely in the environment, prevention of exposure may be particularly challenging, and other strategies in addition to antimicrobial stewardship may need to be explored to reduce infections, including vaccination or prophylaxis during at-risk periods, for example, around healthcare exposures.