Quantitative characterization of Clostridioides difficile population in the gut microbiome of patients with C. difficile infection and their association with clinical factors

Objective was to analyse bacterial composition and abundance of Clostridioides difficile in gut microbiome of patients with C. difficile infection (CDI) in association with clinical characteristics. Whole metagenome sequencing of gut microbiome of 26 CDI patients was performed, and the relative abundance of C. difficile and its toxin genes was measured. Clinical characteristics of the patients were obtained through medical records. A strong correlation between the abundance of C. difficile and tcdB genes in CDI patients was found. The relative abundance of C. difficile in the gut microbiome ranged from undetectable to 2.8% (median 0.089). Patients with fever exhibited low abundance of C. difficile in their gut, and patients with fewer C. difficile organisms required long-term anti-CDI treatment. Abundance of Bifidobacterium and Bacteroides negatively correlated with that of C. difficile at the genus level. CDI patients were clustered using the bacterial composition of the gut: one with high population of Enterococcus (cluster 1, n = 12) and another of Bacteroides or Lactobacillus (cluster 2, n = 14). Cluster1 showed significantly lower bacterial diversity and clinical cure at the end of treatment. Additionally, patients with CDI exhibited increased ARGs; notably, blaTEM, blaSHV and blaCTX-M were enriched. C. difficile existed in variable proportion of the gut microbiome in CDI patients. CDI patients with Enterococcus-rich microbiome in the gut had lower bacterial diversity and poorer clinical cure.

Scientific Reports | (2020) 10:17608 | https://doi.org/10.1038/s41598-020-74090-0 www.nature.com/scientificreports/ in the gut microbiome of patients with CDI 11 , and the influence of abundance of the organisms on the clinical presentation has not been investigated yet. In this study, the population of C. difficile was measured using whole metagenome sequences of gut microbiome in patients with CDI in order to directly analyse the relationship with many clinical and microbiological variables. Firstly, metagenomes were used for analysing the changes in microbial composition in CDI patients. Secondly, we investigated the relationship of each genus or family in gut microbiome with C. difficile by comparing their relative abundance in each patient. Thirdly, despite the low number of CDI patients, the clinical variables such as severity or treatment results and relative abundance of C. difficile in gut microbiome were compared. Fourthly, the patients of CDI were clustered with respect to their metagenome profiles in the gut, and clinical and microbiological characteristics of these two clusters were evaluated. Finally, the distribution of antimicrobial resistance genes (ARGs) in gut of CDI patients was analysed in comparison with healthy individuals.

Results
Demographics and clinical characteristics. During the study period, a total of 26 CDI patients were enrolled. Table 1 presents the demographic and clinical characteristics of the patients. Median age of the patients was 66.5 years, and gender distribution was similar. The patients exhibited a mean score of 3 on the Charlson Comorbidity Index, and diabetes mellitus and malignancy were the common comorbidities. Of the 26 patients, 62% had a history of hospitalization within the previous 2 months and 92% had received antibiotics within 2 months before this episode. As for the clinical parameters at the time of diagnosis, 27% of the patients presented leukocytosis and 7.7% of the patients exhibited elevation of serum creatinine levels (1.5-times baseline). Severity index of CDI was assessed by the two aforementioned methods. Severe CDI was identified in 8 of the 26 patients (31%) based on leukocytosis and acute kidney injury 12 , and in 12 of the 26 patients (46%), the severity was identified based on the four factors described previously 13 . As for treatment of CDI, 24  Changes of microbial taxa correlated to CDI in gut microbiome. Figure 1A shows bacterial composition in the gut microbiome of 26 patients with CDI in comparison with that of 61 healthy individuals. At the genus level, 15 out of 25 detected genera (average proportion > 1%) showed significant decrease, and among them Bifidobacterium, Ruminococcus, Eubacterium, and Faecalibacterium showed a significant decrease in abundance in the gut of patients with CDI (p < 0.001 for all), whereas the abundance of Enterococcus, Lactobacillus, Escherichia, and Klebsiella increased (p < 0.001, p = 0.031, p = 0.002, and p < 0.001, respectively) (Supplementary Table S1). At the family level, 9 out of 16 detected families (average proportion > 1%) showed significant decrease; Ruminococcaceae, Bifidobacteriaceae, Lachnospiraceae, and Eubacteriaceae showed a decrease in abundance (p < 0.001 for all), whereas Enterococcaceae, Lactobacillaceae, and Enterobacteriaceae increased in abundance in the gut microbiome of patients with CDI (p < 0.001, p = 0.012 and p < 0.001, respectively) (Supplementary Table S2). Principle component analysis (PCA) on bacterial composition at genus level showed a clear separation between healthy individuals and patients with CDI (Fig. 1B), and the diversity of the gut microbiome was significantly lower in these patients (Fig. 1C,D).
Relative abundance of toxigenic C. difficile in the gut microbiome of CDI patients. In order to measure the relative abundance of toxigenic C. difficile in the gut microbiome, we measured the relative abundance of tcdB genes in the metagenome sequences of gut microbiome. The median relative abundance of toxigenic C. difficile in the gut microbiome was 0.089%, ranging from 0 to 2.82% ( Fig. 2A). The relative abundance of tcdB genes measured by read mapping on tcdB and RPKM (read per kilobase million reads) showed a strong correlation with the abundance of C. difficile measured by MetaPhlAn clade-specific marker genes (r2 = 0.98) (Fig. 2B). For the reference, the relative abundance of C. difficile in non-CDI population is 0%.
Factors associated with the relative abundance of toxigenic C. difficile in CDI patients. We observed that the relative abundance of toxigenic C. difficile in the gut microbiome shows an agreement with Ct value in real-time PCR of tcdB (rho = − 0.605, p = 0.002) ( Table 2). We analysed the clinical characteristics associated with the abundance of toxigenic C. difficile in the gut microbiome (Table 2). Age, underlying diseases, and the use of antibiotics, proton pump inhibitor, or probiotics within 2 months from the episode of CDI were not associated with the abundance of toxigenic C. difficile. The abundance of toxigenic C. difficile had no effect on the occurrence of leukocytosis, hypoalbuminemia, or acute kidney injury. However, low numbers of toxigenic C. difficile in intestinal metagenomes were associated with fever (rho = − 0.41, p = 0.038), and longer CDI therapy (rho = − 0.405, p = 0.05). Treatment outcomes and the recurrence of CDI were not associated with the abundance of toxigenic C. difficile in the gut.
We observed the relationship between the abundance of microbial families or genera and the abundance of toxigenic C. difficile in the gut microbiome (Supplementary Table S3). At the genus level, the abundance of Bifidobacterium and Bacteroides showed a negative correlation with the abundance of toxigenic C. difficile (rho = − 0.417, p = 0.034; rho = − 0.403, p = 0.041, respectively). As these genera are the main constituents of the families, Bifidobacteriaceae and Bacteroidaceae, their populations also showed a negative correlation with the abundance of toxigenic C. difficile (rho = − 0.411, p = 0.037; rho = − 0.403, p = 0.041, respectively).  Figure 3A presents the abundance of 15 ARG classes, which were significantly different between the healthy individuals and the CDI patients (p value < 0.01 in t-test).
Notably, resistance genes against β-lactam, aminoglycoside, polymyxin, LMS, and glycopeptide were markedly enhanced in patients with CDI (5.1, 4.3, 18.5, 3.1, and 7.7-fold, respectively). Figure 3B-D present the differential prevalence of β-lactam, aminoglycoside, and tetracycline resistance genes in healthy individuals and patients with CDI patients. In particular, class A β-lactamase genes such as bla TEM , bla SHV , and bla CTX-M genes were markedly enhanced; class C plasmid-mediated AmpC genes such as bla CMY and bla DHA were also enhanced in patients   www.nature.com/scientificreports/ with CDI. Notably, KPC and NDM carbapenemase genes were observed in one and two patients, respectively. As for aminoglycoside genes, ANT(3′), AAC (3), and AAC(6′)-APH(2″) were markedly enhanced in the gut microbiome of CDI patients (Fig. 3C). However, tetracycline resistance genes showed only a slight increase (1.5-fold), and the distribution of individual tetracycline resistance genes was significantly different; tet32, tet44, tetM, and tetQ were fewer, but tetO, tetS, and tet34 appeared more in the gut microbiota of patients with CDI than in the gut microbiota of healthy individuals (Fig. 3D).

Distinct bacterial community in two different groups of CDI patients. When the patients with
CDI were clustered with respect to the bacterial composition, two different groups were observed with different major constituents in their bacterial communities (Fig. 4): a cluster of samples with high abundance of Enterococcus (cluster 1, n = 12), and a cluster with high abundance of Bacteroides or Lactobacillus (cluster 2, n = 14). The bacterial diversity was significantly low in cluster 1, compared to cluster 2 (p < 0.001), which was evident in the genus distribution shown in Fig. 4. The proportion of C. difficile was not different between the 2 clusters (p = 0.129). Interestingly, the abundance of ARGs was differentially distributed between the two groups. In particular, aminoglycoside, diaminopyrimidine, and LMS resistance genes were overrepresented in the patients with high abundance of Enterococcus (p value < 0.05). However, the recurrence status, severity scores, and the total abundance of ARGs did not show any significant difference between these two groups. Clinical characteristics were compared between the 2 clusters (Table 3); more patients in Bacteroides group took proton pump inhibitor (p = 0.018) or fluoroquinolone marginally (p = 0.065) within previous 2 months. Disease severity was not different between the 2 groups but clinical cure was achieved in more patients of Bacteroides group (p = 0.031) and all fatal cases came from the Enterococcus group.

Discussion
In this study, we measured the relative abundance of C. difficile in the gut microbiome of CDI patients using metagenome sequences of gut microbiome. The gene tcdB was used as a target gene to estimate the relative abundance of toxigenic C. difficile in the gut microbiome. In general, quantitation of C. difficile culture presented as www.nature.com/scientificreports/ CFU (colony forming unit)/g faeces, or quantitative PCR of stool is used to evaluate the burden of C. difficile 9,10 . Our metagenomic approach applied in this study has an advantage over such existing methods in the sense that it can calculate the proportion of C. difficile population in the gut microbiome. Interestingly, the relative abundance of toxigenic C. difficile in the gut microbiome ranged from undetectable to 2.82% of the organisms in the gut, and in 12 of the 26 patients (46%) the C. difficile population amounted to > 0.1% of the gut microbiome. The mammalian gut is colonized by trillions of microorganisms 14 , and although the number of microorganisms in the gut might be reduced due to antibiotic use that predisposed the patients to CDI, the C. difficile population in the gut of patients with CDI might approach to billions. Several studies have shown that the burden of C. difficile was not associated with the severity of CDI, and that the clinical severity of CDI was associated with the inflammatory response in the gut and the virulence of infecting organisms in human and mice 9,10,15 . The relative abundance of the organisms did not correlate with clinical severity in our study as well. However, unexpectedly, the C. difficile burden showed a negative correlation with the occurrence of fever and the treatment duration. This finding indicates that patients with sufficient immunity might have developed an inflammation to cause fever so as to control the pathogen levels in their gut, and the treatment duration was prolonged due to the inflammation in their gut.
We analysed the taxonomic composition of the intestinal microbiome using whole metagenome sequencing. The change in the bacterial composition of the CDI patients was similar to that reported in previous studies, which used amplification of 16S rRNA gene for identifying microbial taxa 3-5,16-18 . Among many genera that are known to be enhanced in healthy people compared with CDI patients, it is interesting that only the genera of Bifidobacterium and Bacteroides showed a significant negative association with toxigenic C. difficile when analysed using the proportion in the gut microbiome.
Age, gender, and underlying diseases are known to influence the microbiome structures of gut 11,14 , and hospitalization and usage of antibiotics have a huge impact 11 . In this study, the age and underlying diseases were not matched between the patients with CDI and healthy people. In addition, hospitalization and antibiotic usage were not matched because of the inclusion criteria of healthy people, which might contribute to the difference in the structure of gut microbiome and distribution of ARGs between CDI patients and healthy people. www.nature.com/scientificreports/ We found that the bacterial composition is an important discriminator to cluster the patients of CDI into two groups: one group with high abundance of Enterococcus, and the other with high abundance of Bacteroides or Lactobacillus. Compared with the previous report based on 16S rRNA amplicon sequencing data 5 , bacterial composition patterns were generally consistent, but we further characterized the two groups clinically and microbiologically. The bacterial diversity was significantly lower in Enterococcus cluster, which suggests that gut microbiome structure was more destroyed. The total days of antibiotic usage were marginally higher in Enterococcus cluster. Since we counted the antibiotic usage only within 2 months from the enrolment, we suspect that probably more antibiotics might have been used in the cluster. Interestingly, more clinical cure was achieved in Bacteroides cluster, and all fatal cases came from the Enterococcus cluster despite no significant difference in demographics, comorbidities, and clinical severity of the diseases. These findings suggest a poor prognosis of Enterococcus cluster with more destruction of gut microbiome structure.
In terms of ARGs in gut microbiome, a four-fold increase in the number of ARGs was detected in CDI patients relative to that in healthy people. Recent admission history and antibiotic usage in CDI patients would contribute to the enrichment of ARGs in gut microbiome of CDI patients. Above all, class A β-lactamase genes, which include clinically important extended-spectrum β-lactamase genes, were markedly increased with the enhancement of Enterobacteriaceae in patients with CDI. Furthermore, plasmid-mediated carbapenemase genes were detected in three patients. A marked increase in ARGs along with high carriage number of C. difficile organisms in the gut of CDI patients reinforced the necessity of contact precaution of CDI patients.
To the best of our knowledge, this is the first attempt to investigate the association of the abundance of C. difficile with the clinical and microbiological characteristics in gut microbiome. The distribution of ARGs in the gut microbiome of CDI patients was compared with that of healthy individuals for the first time as well. Despite the methodological advantages, there are certain limitations in this study. The number of enrolled patients is relatively small, and healthy controls were not matched with CDI patients in age, underlying diseases, and antibiotic usage.
To summarize, the population of C. difficile in the gut of patients with CDI varied significantly, but did not influence the clinical severity. Regarding the bacterial composition in the gut, the patients of CDI could be discriminated into Enterococcus-rich clusters with low bacterial diversity, and Bacteroides-rich clusters with preserved bacterial diversity, and the patients belonging to the latter cluster led to a better clinical cure. The institutional review boards of Hanyang University Hospital and Hanyang University Guri Hospital approved these protocols, and written informed consent was obtained from all the participants. All methods were performed in accordance with relevant guidelines and regulations.  20 . The severity of CDI was assessed by two methods as described 12,13 . Recurrence was defined as the resurgence of symptoms with diagnosis as CDI after cessation of treatment, at least 10 days after the first episode. Global cure was defined as patients who were cured at the end of treatment and did not have a recurrence 22 Metagenomic analysis of tcdB, taxonomic composition and antibiotic resistance genes. Filtered reads were assembled into contigs using MEGAHIT 26 with default options. Genes were predicted from contigs (> 500 bps) using FragGeneScan 27 . MetaPhlAn 28,29 , which uses clade-specific marker genes to profile bacterial compositions with the whole metagenome sequencing data, was used to find the taxonomic composition of each sample. The relative abundance of C. difficile was also reported based on this profile result. The Partitioning Around Medoids (PAM) algorithm in R package was used to cluster the samples based on bacterial composition. The optimal number of clusters was selected as two after calculating the silhouette score (= 0.36). The relative abundance of tcdB (accession no. NC_009089.1:786021-805508) was measured by mapping reads to the genes using Bowtie to measure toxigenic C. difficile. The relative abundance was reported by RPKM 30 . To identify ARGs, genes predicted in the metagenomic data set were searched against the ARGs annotated in the CARD database 31 using Blastp 32 with the threshold of an e-value less than 1 × 10 -10 , similarity over 70%, and reference coverage over 70%. The resistance genes were classified into 53 ARG subclasses and 20 ARG classes based on the gene ontology 31 (Supplementary Table S4). For normalization, RPKM (read per kilobase million read) was used for measuring the abundance of tcdB; GPM (gene per million genes) was used as a measure of the abundance of ARGs in each sample: The images for figures were generated by using R package.

Statistical analysis.
To compare the demographics and clinical characteristics, SPSS version 24.0 for Windows (SPSS Inc., Armonk, NY, USA) was used. Pearson's chi-square test or Fisher's exact test were used to analyse categorical variables, and Mann-Whitney U-test was used to analyse continuous variables. Spearman's rank correlation test was performed to evaluate the relationship between two variables. A p value of < 0.05 by a two-tailed test was considered to be statistically significant.
Ethics approval and consent to participate. The study protocol was approved by the institutional review boards (IRB No. HYUH 2016-05-031 and HYUH 2017-06-001 from Hanyang University Hospital and GURI 2016-05-003 from Hanyang University Guri hospital), and written informed consent was obtained from all the participants.

GPM =
Number of ARGs annotated Number of genes predicted * 10 6 .