Tag Archives: C. difficile research

Publication: Multicenter Prevalence Study Comparing Molecular and Toxin Assays for Clostridioides difficile Surveillance, Switzerland






Andreas F. Widmer, Reno Frei, Ed J. Kuijper, Mark H. Wilcox, Ruth Schindler, Violeta Spaniol, Daniel Goldenberger, Adrian Egli, Sarah Tschudin-Sutter , and Kuijper
Author affiliations: University Hospital Basel, Basel, Switzerland (A.F. Widmer, R. Frei, R. Schindler, V. Spaniol, D. Goldenberger, A. Egli, S. Tschudin-Sutter)Leiden University Medical Center, Leiden, the Netherlands (E.J. Kuijper)Leeds Institute of Biomedical and Clinical Sciences, University of Leeds, and Leeds Teaching Hospitals, Leeds, UK (M.H. Wilcox)


Public health authorities in the United States and Europe recommend surveillance for Clostridioides difficile infections among hospitalized patients, but differing diagnostic algorithms can hamper comparisons between institutions and countries. We compared surveillance based on the detection of C. difficile by PCR or enzyme immunoassay (EIA) in a nationwide C. difficile prevalence study in Switzerland. We included all routinely collected stool samples from hospitalized patients with diarrhea in 76 hospitals in Switzerland on 2 days, 1 in winter and 1 in summer, in 2015. EIA C. difficile detection rates were 6.4 cases/10,000 patient bed-days in winter and 5.7 cases/10,000 patient bed-days in summer. PCR detection rates were 11.4 cases/10,000 patient bed-days in winter and 7.1 cases/10,000 patient bed-days in summer. We found PCR used alone increased reported C. difficile prevalence rates by <80% compared with a 2-stage EIA-based algorithm.


Since its identification as a cause of antibiotic-associated pseudomembraneous colitis in 1978 (1), Clostridioides difficile has emerged as a major healthcare-associated pathogen worldwide. In the United States, C. difficile infection (CDI) rates doubled during 1996–2003 (2), and rates of CDI were reported to be 76.9 cases/10,000 discharges in 2005 (3). In a more recent national point-prevalence study including US healthcare facility in-patients, 13/1,000 patients were found to be either infected or colonized (4), a higher rate than had been previously estimated. In a national point-prevalence study of nosocomial infections in the United States, C. difficile was the most common causative pathogen overall (5). The increase largely has been attributed to the emergence of the hypervirulent strain, PCR ribotype 027 (RT027), which was identified as causative strain in 82% of cases during CDI outbreaks in Quebec, Canada, during 2001–2003 and accounted for 31% of all cases of healthcare-associated infections in the United States in 2011 (69). In Europe, CDI incidence varies across hospitals and countries with a weighted mean of 4.1 cases/10,000 patient-days per hospital in 2008 (10). The most recent study on CDI prevalence in Europe suggests an increase in the number of cases, reporting a mean of 7.0 cases/10,000 patient-bed days and ranging among countries from 0.7 to 28.7 cases/10,000 patient-bed days (11). The most common ribotype identified was RT027, which was detected in 4 countries: Germany, Hungary, Poland, and Romania (11).

To estimate and compare the burden of CDI across the United States, the US Centers for Disease Control and Prevention (CDC) began population-based CDI surveillance in 10 locations in 2011 (12). The European Centre for Disease Prevention and Control (ECDC) began coordinating CDI surveillance in acute care hospitals in Europe in 2016 (13). Both authorities provide case definitions based on different diagnostic approaches, including detection of C. difficile toxin A and B by enzyme immunoassay (EIA) or detection of toxin-producing C. difficile organisms by PCR. However, the use of different diagnostic algorithms to detect C. difficile might hamper comparisons between institutions and countries. Therefore, in a nationwide C. difficile multicenter prevalence study in Switzerland, we systematically compared surveillance measures based on detection of C. difficile in stool by either PCR as a stand-alone test or by a 2-stage algorithm consisting of an EIA to detect glutamate dehydrogenase (GDH) and toxins A and B.


Study Design

We performed a nationwide multicenter prevalence study of toxigenic C. difficile detected in stool samples routinely collected from hospitalized patients with diarrhea. Our study followed the design of a previous point-prevalence study for maximal comparability between our results and data from Europe (11). University Hospital Basel, a tertiary care center in Switzerland, coordinated the study. All hospitals participating in Swissnoso (https://www.swissnoso.chExternal Link), a national infection prevention network, were asked to participate. The Swissnoso network consists of 85 acute care hospitals that account for a total of 26,341 beds.

The Ethics Committee Northwest and Central Switzerland (Ethikkommission Nordwest-und Zentralschweiz) issued a declaration of no objection for this study. We adhered to STROBE guidelines for reporting on observational studies (14).

Sample Collection

All stool samples collected from inpatients >1 year of age with diarrhea that were submitted to the microbiology laboratory on 2 specified sampling days, 1 in winter and 1 in summer, in 2015 were eligble for inclusion. Only 1 sample per patient was included. In addition, stool samples that tested positive for toxigenic C. difficile <1 week prior to each study day also were collected from all institutions to obtain a more detailed estimate of ribotype distribution in Switzerland.

We collected the following institutional data for all hospitals and their affiliated microbiology laboratories: contact information; detailed information regarding laboratory diagnostics in place; and detailed information on the total number of admissions, number of beds, and number of patients hospitalized on the 2 days of the study. We also collected information on the total number of diagnosed CDI cases at each institution during the study year. For each eligible stool sample, we collected the following data: date of sample collection, age and gender of patient, ward location and clinical specialty, institution, whether a C. difficile test was ordered by the treating physician, and result of the C. difficile test if testing was performed at the local laboratory.


We tested all stool samples at the Division of Clinical Microbiology of the University Hospital Basel by using a 2-stage algorithm consisting of EIA and PCR. We performed EIA to detect GDH and toxins A and B by using C. DIFF QUIK CHEK COMPLETE (Techlab, https://www.techlab.comExternal Link), following the manufacturer’s instructions. We then performed PCR to detect the toxin B gene by using the RealStar PCR Kit (Altona Diagnostics, https://www.altona-diagnostics.comExternal Link). For detected C. difficile, we performed strain typing by using high-resolution capillary gel-based PCR ribotyping according to the method previously described by Stubbs et al. (15).


We calculated reported and measured rates of detected toxigenic C. difficile per 10,000 patient bed-days across participating institutions. We compared differences in testing algorithms for detection of toxigenic C. difficile across institutions in Switzerland and performance characteristics of diagnostic algorithms. We considered the proportion of missed toxigenic C. difficile cases and ribotype distributions as additional outcomes. We further assessed the proportion of laboratories using optimized C. difficile diagnostic tests, which we defined as using an algorithm recommended by the European Society of Clinical Microbiology and Infectious Diseases (16).

Statistical Analyses

We separately calculated rates for each diagnostic algorithm performed in the coordinating center laboratory. In addition, we separately calculated rates for dedicated children’s hospitals. We defined missed C. difficile cases as those in which tests were negative at the participating hospital’s laboratory but positive at our institution. We used descriptive statistics to report ribotypes and differences in diagnostic algorithms across all participating institutions. All analyses were performed in Stata version 15.1 (StataCorp, https://www.stata.comExternal Link).


Figure 1. Distribution of centers participating in a prevalence study comparing molecular and toxin assays for nationwide surveillance of Clostridioides difficile, Switzerland. Red circles represent the location of participating centers.

Participating institutions included 76/85 (89.4%) institutions belonging to the Swissnoso network. Among participating institutions, 5 were academic teaching hospitals, 3 were dedicated children’s hospitals, and 36 were affiliated microbiology laboratories. Participating institutions were distributed across all geographic regions of Switzerland (Figure 1).

Participating institutions reported collecting a fecal sample for microbiological workup in »65% (SD +25%) of all patients with hospital-onset diarrhea. Among participating institutions, 15/76 (19.7%) did not begin CDI treatment before fecal sample collection. Among institutions that initiated treatment before collecting fecal samples, 23/76 (30.3%) began treatment in <2% of patients, 12/76 (15.8%) began treatment in 3%–5% of patients, 8/76 (10.5%) began treatment in 6%–10% of patients, and 1 (1.3%) began treatment in 11%–20% of patients. The other 17 (22%) institutions were not able to provide an estimate of these data.

Overall, 354 stool samples were submitted to the coordinating center, of which 338 were eligible for study inclusion; 16 samples were excluded because they were not liquid, their submitted data were incomplete, or they were duplicate samples from 1 patient. Among 338 samples included, 250 were collected as part of the point-prevalence study. We excluded 8 of these because the samples were collected from patients <1 year of age. In all, we included 242 samples in the point-prevalence study.

Diagnostic Algorithms

Figure 2. Testing algorithms at the 36 laboratories participating in a prevalence study comparing molecular and toxin assays for nationwide surveillance of Clostridioides difficile, Switzerland. EIA, enzyme immunoassay; GDH, glutamate dehydrogenase; NAAT, nucleic…

Among the 36 participating laboratories, 1 routinely tested all diarrheal stool samples for toxigenic C. difficile and 35 tested only if a specific test was requested. Optimized diagnostic tests for detection of toxigenic C. difficile were used by 58% (21/36) of laboratories in the winter sampling period and by 61% (22/36) in the summer sampling period. Among laboratories not following the recommendations of the European Society of Clinical Microbiology and Infectious Diseases (16), 9 in the winter sampling period and 10 in the summer sampling period used a nucleic acid amplification test (NAAT) alone, and 5 in the winter sampling period and 3 in the summer sampling period used EIAs for A and B toxins either as a standalone test or as an initial screening test. Only 1 laboratory reported having established PCR ribotyping methodologies (Figure 2).

Point-Prevalence Analyses

We collected demographic characteristics of patients whose stool samples tested positive by our testing algorithms (Table 1). C. difficile tests were required and performed for 68% (165/242) of stool samples; 6% (27/165) were reported as positive by the affiliated microbiology laboratory.

At the coordinating center, we detected C. difficile in 9% (21/242) of samples by EIA for GDH and A and B toxins and in 12% (30/242) of samples by PCR alone. Among all 27 samples reported as positive by the participating centers, we confirmed 18 (67%) by EIA and 24 (89%) by PCR. Among 138 samples reported as negative by the participating centers, 1 (1%) sample tested positive by EIA and 3 (2%) tested positive by PCR at the coordinating center. Among 77 samples not tested for C. difficile at the participating centers, we detected C. difficile in 2 (3%) by EIA and in 3 (4%) by PCR. Among 21 stool samples that tested positive by EIA at the coordinating center, a C. difficile test was not requested in 2 (10%) cases. Among 30 samples that tested positive by PCR at the coordinating center, a C. difficile test was not requested in 3 cases (10%; Table 2).

Measured detection and testing rates of toxigenic C. difficile were higher than the reported rates across all participating institutions (Table 3). Depending on the diagnostic algorithm applied, the largest difference in prevalence across all institutions was measured during the winter sampling period, which had a prevalence of 6.4 cases/10,000 patient bed-days by EIA and 11.4 cases/10,000 patient bed-days by PCR alone. Thus, across all institutions, rates of toxigenic C. difficile detection by PCR alone were <80% higher than detection rates by EIA for GDH and A and B toxins. In addition, detection rates by PCR alone were <100% higher in dedicated children’s hospitals (Table 3).

Ribotype Distribution

Figure 3. Distribution of PCR ribotypes among 107 samples collected in a prevalence study comparing molecular and toxin assays for nationwide surveillance of Clostridioides difficile, Switzerland. *Unknown ribotype.

We cultured and ribotyped 107 toxigenic C. difficile strains, 29 from the 2 point-prevalence days and 78 collected <1 week before each prevalence day. We identified a large diversity of C. difficile ribotypes, 23 (22%) had not been referenced before. The ribotypes most commonly identified included RT014 (12/107; 11%), presumably hypervirulent RT078 (9/107; 8%), RT001 (7/107; 7%), and RT002 (7/107; 7%) (Figure 3).



In this nationwide multicenter study, we found that PCR as a stand-alone test increased reported C. difficile prevalence rates <80% compared with a 2-stage EIA-based algorithm. At first glance, this finding was not surprising given the higher sensitivity of EIA (16). However, the fact that our results and conclusions are based on a nationwide cohort representing all geographic regions of Switzerland adds to the study’s credibility. In addition, our results strengthen the advice of the European Society of Clinical Microbiology and Infectious Diseases study group for C. difficile against use of a single commercial test for diagnosing CDI because of the low positive predictive values when CDI prevalence is low, 46% at a CDI prevalence of 5% (16). However, CDC and ECDC protocols for CDI surveillance define a case of CDI as the combination of diarrheal stool and a positive PCR (12,13). In addition, the clinical practice guidelines for CDI in adults and children published by the Infectious Diseases Society of America and Society for Healthcare Epidemiology of America recommend testing by different approaches, such as multistep algorithms or NAAT, depending on the degree of clinical suspicion (17). Based on a systematic review and meta-analysis, the American Society of Microbiology also recommends different approaches, including NAAT-only testing, and algorithms that include GDH and NAAT or GDH, toxins, and NAAT (18). Although these recommendations stand to reason for detection of CDI in individual patients, our results challenge their utility for meaningful comparisons in surveillance studies and suggest that uniform definitions should be provided.

On both point-prevalence days, we noted a higher nationwide rate of toxigenic C. difficile than previously reported in incidence studies performed at different institutions in Switzerland (1921). Our findings suggest that CDI rates have increased during the last decade in Switzerland, a finding that is in line with reports from other countries in Europe (11). Using the same diagnostic algorithm, diagnostic test, and a similar study design to the multicenter point-prevalence study of CDI in hospitalized patients with diarrhea in Europe, the nationwide mean prevalence rates are comparable in Switzerland (mean 6.1 cases/10,000 patient bed-days) and Europe (7.0 cases/10,000 patient bed-days) (11). Because we only included liquid stools in our study, our mean prevalence rate of 9.3 cases/10,000 patient bed-days measured by PCR fulfills the ECDC case definition and further shows that CDI is increasing substantially nationwide.

We found a lower proportion of missed detection of toxigenic C. difficile in Switzerland (9.5%), driven by the absence of clinical suspicion, compared with Europe (23%), which equates to 1 missed case of C. difficile per day among the included institutions in Switzerland. False-negative testing accounted for 1 additional missed diagnosis during both point-prevalence days, which extrapolates to »550 missed cases of C. difficile per year among hospitals across the nation.

We detected a variety of different RTs during our study period, 21% of which had not been referenced before. Of note, we did not recover hypervirulent RT027, but RT078 was the third most common strain circulating in Switzerland during our study. In contrast, a point-prevalence study in Europe identified RT027 as the most commonly circulating strain during its study period but did not detect RT078. RT078 has been associated with similarly severe disease manifestations as RT027, but RT078 has been reported to affect younger patients and to be linked more commonly with community-associated disease in the Netherlands (22). RT078 has been isolated from piglets with diarrhea, possibly suggesting ongoing transmission by introduction to the food chain because isolates from humans and pigs were found to be highly genetically related (22). A component of RT078 infections also was reported in Northern Ireland, which has a large pig population and »1:1 ratio of cattle to humans (23). In Switzerland, RT078 has been isolated previously from 6 wastewater treatment plants, suggesting its dissemination in the community (24). Except for both hypervirulent RT027 and RT078, we identified other similarities in RT distribution between Switzerland and the rest of Europe; RT014, RT001, RT002, and RT020 were among the 10 most commonly identified ribotypes in both settings.

Our study has some limitations, most of which are intrinsic to point-prevalence studies. First, our study only reflects frequency of toxigenic C. difficile detected on 2 days in 2015; therefore, we cannot draw solid conclusions regarding incidence. We expanded the timeframe for assessing the distribution of ribotypes circulating in Switzerland by an additional week for each prevalence day, but this still represents a limited collection of the true incidence. Second, we cannot rule out introduction of bias to testing policies at the participating hospitals, which might have increased testing on the 2 point-prevalence days. However, we did not provide any promotional information regarding our study, so alterations in daily clinical practice among treating physicians is unlikely on these 2 days. Third, we only included liquid stool samples for analyses, but we did not consider any other preanalytical factors, such as the use of laxatives, for testing eligibility. Finally, we applied surveillance definitions recommended by CDC and ECDC rather than defintions used for the clinical diagnosis of CDI in individual patients, such as detection of C. difficile in the context of symptoms related to CDI. Therefore, we cannot rule out detection of toxigenic C. difficile from colonization rather than infection in some cases.

In conclusion, this nationwide multicenter study reveals that PCR as a stand-alone test results in an increase of C. difficile prevalence rates of <80% compared with a 2-stage EIA-based algorithm. Our findings underscore the need for consistent testing algorithms for meaningful interinstitutional and nationwide comparisons. Our results also challenge the utility of the current CDC and ECDC case definitions and highlight the need for uniform recommendations on diagnostic approaches.


Dr. Widmer is head of the infection control program at University Hospital Basel, University of Basel, Switzerland. His research interests include all aspects of Clostridioides difficile and the epidemiology and prevention of hospital-acquired infections.



We acknowledge and thank the ESCMID (European Society of Clinical Microbiology and Infectious Diseases) Study group for C. difficile (ESGCD) for professional support. We also thank all participating centers and laboratories (Appendix).

Astellas Pharmaceuticals Europe provided financial support for this study. The funder did not influence the study design and did not contribute to data collection, data analysis, data interpretation, or writing of the report. Astellas Pharma Europe reviewed the report for factual accuracy before submission, in line with the terms of the funding agreement. The corresponding author had full access to all data in the study and had final responsibility for the decision to submit for publication. Alere provided C. DIFF QUIK CHEK COMPLETE test kits for conducting EIAs to detect GDH and toxins A and B.

The authors declare the following possible conflicts of interest: A.W. is a member of the Astellas and Merck Sharp & Dohme Corp. advisory boards for C. difficile and reports grants from the Swiss National Science Foundation. S.T.-S. is a member of the Astellas and Merck Sharp & Dohme Corp. advisory boards for C. difficile and reports grants from the Swiss National Science Foundation (grant nos. NRP72 and 407240_167060), the Gottfried und Julia Bangerter-Rhyner Stiftung, and the Fund for the Promotion of Teaching and Research of the Voluntary Academic Society, Base


  1. Bartlett  JGMoon  NChang  TWTaylor  NOnderdonk  ABRole of Clostridium difficile in antibiotic-associated pseudomembranous colitis. Gastroenterology1978;75:77882DOIExternal LinkPubMedExternal Link
  2. McDonald  LCOwings  MJernigan  DBClostridium difficile infection in patients discharged from US short-stay hospitals, 1996-2003. Emerg Infect Dis2006;12:40915DOIExternal LinkPubMedExternal Link
  3. Elixhauser  AJhung  MClostridium difficile-associated disease in U.S. hospitals, 1993–2005: statistical brief #50; Healthcare Cost and Utilization Project (HCUP). Rockville (MD, USA): Agency for Healthcare Research and Quality; 2008 [cited 2019 May 26]. https://www.hcup-us.ahrq.gov/reports/statbriefs/sb50.jspExternal Link
  4. Jarvis  WRSchlosser  JJarvis  AAChinn  RYNational point prevalence of Clostridium difficile in US health care facility inpatients, 2008. Am J Infect Control2009;37:26370DOIExternal LinkPubMedExternal Link
  5. Magill  SSEdwards  JRBamberg  WBeldavs  ZGDumyati  GKainer  MAet al.Emerging Infections Program Healthcare-Associated Infections and Antimicrobial Use Prevalence Survey TeamMultistate point-prevalence survey of health care–associated infections. N Engl J Med2014;370:1198208DOIExternal LinkPubMedExternal Link
  6. Pépin  JSaheb  NCoulombe  MAAlary  MECorriveau  MPAuthier  Set al. Emergence of fluoroquinolones as the predominant risk factor for Clostridium difficile-associated diarrhea: a cohort study during an epidemic in Quebec. Clin Infect Dis2005;41:125460DOIExternal LinkPubMedExternal Link
  7. Loo  VGPoirier  LMiller  MAOughton  MLibman  MDMichaud  Set al. A predominantly clonal multi-institutional outbreak of Clostridium difficile-associated diarrhea with high morbidity and mortality. N Engl J Med2005;353:24429DOIExternal LinkPubMedExternal Link
  8. O’Donoghue  CKyne  LUpdate on Clostridium difficile infection. Curr Opin Gastroenterol2011;27:3847DOIExternal LinkPubMedExternal Link
  9. Lessa  FCMu  YBamberg  WMBeldavs  ZGDumyati  GKDunn  JRet al. Burden of Clostridium difficile infection in the United States. N Engl J Med2015;372:82534DOIExternal LinkPubMedExternal Link
  10. Bauer  MPNotermans  DWvan Benthem  BHBrazier  JSWilcox  MHRupnik  Met al.ECDIS Study GroupClostridium difficile infection in Europe: a hospital-based survey. Lancet2011;377:6373DOIExternal LinkPubMedExternal Link
  11. Davies  KALongshaw  CMDavis  GLBouza  EBarbut  FBarna  Zet al. Underdiagnosis of Clostridium difficile across Europe: the European, multicentre, prospective, biannual, point-prevalence study of Clostridium difficile infection in hospitalised patients with diarrhoea (EUCLID). Lancet Infect Dis2014;14:120819DOIExternal LinkPubMedExternal Link
  12. US Centers for Disease Control and PreventionClostridioides difficile infection (CDI) tracking. [cited 2019 May 26]. https://www.cdc.gov/hai/eip/cdiff-tracking
  13. European Centre for Disease Prevention and Control. Surveillance and disease data for Clostridium difficile infections. [cited 2019 May 26]. https://ecdc.europa.eu/en/clostridium-difficile-infections/surveillance-and-disease-dataExternal Link
  14. von Elm  EAltman  DGEgger  MPocock  SJGøtzsche  PCVandenbroucke  JPSTROBE InitiativeThe Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet2007;370:14537DOIExternal LinkPubMedExternal Link
  15. Stubbs  SLBrazier  JSO’Neill  GLDuerden  BIPCR targeted to the 16S-23S rRNA gene intergenic spacer region of Clostridium difficile and construction of a library consisting of 116 different PCR ribotypes. J Clin Microbiol1999;37:4613DOIExternal LinkPubMedExternal Link
  16. Crobach  MJPlanche  TEckert  CBarbut  FTerveer  EMDekkers  OMet al. European Society of Clinical Microbiology and Infectious Diseases: update of the diagnostic guidance document for Clostridium difficile infection. Clin Microbiol Infect2016;22(Suppl 4):S6381DOIExternal LinkPubMedExternal Link
  17. McDonald  LCGerding  DNJohnson  SBakken  JSCarroll  KCCoffin  SEet al. Clinical practice guidelines for Clostridium difficile infection in adults and children: 2017 update by the Infectious Diseases Society of America (IDSA) and Society for Healthcare Epidemiology of America (SHEA). Clin Infect Dis2018;66:98794DOIExternal LinkPubMedExternal Link
  18. Kraft  CSParrott  JSCornish  NERubinstein  MLWeissfeld  ASMcNult  Pet al. A laboratory medicine best practices systematic review and meta-analysis of nucleic acid amplification tests (NAATs) and algorithms including NAATs for the diagnosis of Clostridioides (Clostridiumdifficile in adults. Clin Microbiol Rev2019;32:e0003218.PubMedExternal Link
  19. Fenner  LFrei  RGregory  MDangel  MStranden  AWidmer  AFEpidemiology of Clostridium difficile-associated disease at University Hospital Basel including molecular characterisation of the isolates 2006-2007. Eur J Clin Microbiol Infect Dis2008;27:12017DOIExternal LinkPubMedExternal Link
  20. Vernaz  NSax  HPittet  DBonnabry  PSchrenzel  JHarbarth  STemporal effects of antibiotic use and hand rub consumption on the incidence of MRSA and Clostridium difficile. J Antimicrob Chemother2008;62:6017DOIExternal LinkPubMedExternal Link
  21. Kohler  PBregenzer-Witteck  ARafeiner  PSchlegel  MPresumably hospital-transmitted Clostridium difficile infections based on epidemiological linkage. Swiss Med Wkly2013;143:w13824. DOIExternal LinkPubMedExternal Link
  22. Goorhuis  ABakker  DCorver  JDebast  SBHarmanus  CNotermans  DWet al. Emergence of Clostridium difficile infection due to a new hypervirulent strain, polymerase chain reaction ribotype 078. Clin Infect Dis2008;47:116270DOIExternal LinkPubMedExternal Link
  23. Patterson  LWilcox  MHFawley  WNVerlander  NQGeoghegan  LPatel  BCet al. Morbidity and mortality associated with Clostridium difficile ribotype 078: a case–case study. J Hosp Infect2012;82:1258DOIExternal LinkPubMedExternal Link
  24. Romano  VPasquale  VKrovacek  KMauri  FDemarta  ADumontet  SToxigenic Clostridium difficile PCR ribotypes from wastewater treatment plants in southern Switzerland. Appl Environ Microbiol2012;78:66436DOIExternal LinkPubMedExternal Link

Cite This Article

DOI: 10.3201/eid2610.190804

Original Publication Date: September 09, 2020


Resource:  https://wwwnc.cdc.gov/eid/article/26/10/19-0804_article

Researchers Evaluate Isolation, Safety, Probiotic Property of a Novel E. thailandicus Strain, d5b With Effective Antimicrobial Activity Against C. difficile


Authors: Tinghua Li 1Lin Lyu 1Yan Zhang 1Ke Dong 1Qingtian Li 2Xiaokui Guo 3Yongzhang Zhu 4

Colitis induced by C. difficile is one of the most common and costly healthcare-related infections for humans. Probiotics are one of the most promising approaches for controlling CDI. Here, we presented the isolation, safety, and probiotic property evaluation of a novel E. thailandicus strain, d5B, with effective antimicrobial activity against C. difficile.

Strain d5B showed strong bactericidal effects on at least 54C. difficile strains. Safety tests showed that strain d5B was sensitive to clinically important antibiotics, and had no haemolytic and cytotoxic activities. Whole genomic analysis showed strain d5B only contained one aminoglycoside resistance gene located in the chromosome. Moreover, d5B was devoid of functional virulence genes. Finally, strain d5B exhibited probiotic properties, such as tolerance to the gastrointestinal tract, and adhered well to HT-29 cells. In conclusion, the E. thailandicus strain d5B should be investigated further for useful properties as a novel candidate probiotic for controlling CDI.


To review the publication in its entirety please click on the link below to be redirected:


Accessible Database for Clostridioides difficile Genome Sequences Supporting Transmission and Epidemics

  • Martinique Frentrup1​, Zhemin Zhou2​, Matthias Steglich1​,3​, Jan P. Meier-Kolthoff1​, Markus Göker1​, Thomas Riedel1​,3​, Boyke Bunk1​, Cathrin Spröer1​, Jörg Overmann1​,3​,4​, Marion Blaschitz5​, Alexander Indra5​, Lutz von Müller6​, Thomas A. Kohl7​,8​, Stefan Niemann7​,8​, Christian Seyboldt9​, Frank Klawonn10​,11​, Nitin Kumar12​, Trevor D. Lawley12​, Sergio García-Fernández13​,14​, Rafael Cantón13​,14​, Rosa del Campo13​,14​, Ortrud Zimmermann15​, Uwe Groß15​, Mark Achtman2​, Ulrich Nübel1​,3​,4​


Clostridioides difficile is the primary infectious cause of antibiotic-associated diarrhea. Local transmissions and international outbreaks of this pathogen have been previously elucidated by bacterial whole-genome sequencing, but comparative genomic analyses at the global scale were hampered by the lack of specific bioinformatic tools. Here we introduce a publicly accessible database within EnteroBase (http://enterobase.warwick.ac.uk) that automatically retrieves and assembles 

C. difficile short-reads from the public domain, and calls alleles for core-genome multilocus sequence typing (cgMLST). We demonstrate that comparable levels of resolution and precision are attained by EnteroBase cgMLST and single-nucleotide polymorphism analysis. EnteroBase currently contains 18 254 quality-controlled 

C. difficile genomes, which have been assigned to hierarchical sets of single-linkage clusters by cgMLST distances. This hierarchical clustering is used to identify and name populations of C. difficile at all epidemiological levels, from recent transmission chains through to epidemic and endemic strains. Moreover, it puts newly collected isolates into phylogenetic and epidemiological context by identifying related strains among all previously published genome data. For example, HC2 clusters (i.e. chains of genomes with pairwise distances of up to two cgMLST alleles) were statistically associated with specific hospitals (P<10−4) or single wards (P=0.01) within hospitals, indicating they represented local transmission clusters. We also detected several HC2 clusters spanning more than one hospital that by retrospective epidemiological analysis were confirmed to be associated with inter-hospital patient transfers. In contrast, clustering at level HC150 correlated with k-mer-based classification and was largely compatible with PCR ribotyping, thus enabling comparisons to earlier surveillance data. EnteroBase enables contextual interpretation of a growing collection of assembled, quality-controlled C. difficile genome sequences and their associated metadata. Hierarchical clustering rapidly identifies database entries that are related at multiple levels of genetic distance, facilitating communication among researchers, clinicians, and public-health officials who are combatting disease caused by C. difficile.


  • Clostridioides (Clostridium) difficile
  • nosocomial infection, 
  • genomic population structure, 
  • outbreak, 
  • cgMLST, 
  • hierarchical clustering

Author NotesAll supporting data, code and protocols have been provided within the article or through supplementary data files. Three supplementary tables and eight supplementary figures are available with the online version of this article.Sequence accession numbers: PRJEB33768, PRJEB33779, PRJEB33780.†These authors contributed equally to this work‡These authors also contributed equally to this work


CC, cgST complex; CDI, Clostridioides difficile infection; cgMLST, core-genome MLST; cgST, core-genome sequence type; HC150, hierarchical cluster with pairwise distances of up to 150 cgMLST alleles; HC, hierarchial cluster; HierCC, hierarchical clustering; MLST, multilocus sequence typing; ORF, open reading frame; rMLST, ribosomal MLST; rST, ribosomal sequence type; RT, PCR ribotype; SNP, single-nucleotide polymorphism; wgMLST, whole-genome MLST.Impact Statement

Clostridioides difficile is a major cause of healthcare-associated diarrhea and causes large infection outbreaks. Whole-genome sequencing is increasingly applied for genotyping C. difficile, with the objectives to monitor and curb the pathogen’s spread. We present a publicly accessible database for quality-controlled genome sequences from C. difficile that enables contextual interpretation of newly collected isolates by identifying related strains among published data. It also provides a nomenclature for genomic types to facilitate communication about transmission chains, epidemics, and phylogenetic lineages. Finally, we demonstrate that genome-based hierarchical clustering is largely compatible with previously used molecular typing techniques, thus enabling comparisons to earlier surveillance data.

Data Summary

All genome sequencing data were submitted to the European Nucleotide Archive (www.ebi.ac.uk/ena) under study numbers PRJEB33768, PRJEB33779 and PRJEB33780. The Clostridioides database within EnteroBase is publicly accessible at http://enterobase.warwick.ac.uk. In addition, stand-alone versions of all EnteroBase tools are available at https://github.com/zheminzhou/EToKi.


The anaerobic gut bacterium Clostridioides difficile (formerly Clostridium difficile) [1] is the primary cause of antibiotic-associated diarrhea in Europe and North America [2]. Molecular genotyping of C. difficile isolates has demonstrated international dissemination of diverse strains through healthcare systems [3–5], the community [6] and livestock production facilities [7, 8]. Previously, genotyping was commonly performed by PCR ribotyping or DNA macrorestriction. More recent publications have documented that genome-wide single-nucleotide polymorphisms (SNPs) from whole-genome sequences provide improved discrimination, and such analyses have enabled dramatic progress in our understanding of the emergence and spread of epidemic strains [9–12] and the epidemiology of local transmission [13, 14]. Eyre and colleagues have argued that transmission of C. difficile isolates within a hospital environment can be recognized with high probability as chains of genomes, which differ by up to two SNPs whereas genomes, which differ by at least ten genomic SNPs represent unrelated bacteria [13, 15]. However, SNP analyses require sophisticated bioinformatic tools and are difficult to standardize [16, 17]. A convenient alternative to SNP-based genotyping is offered by the commercial software SeqSphere, which implements a core-genome multilocus sequence typing scheme (cgMLST) for the analysis of genomic diversity in 

C. difficile [18] and other organisms. Indeed, cgMLST [18] confirmed the prior conclusion from genomic SNP analyses [19] that a common clone of C. difficile had been isolated over two successive years at a hospital in China [18]. However, a recent quantitative comparison of the two methods showed that SeqSphere’s cgMLST achieved a low predictive value (41 %) for identifying isolate pairs that were closely related by the ≤2 SNPs’ criterion [20]. cgMLST of genomic sequences of a variety of bacterial pathogens can also be performed with EnteroBase (http://enterobase.warwick.ac.uk/), which has been developed over the last few years with the goal of facilitating genomic analyses by microbiologists [21]. EnteroBase automatically retrieves Illumina short-read sequences from public short-read archives. It uses a consistent assembly pipeline to automatically assemble these short-reads into draft genomes consisting of multiple contigs, and presents the assembled genomes together with their metadata for public access [22]. It also performs the same procedures on sequencing data uploaded by its registered users. Assembled genomes that pass quality control are genotyped by MLST at the levels of seven-gene MLST, ribosomal MLST (rMLST), cgMLST and whole-genome MLST (wgMLST) [21, 22]. EnteroBase supports subsequent analyses based on either SNPs or cgMLST alleles using the GrapeTree or Dendrogram visualization tools [23]. EnteroBase also assigns these genotypes to populations by hierarchical clustering (HierCC), which supports the identification of close relatives at the global level [22]. Originally, EnteroBase was restricted to the bacterial genera Salmonella, Escherichia, Yersinia and Moraxella but since January 2018, EnteroBase has included a database for genomes and their metadata for the genus Clostridioides. In June 2020, EnteroBase contained 18 254 draft genomes of C. difficile plus one genome of C. mangenotii. These included over 900 unpublished draft genomes that were sequenced at the Leibniz Institute DSMZ, as well as 80 complete genome sequences based on Pacific Biosciences plus Illumina sequencing technologies. It also included 862 unpublished draft genomes that were sequenced at the Wellcome Sanger Institute.

Here we show that comparable levels of resolution and precision are attained by EnteroBase cgMLST as by SNP analyses. We also summarize the genomic diversity that accumulated during recurring infections within single patients as well as transmission chains within individual hospitals and between neighbouring hospitals in Germany, and show that it can be detected by HierCC. We also demonstrate that HierCC can be used to identify bacterial populations at various epidemiological levels ranging from recent transmission chains through to epidemic and endemic spread, and relate these HierCC clusters to genotypes that were identified by PCR ribotyping and k-mer-based diversity analysis. These observations indicate that cgMLST and HierCC within EnteroBase can provide a common language for communications and interactions by the global community who is combatting disease caused by C. difficile.Results

Implementation of MLST schemes in EnteroBase

cgMLST in EnteroBase consists of a defined subset of genes within a whole-genome MLST scheme that represents all single-copy orthologues within the pan-genome of a representative set of bacterial isolates. To this end, we assembled the draft genomes of 5232 isolates of C. difficile from public short-read archives, and assigned them to ribosomal sequence types (rSTs) according to rMLST, which indexes diversity at 53 loci encoding ribosomal protein subunits on the basis of existing exemplar alleles at PubMLST [24]. We then created a reference set of 442 genomes consisting of one genome of C. mangenotii [1], 18 complete genomes from GenBank, 81 closed genomes from our work and the draft genome with the smallest number of contigs from each of the 343 rSTs (https://tinyurl.com/Cdiff-ref). The Clostridioides pan-genome was calculated with PEPPA [25] and used to define a wgMLST scheme consisting of 13 763 genetic loci (http://enterobase.warwick.ac.uk/species/clostridium/download_data). EnteroBase uses the wgMLST scheme to call loci and alleles from each assembly and extracts the allelic assignments for the subsets corresponding to cgMLST, rMLST and seven-gene MLST from those allelic calls [22]. The cgMLST subset consists of 2556 core genes, which were present in ≥98% of the reference set, intact in ≥94%, and were not excessively divergent (Fig. 1).Fig. 1.Criteria for inclusion in a cgMLST scheme of a subset of wgMLST genes based on their properties in a reference set of 442 genomes (https://tinyurl.com/Cdiff-ref). (a) Numbers of genes versus frequency (% presence) within the reference set. In total, 2634 genes satisfied the cut-off criterion of ≥98 % presence (dashed line). (b) Numbers of genes versus intact ORF (% intact ORF) within the 2634 genes from (a). Overall, 2560 genes satisfied the cut-off criterion of ≥94 % intact ORF (dashed line). (c) Frequency of allelic variants versus gene size among the 2560 genes from (b). The genetic diversity was calculated using the GaussianProcessRegressor function in the sklearn module in Python. This function calculates the Gaussian process regression of the frequency of genetic variants on gene sizes, using a linear combination of a radial basis function kernel (RBF) and a white kernel [57]. The shadowed region shows a single-tailed 99.9% confidence interval (≤3 sigma) of the prediction. Altogether, 2556 loci fell within this area and were retained for the cgMLST scheme, while four were excluded due to excessive numbers of alleles.Click to view

Comparison of cgMLST and SNPs for analyses of transmission chains

We compared the numbers of cgMLST allelic differences and the numbers of non-recombinant SNPs in isolates from multiple epidemiological chains. These included 176 isolates from four patients with recurring CDI (C. difficile infection), 63 isolates from four transmission chains in multiple hospitals [14, 19, 26], and a comprehensive sample of 1158 isolates collected over several years in four hospitals in Oxfordshire, UK [13]. A strong linear relationship (R2, 0.71–0.93) was found in all three analyses between the pairwise differences in cgMLST alleles and non-recombinant SNPs (Fig. S1, available in the online version of this article). The slope of the regression lines was close to 1.0, indicating a 1 : 1 increase in cgMLST allelic differences with numbers of SNPs. The same data were also investigated with cgMLST calculated with the commercial program SeqSphere [18], with similar correlation coefficients but a lower slope due to lesser discriminatory power of the SeqSphere cgMLST scheme (lower panels in Fig. S1).

Eyre et al. [13] concluded that direct transmission between two hospital patients can be detected because their bacterial genomes differ by two SNPs or less. Our analysis indicated that these transmission chains in the Oxfordshire dataset would also have been recognized by cgMLST in EnteroBase. Genomes that differed by two cgMLST alleles usually also differed by ≤2 SNPs according to a binary logistic regression model (probability=89%; 95% confidence interval, 88–89%) (Fig. 2). Of 3807 pairs of genomes with ≤2 allelic differences, 3474 also differed by ≤2 SNPs, yielding a positive predictive value of 91 % for identifying isolate pairs with ≤2 SNPs by EnteroBase cgMLST and a sensitivity of 62 % (≤2 cgMLST allelic differences were found in 3474 of 5707 pairs with ≤2 SNPs). The comparable values for SeqSphere were 78 % positive predictive value and 99% sensitivity.Fig. 2.Binary logistic regression model to determine the probability that two genomes are related at ≤2 SNPs, given a certain difference in their cgMLST allelic profiles, based on the Oxfordshire dataset [13]. The number of SNPs was encoded as a binary dependent variable (1 if ≤2 SNPs, 0 if otherwise) and the number of allelic differences was used as a predictor variable.Click to view

We also compared the genetic distances between 242 genomes from Oxfordshire, which had been isolated during the initial 6 months and 916 genomes from the actual testing period (April 2008 to March 2011) [13]. Overall, 35% (318/916) of the latter genomes matched at least one genome collected earlier by two or less EnteroBase cgMLST alleles and 34% (316/916) matched an earlier genome by ≤2 SNPs. The two sets of genomes were 89% concordant. Thus, cgMLST is equivalent to SNP analysis for detecting inter-patient transmission chains.

Hierarchical clustering for tracing local and regional spread

SNP analyses are computer intensive, and are only feasible with limited numbers of genomes [27]. cgMLST-based relationships can be analysed for up to 100 000 genomes with GrapeTree, but analyses involving more than 10 000 genomes remain computer intensive [23]. EnteroBase implements single-linkage hierarchical clustering (HierCC V1) of cgMLST data in pairwise comparisons at multiple levels of relationship after excluding missing data [22]. These are designated as HC0 for hierarchical clusters of indistinguishable core-genome sequence types (cgSTs), HC2 for clusters with pairwise distances of up to two cgMLST alleles, etc. EnteroBase presents cluster assignments for C. difficile at the levels of HC0, HC2, HC5, HC10, HC20, HC50, HC100, HC150, HC200, HC500, HC950, HC200 and HC2500. Here we address the nature of the genetic relationships that are associated with these multiple levels of HierCC among 13 515 publicly available C. difficile genomes, and examine which levels of pairwise allelic distances correspond to epidemic outbreaks and to endemic populations.

In our analyses of 176 C. difficile isolates from four patients with two recurrent episodes of CDI, multiple genomes were assigned to patient-specific HC2 clusters, some of which were isolated from the initial episode as well as the recurrence 80–153 days later (Fig. 3, patients D, F and G; 4 to 36 isolates had been collected per episode; Table S1). For these patients, relapsing disease likely reflected continued colonization after initially successful therapy. However, some isolates from patient F differed by 12–21 cgMLST alleles from the bulk population (Fig. 3), which indicates that the patient was co-infected simultaneously with multiple related strains. In patient E, the two genomes from the two CDI episodes differed by >2000 allelic differences (Fig. 3), which indicates that the second incident of CDI represented an independent infection with an unrelated strain. Hence, discrimination between relapse and reinfection based on cgMLST appears to be straightforward except that two episodes of CDI might arise by reinfection with identical strains from an environment that is heavily contaminated with C. difficile spores [28]. We note that the time intervals (16–22 weeks) investigated here exceeded the currently recommended threshold of 8 weeks for surveillance-based detection of CDI relapses [29, 30] but still yielded almost identical strains in three of four patients.Fig. 3.Minimum-spanning trees indicating the population structure of C. difficile in four patients with recurrent CDI episodes. Red, first episode; blue, second episode.Click to view

Our examinations of multiple local outbreaks have revealed individual, outbreak-specific HC2 clusters. However, it is also conceivable that multiple HC2 clusters might be isolated from a single epidemiological outbreak due to the accumulation of genetic diversity over time. Alternatively, multiple HC2 clusters within a single outbreak may represent the absence of crucial links due to incomplete sampling. Incomplete sampling of outbreaks is not unlikely because asymptomatic patients are only rarely examined for colonization with C. difficile [31–33] even though they may constitute an important reservoir for transmission. Indeed, some of the outbreaks investigated here did consist of more than one HC2 cluster (Fig. 4). For example, nine isolates from a recently reported ribotype 018 (RT018) outbreak in Germany [26] encompassed four related HC2 clusters, and outbreaks with RT027 and RT106 in a hospital in Spain [14] were each affiliated with two or three HC2 clusters (Fig. 4).Fig. 4.Neighbour-joining trees based on cgMLST showing the phylogenetic relationships among C. difficile isolates from previously published CDI outbreaks as indicated [14, 19, 26]. Nodes are coloured by HC2. CC, cgST complex, i.e. related at level HC150; RT, PCR ribotype. The scale, indicating one allelic difference, applies to all trees.Click to view

We identified 23 HC2 clusters encompassing 133 genome sequences in a dataset of 309 C. difficile genome sequences collected from CDI patients in six neighbouring hospitals in Germany. These HC2 clusters were associated with individual hospitals (Χ2P=8.6×10−5; Shannon entropy, P=4.2×10−5) and even with single wards in these hospitals (Χ2P=0.01; Shannon entropy, P=6.2×10−3). We investigated whether these HC2 clusters reflected the local spread of C. difficile within institutions by retrospective analyses of patient location data. Sixty-six patients (50 %) were found to have had ward contacts with another patient with the same HC2 cluster (median time interval between ward occupancy: 63 days; range, 0 to 521). These results are consistent with the direct transmission on wards of C. difficile isolates of the same HC2 cluster (Fig. 5). For patients such as P1 and P2 where the shared ward contacts were separated in time (Fig. 5), transmission may have occurred indirectly through asymptomatically colonized patients or from a common reservoir, such as environmental spore contamination [14, 31, 32]. We also detected 15 HC2 clusters that included isolates from two or more hospitals in the region. Subsequent analyses of patient location data confirmed that some of these HC2 clusters were associated with patient transferrals between the hospitals (Fig. 5). Hence, hierarchical clustering of C. difficile genome sequences in conjunction with retrospective analysis of patient movements revealed multiple likely nosocomial transmission events, none of which had been detected previously by routine surveillance.Fig. 5. Timelines of two transmission chains discovered retrospectively through inspection of files from CDI patients with closely related C. difficile isolates (HC2). Colours indicate hospital wards, ‘X’ indicates a diagnosis of CDI, and arrows indicate presumed transmission pathways. Minimum-spanning trees indicating genomic distances among C. difficile isolates are shown on the right. Upper panel: patient P1 was diagnosed with CDI in hospital H2 and transferred to hospital H3 15 days later. Another five and 6 days later, respectively, patients P2 in hospital H2 and P3 in hospital H3 got diagnosed with CDI with closely related strains. Both these patients were on the same wards as the initial patient, who probably had been the source for the pathogen. Since there was no temporal overlap between patient P2 and the other patients in hospital H2, the transmission may have occurred indirectly, possibly through environmental contamination. Lower panel: another putative transmission chain involved three patients that had shared time in hospital H2. Patients P4 and P5 got diagnosed with CDI on the same day after they had shared 7 days in this hospital, albeit on different medical wards. The third patient developed CDI with the same C. difficile cgST 4 days after being transferred to another hospital (h5), but had previously stayed at hospital H2 during the time when CDI got diagnosed in the first two patients. Since the three patients stayed on different wards in hospital H2, transmission presumably occurred indirectly.Click to view

Hierarchical clustering for identification of epidemic strains and endemic populations

The international epidemic spread of C. difficile over up to 25 years has been inferred previously on the basis of molecular epidemiology with lower resolution techniques [34]. For multiple representatives of those epidemic strains in EnteroBase, the majority of these epidemic groups corresponded to HC10 clusters, including epidemic RT017 [11] (HC10_17), the two fluoroquinolone-resistant lineages of RT027 [9] (HC10_4, HC10_9), or livestock-associated RT078/126 [35] (HC10_1) (Fig. 6).Fig. 6.Phylogenetic structure of three international C. difficile epidemics, each of which has spread for about 25 years [9, 11]. Within each epidemic, the majority of isolates are related at level HC10, as indicated by the colours. CC, cgST complex, i.e. related at level HC150; RT, PCR ribotype.Click to view

Endemic populations have also been described by ribotyping and phylogenetic analyses, some of which have acted as sources for the emergence of epidemic strains [2, 9]. Many endemic populations seem to be represented by HC150 clusters. Clustering at HC150 was well supported statistically (Fig. S2), and the frequency distribution of pairwise genomic distances indicated that multiple database entries clustered at <150 cgMLST allelic differences (Fig. S3). HC150 clusters also correlated well with k-mer-based classification [36]. When applied to the dataset of 309 C. difficile genomes from six hospitals in Germany, the two methods implemented in EnteroBase and PopPUNK found 51 and 48 clusters, respectively, the majority of which coincided (adjusted Rand coefficient, 0.97).

A cgMLST-based phylogenetic tree of 13515 C. difficile genomes showed 201 well-separated HC150 clusters, each encompassing a set of related isolates, plus 209 singletons (Fig. 7). Because these HC150 clusters are based on cgMLST genetic distances, we refer to them as ‘cgST complexes’, abbreviated as CCs. Genomes from each of the major CCs have been collected over many years in multiple countries, indicating their long-term persistence over wide geographic ranges (Table 1).Fig. 7.Rapid-neighbour-joining phylogenetic tree based on cgMLST variation from 13 515 C. difficile genomes. Colours and numerals indicate CCs (HC150 clusters) with ≥10 entries, and information on predominant PCR ribotypes is provided in brackets.Click to view

Table 1.

Characteristics of cgST complexes (CC) with ≥100 entriesToggle display:Table 1.  Open Table 1. fullscreen 

CC (HC150)PCR RibotypeNumber of entriesSampling yearsNumber of countries% isolates in HC2>21% isolates from animal hosts
1078, 126, 06612221994–2018266117
6020, 4047681995–201714431
22106, 5005311997–20177593
71014, 0203152004–201716401
178018, 3562432006–20177520
141001, 0261072007–2016270
49011, 056, 4461032001–20175350
1isolates in HC2 clusters with >2 entries.

We compared HC150 clustering with PCR ribotyping for 2263 genomes spanning 84 PCR ribotypes for which PCR ribotyping data were available in EnteroBase. These included 905 genomes, which we ribotyped (Table S2), as well as several hundred other genomes for which ribotype information was manually retrieved from published data. The correlation between HC150 clustering and ribotyping was high (adjusted Rand coefficient, 0.92; 95% confidence interval, 0.90–0.93). However, our analysis also revealed that PCR ribotypes did not always correspond to phylogenetically coherent groupings. PCR ribotypes 002, 015 and 018 were each distributed across multiple phylogenetic branches (Fig. 8). Furthermore, some genomes with indistinguishable cgMLST alleles were assigned to multiple ribotypes, including RT001/RT241, RT106/RT500 and RT126/RT078 (Fig. 8Table 1). In these cases, both ribotypes occurred in several, closely related clades (Fig. 8), indicating that similar ribotype banding patterns had evolved multiple times. In contrast, HC150 clusters corresponded to clear-cut phylogenetic groupings within a phylogenetic tree of core genes (Fig. 8b).Fig. 8.Rapid-neighbour-joining phylogenetic tree based on cgMLST variation from 2263 C. difficile genomes, for which PCR ribotyping information is available. Upper panel: nodes are coloured by PCR ribotype as indicated. Lower panel: nodes are coloured by CC (HC150 clusters).Click to view

Higher population levels

HierCC can also identify clusters at still higher taxonomic levels, up to the levels of species and sub-species [22]. In C. difficile, HC950 clusters seem to correspond to deep evolutionary branches (Fig. S4) and HC2000 clusters were congruent with the major clades reported previously [37], except that cluster HC2000_2 encompassed clade 1 plus clade 2 (Fig. S5). Finally, HC2500 may correspond to the subspecies level, because it distinguished between C. difficile and distantly related ‘cryptic clades’ (Fig. S6).Discussion

Infectious disease epidemiologists frequently seek to know if new isolates of bacterial pathogens are closely related to others from the different geographical origin, i.e. if they are part of a widespread outbreak. Unlike a previous cgMLST implementation [18], EnteroBase supports this goal by taking full advantage of rapidly growing, public repositories of short-read genome sequences [22]. In contrast to short-read archives, however, where stored sequence data are not readily interpretable without specialized bioinformatic tools [38], EnteroBase enables contextual interpretation of a growing collection (18 254 entries as of June 2020) of assembled, quality-controlled C. difficile genome sequences and their associated metadata. At least the collection date (year), the geographic origin (country), and the source (host species) are available for the majority of database entries. Importantly, phylogenetic trees based on cgMLST allelic profiles from many thousand bacterial genomes can be reconstructed within a few minutes, whereas such calculations are currently prohibitively slow based on SNP alignments [22]. Genome-sequencing reads from newly sampled C. difficile isolates can be uploaded to EnteroBase and compared to all publicly available genome data within hours, without requiring any command-line skills.

We demonstrate that the application of cgMLST to investigations of local C. difficile epidemiology yields results that are quantitatively equivalent to those from SNP analyses. This is a major advance because SNP analyses require specific bioinformatic skills and infrastructure, are time-consuming and not easily standardized [16]. A web platform for centralized, automated SNP analyses on bacterial genomes is limited to food pathogens currently and does not offer any analyses on C. difficile genomes [39]. Even though a cgMLST scheme for C. difficile had been published recently [18], its ability to identify closely related isolates and the inferred genomic distances was shown to be inferior to SNP analyses due to an excess of errors introduced by the de novo assembly of sequencing reads and a lack of per-base quality control [20]. In EnteroBase, cgMLST is also based on de novo assembly, but EnteroBase uses Pilon [40] to polish the assembled scaffolds and evaluate the reliability of consensus bases of the scaffolds, thereby achieving comparable accuracy to mapping-based SNP analyses. When applied to a large dataset of C. difficile genomes from hospital patients in the Oxfordshire region (UK), cgMLST and SNP analysis were largely consistent (89% match) at discriminating between isolates that were sufficiently closely related to have arisen during transmissions chains from others that were epidemiologically unrelated.

After assembly, draft genomes contain missing data and many cgSTs have unique cgST numbers but are identical to other cgSTs, except for missing data. Hence, individual cgST numbers are only rarely informative. However, indistinguishable cgSTs are clustered in common HierCC HC0 clusters, which ignore missing data. In June 2020, the Clostridioides database contained >12 000 HC0 clusters, indicating that the majority of genomes was unique. Similarly, EnteroBase provides cluster designations at multiple levels of HierCC, enabling rapid identification of all cgSTs that are related at multiple levels of genetic distance. The data presented here shows that HierCC designations can facilitate communications between researchers, clinicians and public-health officials about transmission chains, epidemic outbreaks, endemic populations and higher phylogenetic lineages up to the level of subspecies.

EnteroBase cgMLST identified numerous HC2 clusters of strains in C. difficile isolates that seem to have arisen during transmission chains in six neighboring hospitals in Germany. These assignments were in part consistent with retrospective investigation of patient location data, although none of the nosocomial outbreaks (defined by German law as two or more infections with likely epidemiological connections [http://www.gesetze-im-internet.de/ifsg/]) had been detected previously by standard epidemiological surveillance by skilled clinical microbiologists. Recent publications propose that prospective genome sequencing of nosocomial pathogens should be applied routinely at the hospital level to guide epidemiological surveillance [41]. Our data indicates that the combination of genome sequencing with cgMLST and HierCC may identify nosocomial transmission routes of C. difficile more effectively than presently common practice, and hence could help to reduce pathogen spread and the burden of disease. Reliable identification of transmission chains requires interpretation of pathogen genome sequence data in its epidemiological context, however [42].

HierCC will also enable comparisons to previously published data because we have provided a correspondence table between HC150 clusters and PCR ribotypes (Table 1). Rarefaction analysis indicated that the currently available genome sequences represent about two-thirds of extant HC150 (CC) diversity, which extrapolated to about 600 CCs (Fig. S7). At least some of this enormous diversity may be due to the occupation of multiple, distinct ecological niches, as exemplified by differential propensities for colonizing non-human host species (Table 1) [43, 44]. Individual CCs may also differ in their aptitudes for epidemic spread, as indicated by drastically different proportions of genomes assigned to HC2 chains: only 7% of CC141 were assigned to HC2 clusters versus 35% of CC34 and 77% of CC4 (Table 1). A full understanding of the population structure of C. difficile and its relationship to epidemiological patterns will require additional study because many of the clusters described here have not yet been studied or described. However, this task can be addressed by the global community due to the free public access to such an unprecedented amount of genomic data from this important pathogen.Methods


In total, 309 C. difficile isolates were collected at a diagnostic laboratory providing clinical microbiology services to several hospitals in central Germany. To assemble a representative sample, we included the first 20 isolates from each of six hospitals from each of three consecutive calendar years (Table S2). For the investigation of recurrent CDI, a set of 176 C. difficile isolates were collected in a diagnostic laboratory in Saarland, Germany. Here, primary stool culture agar plates were stored at 4 °C for 5 months to eventually enable the analysis of multiple plates representing episodes of recurrent C. difficile infection from individual patients, who had developed the recurrent disease by then and could be chosen with hindsight. It was attempted to pick and cultivate as many bacterial colonies from each selected plate as possible, resulting in 6 to 36 isolates per CDI episode (Table S1). In addition, we sequenced the genomes from 383 isolates that had been characterized by PCR ribotyping previously, including 184 isolates sampled from piglets [8], 71 isolates from various hospitals in Germany [3], and 108 isolates from stool samples collected from nursing home residents (unpublished; Table S2).

PCR ribotyping

PCR ribotyping was performed as described previously [45], applying an ABI Prism 3100 apparatus for capillary electrophoresis and comparing banding patterns to the Webribo database (https://webribo.ages.at/).Whole-genome sequencing

For Illumina sequencing, genomic DNA was extracted from bacterial isolates by using the DNeasy Blood and Tissue kit (Qiagen), and libraries were prepared as described previously [46] and sequenced on an Illumina NextSeq 500 machine using a Mid-Output kit (Illumina) with 300 cycles. For generating complete genome sequences, we applied SMRT long-read sequencing on an RSII instrument (Pacific Biosciences) in combination with Illumina sequencing as reported previously [46]. All genome sequencing data were submitted to the European Nucleotide Archive (www.ebi.ac.uk/ena) under study numbers PRJEB33768, PRJEB33779 and PRJEB33780

.SNP detection and phylogenetic analysis

Sequencing reads were mapped to the reference genome sequence from C. difficile strain R20291 (sequence accession number FN545816) by using BWA-MEM and sequence variation was detected by applying VarScan2 as reported previously [46]. Sequence variation likely generated by recombination was detected through analysis with ClonalFrameML [47] and removed prior to determination of pairwise sequence distances [15] and to construction of maximum-likelihood phylogenetic trees with RAxML (version 8.2.9) [48].Genome assembly, quality control and wgMLST allele calling

Genomic data was processed by automated pipelines within EnteroBase, which were described in detail previously [22]. Briefly, Illumina sequencing reads were assembled by using Spades v3.10 [49] and assemblies were improved by applying Pilon [40]. To pass quality control, assemblies were required to comply with the following thresholds: total length, 3.6 to 4.8 Mbp; N50, ≥20 000; the number of contigs, ≤600; number of unresolved nucleotides, ≤3%; the proportion of Clostridioides sequences, >65 % (as determined by Kraken with MiniKraken database [50]). Assemblies were aligned to exemplar alleles by using blastn [51] and the usearch module UblastP [52], and allele numbers, STs, and HC numbers assigned by using the EnteroBase module MLSType [22]. All EnteroBase tools are available at https://github.com/zheminzhou/EToKi.

To determine the probability that two genomes are related at ≤2 SNPs, given a certain difference in their cgMLST allelic profiles, we inferred a logistic regression model using R ([53], pp. 593–609). Genomic relatedness was encoded as a binary response variable (1 if ≤2 SNPs, 0 if otherwise) and the number of core-genome allelic differences was used as a predictor variable. We applied this model to a dataset of 1158 genome sequences from a previous study, representing almost all symptomatic CDI patients in Oxfordshire, UK, from 2007 through 2011 [13]. While that original study had encompassed a slightly larger number of sequences, we restricted our analysis to the data (95 %) that had passed quality control as implemented in EnteroBase [21]. We used the SNP data from Eyre’s report [13].

The hierarchical single-linkage clustering of cgMLST sequence types was carried out as described [22] for all levels of allelic distances between 0 and 2556. We searched for stable levels of differentiation by HierCC according to the Silhouette index [54], a measure of uniformity of the divergence within clusters. The Silhouette index was calculated based on d^’, a normalized genetic distance between pairs of STs, which was calculated from their allelic distance d as follows: d^’=1-(1-d)^(1/l), where l is the average length (937 bp) of the genes in the cgMLST scheme.

We further evaluated the ‘stability’ of hierarchical clustering using two other criteria. The Shannon index is a measure of diversity in a given population. The Shannon index drops from nearly 1 in HC0, because most cgSTs are assigned to a unique HC0 cluster, to 0 in HC2500, which assigns all sequence types to one cluster. The gradient of the Shannon index between the two extremes reflects the frequencies of coalescence of multiple clusters at a lower HC level. Thus, the plateaus in the curve correspond to stable hierarchical levels, where the Shannon index does not change dramatically with HC level. We also evaluated the stability of hierarchical clustering by pairwise comparison of the results from different levels based on the normalized mutual information score [55] (Fig. S3).

For clustering C. difficile diversity with PopPUNK [36], we used a sketch size of 105 and a K value (maximum number of mixture components) of 15. Of note, the resulting number of clusters for the tested dataset was identical for all K between 15 and 30.

To estimate concordance between cgMLST-based hierarchical clustering and PCR ribotyping or PopPUNK clustering, respectively, we calculated the adjusted Rand coefficient [56] by using the online tool available at http://www.comparingpartitions.info/. To test statistical associations of HC2 clusters with specific hospitals and hospital wards, respectively, we compared Χ2 values and normalized Shannon entropy values (R package ‘entropy’ v.1.2.1) from contingency tables containing real isolate distributions (Table S3) and randomly permuted distributions (n=1000), by using the non-parametric, two-sided Mann–Whitney U test (R package ‘stats’ v.3.5.0).

This work was partially funded by the German Center for Infection Research (DZIF), by the Federal State of Lower Saxony (Niedersächsisches Vorab VWZN2889/3215/3266), by the EU Horizon 2020 programme (grant agreement number 643476), the Wellcome Trust (098051), and the UK Medical Research Council (PF451). EnteroBase development was funded by the BBSRC (BB/L020319/1) and the Wellcome Trust (202792/Z/16/Z), and the salary of Z.Z. was also provided by The Wellcome Trust. The funders had no role in the study design, preparation of the article, or decision to publish.


We thank Vera Junker, Simone Severitt, Nicole Heyer and Carola Berg for excellent technical assistance, Johannes Sikorski for help with R, and David Eyre for supplying SNP data from his 2013 paper in tabular format.Author contributions

M.A. and U.N. designed the study. M.F., Z.Z., M.S., J.P.M., M.G., T.R., B.B., C.S., J.O., M.B., A.I., L.v.M., F.K., S.G.F., O.Z., U.G., M.A., and U.N. analyzed the data. L.v.M., C.S., O.Z., U.G., R.C., R.d.C., N.K., T.D.L., T.A.K. and S.N. acquired bacterial isolates and sequence data. M.F., Z.Z., M.A., and U.N. wrote the manuscript and all authors reviewed and edited the manuscript and approved the final version.

to review the article in its entirety please click on the following link to be redirected. Thank you. https://www.microbiologyresearch.org/content/journal/mgen/10.1099/mgen.0.000410

Study Hypothesized Commensal Clostridia are Important for Providing Colonization Resistance Against C. difficile Due to Their Ability to Produce Secondary Bile Acids


Clostridioides difficile is one of the leading causes of antibiotic-associated diarrhea.

Gut microbiota-derived secondary bile acids and commensal Clostridia that carry the bile acid-inducible (bai) operon are associated with protection from C. difficile infection (CDI), although the mechanism is not known.

In this study, we hypothesized that commensal Clostridia are important for providing colonization resistance against C. difficile due to their ability to produce secondary bile acids, as well as potentially competing against C. difficile for similar nutrients.

To test this hypothesis, we examined the abilities of four commensal Clostridia carrying the bai operon (Clostridium scindens VPI 12708, C. scindens ATCC 35704, Clostridium hiranonis, and Clostridium hylemonae) to convert cholate (CA) to deoxycholate (DCA) in vitro, and we determined whether the amount of DCA produced was sufficient to inhibit the growth of a clinically relevant C. difficile strain.

We also investigated the competitive relationships between these commensals and
C. difficile using an in vitro coculture system.

We found that inhibition of C. difficile growth by commensal Clostridia supplemented with CA was strain dependent, correlated with the production of ∼2 mM  DCA, and increased the expression of bai operon genes.

We also found that C. difficile was able to outcompete all four commensal Clostridia in an in vitro coculture system. These studies are instrumental in understanding the relationship between commensal Clostridia and C. difficile in the gut, which is vital for designing targeted bacterial therapeutics. Future studies dissecting the regulation of the bai operon in vitro and in vivo and how this affects CDI will be important.

IMPORTANCE : Commensal Clostridia carrying the bai operon, such as C. scindens, have been associated with protection against CDI; however, the mechanism for this protection is unknown. Herein, we show four commensal Clostridia that carry the bai operon and affect                                C. difficile growth in a strain-dependent manner, with and without the addition of cholate. Inhibition of C. difficile by commensals correlated with the efficient conversion of cholate to deoxycholate, a secondary bile acid that inhibits C. difficile germination, growth, and toxin production.

Competition studies also revealed that C. difficile was able to outcompete the commensals in an in vitro coculture system.

These studies are instrumental in understanding the relationship between commensal Clostridia and C. difficile in the gut, which is vital for designing targeted bacterial therapeutics.







SOURCE: https://jb.asm.org/content/202/11/e00039-20

Study of Hospital-Associated Infection – Clostridioides difficile Identified NpmA In the Genome of a Clinical Isolate

Clostridioides difficile: a potential source of NpmA in the clinical environment

Aminoglycosides are widely used to treat MDR Gram-negative bacterial infections with bactericidal activity mediated by binding the 16S rRNA aminoacyl-tRNA recognition site to prevent protein synthesis. Multiple aminoglycoside resistance mechanisms have been documented including 16S rRNA modification. NpmA, an uncommon 16S rRNA methyltransferase originally identified in a clinical Escherichia coli isolate confers pan-aminoglycoside resistance.

In this study, routine WGS of hospital-acquired Clostridioides (Clostridiumdifficile identified npmA in the genome of a clinical isolate (CD7814).

C. difficile is a Gram-positive, spore-forming enteric pathogen and the cause of most hospital-acquired, antibiotic-associated diarrhoea. The epidemiology of C. difficile has changed in the past several decades with infections increasingly being reported outside of acute care settings.

The discovery of npmA in the genome of a clinical C. difficile isolate has implications for the spread of aminoglycoside resistance.

Clinical C. difficile isolates are routinely sequenced using Illumina NextSeq500 and Nextera libraries as part of an infection prevention initiative at our hospital. Genomes are assembled using SPAdes, annotated with Prokka and characterized by searches of ResFinder and pubMLST databases (https://pubmlst.org/cdifficile/).5–7 This analysis identified NpmA-coding sequences in C. difficile isolate CD7814 belonging to ST11. CD7814 carries two additional aminoglycoside resistance determinants: (i) aph(3′)-III, which encodes an aminoglycoside phosphotransferase; and (ii) ant(6)-Ia, which encodes an aminoglycoside nucleotidyltransferase.

Aminoglycoside susceptibility testing was performed by Etest on CD7814 and two additional ST11 isolates from our hospital that lack npmA (CD7861 and CD7786). Cell suspensions corresponding to 0.5 McFarland were prepared and plated onto Brucella blood agar plates supplemented with vitamin K1 and haemin (Anaerobe Systems, Morgan Hill, CA, USA). Etest strips (bioMérieux, Durham, NC, USA) were applied and the plates were incubated anaerobically at 37°C for 48 h as previously described. Etests were read according to the manufacturer’s instructions. CD7814 demonstrated high-level resistance to gentamicin (>256 mg/L) relative to CD7861 (64 mg/L) and CD7786 (24 mg/L). High-level resistance to tobramycin and amikacin was observed in all ST11 C. difficile isolates tested. Although CLSI breakpoints for C. difficile are not defined for aminoglycosides, these data suggest that NpmA is expressed and associated with increased gentamicin resistance in CD7814.

Genomic analysis of the CD7814 assembly identified a large, presumably chromosomal contig of ∼150 kb containing the npmA gene. The majority of predicted ORFs surrounding npmA in CD7814 are hypothetical proteins whereas others encode proteins involved in recombination suggesting that npmA was acquired via horizontal gene transfer, which is consistent with the mosaic structure of the C. difficile chromosome (Figure 1a). Five additional C. difficile genomes bearing npmA gene sequences that show 99% nucleotide identity to npmA from CD7814 and pARS3 were identified through BLAST and PubMed literature searches (Figure 1a). The DNA sequence flanking npmA in CD7814 shows little or no nucleotide identity to either E. coli pARS3 or the five C. difficile npmA flanking regions. The five C. difficile genomes are of human and animal origin and were collected from three different continents over a period of at least 10 years. Interestingly, these C. difficile isolates belong to three different STs but their genomes share 99% nucleotide identity across ∼3kb of the npmA region (Figure 1a).

None of the five C. difficile genomes share any other sequence similarity to pARS3 outside of npmA. Together, these genomic data suggest that npmA is carried on a conserved element in the five C. difficile genomes and that the mechanism of npmA acquisition in CD7814 is different. In addition, all five C. difficile genomes encode a missense mutation in npmA resulting in a K131N substitution in NpmA relative to CD7814 and pARS3 sequences (Figure 1b). To maintain the established nomenclature for 16S rRNA methyltransferase genes, the CD7814 and, by default, the E. coli pARS3 npmA genes can be re-designated as npmA1 while npmA sequences containing the K131N mutation can be designated npmA2. CD7814 npmA1 has been deposited in GenBank under accession number MH249957.

(a) Predicted ORFs surrounding the npmA gene (dark grey) in CD7814 (light grey), the E. coli pARS3 plasmid (black) and five C. difficile genomes of human and animal origin (white). Shaded areas highlight regions of sequence homology. Chequered arrows indicate ORFs encoding recombinases potentially associated with horizontal transfer of npmA into CD7814. hu, human; sw, swine; bv, bovine; US, USA, JP, Japan; AU, Australia; CA, Canada. (b) NpmA protein alignment depicting the K131N substitution in CD7814 and E. coli pARS3.

To the best of our knowledge, this is the first description of npmA and high-level aminoglycoside resistance in a hospital-acquired C. difficile isolate. Because of strict anaerobic growth conditions, Etest is the most practical method to measure antibiotic susceptibility in C. difficile. As Etests for aramycin and neomycin are unavailable, a limitation of this study is our inability to demonstrate the specificity of NpmA methyltransferase activity for the N1-A1408 16S rRNA. However, the nucleotide identity between CD7814 npmA and the original E. coli sequence and the high-level aminoglycoside resistance observed support NpmA-mediated resistance in this clinical C. difficile isolate. To the best of our knowledge, no evidence to support high-level gentamicin resistance in the presence of ant(6)-Ia and aph(3′)-III has been reported.

In conclusion, this study demonstrates the presence of npmA and high-level aminoglycoside resistance in a clinical C. difficile isolate. The ability of C. difficile to persist in the environment as a spore former may facilitate acquisition of novel antibiotic resistance determinants. These data suggest that hospital-acquired C. difficile may be a reservoir for uncommon antibiotic resistance determinants such as npmA.


This work was supported by a research grant from the National Institute of Allergy and Infectious Diseases (R01AI127472 to L. H. H.).

Transparency declarations

None to declare.


1. Doi Y, Wachino JI, Arakawa Y.. Aminoglycoside resistance: the emergence of acquired 16S ribosomal RNA methyltransferasesInfect Dis Clin North Am 2016; 30: 523–37. [PMC free article] [PubMed[]
2. Wachino J, Shibayama K, Kurokawa H. et al. Novel plasmid-mediated 16S rRNA m1A1408 methyltransferase, NpmA, found in a clinically isolated Escherichia coli strain resistant to structurally diverse aminoglycosidesAntimicrob Agents Chemother 2007; 51: 4401–9. [PMC free article] [PubMed[]
3. Lessa FC, Winston LG, McDonald LC. et al. Burden of Clostridium difficile infection in the United StatesN Engl J Med 2015; 372: 2369–70. [PubMed[]
4. Baym M, Kryazhimskiy S, Lieberman TD. et al. Inexpensive multiplexed library preparation for megabase-sized genomesPLoS One 2015; 10: e0128036.. [PMC free article] [PubMed[]
5. Bankevich A, Nurk S, Antipov D. et al. SPAdes: a new genome assembly algorithm and its applications to single-cell sequencingJ Comput Biol 2012; 19: 455–77. [PMC free article] [PubMed[]
6. Seemann T. Prokka: rapid prokaryotic genome annotationBioinformatics 2014; 30: 2068–9. [PubMed[]
7. Zankari E, Hasman H, Cosentino S. et al. Identification of acquired antimicrobial resistance genesJ Antimicrob Chemother 2012; 67: 2640–4. [PMC free article] [PubMed[]
8. Citron DM, Ostovari MI, Karlsson A. et al. Evaluation of the E test for susceptibility testing of anaerobic bacteriaJ Clin Microbiol 1991; 29: 2197–203. [PMC free article] [PubMed[]
9. Amy J, Johanesen P, Lyras D.. Extrachromosomal and integrated genetic elements in Clostridium difficilePlasmid 2015; 80: 97–110. [PubMed[]
10. Doi Y, Wachino J, Arakawa Y.. Nomenclature of plasmid-mediated 16S rRNA methylases responsible for panaminoglycoside resistanceAntimicrob Agents Chemother 2008; 52: 2287–8. [PMC free article] [PubMed[]