Molecular Epidemiology and Antimicrobial Susceptibility of Clostridium difficile Isolates from a University Teaching Hospital in China

While the developed world has seen a significant increase in the number of scientific articles on Clostridium difficile infection (CDI), the developing world still lags behind on this subject due to limited laboratory capacity, low awareness, and limited surveillance of this problem. As such, CDI is considered a neglected but potentially huge problem in developing countries. The major aim of this study was to systemically evaluate the utility of several molecular typing tools for CDI, including their relevance in epidemiological studies in developing countries such as China. A total of 116 non-repetitive toxigenic C. difficile isolates from Chinese patients, were studied. The isolates comprised 83 (71.6%) A+B+CDT- isolates, 27 (23.3%) A-B+CDT- isolates, and 6 (5.1%) A+B+CDT+ isolates. Typing methods evaluated included multilocus variable-number tandem-repeat analysis, PCR ribotyping, multilocus sequence typing, and sequencing of slpA and tcdC genes, which identified 113, 30, 22, 18, and 8 genotypes each and exhibited discriminatory powers of 0.999, 0.916, 0.907, 0.883, and 0.765, respectively. Compared to A+B+ strains, A-B+ strains exhibited higher prevalence of drug resistance to clindamycin, erythromycin, levofloxacin, rifampicin, rifaximin, and tetracycline. Furthermore, drug resistance rates of strains with different PCR ribotypes differed, supporting the importance of molecular typing in management and control of CDI. Based on our earlier suggestion to improve the diagnostic laboratory capacity of CDI in developing countries, setting up efficient surveillance programs complemented by relevant molecular typing methods is warranted.


INTRODUCTION
Clostridium difficile is a leading cause of both nosocomial and antibiotic-associated diarrhea. The clinical manifestation of C. difficile infection (CDI) ranges from asymptomatic colonization to mild diarrhea to toxic megacolon and fulminant colitis. CDI is now considered a public health threat, with a dramatic rise in the incidence and severity of CDI observed in Europe and North America in the past two decades (Rupnik et al., 2009).
Most CDI cases have largely been attributed to the presence of the hyper-virulent C. difficile strain BI/NAP1/027 [restriction endonuclease analysis (REA) group BI, North American pulsefield type 1, PCR ribotype 027; McDonald et al., 2005]. Although ribotype 027 strains have caused major epidemics in North America and Europe, only sporadic cases have been reported in Asia (Collins et al., 2013). The major virulence factors for C. difficile are enterotoxin A (TcdA) and cytotoxin B (TcdB) (George et al., 1978). However, some strains that produce a binary toxin called C. difficile binary toxin (CDT), and whose role in causing disease is not yet clear, have been described and are associated with increased virulence and recurrence rates (Popoff et al., 1988).
Knowledge of the antimicrobial susceptibility profiles and molecular types of C. difficile is an important first step for monitoring and understanding the epidemiology of this organism. A variety of molecular methods have been applied for genotyping of C. difficile, including PCR ribotyping, pulsed-field gel electrophoresis (PFGE), multilocus sequence typing (MLST), multilocus variable-number tandem-repeat analysis (MLVA), and sequencing of functional genes such as slpA and tcdC (Collins et al., 2015). Each of these typing methods has its own merits and disadvantages, and may be applied in different occasions according to the aim and scale of the study (Killgore et al., 2008;Collins et al., 2015).
In China, there are limited studies on the molecular epidemiology and antibiotic susceptibility profiles of C. difficile, possibly due to insufficient laboratory diagnostic capacity, low awareness, and lack of high-quality surveillance systems (Collins et al., 2013;Hawkey et al., 2013;Cheng et al., 2015). However, in two previous retrospective studies, PCR ribotype 027 isolates were detected, suggesting that the threat of CDI in China is generally neglected and possibly underestimated (Cheng et al., 2016). To partly address this problem, we previously proposed a glutamate dehydrogenase-based algorithm for improving the clinical laboratory diagnostic capacity for CDI in China (Cheng et al., 2015).
The objective of this study was to evaluate the utility of different molecular typing assays for C. difficile, in relation to toxigenicity and antimicrobial susceptibility, using isolates obtained from one hospital in China. The usefulness of the different assays in different situations was considered, to inform decision making for improving the capacity of clinical laboratories in the management of CDI in developing countries like China.

Ethics
The study was approved by the Human Research Ethics Committee of Peking Union Medical College Hospital (PUMCH) (No. PUMCHBC-C-4). Written informed consents were obtained from patients for use of the samples in research.
All specimens were initially tested for toxin A/B using enzyme immunoassay (EIA; VIDAS C. difficile Toxin A&B, bioMérieux, Marcy l'Etiole, France) and cultured on selective cycloserine-cefoxitin-fructose agar (CCFA) plates. Typical colonies on CCFA were identified as C. difficile by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS; Bruker Daltonics GmbH, Bremen, Germany). Only C. difficile isolates obtained from stool specimens with positive EIA results, and confirmed by MALDI-TOF MS, were included in the study.
DNA Extraction, Toxin Gene Detection, and Sequencing of tcdC Gene DNA extraction and subsequent toxin gene detection was carried out as previously described (Cheng et al., 2015). Genotype of tcdC gene was determined by comparing the obtained sequences with previous published sequences as described by Curry et al. (2007).

Molecular Type Assays
Capillary sequencer-based PCR ribotyping was performed as described by Indra et al. (2008), and ribotypes were assigned by querying the results against WEBRIBO database 1 . Novel ribotypes were named as "PUR" plus two Arabic numbers (e.g., PUR01).
Multilocus sequence typing was performed by sequencing seven house-keeping gene loci as previously described by Griffiths et al. (2010), and sequence type (ST) and clades of C. difficile isolates were determined by querying on official website 2 .
Multilocus variable-number tandem-repeat analysis was performed using the set of seven loci as proposed by van den Berg et al. (2007). Repeat numbers were analyzed using BioNumerics software v6.5 (Applied Maths, Texas, USA) for cluster analysis.
Sequence typing of the slpA gene was performed as described previously by Kato et al. (2005). However, due to lack of a consistent nomenclature for slpA genotypes, different slpA genotypes and subtypes were defined as per NCBI database GenBank accession entries for slpA genotype descriptions (Joost et al., 2009).

Antimicrobial Susceptibility Testing
Antimicrobial susceptibility testing was performed by the agar dilution method according to Clinical and Laboratory Standards Institute (CLSI) guidelines (document M11-A8; CLSI, 2012). The following 11 antimicrobial agents were used: ciprofloxacin, clindamycin, erythromycin, levofloxacin, meropenem, metronidazole, piperacillin/tazobactam, rifampicin, rifaximin, tetracycline, and vancomycin. Interpretation of testing results was based on CLSI M100-S25 (CLSI, 2015), or according to the criteria suggested by Huang et al. (2009) for the drugs whose breakpoints were not available in CLSI documents, as summarized in Supplementary Table S1. Bacteroides fragilis ATCC 25285 was used for quality control.

Data Analysis
The genetic relationships of the isolates was determined by cluster analysis using the minimum-spanning tree available in the BioNumerics software v 6.5 (Applied Maths).
To compare the discriminatory power of different molecular methods, we used an index of discriminatory power (D) based on Simpson's index of diversity (Fawley et al., 2015).
Statistical analyses were performed using SPSS software (version 17.0, IBM, New York, NY, USA). The chi-square test was applied to compare categorical variables. The level of statistical significance was defined as P < 0.05.

Relatedness of Genotypes Determined by Different Typing Methods
Of the different typing methods used, MLVA exhibited the greatest discriminatory power, generating a D-value of 0.999, followed by PCR ribotyping (D-value 0.916) and MLST (D-value 0.907). In comparison to the above, slpA and tcdC sequencing exhibited lower discriminatory power (D-value 0.883 and 0.765, respectively; Table 2).
Of note, there was a strong correlation between toxigenic types and MLST clades: all 82 MLST clade 1 isolates were A+B+CDT−. Among 28 MLST clade 4 isolates, 27 (96.4%) were of toxigenic type A−B+CDT− and only one isolate (3.6%) was A+B+CDT−. Furthermore, all A+B+CDT+ isolates were only found in MLST clades 2 and 3 ( Table 1).
Multilocus variable-number tandem-repeat analysis exhibited the highest discriminatory power amongst all typing methods used ( Figure 1D). The minimum spanning tree generated by MLVA also indicated a general phylogenetic relatedness of MLVA typing results and MLST clades ( Figure 1D). As can be seen clearly in Figure 1D, using the MLVA spanning tree, MLST clade 1 was separated into two major branches by MLST clade 4. However, discrepancies between MLVA and MLST were also observed. For example, by MLVA analysis, the three MLST  Figure 1D) were classified into two different branches: two isolates were more phylogenetically related to MLST clade 3, but one isolate was more closely related to MLST ST2 of MLST clade 1 (Figure 1D).

Antimicrobial Susceptibilities
The minimum inhibitory concentrations (MICs) of 11 antimicrobial agents for the 116 isolates are summarized in Table 3. All the isolates were susceptible to metronidazole, Frontiers in Microbiology | www.frontiersin.org vancomycin, meropenem, piperacillin/tazobactam, but resistant to ciprofloxacin. Co-resistance to erythromycin, clindamycin, and levofloxacin was observed in 41.3% of the isolates. High-level resistance to erythromycin (MIC > 128 mg/L) and clindamycin (MIC > 128 mg/L) was detected in 63.8% and 55.2% of the 116 isolates, respectively. The MICs of rifampin were consistent with those of rifaximin, and were either ≤0.064 mg/L or >256 mg/L. Moreover, 25.9% of the isolates were resistant to tetracycline. Drug resistance rates for A−B+ isolates for the antibiotics levofloxacin, rifampicin, rifaximin, and tetracycline, were significantly higher than that of A+B+ isolates (P < 0.001; Table 3). In addition, varied antimicrobial resistance rates were observed amongst different molecular types. For instance, ribotypes 012, 017, and PUR09 isolates exhibited higher resistance rates to clindamycin and erythromycin compared to ribotype 001 and other ribotype isolates (Figure 2). Moreover, ribotype 017 and PUR09 isolates showed higher prevalence rate of resistance to levofloxacin and tetracycline, compared to other ribotype isolates. Ribotype 017 isolates also had higher rate of resistance to rifampicin and rifaximin (Figure 2).

DISCUSSION
Clostridium difficile infection has become a significant public health threat in the developed world, with substantial increase in morbidity and mortality reported since the early 2000s. However, due to inadequate laboratory diagnostic capacity, little is known about the magnitude of the CDI problem in developing countries like China, and hence the problem may be much underestimated (Collins et al., 2013;Hawkey et al., 2013;Cheng et al., 2016). In this regard, our group has previously attempted to improve the diagnosis of CDI in China by proposing a practical workflow for future CDI laboratory diagnosis (Cheng et al., 2015). Furthermore, the lack of high-quality CDI surveillance systems and epidemiology studies in developing countries contributes to the low awareness of CDI. Thus further studies are needed to address this problem.
Molecular typing methods are essential for understanding the epidemiology of C. difficile, and are especially important as part of an active surveillance and hospital infection control strategy for this organism. To date, several typing methods have been used to study the epidemiology of C. difficile, and there are pros and cons for each method. Most studies evaluating the utility of different molecular typing methods for CDI are from the developed world. In the present study, we evaluated the performance and utility   of the most commonly used genotypic assays for C. difficile, including toxigenic typing, tcdC genotyping, slpA genotyping, MLST, PCR-ribotyping, and MLVA.
Although the dominant C. difficile toxigenic type is A+B+, there has been an increase in the prevalence and significance of the A-B+ type in some regions of the world (Kim et al., 2008;King et al., 2015). In this study, 23.3% of the isolates were A−B+CDT−, which is in agreement with previous studies in China in which the average prevalence rate of A−B+ strains is 21.4% (201/937), ranging from 0 to 53.6% by geographic distribution (Huang et al., 2009(Huang et al., , 2010Dong et al., 2013;Hawkey et al., 2013;Yan et al., 2013;Chen et al., 2014;Fang et al., 2014;Wang et al., 2014;Zhou et al., 2014;Cheng et al., 2015;Ye et al., 2015;Tian et al., 2016). Interestingly, infection with CDTpositive C. difficile strains has been reported to be associated with higher mortality and recurrence rates (Stewart et al., 2013). In this study, 5.1% (6/116) isolates were A+B+CDT+, which is slightly higher than in previous studies in Shanghai (1.6%, 2/110) and Zhejiang provinces (0%, 0/82; Huang et al., 2010;Fang et al., 2014). However, the overall prevalence of A+B+CDT+ C. difficile strains in China is much lower than in North America and Europe (Bassetti et al., 2012), possibly due to low lab diagnostic capacity and subsequent lack of awareness amongst medical personnel. By tcdC sequencing, five of six A+B+CDT+ isolates detected in this study had an 18-bp deletion at nucleotides 330-347, but only two ribotype 027 isolates, characterized by a single deletion at nucleotide 117, were detected.
The predominant ribotype and ST identified in the present study was ribotype 012/ST54 (16.4%), followed by ribotypes 001/ST3 (14.7%) and 017/ST37 (13.8%), whilst the commonest ribotype is ribotype 012 reported overall in China (100 of 589 isolates, 17.0%), and ST54, ST3,ST35,and ST37 (prevalence of 21.4,12.6,11.0,and 9.4 of 682 isolates tested, respectively) rank the top four STs (Huang et al., 2009(Huang et al., , 2010Dong et al., 2013;Hawkey et al., 2013;Yan et al., 2013;Chen et al., 2014;Fang et al., 2014;Wang et al., 2014;Zhou et al., 2014;Cheng et al., 2015;Ye et al., 2015;Tian et al., 2016). The remaining ribotypes/STs were distributed sporadically among different departments and with a lower prevalence. Generally, there was no obvious correlation between clonal clusters of the organism generated from the different departments during the study period, suggesting that a wide variety of strains are implicated in this hospital. However, due to lack of consistent data and limited lab diagnostic capacity for CDI at the hospital concerned, these findings may be biased. Although the major molecular types identified in different C. difficile studies were similar, the prevalence of different molecular types varied amongst studies in China. In agreement to our findings, ribotype 017 C. difficile strains have been reported in several Asian countries (Collins et al., 2013). However, in North America and Europe, many CDI cases have been attributed to C. difficile ribotype 027 and 078 strains which have been described in many hospitals, and were particularly associated with outbreaks or epidemics (Bassetti et al., 2012). The major clone implicated in epidemics varies according to geographic locale, highlighting the significance of establishing a local surveillance network for CDI.
The emergence of increased antibiotic resistance in different C. difficile strain types worldwide may be a contributing factor to increased occurrence of CDI outbreaks in several hospitals.
Although several studies have reported on the increasing MIC for metronidazole and vancomycin in C. difficile (Baines et al., 2008;Adler et al., 2015), both drugs showed high in vitro activity against all the isolates in our study. Furthermore, all strains were also susceptible to meropenem and piperacillin/tazobactam. Although a small number (12.9%) of C. difficile strains with highlevel resistance to rifampin and rifaximin was observed in the present study, the rate was lower than the 25-29% reported in Shanghai, China (Huang et al., 2009(Huang et al., , 2010. The high rate of drug resistance for ciprofloxacin (100%), erythromycin (69%), and clindamycin (87.9%) among the studied isolates suggests that use of these antibiotics may be a risk factor for the development of CDI in this geographic locale (Bassetti et al., 2012). In addition, A−B+ strains in the present study exhibited higher drug resistance rates compared to A+B+ strains for the antibiotics clindamycin, erythromycin, levofloxacin, rifampicin, rifaximin, and tetracycline. Thus toxigenic typing may be valuable for understanding antibiotic resistance in the management and control of CDI. Furthermore, the drug resistance rates of strains of different PCR ribotypes differed. These findings emphasize the importance of implementing active surveillance and molecular epidemiologic studies on CDI.
All the typing methods used in this study have the advantage of high reproducibility (Table 4). MLVA exhibited extremely high discriminatory power, and thus may be a potential valuable tool for investigating CDI outbreaks (Eyre et al., 2013b). However, a reasonable cut-off value is needed since C. difficile isolates from the same outbreak may belong to genetically closely related but different MLVA types ( Table 4) (Marsh et al., 2006).
Based on our findings, we recommend PCR ribotyping as the method of choice for epidemiological surveillance and infection prevention and control, as it delivers high discrimination, accuracy, and reproducibility. However, more work is needed for its standardization and database construction ( Table 4) (Indra et al., 2008;Fawley et al., 2015). Although MLST also offers high discriminatory power index and easy inter-laboratory data comparison which may be useful for epidemiology surveillance and phylogenetic studies, the high cost is prohibitive (Table 4) (Killgore et al., 2008;Collins et al., 2015;Tian et al., 2016).
Sequencing of the slpA gene is valuable for comparison of C. difficile strains implicated in epidemics in diverse areas because the typing results are reproducible and can easily be shared. In addition, as the slpA gene is related to the strain's serogroup, this typing method could be useful in vaccine development (Kato et al., 2005;Tian et al., 2016). tcdC genotyping had the lowest D-value, but sequencing of this gene may be used as an indicator for the hyper-virulent C. difficile clones, e.g., PCR ribotype 027/078 isolates (Curry et al., 2007).
Several other C. difficile typing methods apart from the ones applied in this study have been used, including PFGE, REA, and PCR-restriction fragment length polymorphism (RFLP). Unfortunately, these restriction enzyme-agarose gel electrophoresis-based methods have poor reproducibility, and are difficult to standardize for result interpretation and inter-laboratory data exchange (Killgore et al., 2008;Collins et al., 2015). Recent advances in whole-genome sequencing of C. difficile has the potential to provide even greater epidemiological information, but is still premature for use in large scale epidemiologic studies (Eyre et al., 2013a;Knight et al., 2015).

CONCLUSION
Our study is the most systematic study to integrate molecular epidemiology and antibiotic susceptibility testing, and provides comprehensive data for a better understanding of CDI in China. The findings highlight the importance of active surveillance using molecular typing techniques for better management and control of CDI in developing countries. Different molecular typing assays could be used depending on the scale and aim of the surveillance.